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
);
60 * sg_nents_for_len - return total count of entries in scatterlist
61 * needed to satisfy the supplied length
62 * @sg: The scatterlist
63 * @len: The total required length
66 * Determines the number of entries in sg that are required to meet
67 * the supplied length, taking into acount chaining as well
70 * the number of sg entries needed, negative error on failure
73 int sg_nents_for_len(struct scatterlist
*sg
, u64 len
)
81 for (nents
= 0, total
= 0; sg
; sg
= sg_next(sg
)) {
90 EXPORT_SYMBOL(sg_nents_for_len
);
93 * sg_last - return the last scatterlist entry in a list
94 * @sgl: First entry in the scatterlist
95 * @nents: Number of entries in the scatterlist
98 * Should only be used casually, it (currently) scans the entire list
99 * to get the last entry.
101 * Note that the @sgl@ pointer passed in need not be the first one,
102 * the important bit is that @nents@ denotes the number of entries that
106 struct scatterlist
*sg_last(struct scatterlist
*sgl
, unsigned int nents
)
108 struct scatterlist
*sg
, *ret
= NULL
;
111 for_each_sg(sgl
, sg
, nents
, i
)
114 #ifdef CONFIG_DEBUG_SG
115 BUG_ON(sgl
[0].sg_magic
!= SG_MAGIC
);
116 BUG_ON(!sg_is_last(ret
));
120 EXPORT_SYMBOL(sg_last
);
123 * sg_init_table - Initialize SG table
125 * @nents: Number of entries in table
128 * If this is part of a chained sg table, sg_mark_end() should be
129 * used only on the last table part.
132 void sg_init_table(struct scatterlist
*sgl
, unsigned int nents
)
134 memset(sgl
, 0, sizeof(*sgl
) * nents
);
135 #ifdef CONFIG_DEBUG_SG
138 for (i
= 0; i
< nents
; i
++)
139 sgl
[i
].sg_magic
= SG_MAGIC
;
142 sg_mark_end(&sgl
[nents
- 1]);
144 EXPORT_SYMBOL(sg_init_table
);
147 * sg_init_one - Initialize a single entry sg list
149 * @buf: Virtual address for IO
153 void sg_init_one(struct scatterlist
*sg
, const void *buf
, unsigned int buflen
)
155 sg_init_table(sg
, 1);
156 sg_set_buf(sg
, buf
, buflen
);
158 EXPORT_SYMBOL(sg_init_one
);
161 * The default behaviour of sg_alloc_table() is to use these kmalloc/kfree
164 static struct scatterlist
*sg_kmalloc(unsigned int nents
, gfp_t gfp_mask
)
166 if (nents
== SG_MAX_SINGLE_ALLOC
) {
168 * Kmemleak doesn't track page allocations as they are not
169 * commonly used (in a raw form) for kernel data structures.
170 * As we chain together a list of pages and then a normal
171 * kmalloc (tracked by kmemleak), in order to for that last
172 * allocation not to become decoupled (and thus a
173 * false-positive) we need to inform kmemleak of all the
174 * intermediate allocations.
176 void *ptr
= (void *) __get_free_page(gfp_mask
);
177 kmemleak_alloc(ptr
, PAGE_SIZE
, 1, gfp_mask
);
180 return kmalloc(nents
* sizeof(struct scatterlist
), gfp_mask
);
183 static void sg_kfree(struct scatterlist
*sg
, unsigned int nents
)
185 if (nents
== SG_MAX_SINGLE_ALLOC
) {
187 free_page((unsigned long) sg
);
193 * __sg_free_table - Free a previously mapped sg table
194 * @table: The sg table header to use
195 * @max_ents: The maximum number of entries per single scatterlist
196 * @skip_first_chunk: don't free the (preallocated) first scatterlist chunk
197 * @free_fn: Free function
200 * Free an sg table previously allocated and setup with
201 * __sg_alloc_table(). The @max_ents value must be identical to
202 * that previously used with __sg_alloc_table().
