2 * Copyright (c) 2008 Yahoo!, Inc.
4 * Written by: John Baldwin <jhb@FreeBSD.org>
6 * Redistribution and use in source and binary forms, with or without
7 * modification, are permitted provided that the following conditions
9 * 1. Redistributions of source code must retain the above copyright
10 * notice, this list of conditions and the following disclaimer.
11 * 2. Redistributions in binary form must reproduce the above copyright
12 * notice, this list of conditions and the following disclaimer in the
13 * documentation and/or other materials provided with the distribution.
14 * 3. Neither the name of the author nor the names of any co-contributors
15 * may be used to endorse or promote products derived from this software
16 * without specific prior written permission.
18 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
19 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
20 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
21 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
22 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
23 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
24 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
25 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
26 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
27 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 #include <sys/cdefs.h>
32 __FBSDID("$FreeBSD$");
34 #include <sys/param.h>
35 #include <sys/kernel.h>
37 #include <sys/malloc.h>
40 #include <sys/sglist.h>
44 #include <vm/vm_page.h>
46 #include <vm/vm_map.h>
50 static MALLOC_DEFINE(M_SGLIST, "sglist", "scatter/gather lists");
53 * Convenience macros to save the state of an sglist so it can be restored
54 * if an append attempt fails. Since sglist's only grow we only need to
55 * save the current count of segments and the length of the ending segment.
56 * Earlier segments will not be changed by an append, and the only change
57 * that can occur to the ending segment is that it can be extended.
64 #define SGLIST_SAVE(sg, sgsave) do { \
65 (sgsave).sg_nseg = (sg)->sg_nseg; \
66 if ((sgsave).sg_nseg > 0) \
67 (sgsave).ss_len = (sg)->sg_segs[(sgsave).sg_nseg - 1].ss_len; \
69 (sgsave).ss_len = 0; \
72 #define SGLIST_RESTORE(sg, sgsave) do { \
73 (sg)->sg_nseg = (sgsave).sg_nseg; \
74 if ((sgsave).sg_nseg > 0) \
75 (sg)->sg_segs[(sgsave).sg_nseg - 1].ss_len = (sgsave).ss_len; \
79 * Append a single (paddr, len) to a sglist. sg is the list and ss is
80 * the current segment in the list. If we run out of segments then
81 * EFBIG will be returned.
84 _sglist_append_range(struct sglist *sg, struct sglist_seg **ssp,
85 vm_paddr_t paddr, size_t len)
87 struct sglist_seg *ss;
90 if (ss->ss_paddr + ss->ss_len == paddr)
93 if (sg->sg_nseg == sg->sg_maxseg)
105 * Worker routine to append a virtual address range (either kernel or
106 * user) to a scatter/gather list.
109 _sglist_append_buf(struct sglist *sg, void *buf, size_t len, pmap_t pmap,
112 struct sglist_seg *ss;
113 vm_offset_t vaddr, offset;
123 /* Do the first page. It may have an offset. */
124 vaddr = (vm_offset_t)buf;
125 offset = vaddr & PAGE_MASK;
127 paddr = pmap_extract(pmap, vaddr);
129 paddr = pmap_kextract(vaddr);
130 seglen = MIN(len, PAGE_SIZE - offset);
131 if (sg->sg_nseg == 0) {
133 ss->ss_paddr = paddr;
137 ss = &sg->sg_segs[sg->sg_nseg - 1];
138 error = _sglist_append_range(sg, &ss, paddr, seglen);
148 seglen = MIN(len, PAGE_SIZE);
150 paddr = pmap_extract(pmap, vaddr);
152 paddr = pmap_kextract(vaddr);
153 error = _sglist_append_range(sg, &ss, paddr, seglen);
166 * Determine the number of scatter/gather list elements needed to
167 * describe a kernel virtual address range.
170 sglist_count(void *buf, size_t len)
172 vm_offset_t vaddr, vendaddr;
173 vm_paddr_t lastaddr, paddr;
179 vaddr = trunc_page((vm_offset_t)buf);
180 vendaddr = (vm_offset_t)buf + len;
182 lastaddr = pmap_kextract(vaddr);
184 while (vaddr < vendaddr) {
185 paddr = pmap_kextract(vaddr);
186 if (lastaddr + PAGE_SIZE != paddr)
195 * Allocate a scatter/gather list along with 'nsegs' segments. The
196 * 'mflags' parameters are the same as passed to malloc(9). The caller
197 * should use sglist_free() to free this list.
