blob: 418677dcec60f58d9b8a2d53cbfd8cc611f7032c [file] [log] [blame]
Greg Kroah-Hartmanb2441312017-11-01 15:07:57 +01001// SPDX-License-Identifier: GPL-2.0
David Howells831058d2006-08-29 19:06:00 +01002/* bounce buffer handling for block devices
3 *
4 * - Split from highmem.c
5 */
6
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -07007#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8
David Howells831058d2006-08-29 19:06:00 +01009#include <linux/mm.h>
Paul Gortmakerb95f1b312011-10-16 02:01:52 -040010#include <linux/export.h>
David Howells831058d2006-08-29 19:06:00 +010011#include <linux/swap.h>
Tejun Heo5a0e3ad2010-03-24 17:04:11 +090012#include <linux/gfp.h>
David Howells831058d2006-08-29 19:06:00 +010013#include <linux/bio.h>
14#include <linux/pagemap.h>
15#include <linux/mempool.h>
16#include <linux/blkdev.h>
Tejun Heo66114ca2015-05-22 17:13:32 -040017#include <linux/backing-dev.h>
David Howells831058d2006-08-29 19:06:00 +010018#include <linux/init.h>
19#include <linux/hash.h>
20#include <linux/highmem.h>
David Vrabel3bcfeaf2011-10-20 21:24:30 +020021#include <linux/bootmem.h>
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -070022#include <linux/printk.h>
David Howells831058d2006-08-29 19:06:00 +010023#include <asm/tlbflush.h>
24
Li Zefan55782132009-06-09 13:43:05 +080025#include <trace/events/block.h>
Christoph Hellwig3bce0162017-06-19 09:26:21 +020026#include "blk.h"
Li Zefan55782132009-06-09 13:43:05 +080027
David Howells831058d2006-08-29 19:06:00 +010028#define POOL_SIZE 64
29#define ISA_POOL_SIZE 16
30
Kent Overstreet338aa962018-05-20 18:25:47 -040031static struct bio_set bounce_bio_set, bounce_bio_split;
32static mempool_t page_pool, isa_page_pool;
David Howells831058d2006-08-29 19:06:00 +010033
Jens Axboecf8d0972018-10-21 12:02:36 -060034static void init_bounce_bioset(void)
35{
36 static bool bounce_bs_setup;
37 int ret;
38
39 if (bounce_bs_setup)
40 return;
41
42 ret = bioset_init(&bounce_bio_set, BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
43 BUG_ON(ret);
44 if (bioset_integrity_create(&bounce_bio_set, BIO_POOL_SIZE))
45 BUG_ON(1);
46
47 ret = bioset_init(&bounce_bio_split, BIO_POOL_SIZE, 0, 0);
48 BUG_ON(ret);
49 bounce_bs_setup = true;
50}
51
Arnd Bergmanna687a532018-03-07 23:30:54 +010052#if defined(CONFIG_HIGHMEM)
David Howells831058d2006-08-29 19:06:00 +010053static __init int init_emergency_pool(void)
54{
Kent Overstreet338aa962018-05-20 18:25:47 -040055 int ret;
Chris Metcalff1006252012-06-16 16:41:05 -040056#if defined(CONFIG_HIGHMEM) && !defined(CONFIG_MEMORY_HOTPLUG)
David Vrabel3bcfeaf2011-10-20 21:24:30 +020057 if (max_pfn <= max_low_pfn)
David Howells831058d2006-08-29 19:06:00 +010058 return 0;
David Vrabel3bcfeaf2011-10-20 21:24:30 +020059#endif
David Howells831058d2006-08-29 19:06:00 +010060
Kent Overstreet338aa962018-05-20 18:25:47 -040061 ret = mempool_init_page_pool(&page_pool, POOL_SIZE, 0);
62 BUG_ON(ret);
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -070063 pr_info("pool size: %d pages\n", POOL_SIZE);
David Howells831058d2006-08-29 19:06:00 +010064
