2 * Copyright (C) 2007 Jens Axboe <jens.axboe@oracle.com>
4 * Scatterlist handling helpers.
6 * This source code is licensed under the GNU General Public License,
7 * Version 2. See the file COPYING for more details.
9 #include <linux/export.h>
10 #include <linux/slab.h>
11 #include <linux/scatterlist.h>
12 #include <linux/highmem.h>
13 #include <linux/kmemleak.h>
16 * sg_next - return the next scatterlist entry in a list
17 * @sg: The current sg entry
20 * Usually the next entry will be @sg@ + 1, but if this sg element is part
21 * of a chained scatterlist, it could jump to the start of a new
25 struct scatterlist *sg_next(struct scatterlist *sg)
27 #ifdef CONFIG_DEBUG_SG
28 BUG_ON(sg->sg_magic != SG_MAGIC);
34 if (unlikely(sg_is_chain(sg)))
35 sg = sg_chain_ptr(sg);
39 EXPORT_SYMBOL(sg_next);
42 * sg_nents - return total count of entries in scatterlist
43 * @sg: The scatterlist
46 * Allows to know how many entries are in sg, taking into acount
50 int sg_nents(struct scatterlist *sg)
53 for (nents = 0; sg; sg = sg_next(sg))
57 EXPORT_SYMBOL(sg_nents);
61 * sg_last - return the last scatterlist entry in a list
62 * @sgl: First entry in the scatterlist
63 * @nents: Number of entries in the scatterlist
66 * Should only be used casually, it (currently) scans the entire list
67 * to get the last entry.
69 * Note that the @sgl@ pointer passed in need not be the first one,
70 * the important bit is that @nents@ denotes the number of entries that
74 struct scatterlist *sg_last(struct scatterlist *sgl, unsigned int nents)
76 #ifndef CONFIG_ARCH_HAS_SG_CHAIN
77 struct scatterlist *ret = &sgl[nents - 1];
79 struct scatterlist *sg, *ret = NULL;
82 for_each_sg(sgl, sg, nents, i)
86 #ifdef CONFIG_DEBUG_SG
87 BUG_ON(sgl[0].sg_magic != SG_MAGIC);
88 BUG_ON(!sg_is_last(ret));
92 EXPORT_SYMBOL(sg_last);
95 * sg_init_table - Initialize SG table
97 * @nents: Number of entries in table
100 * If this is part of a chained sg table, sg_mark_end() should be
101 * used only on the last table part.
104 void sg_init_table(struct scatterlist *sgl, unsigned int nents)
106 memset(sgl, 0, sizeof(*sgl) * nents);
107 #ifdef CONFIG_DEBUG_SG
110 for (i = 0; i < nents; i++)
111 sgl[i].sg_magic = SG_MAGIC;
114 sg_mark_end(&sgl[nents - 1]);
116 EXPORT_SYMBOL(sg_init_table);
119 * sg_init_one - Initialize a single entry sg list
121 * @buf: Virtual address for IO
125 void sg_init_one(struct scatterlist *sg, const void *buf, unsigned int buflen)
127 sg_init_table(sg, 1);
128 sg_set_buf(sg, buf, buflen);
130 EXPORT_SYMBOL(sg_init_one);
133 * The default behaviour of sg_alloc_table() is to use these kmalloc/kfree
136 static struct scatterlist *sg_kmalloc(unsigned int nents, gfp_t gfp_mask)
138 if (nents == SG_MAX_SINGLE_ALLOC) {
140 * Kmemleak doesn't track page allocations as they are not
141 * commonly used (in a raw form) for kernel data structures.
142 * As we chain together a list of pages and then a normal
143 * kmalloc (tracked by kmemleak), in order to for that last
144 * allocation not to become decoupled (and thus a
145 * false-positive) we need to inform kmemleak of all the
146 * intermediate allocations.
