Loading...
1// SPDX-License-Identifier: GPL-2.0
2/* bounce buffer handling for block devices
3 *
4 * - Split from highmem.c
5 */
6
7#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8
9#include <linux/mm.h>
10#include <linux/export.h>
11#include <linux/swap.h>
12#include <linux/gfp.h>
13#include <linux/bio.h>
14#include <linux/pagemap.h>
15#include <linux/mempool.h>
16#include <linux/blkdev.h>
17#include <linux/backing-dev.h>
18#include <linux/init.h>
19#include <linux/hash.h>
20#include <linux/highmem.h>
21#include <linux/bootmem.h>
22#include <linux/printk.h>
23#include <asm/tlbflush.h>
24
25#include <trace/events/block.h>
26#include "blk.h"
27
28#define POOL_SIZE 64
29#define ISA_POOL_SIZE 16
30
31static struct bio_set *bounce_bio_set, *bounce_bio_split;
32static mempool_t *page_pool, *isa_page_pool;
33
34#if defined(CONFIG_HIGHMEM)
35static __init int init_emergency_pool(void)
36{
37#if defined(CONFIG_HIGHMEM) && !defined(CONFIG_MEMORY_HOTPLUG)
38 if (max_pfn <= max_low_pfn)
39 return 0;
40#endif
41
42 page_pool = mempool_create_page_pool(POOL_SIZE, 0);
43 BUG_ON(!page_pool);
44 pr_info("pool size: %d pages\n", POOL_SIZE);
45
46 bounce_bio_set = bioset_create(BIO_POOL_SIZE, 0, BIOSET_NEED_BVECS);
47 BUG_ON(!bounce_bio_set);
48 if (bioset_integrity_create(bounce_bio_set, BIO_POOL_SIZE))
49 BUG_ON(1);
50
51 bounce_bio_split = bioset_create(BIO_POOL_SIZE, 0, 0);
52 BUG_ON(!bounce_bio_split);
53
54 return 0;
55}
56
57__initcall(init_emergency_pool);
58#endif
59
60#ifdef CONFIG_HIGHMEM
61/*
62 * highmem version, map in to vec
63 */
64static void bounce_copy_vec(struct bio_vec *to, unsigned char *vfrom)
65{
66 unsigned long flags;
67 unsigned char *vto;
68
69 local_irq_save(flags);
70 vto = kmap_atomic(to->bv_page);
71 memcpy(vto + to->bv_offset, vfrom, to->bv_len);
72 kunmap_atomic(vto);
73 local_irq_restore(flags);
74}
75
76#else /* CONFIG_HIGHMEM */
77
78#define bounce_copy_vec(to, vfrom) \
79 memcpy(page_address((to)->bv_page) + (to)->bv_offset, vfrom, (to)->bv_len)
80
81#endif /* CONFIG_HIGHMEM */
82
83/*
84 * allocate pages in the DMA region for the ISA pool
85 */
86static void *mempool_alloc_pages_isa(gfp_t gfp_mask, void *data)
87{
88 return mempool_alloc_pages(gfp_mask | GFP_DMA, data);
89}
90
91/*
92 * gets called "every" time someone init's a queue with BLK_BOUNCE_ISA
93 * as the max address, so check if the pool has already been created.
94 */
95int init_emergency_isa_pool(void)
96{
97 if (isa_page_pool)
98 return 0;
99
100 isa_page_pool = mempool_create(ISA_POOL_SIZE, mempool_alloc_pages_isa,
101 mempool_free_pages, (void *) 0);
102 BUG_ON(!isa_page_pool);
103
104 pr_info("isa pool size: %d pages\n", ISA_POOL_SIZE);
105 return 0;
106}
107
108/*
109 * Simple bounce buffer support for highmem pages. Depending on the
110 * queue gfp mask set, *to may or may not be a highmem page. kmap it
111 * always, it will do the Right Thing
112 */
113static void copy_to_high_bio_irq(struct bio *to, struct bio *from)
114{
115 unsigned char *vfrom;
116 struct bio_vec tovec, fromvec;
117 struct bvec_iter iter;
118 /*
119 * The bio of @from is created by bounce, so we can iterate
120 * its bvec from start to end, but the @from->bi_iter can't be
121 * trusted because it might be changed by splitting.