Jens Axboecf8d0972018-10-21 12:02:36 -060065 init_bounce_bioset();
David Howells831058d2006-08-29 19:06:00 +010066 return 0;
67}
68
69__initcall(init_emergency_pool);
Chris Metcalff1006252012-06-16 16:41:05 -040070#endif
David Howells831058d2006-08-29 19:06:00 +010071
Chris Metcalff1006252012-06-16 16:41:05 -040072#ifdef CONFIG_HIGHMEM
David Howells831058d2006-08-29 19:06:00 +010073/*
74 * highmem version, map in to vec
75 */
76static void bounce_copy_vec(struct bio_vec *to, unsigned char *vfrom)
77{
David Howells831058d2006-08-29 19:06:00 +010078 unsigned char *vto;
79
Cong Wang9b04c5f2011-11-25 23:14:39 +080080 vto = kmap_atomic(to->bv_page);
David Howells831058d2006-08-29 19:06:00 +010081 memcpy(vto + to->bv_offset, vfrom, to->bv_len);
Cong Wang9b04c5f2011-11-25 23:14:39 +080082 kunmap_atomic(vto);
David Howells831058d2006-08-29 19:06:00 +010083}
84
85#else /* CONFIG_HIGHMEM */
86
87#define bounce_copy_vec(to, vfrom) \
88 memcpy(page_address((to)->bv_page) + (to)->bv_offset, vfrom, (to)->bv_len)
89
90#endif /* CONFIG_HIGHMEM */
91
92/*
93 * allocate pages in the DMA region for the ISA pool
94 */
95static void *mempool_alloc_pages_isa(gfp_t gfp_mask, void *data)
96{
97 return mempool_alloc_pages(gfp_mask | GFP_DMA, data);
98}
99
Jens Axboecf8d0972018-10-21 12:02:36 -0600100static DEFINE_MUTEX(isa_mutex);
101
David Howells831058d2006-08-29 19:06:00 +0100102/*
103 * gets called "every" time someone init's a queue with BLK_BOUNCE_ISA
104 * as the max address, so check if the pool has already been created.
105 */
106int init_emergency_isa_pool(void)
107{
Kent Overstreet338aa962018-05-20 18:25:47 -0400108 int ret;
109
Jens Axboecf8d0972018-10-21 12:02:36 -0600110 mutex_lock(&isa_mutex);
111
112 if (mempool_initialized(&isa_page_pool)) {
113 mutex_unlock(&isa_mutex);
David Howells831058d2006-08-29 19:06:00 +0100114 return 0;
Jens Axboecf8d0972018-10-21 12:02:36 -0600115 }
David Howells831058d2006-08-29 19:06:00 +0100116
Kent Overstreet338aa962018-05-20 18:25:47 -0400117 ret = mempool_init(&isa_page_pool, ISA_POOL_SIZE, mempool_alloc_pages_isa,
118 mempool_free_pages, (void *) 0);
119 BUG_ON(ret);
David Howells831058d2006-08-29 19:06:00 +0100120
Mitchel Humpherysb1de0d12014-06-06 14:38:30 -0700121 pr_info("isa pool size: %d pages\n", ISA_POOL_SIZE);
Jens Axboecf8d0972018-10-21 12:02:36 -0600122 init_bounce_bioset();
123 mutex_unlock(&isa_mutex);
David Howells831058d2006-08-29 19:06:00 +0100124 return 0;
125}
126
127/*
128 * Simple bounce buffer support for highmem pages. Depending on the
129 * queue gfp mask set, *to may or may not be a highmem page. kmap it
130 * always, it will do the Right Thing
131 */
132static void copy_to_high_bio_irq(struct bio *to, struct bio *from)
133{
134 unsigned char *vfrom;
Ming Lei3c892a02017-12-18 20:22:07 +0800135 struct bio_vec tovec, fromvec;
Kent Overstreet79886132013-11-23 17:19:00 -0800136 struct bvec_iter iter;
Ming Lei3c892a02017-12-18 20:22:07 +0800137 /*
138 * The bio of @from is created by bounce, so we can iterate
139 * its bvec from start to end, but the @from->bi_iter can't be
140 * trusted because it might be changed by splitting.