148 void *ptr = (void *) __get_free_page(gfp_mask);
149 kmemleak_alloc(ptr, PAGE_SIZE, 1, gfp_mask);
152 return kmalloc(nents * sizeof(struct scatterlist), gfp_mask);
155 static void sg_kfree(struct scatterlist *sg, unsigned int nents)
157 if (nents == SG_MAX_SINGLE_ALLOC) {
159 free_page((unsigned long) sg);
165 * __sg_free_table - Free a previously mapped sg table
166 * @table: The sg table header to use
167 * @max_ents: The maximum number of entries per single scatterlist
168 * @skip_first_chunk: don't free the (preallocated) first scatterlist chunk
169 * @free_fn: Free function
172 * Free an sg table previously allocated and setup with
173 * __sg_alloc_table(). The @max_ents value must be identical to
174 * that previously used with __sg_alloc_table().
177 void __sg_free_table(struct sg_table *table, unsigned int max_ents,
178 bool skip_first_chunk, sg_free_fn *free_fn)
180 struct scatterlist *sgl, *next;
182 if (unlikely(!table->sgl))
186 while (table->orig_nents) {
187 unsigned int alloc_size = table->orig_nents;
188 unsigned int sg_size;
191 * If we have more than max_ents segments left,
192 * then assign 'next' to the sg table after the current one.
193 * sg_size is then one less than alloc size, since the last
194 * element is the chain pointer.
196 if (alloc_size > max_ents) {
197 next = sg_chain_ptr(&sgl[max_ents - 1]);
198 alloc_size = max_ents;
199 sg_size = alloc_size - 1;
201 sg_size = alloc_size;
205 table->orig_nents -= sg_size;
206 if (skip_first_chunk)
207 skip_first_chunk = false;
209 free_fn(sgl, alloc_size);
215 EXPORT_SYMBOL(__sg_free_table);
218 * sg_free_table - Free a previously allocated sg table
219 * @table: The mapped sg table header
222 void sg_free_table(struct sg_table *table)
224 __sg_free_table(table, SG_MAX_SINGLE_ALLOC, false, sg_kfree);
226 EXPORT_SYMBOL(sg_free_table);
229 * __sg_alloc_table - Allocate and initialize an sg table with given allocator
230 * @table: The sg table header to use
231 * @nents: Number of entries in sg list
232 * @max_ents: The maximum number of entries the allocator returns per call
233 * @gfp_mask: GFP allocation mask
234 * @alloc_fn: Allocator to use
237 * This function returns a @table @nents long. The allocator is
238 * defined to return scatterlist chunks of maximum size @max_ents.
239 * Thus if @nents is bigger than @max_ents, the scatterlists will be
240 * chained in units of @max_ents.
243 * If this function returns non-0 (eg failure), the caller must call
244 * __sg_free_table() to cleanup any leftover allocations.
247 int __sg_alloc_table(struct sg_table *table, unsigned int nents,
248 unsigned int max_ents, struct scatterlist *first_chunk,
249 gfp_t gfp_mask, sg_alloc_fn *alloc_fn)
251 struct scatterlist *sg, *prv;
254 memset(table, 0, sizeof(*table));
258 #ifndef CONFIG_ARCH_HAS_SG_CHAIN
259 if (WARN_ON_ONCE(nents > max_ents))
266 unsigned int sg_size, alloc_size = left;
268 if (alloc_size > max_ents) {
269 alloc_size = max_ents;
270 sg_size = alloc_size - 1;
272 sg_size = alloc_size;
280 sg = alloc_fn(alloc_size, gfp_mask);
284 * Adjust entry count to reflect that the last
285 * entry of the previous table won't be used for
286 * linkage. Without this, sg_kfree() may get
290 table->nents = ++table->orig_nents;
295 sg_init_table(sg, alloc_size);
296 table->nents = table->orig_nents += sg_size;
299 * If this is the first mapping, assign the sg table header.
300 * If this is not the first mapping, chain previous part.