200 sglist_alloc(int nsegs, int mflags)
204 sg = malloc(sizeof(struct sglist) + nsegs * sizeof(struct sglist_seg),
208 sglist_init(sg, nsegs, (struct sglist_seg *)(sg + 1));
213 * Free a scatter/gather list allocated via sglist_allc().
216 sglist_free(struct sglist *sg)
219 if (refcount_release(&sg->sg_refs))
224 * Append the segments to describe a single kernel virtual address
225 * range to a scatter/gather list. If there are insufficient
226 * segments, then this fails with EFBIG.
229 sglist_append(struct sglist *sg, void *buf, size_t len)
234 if (sg->sg_maxseg == 0)
236 SGLIST_SAVE(sg, save);
237 error = _sglist_append_buf(sg, buf, len, NULL, NULL);
239 SGLIST_RESTORE(sg, save);
244 * Append the segments to describe a bio's data to a scatter/gather list.
245 * If there are insufficient segments, then this fails with EFBIG.
247 * NOTE: This function expects bio_bcount to be initialized.
250 sglist_append_bio(struct sglist *sg, struct bio *bp)
255 int error, i, ma_offs;
257 if ((bp->bio_flags & BIO_UNMAPPED) == 0) {
258 error = sglist_append(sg, bp->bio_data, bp->bio_bcount);
262 if (sg->sg_maxseg == 0)
265 SGLIST_SAVE(sg, save);
266 tlen = bp->bio_bcount;
267 ma_offs = bp->bio_ma_offset;
268 for (i = 0; tlen > 0; i++, tlen -= len) {
269 len = min(PAGE_SIZE - ma_offs, tlen);
270 paddr = VM_PAGE_TO_PHYS(bp->bio_ma[i]) + ma_offs;
271 error = sglist_append_phys(sg, paddr, len);
273 SGLIST_RESTORE(sg, save);
282 * Append a single physical address range to a scatter/gather list.
283 * If there are insufficient segments, then this fails with EFBIG.
286 sglist_append_phys(struct sglist *sg, vm_paddr_t paddr, size_t len)
288 struct sglist_seg *ss;
292 if (sg->sg_maxseg == 0)
297 if (sg->sg_nseg == 0) {
298 sg->sg_segs[0].ss_paddr = paddr;
299 sg->sg_segs[0].ss_len = len;
303 ss = &sg->sg_segs[sg->sg_nseg - 1];
304 SGLIST_SAVE(sg, save);
305 error = _sglist_append_range(sg, &ss, paddr, len);
307 SGLIST_RESTORE(sg, save);
312 * Append the segments that describe a single mbuf chain to a
313 * scatter/gather list. If there are insufficient segments, then this
317 sglist_append_mbuf(struct sglist *sg, struct mbuf *m0)
323 if (sg->sg_maxseg == 0)
327 SGLIST_SAVE(sg, save);
328 for (m = m0; m != NULL; m = m->m_next) {
330 error = sglist_append(sg, m->m_data, m->m_len);
332 SGLIST_RESTORE(sg, save);
341 * Append the segments that describe a single user address range to a
342 * scatter/gather list. If there are insufficient segments, then this
346 sglist_append_user(struct sglist *sg, void *buf, size_t len, struct thread *td)
351 if (sg->sg_maxseg == 0)
353 SGLIST_SAVE(sg, save);
354 error = _sglist_append_buf(sg, buf, len,
355 vmspace_pmap(td->td_proc->p_vmspace), NULL);
357 SGLIST_RESTORE(sg, save);
362 * Append the segments that describe a single uio to a scatter/gather
363 * list. If there are insufficient segments, then this fails with
367 sglist_append_uio(struct sglist *sg, struct uio *uio)
371 size_t resid, minlen;
375 if (sg->sg_maxseg == 0)
378 resid = uio->uio_resid;
381 if (uio->uio_segflg == UIO_USERSPACE) {
382 KASSERT(uio->uio_td != NULL,
383 ("sglist_append_uio: USERSPACE but no thread"));
384 pmap = vmspace_pmap(uio->uio_td->td_proc->p_vmspace);
389 SGLIST_SAVE(sg, save);
390 for (i = 0; i < uio->uio_iovcnt && resid != 0; i++) {
392 * Now at the first iovec to load. Load each iovec
393 * until we have exhausted the residual count.