122 */
123 struct bvec_iter from_iter = BVEC_ITER_ALL_INIT;
124
125 bio_for_each_segment(tovec, to, iter) {
126 fromvec = bio_iter_iovec(from, from_iter);
127 if (tovec.bv_page != fromvec.bv_page) {
128 /*
129 * fromvec->bv_offset and fromvec->bv_len might have
130 * been modified by the block layer, so use the original
131 * copy, bounce_copy_vec already uses tovec->bv_len
132 */
133 vfrom = page_address(fromvec.bv_page) +
134 tovec.bv_offset;
135
136 bounce_copy_vec(&tovec, vfrom);
137 flush_dcache_page(tovec.bv_page);
138 }
139 bio_advance_iter(from, &from_iter, tovec.bv_len);
140 }
141}
142
143static void bounce_end_io(struct bio *bio, mempool_t *pool)
144{
145 struct bio *bio_orig = bio->bi_private;
146 struct bio_vec *bvec, orig_vec;
147 int i;
148 struct bvec_iter orig_iter = bio_orig->bi_iter;
149
150 /*
151 * free up bounce indirect pages used
152 */
153 bio_for_each_segment_all(bvec, bio, i) {
154 orig_vec = bio_iter_iovec(bio_orig, orig_iter);
155 if (bvec->bv_page != orig_vec.bv_page) {
156 dec_zone_page_state(bvec->bv_page, NR_BOUNCE);
157 mempool_free(bvec->bv_page, pool);
158 }
159 bio_advance_iter(bio_orig, &orig_iter, orig_vec.bv_len);
160 }
161
162 bio_orig->bi_status = bio->bi_status;
163 bio_endio(bio_orig);
164 bio_put(bio);
165}
166
167static void bounce_end_io_write(struct bio *bio)
168{
169 bounce_end_io(bio, page_pool);
170}
171
172static void bounce_end_io_write_isa(struct bio *bio)
173{
174
175 bounce_end_io(bio, isa_page_pool);
176}
177
178static void __bounce_end_io_read(struct bio *bio, mempool_t *pool)
179{
180 struct bio *bio_orig = bio->bi_private;
181
182 if (!bio->bi_status)
183 copy_to_high_bio_irq(bio_orig, bio);
184
185 bounce_end_io(bio, pool);
186}
187
188static void bounce_end_io_read(struct bio *bio)
189{
190 __bounce_end_io_read(bio, page_pool);
191}
192
193static void bounce_end_io_read_isa(struct bio *bio)
194{
195 __bounce_end_io_read(bio, isa_page_pool);
196}
197
198static void __blk_queue_bounce(struct request_queue *q, struct bio **bio_orig,
199 mempool_t *pool)
200{
201 struct bio *bio;
202 int rw = bio_data_dir(*bio_orig);
203 struct bio_vec *to, from;
204 struct bvec_iter iter;
205 unsigned i = 0;
206 bool bounce = false;
207 int sectors = 0;
208 bool passthrough = bio_is_passthrough(*bio_orig);
209
210 bio_for_each_segment(from, *bio_orig, iter) {
211 if (i++ < BIO_MAX_PAGES)
212 sectors += from.bv_len >> 9;
213 if (page_to_pfn(from.bv_page) > q->limits.bounce_pfn)
214 bounce = true;
215 }
216 if (!bounce)
217 return;
218
219 if (!passthrough && sectors < bio_sectors(*bio_orig)) {
220 bio = bio_split(*bio_orig, sectors, GFP_NOIO, bounce_bio_split);
221 bio_chain(bio, *bio_orig);
222 generic_make_request(*bio_orig);
223 *bio_orig = bio;
224 }
225 bio = bio_clone_bioset(*bio_orig, GFP_NOIO, passthrough ? NULL :
226 bounce_bio_set);
227
228 bio_for_each_segment_all(to, bio, i) {
229 struct page *page = to->bv_page;
230
231 if (page_to_pfn(page) <= q->limits.bounce_pfn)
232 continue;
233
234 to->bv_page = mempool_alloc(pool, q->bounce_gfp);
235 inc_zone_page_state(to->bv_page, NR_BOUNCE);