205 void __sg_free_table(struct sg_table
*table
, unsigned int max_ents
,
206 bool skip_first_chunk
, sg_free_fn
*free_fn
)
208 struct scatterlist
*sgl
, *next
;
210 if (unlikely(!table
->sgl
))
214 while (table
->orig_nents
) {
215 unsigned int alloc_size
= table
->orig_nents
;
216 unsigned int sg_size
;
219 * If we have more than max_ents segments left,
220 * then assign 'next' to the sg table after the current one.
221 * sg_size is then one less than alloc size, since the last
222 * element is the chain pointer.
224 if (alloc_size
> max_ents
) {
225 next
= sg_chain_ptr(&sgl
[max_ents
- 1]);
226 alloc_size
= max_ents
;
227 sg_size
= alloc_size
- 1;
229 sg_size
= alloc_size
;
233 table
->orig_nents
-= sg_size
;
234 if (skip_first_chunk
)
235 skip_first_chunk
= false;
237 free_fn(sgl
, alloc_size
);
243 EXPORT_SYMBOL(__sg_free_table
);
246 * sg_free_table - Free a previously allocated sg table
247 * @table: The mapped sg table header
250 void sg_free_table(struct sg_table
*table
)
252 __sg_free_table(table
, SG_MAX_SINGLE_ALLOC
, false, sg_kfree
);
254 EXPORT_SYMBOL(sg_free_table
);
257 * __sg_alloc_table - Allocate and initialize an sg table with given allocator
258 * @table: The sg table header to use
259 * @nents: Number of entries in sg list
260 * @max_ents: The maximum number of entries the allocator returns per call
261 * @gfp_mask: GFP allocation mask
262 * @alloc_fn: Allocator to use
265 * This function returns a @table @nents long. The allocator is
266 * defined to return scatterlist chunks of maximum size @max_ents.
267 * Thus if @nents is bigger than @max_ents, the scatterlists will be
268 * chained in units of @max_ents.
271 * If this function returns non-0 (eg failure), the caller must call
272 * __sg_free_table() to cleanup any leftover allocations.
275 int __sg_alloc_table(struct sg_table
*table
, unsigned int nents
,
276 unsigned int max_ents
, struct scatterlist
*first_chunk
,
277 gfp_t gfp_mask
, sg_alloc_fn
*alloc_fn
)
279 struct scatterlist
*sg
, *prv
;
282 memset(table
, 0, sizeof(*table
));
286 #ifndef CONFIG_ARCH_HAS_SG_CHAIN
287 if (WARN_ON_ONCE(nents
> max_ents
))
294 unsigned int sg_size
, alloc_size
= left
;
296 if (alloc_size
> max_ents
) {
297 alloc_size
= max_ents
;
298 sg_size
= alloc_size
- 1;
300 sg_size
= alloc_size
;
308 sg
= alloc_fn(alloc_size
, gfp_mask
);
312 * Adjust entry count to reflect that the last
313 * entry of the previous table won't be used for
314 * linkage. Without this, sg_kfree() may get
318 table
->nents
= ++table
->orig_nents
;
323 sg_init_table(sg
, alloc_size
);
324 table
->nents
= table
->orig_nents
+= sg_size
;
327 * If this is the first mapping, assign the sg table header.
328 * If this is not the first mapping, chain previous part.
331 sg_chain(prv
, max_ents
, sg
);
336 * If no more entries after this one, mark the end
339 sg_mark_end(&sg
[sg_size
- 1]);
346 EXPORT_SYMBOL(__sg_alloc_table
);
349 * sg_alloc_table - Allocate and initialize an sg table
350 * @table: The sg table header to use
351 * @nents: Number of entries in sg list
352 * @gfp_mask: GFP allocation mask
355 * Allocate and initialize an sg table. If @nents@ is larger than
356 * SG_MAX_SINGLE_ALLOC a chained sg table will be setup.