303 sg_chain(prv, max_ents, sg);
308 * If no more entries after this one, mark the end
311 sg_mark_end(&sg[sg_size - 1]);
318 EXPORT_SYMBOL(__sg_alloc_table);
321 * sg_alloc_table - Allocate and initialize an sg table
322 * @table: The sg table header to use
323 * @nents: Number of entries in sg list
324 * @gfp_mask: GFP allocation mask
327 * Allocate and initialize an sg table. If @nents@ is larger than
328 * SG_MAX_SINGLE_ALLOC a chained sg table will be setup.
331 int sg_alloc_table(struct sg_table *table, unsigned int nents, gfp_t gfp_mask)
335 ret = __sg_alloc_table(table, nents, SG_MAX_SINGLE_ALLOC,
336 NULL, gfp_mask, sg_kmalloc);
338 __sg_free_table(table, SG_MAX_SINGLE_ALLOC, false, sg_kfree);
342 EXPORT_SYMBOL(sg_alloc_table);
345 * sg_alloc_table_from_pages - Allocate and initialize an sg table from
347 * @sgt: The sg table header to use
348 * @pages: Pointer to an array of page pointers
349 * @n_pages: Number of pages in the pages array
350 * @offset: Offset from start of the first page to the start of a buffer
351 * @size: Number of valid bytes in the buffer (after offset)
352 * @gfp_mask: GFP allocation mask
355 * Allocate and initialize an sg table from a list of pages. Contiguous
356 * ranges of the pages are squashed into a single scatterlist node. A user
357 * may provide an offset at a start and a size of valid data in a buffer
358 * specified by the page array. The returned sg table is released by
362 * 0 on success, negative error on failure
364 int sg_alloc_table_from_pages(struct sg_table *sgt,
365 struct page **pages, unsigned int n_pages,
366 unsigned long offset, unsigned long size,
371 unsigned int cur_page;
373 struct scatterlist *s;
375 /* compute number of contiguous chunks */
377 for (i = 1; i < n_pages; ++i)
378 if (page_to_pfn(pages[i]) != page_to_pfn(pages[i - 1]) + 1)
381 ret = sg_alloc_table(sgt, chunks, gfp_mask);
385 /* merging chunks and putting them into the scatterlist */
387 for_each_sg(sgt->sgl, s, sgt->orig_nents, i) {
388 unsigned long chunk_size;
391 /* look for the end of the current chunk */
392 for (j = cur_page + 1; j < n_pages; ++j)
393 if (page_to_pfn(pages[j]) !=
394 page_to_pfn(pages[j - 1]) + 1)
397 chunk_size = ((j - cur_page) << PAGE_SHIFT) - offset;
398 sg_set_page(s, pages[cur_page], min(size, chunk_size), offset);
406 EXPORT_SYMBOL(sg_alloc_table_from_pages);
408 void __sg_page_iter_start(struct sg_page_iter *piter,
409 struct scatterlist *sglist, unsigned int nents,
410 unsigned long pgoffset)
412 piter->__pg_advance = 0;
413 piter->__nents = nents;
416 piter->sg_pgoffset = pgoffset;
418 EXPORT_SYMBOL(__sg_page_iter_start);
420 static int sg_page_count(struct scatterlist *sg)
422 return PAGE_ALIGN(sg->offset + sg->length) >> PAGE_SHIFT;
425 bool __sg_page_iter_next(struct sg_page_iter *piter)
427 if (!piter->__nents || !piter->sg)
430 piter->sg_pgoffset += piter->__pg_advance;
431 piter->__pg_advance = 1;
433 while (piter->sg_pgoffset >= sg_page_count(piter->sg)) {
434 piter->sg_pgoffset -= sg_page_count(piter->sg);
435 piter->sg = sg_next(piter->sg);
436 if (!--piter->__nents || !piter->sg)
442 EXPORT_SYMBOL(__sg_page_iter_next);
445 * sg_miter_start - start mapping iteration over a sg list
446 * @miter: sg mapping iter to be started
447 * @sgl: sg list to iterate over
448 * @nents: number of sg entries
451 * Starts mapping iterator @miter.