236
237 if (rw == WRITE) {
238 char *vto, *vfrom;
239
240 flush_dcache_page(page);
241
242 vto = page_address(to->bv_page) + to->bv_offset;
243 vfrom = kmap_atomic(page) + to->bv_offset;
244 memcpy(vto, vfrom, to->bv_len);
245 kunmap_atomic(vfrom);
246 }
247 }
248
249 trace_block_bio_bounce(q, *bio_orig);
250
251 bio->bi_flags |= (1 << BIO_BOUNCED);
252
253 if (pool == page_pool) {
254 bio->bi_end_io = bounce_end_io_write;
255 if (rw == READ)
256 bio->bi_end_io = bounce_end_io_read;
257 } else {
258 bio->bi_end_io = bounce_end_io_write_isa;
259 if (rw == READ)
260 bio->bi_end_io = bounce_end_io_read_isa;
261 }
262
263 bio->bi_private = *bio_orig;
264 *bio_orig = bio;
265}
266
267void blk_queue_bounce(struct request_queue *q, struct bio **bio_orig)
268{
269 mempool_t *pool;
270
271 /*
272 * Data-less bio, nothing to bounce
273 */
274 if (!bio_has_data(*bio_orig))
275 return;
276
277 /*
278 * for non-isa bounce case, just check if the bounce pfn is equal
279 * to or bigger than the highest pfn in the system -- in that case,
280 * don't waste time iterating over bio segments
281 */
282 if (!(q->bounce_gfp & GFP_DMA)) {
283 if (q->limits.bounce_pfn >= blk_max_pfn)
284 return;
285 pool = page_pool;
286 } else {
287 BUG_ON(!isa_page_pool);
288 pool = isa_page_pool;
289 }
290
291 /*
292 * slow path
293 */
294 __blk_queue_bounce(q, bio_orig, pool);
295}
1// SPDX-License-Identifier: GPL-2.0
2/* bounce buffer handling for block devices
3 *
4 * - Split from highmem.c
5 */
6
7#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
8
9#include <linux/mm.h>
10#include <linux/export.h>
11#include <linux/swap.h>
12#include <linux/gfp.h>
13#include <linux/bio.h>
14#include <linux/pagemap.h>
15#include <linux/mempool.h>
16#include <linux/blkdev.h>
17#include <linux/backing-dev.h>
18#include <linux/init.h>
19#include <linux/hash.h>
20#include <linux/highmem.h>
21#include <linux/printk.h>
22#include <asm/tlbflush.h>
23
24#include <trace/events/block.h>
25#include "blk.h"
26#include "blk-cgroup.h"
27
28#define POOL_SIZE 64
29#define ISA_POOL_SIZE 16
30
31static struct bio_set bounce_bio_set, bounce_bio_split;
32static mempool_t page_pool;
33
34static 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
52static __init int init_emergency_pool(void)
53{
54 int ret;
55
56#ifndef CONFIG_MEMORY_HOTPLUG
57 if (max_pfn <= max_low_pfn)
58 return 0;
59#endif
60
61 ret = mempool_init_page_pool(&page_pool, POOL_SIZE, 0);
62 BUG_ON(ret);
63 pr_info("pool size: %d pages\n", POOL_SIZE);
64
65 init_bounce_bioset();
66 return 0;
67}
68
69__initcall(init_emergency_pool);
70
71/*
72 * Simple bounce buffer support for highmem pages. Depending on the
73 * queue gfp mask set, *to may or may not be a highmem page. kmap it
74 * always, it will do the Right Thing
75 */
76static void copy_to_high_bio_irq(struct bio *to, struct bio *from)
77{
78 struct bio_vec tovec, fromvec;
79 struct bvec_iter iter;
80 /*
81 * The bio of @from is created by bounce, so we can iterate
82 * its bvec from start to end, but the @from->bi_iter can't be
83 * trusted because it might be changed by splitting.