359 int sg_alloc_table(struct sg_table
*table
, unsigned int nents
, gfp_t gfp_mask
)
363 ret
= __sg_alloc_table(table
, nents
, SG_MAX_SINGLE_ALLOC
,
364 NULL
, gfp_mask
, sg_kmalloc
);
366 __sg_free_table(table
, SG_MAX_SINGLE_ALLOC
, false, sg_kfree
);
370 EXPORT_SYMBOL(sg_alloc_table
);
373 * sg_alloc_table_from_pages - Allocate and initialize an sg table from
375 * @sgt: The sg table header to use
376 * @pages: Pointer to an array of page pointers
377 * @n_pages: Number of pages in the pages array
378 * @offset: Offset from start of the first page to the start of a buffer
379 * @size: Number of valid bytes in the buffer (after offset)
380 * @gfp_mask: GFP allocation mask
383 * Allocate and initialize an sg table from a list of pages. Contiguous
384 * ranges of the pages are squashed into a single scatterlist node. A user
385 * may provide an offset at a start and a size of valid data in a buffer
386 * specified by the page array. The returned sg table is released by
390 * 0 on success, negative error on failure
392 int sg_alloc_table_from_pages(struct sg_table
*sgt
,
393 struct page
**pages
, unsigned int n_pages
,
394 unsigned long offset
, unsigned long size
,
399 unsigned int cur_page
;
401 struct scatterlist
*s
;
403 /* compute number of contiguous chunks */
405 for (i
= 1; i
< n_pages
; ++i
)
406 if (page_to_pfn(pages
[i
]) != page_to_pfn(pages
[i
- 1]) + 1)
409 ret
= sg_alloc_table(sgt
, chunks
, gfp_mask
);
413 /* merging chunks and putting them into the scatterlist */
415 for_each_sg(sgt
->sgl
, s
, sgt
->orig_nents
, i
) {
416 unsigned long chunk_size
;
419 /* look for the end of the current chunk */
420 for (j
= cur_page
+ 1; j
< n_pages
; ++j
)
421 if (page_to_pfn(pages
[j
]) !=
422 page_to_pfn(pages
[j
- 1]) + 1)
425 chunk_size
= ((j
- cur_page
) << PAGE_SHIFT
) - offset
;
426 sg_set_page(s
, pages
[cur_page
], min(size
, chunk_size
), offset
);
434 EXPORT_SYMBOL(sg_alloc_table_from_pages
);
436 #ifdef CONFIG_SGL_ALLOC
439 * sgl_alloc_order - allocate a scatterlist and its pages
440 * @length: Length in bytes of the scatterlist. Must be at least one
441 * @order: Second argument for alloc_pages()
442 * @chainable: Whether or not to allocate an extra element in the scatterlist
443 * for scatterlist chaining purposes
444 * @gfp: Memory allocation flags
445 * @nent_p: [out] Number of entries in the scatterlist that have pages
447 * Returns: A pointer to an initialized scatterlist or %NULL upon failure.
449 struct scatterlist
*sgl_alloc_order(unsigned long long length
,
450 unsigned int order
, bool chainable
,
451 gfp_t gfp
, unsigned int *nent_p
)
453 struct scatterlist
*sgl
, *sg
;
455 unsigned int nent
, nalloc
;
458 nent
= round_up(length
, PAGE_SIZE
<< order
) >> (PAGE_SHIFT
+ order
);
459 /* Check for integer overflow */
460 if (length
> (nent
<< (PAGE_SHIFT
+ order
)))
464 /* Check for integer overflow */
465 if (nalloc
+ 1 < nalloc
)
469 sgl
= kmalloc_array(nalloc
, sizeof(struct scatterlist
),
470 (gfp
& ~GFP_DMA
) | __GFP_ZERO
);
474 sg_init_table(sgl
, nent
);
477 elem_len
= min_t(u64
, length
, PAGE_SIZE
<< order
);
478 page
= alloc_pages(gfp
, order
);
484 sg_set_page(sg
, page
, elem_len
, 0);
493 EXPORT_SYMBOL(sgl_alloc_order
);
496 * sgl_alloc - allocate a scatterlist and its pages
497 * @length: Length in bytes of the scatterlist
498 * @gfp: Memory allocation flags
499 * @nent_p: [out] Number of entries in the scatterlist
501 * Returns: A pointer to an initialized scatterlist or %NULL upon failure.