141 */
142 struct bvec_iter from_iter = BVEC_ITER_ALL_INIT;
David Howells831058d2006-08-29 19:06:00 +0100143
Kent Overstreet79886132013-11-23 17:19:00 -0800144 bio_for_each_segment(tovec, to, iter) {
Ming Lei3c892a02017-12-18 20:22:07 +0800145 fromvec = bio_iter_iovec(from, from_iter);
146 if (tovec.bv_page != fromvec.bv_page) {
Kent Overstreet79886132013-11-23 17:19:00 -0800147 /*
148 * fromvec->bv_offset and fromvec->bv_len might have
149 * been modified by the block layer, so use the original
150 * copy, bounce_copy_vec already uses tovec->bv_len
151 */
Ming Lei3c892a02017-12-18 20:22:07 +0800152 vfrom = page_address(fromvec.bv_page) +
Kent Overstreet79886132013-11-23 17:19:00 -0800153 tovec.bv_offset;
David Howells831058d2006-08-29 19:06:00 +0100154
Kent Overstreet79886132013-11-23 17:19:00 -0800155 bounce_copy_vec(&tovec, vfrom);
156 flush_dcache_page(tovec.bv_page);
157 }
Ming Lei3c892a02017-12-18 20:22:07 +0800158 bio_advance_iter(from, &from_iter, tovec.bv_len);
David Howells831058d2006-08-29 19:06:00 +0100159 }
160}
161
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200162static void bounce_end_io(struct bio *bio, mempool_t *pool)
David Howells831058d2006-08-29 19:06:00 +0100163{
164 struct bio *bio_orig = bio->bi_private;
Ming Lei7891f052017-12-18 20:22:06 +0800165 struct bio_vec *bvec, orig_vec;
David Howells831058d2006-08-29 19:06:00 +0100166 int i;
Ming Lei7891f052017-12-18 20:22:06 +0800167 struct bvec_iter orig_iter = bio_orig->bi_iter;
David Howells831058d2006-08-29 19:06:00 +0100168
David Howells831058d2006-08-29 19:06:00 +0100169 /*
170 * free up bounce indirect pages used
171 */
Kent Overstreetd74c6d52013-02-06 12:23:11 -0800172 bio_for_each_segment_all(bvec, bio, i) {
Ming Lei7891f052017-12-18 20:22:06 +0800173 orig_vec = bio_iter_iovec(bio_orig, orig_iter);
174 if (bvec->bv_page != orig_vec.bv_page) {
175 dec_zone_page_state(bvec->bv_page, NR_BOUNCE);
176 mempool_free(bvec->bv_page, pool);
177 }
178 bio_advance_iter(bio_orig, &orig_iter, orig_vec.bv_len);
David Howells831058d2006-08-29 19:06:00 +0100179 }
180
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200181 bio_orig->bi_status = bio->bi_status;
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200182 bio_endio(bio_orig);
David Howells831058d2006-08-29 19:06:00 +0100183 bio_put(bio);
184}
185
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200186static void bounce_end_io_write(struct bio *bio)
David Howells831058d2006-08-29 19:06:00 +0100187{
Kent Overstreet338aa962018-05-20 18:25:47 -0400188 bounce_end_io(bio, &page_pool);
David Howells831058d2006-08-29 19:06:00 +0100189}
190
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200191static void bounce_end_io_write_isa(struct bio *bio)
David Howells831058d2006-08-29 19:06:00 +0100192{
David Howells831058d2006-08-29 19:06:00 +0100193
Kent Overstreet338aa962018-05-20 18:25:47 -0400194 bounce_end_io(bio, &isa_page_pool);
David Howells831058d2006-08-29 19:06:00 +0100195}
196
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200197static void __bounce_end_io_read(struct bio *bio, mempool_t *pool)
David Howells831058d2006-08-29 19:06:00 +0100198{
199 struct bio *bio_orig = bio->bi_private;
200
Christoph Hellwig4e4cbee2017-06-03 09:38:06 +0200201 if (!bio->bi_status)
David Howells831058d2006-08-29 19:06:00 +0100202 copy_to_high_bio_irq(bio_orig, bio);
203
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200204 bounce_end_io(bio, pool);
David Howells831058d2006-08-29 19:06:00 +0100205}
206
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200207static void bounce_end_io_read(struct bio *bio)
David Howells831058d2006-08-29 19:06:00 +0100208{
Kent Overstreet338aa962018-05-20 18:25:47 -0400209 __bounce_end_io_read(bio, &page_pool);
David Howells831058d2006-08-29 19:06:00 +0100210}
211
Christoph Hellwig4246a0b2015-07-20 15:29:37 +0200212static void bounce_end_io_read_isa(struct bio *bio)
David Howells831058d2006-08-29 19:06:00 +0100213{
Kent Overstreet338aa962018-05-20 18:25:47 -0400214 __bounce_end_io_read(bio, &isa_page_pool);
David Howells831058d2006-08-29 19:06:00 +0100215}
216
Christoph Hellwigc55183c2018-07-24 09:52:34 +0200217static struct bio *bounce_clone_bio(struct bio *bio_src, gfp_t gfp_mask,
218 struct bio_set *bs)
219{
220 struct bvec_iter iter;
221 struct bio_vec bv;
222 struct bio *bio;
223
224 /*
225 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
226 * bio_src->bi_io_vec to bio->bi_io_vec.