456 void sg_miter_start(struct sg_mapping_iter *miter, struct scatterlist *sgl,
457 unsigned int nents, unsigned int flags)
459 memset(miter, 0, sizeof(struct sg_mapping_iter));
461 __sg_page_iter_start(&miter->piter, sgl, nents, 0);
462 WARN_ON(!(flags & (SG_MITER_TO_SG | SG_MITER_FROM_SG)));
463 miter->__flags = flags;
465 EXPORT_SYMBOL(sg_miter_start);
467 static bool sg_miter_get_next_page(struct sg_mapping_iter *miter)
469 if (!miter->__remaining) {
470 struct scatterlist *sg;
471 unsigned long pgoffset;
473 if (!__sg_page_iter_next(&miter->piter))
476 sg = miter->piter.sg;
477 pgoffset = miter->piter.sg_pgoffset;
479 miter->__offset = pgoffset ? 0 : sg->offset;
480 miter->__remaining = sg->offset + sg->length -
481 (pgoffset << PAGE_SHIFT) - miter->__offset;
482 miter->__remaining = min_t(unsigned long, miter->__remaining,
483 PAGE_SIZE - miter->__offset);
490 * sg_miter_skip - reposition mapping iterator
491 * @miter: sg mapping iter to be skipped
492 * @offset: number of bytes to plus the current location
495 * Sets the offset of @miter to its current location plus @offset bytes.
496 * If mapping iterator @miter has been proceeded by sg_miter_next(), this
500 * Don't care if @miter is stopped, or not proceeded yet.
501 * Otherwise, preemption disabled if the SG_MITER_ATOMIC is set.
504 * true if @miter contains the valid mapping. false if end of sg
507 bool sg_miter_skip(struct sg_mapping_iter *miter, off_t offset)
509 sg_miter_stop(miter);
514 if (!sg_miter_get_next_page(miter))
517 consumed = min_t(off_t, offset, miter->__remaining);
518 miter->__offset += consumed;
519 miter->__remaining -= consumed;
525 EXPORT_SYMBOL(sg_miter_skip);
528 * sg_miter_next - proceed mapping iterator to the next mapping
529 * @miter: sg mapping iter to proceed
532 * Proceeds @miter to the next mapping. @miter should have been started
533 * using sg_miter_start(). On successful return, @miter->page,
534 * @miter->addr and @miter->length point to the current mapping.
537 * Preemption disabled if SG_MITER_ATOMIC. Preemption must stay disabled
538 * till @miter is stopped. May sleep if !SG_MITER_ATOMIC.
541 * true if @miter contains the next mapping. false if end of sg
544 bool sg_miter_next(struct sg_mapping_iter *miter)
546 sg_miter_stop(miter);
549 * Get to the next page if necessary.
550 * __remaining, __offset is adjusted by sg_miter_stop
552 if (!sg_miter_get_next_page(miter))
555 miter->page = sg_page_iter_page(&miter->piter);
556 miter->consumed = miter->length = miter->__remaining;
558 if (miter->__flags & SG_MITER_ATOMIC)
559 miter->addr = kmap_atomic(miter->page) + miter->__offset;
561 miter->addr = kmap(miter->page) + miter->__offset;
565 EXPORT_SYMBOL(sg_miter_next);
568 * sg_miter_stop - stop mapping iteration
569 * @miter: sg mapping iter to be stopped
572 * Stops mapping iterator @miter. @miter should have been started
573 * started using sg_miter_start(). A stopped iteration can be
574 * resumed by calling sg_miter_next() on it. This is useful when
575 * resources (kmap) need to be released during iteration.