84 */
85 struct bvec_iter from_iter = BVEC_ITER_ALL_INIT;
86
87 bio_for_each_segment(tovec, to, iter) {
88 fromvec = bio_iter_iovec(from, from_iter);
89 if (tovec.bv_page != fromvec.bv_page) {
90 /*
91 * fromvec->bv_offset and fromvec->bv_len might have
92 * been modified by the block layer, so use the original
93 * copy, bounce_copy_vec already uses tovec->bv_len
94 */
95 memcpy_to_bvec(&tovec, page_address(fromvec.bv_page) +
96 tovec.bv_offset);
97 }
98 bio_advance_iter(from, &from_iter, tovec.bv_len);
99 }
100}
101
102static void bounce_end_io(struct bio *bio)
103{
104 struct bio *bio_orig = bio->bi_private;
105 struct bio_vec *bvec, orig_vec;
106 struct bvec_iter orig_iter = bio_orig->bi_iter;
107 struct bvec_iter_all iter_all;
108
109 /*
110 * free up bounce indirect pages used
111 */
112 bio_for_each_segment_all(bvec, bio, iter_all) {
113 orig_vec = bio_iter_iovec(bio_orig, orig_iter);
114 if (bvec->bv_page != orig_vec.bv_page) {
115 dec_zone_page_state(bvec->bv_page, NR_BOUNCE);
116 mempool_free(bvec->bv_page, &page_pool);
117 }
118 bio_advance_iter(bio_orig, &orig_iter, orig_vec.bv_len);
119 }
120
121 bio_orig->bi_status = bio->bi_status;
122 bio_endio(bio_orig);
123 bio_put(bio);
124}
125
126static void bounce_end_io_write(struct bio *bio)
127{
128 bounce_end_io(bio);
129}
130
131static void bounce_end_io_read(struct bio *bio)
132{
133 struct bio *bio_orig = bio->bi_private;
134
135 if (!bio->bi_status)
136 copy_to_high_bio_irq(bio_orig, bio);
137
138 bounce_end_io(bio);
139}
140
141static struct bio *bounce_clone_bio(struct bio *bio_src)
142{
143 struct bvec_iter iter;
144 struct bio_vec bv;
145 struct bio *bio;
146
147 /*
148 * Pre immutable biovecs, __bio_clone() used to just do a memcpy from
149 * bio_src->bi_io_vec to bio->bi_io_vec.
150 *
151 * We can't do that anymore, because:
152 *
153 * - The point of cloning the biovec is to produce a bio with a biovec
154 * the caller can modify: bi_idx and bi_bvec_done should be 0.
155 *
156 * - The original bio could've had more than BIO_MAX_VECS biovecs; if
157 * we tried to clone the whole thing bio_alloc_bioset() would fail.
158 * But the clone should succeed as long as the number of biovecs we
159 * actually need to allocate is fewer than BIO_MAX_VECS.
160 *
161 * - Lastly, bi_vcnt should not be looked at or relied upon by code
162 * that does not own the bio - reason being drivers don't use it for
163 * iterating over the biovec anymore, so expecting it to be kept up
164 * to date (i.e. for clones that share the parent biovec) is just
165 * asking for trouble and would force extra work.