227 *
228 * We can't do that anymore, because:
229 *
230 * - The point of cloning the biovec is to produce a bio with a biovec
231 * the caller can modify: bi_idx and bi_bvec_done should be 0.
232 *
233 * - The original bio could've had more than BIO_MAX_PAGES biovecs; if
234 * we tried to clone the whole thing bio_alloc_bioset() would fail.
235 * But the clone should succeed as long as the number of biovecs we
236 * actually need to allocate is fewer than BIO_MAX_PAGES.
237 *
238 * - Lastly, bi_vcnt should not be looked at or relied upon by code
239 * that does not own the bio - reason being drivers don't use it for
240 * iterating over the biovec anymore, so expecting it to be kept up
241 * to date (i.e. for clones that share the parent biovec) is just
242 * asking for trouble and would force extra work on
243 * __bio_clone_fast() anyways.
244 */
245
246 bio = bio_alloc_bioset(gfp_mask, bio_segments(bio_src), bs);
247 if (!bio)
248 return NULL;
249 bio->bi_disk = bio_src->bi_disk;
250 bio->bi_opf = bio_src->bi_opf;
251 bio->bi_write_hint = bio_src->bi_write_hint;
252 bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector;
253 bio->bi_iter.bi_size = bio_src->bi_iter.bi_size;
254
255 switch (bio_op(bio)) {
256 case REQ_OP_DISCARD:
257 case REQ_OP_SECURE_ERASE:
258 case REQ_OP_WRITE_ZEROES:
259 break;
260 case REQ_OP_WRITE_SAME:
261 bio->bi_io_vec[bio->bi_vcnt++] = bio_src->bi_io_vec[0];
262 break;
263 default:
264 bio_for_each_segment(bv, bio_src, iter)
265 bio->bi_io_vec[bio->bi_vcnt++] = bv;
266 break;
267 }
268
269 if (bio_integrity(bio_src)) {
270 int ret;
271
272 ret = bio_integrity_clone(bio, bio_src, gfp_mask);
273 if (ret < 0) {
274 bio_put(bio);
275 return NULL;
276 }
277 }
278
279 bio_clone_blkcg_association(bio, bio_src);
280
281 return bio;
282}
283
Jens Axboe165125e2007-07-24 09:28:11 +0200284static void __blk_queue_bounce(struct request_queue *q, struct bio **bio_orig,
Jan Karaa3ad0a92015-06-18 17:19:14 +0200285 mempool_t *pool)
David Howells831058d2006-08-29 19:06:00 +0100286{
Kent Overstreet6bc454d2012-09-10 14:30:37 -0700287 struct bio *bio;
288 int rw = bio_data_dir(*bio_orig);
Kent Overstreet79886132013-11-23 17:19:00 -0800289 struct bio_vec *to, from;
290 struct bvec_iter iter;
NeilBrowna8821f32017-06-18 14:38:58 +1000291 unsigned i = 0;
292 bool bounce = false;
293 int sectors = 0;
Ming Lei14cb0dc2017-12-18 15:40:43 +0800294 bool passthrough = bio_is_passthrough(*bio_orig);
David Howells831058d2006-08-29 19:06:00 +0100295
NeilBrowna8821f32017-06-18 14:38:58 +1000296 bio_for_each_segment(from, *bio_orig, iter) {
297 if (i++ < BIO_MAX_PAGES)
298 sectors += from.bv_len >> 9;
Christoph Hellwig1c4bc3a2017-06-19 09:26:22 +0200299 if (page_to_pfn(from.bv_page) > q->limits.bounce_pfn)
NeilBrowna8821f32017-06-18 14:38:58 +1000300 bounce = true;
301 }
302 if (!bounce)
303 return;
David Howells831058d2006-08-29 19:06:00 +0100304
Ming Lei14cb0dc2017-12-18 15:40:43 +0800305 if (!passthrough && sectors < bio_sectors(*bio_orig)) {
Kent Overstreet338aa962018-05-20 18:25:47 -0400306 bio = bio_split(*bio_orig, sectors, GFP_NOIO, &bounce_bio_split);
NeilBrowna8821f32017-06-18 14:38:58 +1000307 bio_chain(bio, *bio_orig);
308 generic_make_request(*bio_orig);
309 *bio_orig = bio;
310 }
Christoph Hellwigc55183c2018-07-24 09:52:34 +0200311 bio = bounce_clone_bio(*bio_orig, GFP_NOIO, passthrough ? NULL :