503 struct scatterlist
*sgl_alloc(unsigned long long length
, gfp_t gfp
,
504 unsigned int *nent_p
)
506 return sgl_alloc_order(length
, 0, false, gfp
, nent_p
);
508 EXPORT_SYMBOL(sgl_alloc
);
511 * sgl_free_order - free a scatterlist and its pages
512 * @sgl: Scatterlist with one or more elements
513 * @order: Second argument for __free_pages()
515 void sgl_free_order(struct scatterlist
*sgl
, int order
)
517 struct scatterlist
*sg
;
520 for (sg
= sgl
; sg
; sg
= sg_next(sg
)) {
523 __free_pages(page
, order
);
527 EXPORT_SYMBOL(sgl_free_order
);
530 * sgl_free - free a scatterlist and its pages
531 * @sgl: Scatterlist with one or more elements
533 void sgl_free(struct scatterlist
*sgl
)
535 sgl_free_order(sgl
, 0);
537 EXPORT_SYMBOL(sgl_free
);
539 #endif /* CONFIG_SGL_ALLOC */
541 void __sg_page_iter_start(struct sg_page_iter
*piter
,
542 struct scatterlist
*sglist
, unsigned int nents
,
543 unsigned long pgoffset
)
545 piter
->__pg_advance
= 0;
546 piter
->__nents
= nents
;
549 piter
->sg_pgoffset
= pgoffset
;
551 EXPORT_SYMBOL(__sg_page_iter_start
);
553 static int sg_page_count(struct scatterlist
*sg
)
555 return PAGE_ALIGN(sg
->offset
+ sg
->length
) >> PAGE_SHIFT
;
558 bool __sg_page_iter_next(struct sg_page_iter
*piter
)
560 if (!piter
->__nents
|| !piter
->sg
)
563 piter
->sg_pgoffset
+= piter
->__pg_advance
;
564 piter
->__pg_advance
= 1;
566 while (piter
->sg_pgoffset
>= sg_page_count(piter
->sg
)) {
567 piter
->sg_pgoffset
-= sg_page_count(piter
->sg
);
568 piter
->sg
= sg_next(piter
->sg
);
569 if (!--piter
->__nents
|| !piter
->sg
)
575 EXPORT_SYMBOL(__sg_page_iter_next
);
578 * sg_miter_start - start mapping iteration over a sg list
579 * @miter: sg mapping iter to be started
580 * @sgl: sg list to iterate over
581 * @nents: number of sg entries
584 * Starts mapping iterator @miter.
589 void sg_miter_start(struct sg_mapping_iter
*miter
, struct scatterlist
*sgl
,
590 unsigned int nents
, unsigned int flags
)
592 memset(miter
, 0, sizeof(struct sg_mapping_iter
));
594 __sg_page_iter_start(&miter
->piter
, sgl
, nents
, 0);
595 WARN_ON(!(flags
& (SG_MITER_TO_SG
| SG_MITER_FROM_SG
)));
596 miter
->__flags
= flags
;
598 EXPORT_SYMBOL(sg_miter_start
);
600 static bool sg_miter_get_next_page(struct sg_mapping_iter
*miter
)
602 if (!miter
->__remaining
) {
603 struct scatterlist
*sg
;
605 if (!__sg_page_iter_next(&miter
->piter
))
608 sg
= miter
->piter
.sg
;
610 miter
->__offset
= miter
->piter
.sg_pgoffset
? 0 : sg
->offset
;
611 miter
->piter
.sg_pgoffset
+= miter
->__offset
>> PAGE_SHIFT
;
612 miter
->__offset
&= PAGE_SIZE
- 1;
613 miter
->__remaining
= sg
->offset
+ sg
->length
-
614 (miter
->piter
.sg_pgoffset
<< PAGE_SHIFT
) -
616 miter
->__remaining
= min_t(unsigned long, miter
->__remaining
,
617 PAGE_SIZE
- miter
->__offset
);
624 * sg_miter_skip - reposition mapping iterator
625 * @miter: sg mapping iter to be skipped
626 * @offset: number of bytes to plus the current location
629 * Sets the offset of @miter to its current location plus @offset bytes.