166 */
167 bio = bio_alloc_bioset(bio_src->bi_bdev, bio_segments(bio_src),
168 bio_src->bi_opf, GFP_NOIO, &bounce_bio_set);
169 if (bio_flagged(bio_src, BIO_REMAPPED))
170 bio_set_flag(bio, BIO_REMAPPED);
171 bio->bi_ioprio = bio_src->bi_ioprio;
172 bio->bi_write_hint = bio_src->bi_write_hint;
173 bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector;
174 bio->bi_iter.bi_size = bio_src->bi_iter.bi_size;
175
176 switch (bio_op(bio)) {
177 case REQ_OP_DISCARD:
178 case REQ_OP_SECURE_ERASE:
179 case REQ_OP_WRITE_ZEROES:
180 break;
181 default:
182 bio_for_each_segment(bv, bio_src, iter)
183 bio->bi_io_vec[bio->bi_vcnt++] = bv;
184 break;
185 }
186
187 if (bio_crypt_clone(bio, bio_src, GFP_NOIO) < 0)
188 goto err_put;
189
190 if (bio_integrity(bio_src) &&
191 bio_integrity_clone(bio, bio_src, GFP_NOIO) < 0)
192 goto err_put;
193
194 bio_clone_blkg_association(bio, bio_src);
195
196 return bio;
197
198err_put:
199 bio_put(bio);
200 return NULL;
201}
202
203struct bio *__blk_queue_bounce(struct bio *bio_orig, struct request_queue *q)
204{
205 struct bio *bio;
206 int rw = bio_data_dir(bio_orig);
207 struct bio_vec *to, from;
208 struct bvec_iter iter;
209 unsigned i = 0, bytes = 0;
210 bool bounce = false;
211 int sectors;
212
213 bio_for_each_segment(from, bio_orig, iter) {
214 if (i++ < BIO_MAX_VECS)
215 bytes += from.bv_len;
216 if (PageHighMem(from.bv_page))
217 bounce = true;
218 }
219 if (!bounce)
220 return bio_orig;
221
222 /*
223 * Individual bvecs might not be logical block aligned. Round down
224 * the split size so that each bio is properly block size aligned,
225 * even if we do not use the full hardware limits.
226 */
227 sectors = ALIGN_DOWN(bytes, queue_logical_block_size(q)) >>
228 SECTOR_SHIFT;
229 if (sectors < bio_sectors(bio_orig)) {
230 bio = bio_split(bio_orig, sectors, GFP_NOIO, &bounce_bio_split);
231 bio_chain(bio, bio_orig);
232 submit_bio_noacct(bio_orig);
233 bio_orig = bio;
234 }
235 bio = bounce_clone_bio(bio_orig);
236
237 /*
238 * Bvec table can't be updated by bio_for_each_segment_all(),
239 * so retrieve bvec from the table directly. This way is safe
240 * because the 'bio' is single-page bvec.
241 */
242 for (i = 0, to = bio->bi_io_vec; i < bio->bi_vcnt; to++, i++) {
243 struct page *bounce_page;
244
245 if (!PageHighMem(to->bv_page))
246 continue;
247
248 bounce_page = mempool_alloc(&page_pool, GFP_NOIO);
249 inc_zone_page_state(bounce_page, NR_BOUNCE);
250
251 if (rw == WRITE) {
252 flush_dcache_page(to->bv_page);
253 memcpy_from_bvec(page_address(bounce_page), to);
254 }
255 to->bv_page = bounce_page;
256 }
257
258 trace_block_bio_bounce(bio_orig);
259
260 bio->bi_flags |= (1 << BIO_BOUNCED);
261
262 if (rw == READ)
263 bio->bi_end_io = bounce_end_io_read;
264 else
265 bio->bi_end_io = bounce_end_io_write;
266
267 bio->bi_private = bio_orig;
268 return bio;
269}