395 minlen = MIN(resid, iov[i].iov_len);
397 error = _sglist_append_buf(sg, iov[i].iov_base, minlen,
400 SGLIST_RESTORE(sg, save);
410 * Append the segments that describe at most 'resid' bytes from a
411 * single uio to a scatter/gather list. If there are insufficient
412 * segments, then only the amount that fits is appended.
415 sglist_consume_uio(struct sglist *sg, struct uio *uio, size_t resid)
422 if (sg->sg_maxseg == 0)
425 if (uio->uio_segflg == UIO_USERSPACE) {
426 KASSERT(uio->uio_td != NULL,
427 ("sglist_consume_uio: USERSPACE but no thread"));
428 pmap = vmspace_pmap(uio->uio_td->td_proc->p_vmspace);
433 while (resid > 0 && uio->uio_resid) {
445 * Try to append this iovec. If we run out of room,
446 * then break out of the loop.
448 error = _sglist_append_buf(sg, iov->iov_base, len, pmap, &done);
449 iov->iov_base = (char *)iov->iov_base + done;
450 iov->iov_len -= done;
451 uio->uio_resid -= done;
452 uio->uio_offset += done;
461 * Allocate and populate a scatter/gather list to describe a single
462 * kernel virtual address range.
465 sglist_build(void *buf, size_t len, int mflags)
473 nsegs = sglist_count(buf, len);
474 sg = sglist_alloc(nsegs, mflags);
477 if (sglist_append(sg, buf, len) != 0) {
485 * Clone a new copy of a scatter/gather list.
488 sglist_clone(struct sglist *sg, int mflags)
494 new = sglist_alloc(sg->sg_maxseg, mflags);
497 new->sg_nseg = sg->sg_nseg;
498 bcopy(sg->sg_segs, new->sg_segs, sizeof(struct sglist_seg) *
504 * Calculate the total length of the segments described in a
505 * scatter/gather list.
508 sglist_length(struct sglist *sg)
514 for (i = 0; i < sg->sg_nseg; i++)
515 space += sg->sg_segs[i].ss_len;
520 * Split a scatter/gather list into two lists. The scatter/gather
521 * entries for the first 'length' bytes of the 'original' list are
522 * stored in the '*head' list and are removed from 'original'.
524 * If '*head' is NULL, then a new list will be allocated using
525 * 'mflags'. If M_NOWAIT is specified and the allocation fails,
526 * ENOMEM will be returned.
528 * If '*head' is not NULL, it should point to an empty sglist. If it
529 * does not have enough room for the remaining space, then EFBIG will
530 * be returned. If '*head' is not empty, then EINVAL will be
533 * If 'original' is shared (refcount > 1), then EDOOFUS will be
537 sglist_split(struct sglist *original, struct sglist **head, size_t length,
544 if (original->sg_refs > 1)
547 /* Figure out how big of a sglist '*head' has to hold. */
551 for (i = 0; i < original->sg_nseg; i++) {
552 space += original->sg_segs[i].ss_len;
554 if (space >= length) {
556 * If 'length' falls in the middle of a
557 * scatter/gather list entry, then 'split'
558 * holds how much of that entry will remain in
561 split = space - length;
566 /* Nothing to do, so leave head empty. */
571 sg = sglist_alloc(count, mflags);
577 if (sg->sg_maxseg < count)
579 if (sg->sg_nseg != 0)
583 /* Copy 'count' entries to 'sg' from 'original'. */
584 bcopy(original->sg_segs, sg->sg_segs, count *
585 sizeof(struct sglist_seg));
589 * If we had to split a list entry, fixup the last entry in
590 * 'sg' and the new first entry in 'original'. We also
591 * decrement 'count' by 1 since we will only be removing
592 * 'count - 1' segments from 'original' now.