578 * Preemption disabled if the SG_MITER_ATOMIC is set. Don't care
581 void sg_miter_stop(struct sg_mapping_iter *miter)
583 WARN_ON(miter->consumed > miter->length);
585 /* drop resources from the last iteration */
587 miter->__offset += miter->consumed;
588 miter->__remaining -= miter->consumed;
590 if ((miter->__flags & SG_MITER_TO_SG) &&
591 !PageSlab(miter->page))
592 flush_kernel_dcache_page(miter->page);
594 if (miter->__flags & SG_MITER_ATOMIC) {
595 WARN_ON_ONCE(preemptible());
596 kunmap_atomic(miter->addr);
606 EXPORT_SYMBOL(sg_miter_stop);
609 * sg_copy_buffer - Copy data between a linear buffer and an SG list
611 * @nents: Number of SG entries
612 * @buf: Where to copy from
613 * @buflen: The number of bytes to copy
614 * @skip: Number of bytes to skip before copying
615 * @to_buffer: transfer direction (true == from an sg list to a
616 * buffer, false == from a buffer to an sg list
618 * Returns the number of copied bytes.
621 static size_t sg_copy_buffer(struct scatterlist *sgl, unsigned int nents,
622 void *buf, size_t buflen, off_t skip,
625 unsigned int offset = 0;
626 struct sg_mapping_iter miter;
628 unsigned int sg_flags = SG_MITER_ATOMIC;
631 sg_flags |= SG_MITER_FROM_SG;
633 sg_flags |= SG_MITER_TO_SG;
635 sg_miter_start(&miter, sgl, nents, sg_flags);
637 if (!sg_miter_skip(&miter, skip))
640 local_irq_save(flags);
642 while (sg_miter_next(&miter) && offset < buflen) {
645 len = min(miter.length, buflen - offset);
648 memcpy(buf + offset, miter.addr, len);
650 memcpy(miter.addr, buf + offset, len);
655 sg_miter_stop(&miter);
657 local_irq_restore(flags);
662 * sg_copy_from_buffer - Copy from a linear buffer to an SG list
664 * @nents: Number of SG entries
665 * @buf: Where to copy from
666 * @buflen: The number of bytes to copy
668 * Returns the number of copied bytes.
671 size_t sg_copy_from_buffer(struct scatterlist *sgl, unsigned int nents,
672 void *buf, size_t buflen)
674 return sg_copy_buffer(sgl, nents, buf, buflen, 0, false);
676 EXPORT_SYMBOL(sg_copy_from_buffer);
679 * sg_copy_to_buffer - Copy from an SG list to a linear buffer
681 * @nents: Number of SG entries
682 * @buf: Where to copy to
683 * @buflen: The number of bytes to copy
685 * Returns the number of copied bytes.
688 size_t sg_copy_to_buffer(struct scatterlist *sgl, unsigned int nents,
689 void *buf, size_t buflen)
691 return sg_copy_buffer(sgl, nents, buf, buflen, 0, true);
693 EXPORT_SYMBOL(sg_copy_to_buffer);
696 * sg_pcopy_from_buffer - Copy from a linear buffer to an SG list
698 * @nents: Number of SG entries
699 * @buf: Where to copy from
700 * @skip: Number of bytes to skip before copying
701 * @buflen: The number of bytes to copy
703 * Returns the number of copied bytes.
706 size_t sg_pcopy_from_buffer(struct scatterlist *sgl, unsigned int nents,
707 void *buf, size_t buflen, off_t skip)
709 return sg_copy_buffer(sgl, nents, buf, buflen, skip, false);
711 EXPORT_SYMBOL(sg_pcopy_from_buffer);
714 * sg_pcopy_to_buffer - Copy from an SG list to a linear buffer
716 * @nents: Number of SG entries
717 * @buf: Where to copy to
718 * @skip: Number of bytes to skip before copying
719 * @buflen: The number of bytes to copy
721 * Returns the number of copied bytes.
724 size_t sg_pcopy_to_buffer(struct scatterlist *sgl, unsigned int nents,
725 void *buf, size_t buflen, off_t skip)
727 return sg_copy_buffer(sgl, nents, buf, buflen, skip, true);
729 EXPORT_SYMBOL(sg_pcopy_to_buffer);