630 * If mapping iterator @miter has been proceeded by sg_miter_next(), this
634 * Don't care if @miter is stopped, or not proceeded yet.
635 * Otherwise, preemption disabled if the SG_MITER_ATOMIC is set.
638 * true if @miter contains the valid mapping. false if end of sg
641 bool sg_miter_skip(struct sg_mapping_iter
*miter
, off_t offset
)
643 sg_miter_stop(miter
);
648 if (!sg_miter_get_next_page(miter
))
651 consumed
= min_t(off_t
, offset
, miter
->__remaining
);
652 miter
->__offset
+= consumed
;
653 miter
->__remaining
-= consumed
;
659 EXPORT_SYMBOL(sg_miter_skip
);
662 * sg_miter_next - proceed mapping iterator to the next mapping
663 * @miter: sg mapping iter to proceed
666 * Proceeds @miter to the next mapping. @miter should have been started
667 * using sg_miter_start(). On successful return, @miter->page,
668 * @miter->addr and @miter->length point to the current mapping.
671 * Preemption disabled if SG_MITER_ATOMIC. Preemption must stay disabled
672 * till @miter is stopped. May sleep if !SG_MITER_ATOMIC.
675 * true if @miter contains the next mapping. false if end of sg
678 bool sg_miter_next(struct sg_mapping_iter
*miter
)
680 sg_miter_stop(miter
);
683 * Get to the next page if necessary.
684 * __remaining, __offset is adjusted by sg_miter_stop
686 if (!sg_miter_get_next_page(miter
))
689 miter
->page
= sg_page_iter_page(&miter
->piter
);
690 miter
->consumed
= miter
->length
= miter
->__remaining
;
692 if (miter
->__flags
& SG_MITER_ATOMIC
)
693 miter
->addr
= kmap_atomic(miter
->page
) + miter
->__offset
;
695 miter
->addr
= kmap(miter
->page
) + miter
->__offset
;
699 EXPORT_SYMBOL(sg_miter_next
);
702 * sg_miter_stop - stop mapping iteration
703 * @miter: sg mapping iter to be stopped
706 * Stops mapping iterator @miter. @miter should have been started
707 * using sg_miter_start(). A stopped iteration can be resumed by
708 * calling sg_miter_next() on it. This is useful when resources (kmap)
709 * need to be released during iteration.
712 * Preemption disabled if the SG_MITER_ATOMIC is set. Don't care
715 void sg_miter_stop(struct sg_mapping_iter
*miter
)
717 WARN_ON(miter
->consumed
> miter
->length
);
719 /* drop resources from the last iteration */
721 miter
->__offset
+= miter
->consumed
;
722 miter
->__remaining
-= miter
->consumed
;
724 if ((miter
->__flags
& SG_MITER_TO_SG
) &&
725 !PageSlab(miter
->page
))
726 flush_kernel_dcache_page(miter
->page
);
728 if (miter
->__flags
& SG_MITER_ATOMIC
) {
729 WARN_ON_ONCE(preemptible());
730 kunmap_atomic(miter
->addr
);
740 EXPORT_SYMBOL(sg_miter_stop
);
743 * sg_copy_buffer - Copy data between a linear buffer and an SG list
745 * @nents: Number of SG entries
746 * @buf: Where to copy from
747 * @buflen: The number of bytes to copy
748 * @skip: Number of bytes to skip before copying
749 * @to_buffer: transfer direction (true == from an sg list to a
750 * buffer, false == from a buffer to an sg list
752 * Returns the number of copied bytes.