Kent Overstreet338aa962018-05-20 18:25:47 -0400312 &bounce_bio_set);
Kent Overstreet6bc454d2012-09-10 14:30:37 -0700313
Kent Overstreetcb34e052012-09-05 15:22:02 -0700314 bio_for_each_segment_all(to, bio, i) {
Kent Overstreet6bc454d2012-09-10 14:30:37 -0700315 struct page *page = to->bv_page;
316
Christoph Hellwig1c4bc3a2017-06-19 09:26:22 +0200317 if (page_to_pfn(page) <= q->limits.bounce_pfn)
David Howells831058d2006-08-29 19:06:00 +0100318 continue;
319
Kent Overstreet6bc454d2012-09-10 14:30:37 -0700320 to->bv_page = mempool_alloc(pool, q->bounce_gfp);
Wang YanQing393a3392015-04-26 16:43:31 +0800321 inc_zone_page_state(to->bv_page, NR_BOUNCE);
David Howells831058d2006-08-29 19:06:00 +0100322
323 if (rw == WRITE) {
324 char *vto, *vfrom;
325
Kent Overstreet6bc454d2012-09-10 14:30:37 -0700326 flush_dcache_page(page);
327
David Howells831058d2006-08-29 19:06:00 +0100328 vto = page_address(to->bv_page) + to->bv_offset;
Kent Overstreet6bc454d2012-09-10 14:30:37 -0700329 vfrom = kmap_atomic(page) + to->bv_offset;
David Howells831058d2006-08-29 19:06:00 +0100330 memcpy(vto, vfrom, to->bv_len);
Kent Overstreet6bc454d2012-09-10 14:30:37 -0700331 kunmap_atomic(vfrom);
David Howells831058d2006-08-29 19:06:00 +0100332 }
333 }
334
Arnaldo Carvalho de Melo5f3ea372008-10-30 08:34:33 +0100335 trace_block_bio_bounce(q, *bio_orig);
Jens Axboec43a5082007-01-12 12:20:26 +0100336
David Howells831058d2006-08-29 19:06:00 +0100337 bio->bi_flags |= (1 << BIO_BOUNCED);
David Howells831058d2006-08-29 19:06:00 +0100338
Kent Overstreet338aa962018-05-20 18:25:47 -0400339 if (pool == &page_pool) {
David Howells831058d2006-08-29 19:06:00 +0100340 bio->bi_end_io = bounce_end_io_write;
341 if (rw == READ)
342 bio->bi_end_io = bounce_end_io_read;
343 } else {
344 bio->bi_end_io = bounce_end_io_write_isa;
345 if (rw == READ)
346 bio->bi_end_io = bounce_end_io_read_isa;
347 }
348
349 bio->bi_private = *bio_orig;
350 *bio_orig = bio;
351}
352
Jens Axboe165125e2007-07-24 09:28:11 +0200353void blk_queue_bounce(struct request_queue *q, struct bio **bio_orig)
David Howells831058d2006-08-29 19:06:00 +0100354{
355 mempool_t *pool;
356
357 /*
Jens Axboebf2de6f2007-09-27 13:01:25 +0200358 * Data-less bio, nothing to bounce
359 */
Jens Axboe36144072008-08-14 13:12:15 +0200360 if (!bio_has_data(*bio_orig))
Jens Axboebf2de6f2007-09-27 13:01:25 +0200361 return;
362
363 /*
David Howells831058d2006-08-29 19:06:00 +0100364 * for non-isa bounce case, just check if the bounce pfn is equal
365 * to or bigger than the highest pfn in the system -- in that case,
366 * don't waste time iterating over bio segments
367 */
368 if (!(q->bounce_gfp & GFP_DMA)) {
Christoph Hellwig1c4bc3a2017-06-19 09:26:22 +0200369 if (q->limits.bounce_pfn >= blk_max_pfn)
David Howells831058d2006-08-29 19:06:00 +0100370 return;
Kent Overstreet338aa962018-05-20 18:25:47 -0400371 pool = &page_pool;
David Howells831058d2006-08-29 19:06:00 +0100372 } else {
Kent Overstreet338aa962018-05-20 18:25:47 -0400373 BUG_ON(!mempool_initialized(&isa_page_pool));
374 pool = &isa_page_pool;
David Howells831058d2006-08-29 19:06:00 +0100375 }
376
David Howells831058d2006-08-29 19:06:00 +0100377 /*
378 * slow path
379 */
Jan Karaa3ad0a92015-06-18 17:19:14 +0200380 __blk_queue_bounce(q, bio_orig, pool);
David Howells831058d2006-08-29 19:06:00 +0100381}