596 sg->sg_segs[count].ss_len -= split;
597 original->sg_segs[count].ss_paddr =
598 sg->sg_segs[count].ss_paddr + split;
599 original->sg_segs[count].ss_len = split;
602 /* Trim 'count' entries from the front of 'original'. */
603 original->sg_nseg -= count;
604 bcopy(original->sg_segs + count, original->sg_segs, count *
605 sizeof(struct sglist_seg));
610 * Append the scatter/gather list elements in 'second' to the
611 * scatter/gather list 'first'. If there is not enough space in
612 * 'first', EFBIG is returned.
615 sglist_join(struct sglist *first, struct sglist *second)
617 struct sglist_seg *flast, *sfirst;
620 /* If 'second' is empty, there is nothing to do. */
621 if (second->sg_nseg == 0)
625 * If the first entry in 'second' can be appended to the last entry
626 * in 'first' then set append to '1'.
629 flast = &first->sg_segs[first->sg_nseg - 1];
630 sfirst = &second->sg_segs[0];
631 if (first->sg_nseg != 0 &&
632 flast->ss_paddr + flast->ss_len == sfirst->ss_paddr)
635 /* Make sure 'first' has enough room. */
636 if (first->sg_nseg + second->sg_nseg - append > first->sg_maxseg)
639 /* Merge last in 'first' and first in 'second' if needed. */
641 flast->ss_len += sfirst->ss_len;
643 /* Append new segments from 'second' to 'first'. */
644 bcopy(first->sg_segs + first->sg_nseg, second->sg_segs + append,
645 (second->sg_nseg - append) * sizeof(struct sglist_seg));
646 first->sg_nseg += second->sg_nseg - append;
647 sglist_reset(second);
652 * Generate a new scatter/gather list from a range of an existing
653 * scatter/gather list. The 'offset' and 'length' parameters specify
654 * the logical range of the 'original' list to extract. If that range
655 * is not a subset of the length of 'original', then EINVAL is
656 * returned. The new scatter/gather list is stored in '*slice'.
658 * If '*slice' is NULL, then a new list will be allocated using
659 * 'mflags'. If M_NOWAIT is specified and the allocation fails,
660 * ENOMEM will be returned.
662 * If '*slice' is not NULL, it should point to an empty sglist. If it
663 * does not have enough room for the remaining space, then EFBIG will
664 * be returned. If '*slice' is not empty, then EINVAL will be
668 sglist_slice(struct sglist *original, struct sglist **slice, size_t offset,
669 size_t length, int mflags)
672 size_t space, end, foffs, loffs;
679 /* Figure out how many segments '*slice' needs to have. */
680 end = offset + length;
685 for (i = 0; i < original->sg_nseg; i++) {
686 space += original->sg_segs[i].ss_len;
687 if (space > offset) {
689 * When we hit the first segment, store its index
690 * in 'fseg' and the offset into the first segment
691 * of 'offset' in 'foffs'.
695 foffs = offset - (space -
696 original->sg_segs[i].ss_len);
697 CTR1(KTR_DEV, "sglist_slice: foffs = %08lx",
703 * When we hit the last segment, break out of
704 * the loop. Store the amount of extra space
705 * at the end of this segment in 'loffs'.
709 CTR1(KTR_DEV, "sglist_slice: loffs = %08lx",
716 /* If we never hit 'end', then 'length' ran off the end, so fail. */
720 if (*slice == NULL) {
721 sg = sglist_alloc(count, mflags);
727 if (sg->sg_maxseg < count)
729 if (sg->sg_nseg != 0)
734 * Copy over 'count' segments from 'original' starting at
737 bcopy(original->sg_segs + fseg, sg->sg_segs,
738 count * sizeof(struct sglist_seg));
741 /* Fixup first and last segments if needed. */
743 sg->sg_segs[0].ss_paddr += foffs;
744 sg->sg_segs[0].ss_len -= foffs;
745 CTR2(KTR_DEV, "sglist_slice seg[0]: %08lx:%08lx",
746 (long)sg->sg_segs[0].ss_paddr, sg->sg_segs[0].ss_len);
749 sg->sg_segs[count - 1].ss_len -= loffs;
750 CTR2(KTR_DEV, "sglist_slice seg[%d]: len %08x", count - 1,
751 sg->sg_segs[count - 1].ss_len);