755 size_t sg_copy_buffer(struct scatterlist
*sgl
, unsigned int nents
, void *buf
,
756 size_t buflen
, off_t skip
, bool to_buffer
)
758 unsigned int offset
= 0;
759 struct sg_mapping_iter miter
;
760 unsigned int sg_flags
= SG_MITER_ATOMIC
;
763 sg_flags
|= SG_MITER_FROM_SG
;
765 sg_flags
|= SG_MITER_TO_SG
;
767 sg_miter_start(&miter
, sgl
, nents
, sg_flags
);
769 if (!sg_miter_skip(&miter
, skip
))
772 while ((offset
< buflen
) && sg_miter_next(&miter
)) {
775 len
= min(miter
.length
, buflen
- offset
);
778 memcpy(buf
+ offset
, miter
.addr
, len
);
780 memcpy(miter
.addr
, buf
+ offset
, len
);
785 sg_miter_stop(&miter
);
789 EXPORT_SYMBOL(sg_copy_buffer
);
792 * sg_copy_from_buffer - Copy from a linear buffer to an SG list
794 * @nents: Number of SG entries
795 * @buf: Where to copy from
796 * @buflen: The number of bytes to copy
798 * Returns the number of copied bytes.
801 size_t sg_copy_from_buffer(struct scatterlist
*sgl
, unsigned int nents
,
802 const void *buf
, size_t buflen
)
804 return sg_copy_buffer(sgl
, nents
, (void *)buf
, buflen
, 0, false);
806 EXPORT_SYMBOL(sg_copy_from_buffer
);
809 * sg_copy_to_buffer - Copy from an SG list to a linear buffer
811 * @nents: Number of SG entries
812 * @buf: Where to copy to
813 * @buflen: The number of bytes to copy
815 * Returns the number of copied bytes.
818 size_t sg_copy_to_buffer(struct scatterlist
*sgl
, unsigned int nents
,
819 void *buf
, size_t buflen
)
821 return sg_copy_buffer(sgl
, nents
, buf
, buflen
, 0, true);
823 EXPORT_SYMBOL(sg_copy_to_buffer
);
826 * sg_pcopy_from_buffer - Copy from a linear buffer to an SG list
828 * @nents: Number of SG entries
829 * @buf: Where to copy from
830 * @buflen: The number of bytes to copy
831 * @skip: Number of bytes to skip before copying
833 * Returns the number of copied bytes.
836 size_t sg_pcopy_from_buffer(struct scatterlist
*sgl
, unsigned int nents
,
837 const void *buf
, size_t buflen
, off_t skip
)
839 return sg_copy_buffer(sgl
, nents
, (void *)buf
, buflen
, skip
, false);
841 EXPORT_SYMBOL(sg_pcopy_from_buffer
);
844 * sg_pcopy_to_buffer - Copy from an SG list to a linear buffer
846 * @nents: Number of SG entries
847 * @buf: Where to copy to
848 * @buflen: The number of bytes to copy
849 * @skip: Number of bytes to skip before copying
851 * Returns the number of copied bytes.
854 size_t sg_pcopy_to_buffer(struct scatterlist
*sgl
, unsigned int nents
,
855 void *buf
, size_t buflen
, off_t skip
)
857 return sg_copy_buffer(sgl
, nents
, buf
, buflen
, skip
, true);
859 EXPORT_SYMBOL(sg_pcopy_to_buffer
);
862 * sg_zero_buffer - Zero-out a part of a SG list
864 * @nents: Number of SG entries
865 * @buflen: The number of bytes to zero out
866 * @skip: Number of bytes to skip before zeroing
868 * Returns the number of bytes zeroed.
870 size_t sg_zero_buffer(struct scatterlist
*sgl
, unsigned int nents
,
871 size_t buflen
, off_t skip
)
873 unsigned int offset
= 0;
874 struct sg_mapping_iter miter
;
875 unsigned int sg_flags
= SG_MITER_ATOMIC
| SG_MITER_TO_SG
;
877 sg_miter_start(&miter
, sgl
, nents
, sg_flags
);
879 if (!sg_miter_skip(&miter
, skip
))
882 while (offset
< buflen
&& sg_miter_next(&miter
)) {
885 len
= min(miter
.length
, buflen
- offset
);
886 memset(miter
.addr
, 0, len
);
891 sg_miter_stop(&miter
);
894 EXPORT_SYMBOL(sg_zero_buffer
);