1/* bounce buffer handling for block devices
2 *
3 * - Split from highmem.c
4 */
5
6#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
7
8#include <linux/mm.h>
9#include <linux/export.h>
10#include <linux/swap.h>
11#include <linux/gfp.h>
12#include <linux/bio.h>
13#include <linux/pagemap.h>
14#include <linux/mempool.h>
15#include <linux/blkdev.h>
16#include <linux/init.h>
17#include <linux/hash.h>
18#include <linux/highmem.h>
19#include <linux/bootmem.h>
20#include <linux/printk.h>
21#include <asm/tlbflush.h>
22
23#include <trace/events/block.h>
24
25#define POOL_SIZE	64
26#define ISA_POOL_SIZE	16
27
28static mempool_t *page_pool, *isa_page_pool;
29
30#if defined(CONFIG_HIGHMEM) || defined(CONFIG_NEED_BOUNCE_POOL)
31static __init int init_emergency_pool(void)
32{
33#if defined(CONFIG_HIGHMEM) && !defined(CONFIG_MEMORY_HOTPLUG)
34	if (max_pfn <= max_low_pfn)
35		return 0;
36#endif
37
38	page_pool = mempool_create_page_pool(POOL_SIZE, 0);
39	BUG_ON(!page_pool);
40	pr_info("pool size: %d pages\n", POOL_SIZE);
41
42	return 0;
43}
44
45__initcall(init_emergency_pool);
46#endif
47
48#ifdef CONFIG_HIGHMEM
49/*
50 * highmem version, map in to vec
51 */
52static void bounce_copy_vec(struct bio_vec *to, unsigned char *vfrom)
53{
54	unsigned long flags;
55	unsigned char *vto;
56
57	local_irq_save(flags);
58	vto = kmap_atomic(to->bv_page);
59	memcpy(vto + to->bv_offset, vfrom, to->bv_len);
60	kunmap_atomic(vto);
61	local_irq_restore(flags);
62}
63
64#else /* CONFIG_HIGHMEM */
65
66#define bounce_copy_vec(to, vfrom)	\
67	memcpy(page_address((to)->bv_page) + (to)->bv_offset, vfrom, (to)->bv_len)
68
69#endif /* CONFIG_HIGHMEM */
70
71/*
72 * allocate pages in the DMA region for the ISA pool
73 */
74static void *mempool_alloc_pages_isa(gfp_t gfp_mask, void *data)
75{
76	return mempool_alloc_pages(gfp_mask | GFP_DMA, data);
77}
78
79/*
80 * gets called "every" time someone init's a queue with BLK_BOUNCE_ISA
81 * as the max address, so check if the pool has already been created.
82 */
83int init_emergency_isa_pool(void)
84{
85	if (isa_page_pool)
86		return 0;
87
88	isa_page_pool = mempool_create(ISA_POOL_SIZE, mempool_alloc_pages_isa,
89				       mempool_free_pages, (void *) 0);
90	BUG_ON(!isa_page_pool);
91
92	pr_info("isa pool size: %d pages\n", ISA_POOL_SIZE);
93	return 0;
94}
95
96/*
97 * Simple bounce buffer support for highmem pages. Depending on the
98 * queue gfp mask set, *to may or may not be a highmem page. kmap it
99 * always, it will do the Right Thing
100 */
101static void copy_to_high_bio_irq(struct bio *to, struct bio *from)
102{
103	unsigned char *vfrom;
104	struct bio_vec tovec, *fromvec = from->bi_io_vec;
105	struct bvec_iter iter;
106
107	bio_for_each_segment(tovec, to, iter) {
108		if (tovec.bv_page != fromvec->bv_page) {
109			/*
110			 * fromvec->bv_offset and fromvec->bv_len might have
111			 * been modified by the block layer, so use the original
112			 * copy, bounce_copy_vec already uses tovec->bv_len
113			 */
114			vfrom = page_address(fromvec->bv_page) +
115				tovec.bv_offset;
116
117			bounce_copy_vec(&tovec, vfrom);
118			flush_dcache_page(tovec.bv_page);
119		}
120
121		fromvec++;
122	}
123}
124
125static void bounce_end_io(struct bio *bio, mempool_t *pool, int err)
126{
127	struct bio *bio_orig = bio->bi_private;
128	struct bio_vec *bvec, *org_vec;
129	int i;
130
131	if (test_bit(BIO_EOPNOTSUPP, &bio->bi_flags))
132		set_bit(BIO_EOPNOTSUPP, &bio_orig->bi_flags);
133
134	/*
135	 * free up bounce indirect pages used
136	 */
137	bio_for_each_segment_all(bvec, bio, i) {
138		org_vec = bio_orig->bi_io_vec + i;
139		if (bvec->bv_page == org_vec->bv_page)
140			continue;
141
142		dec_zone_page_state(bvec->bv_page, NR_BOUNCE);
143		mempool_free(bvec->bv_page, pool);
144	}
145
146	bio_endio(bio_orig, err);
147	bio_put(bio);
148}
149
150static void bounce_end_io_write(struct bio *bio, int err)
151{
152	bounce_end_io(bio, page_pool, err);
153}
154
155static void bounce_end_io_write_isa(struct bio *bio, int err)
156{
157
158	bounce_end_io(bio, isa_page_pool, err);
159}
160
161static void __bounce_end_io_read(struct bio *bio, mempool_t *pool, int err)
162{
163	struct bio *bio_orig = bio->bi_private;
164
165	if (test_bit(BIO_UPTODATE, &bio->bi_flags))
166		copy_to_high_bio_irq(bio_orig, bio);
167
168	bounce_end_io(bio, pool, err);
169}
170
171static void bounce_end_io_read(struct bio *bio, int err)
172{
173	__bounce_end_io_read(bio, page_pool, err);
174}
175
176static void bounce_end_io_read_isa(struct bio *bio, int err)
177{
178	__bounce_end_io_read(bio, isa_page_pool, err);
179}
180
181#ifdef CONFIG_NEED_BOUNCE_POOL
182static int must_snapshot_stable_pages(struct request_queue *q, struct bio *bio)
183{
184	if (bio_data_dir(bio) != WRITE)
185		return 0;
186
187	if (!bdi_cap_stable_pages_required(&q->backing_dev_info))
188		return 0;
189
190	return test_bit(BIO_SNAP_STABLE, &bio->bi_flags);
191}
192#else
193static int must_snapshot_stable_pages(struct request_queue *q, struct bio *bio)
194{
195	return 0;
196}
197#endif /* CONFIG_NEED_BOUNCE_POOL */
198
199static void __blk_queue_bounce(struct request_queue *q, struct bio **bio_orig,
200			       mempool_t *pool, int force)
201{
202	struct bio *bio;
203	int rw = bio_data_dir(*bio_orig);
204	struct bio_vec *to, from;
205	struct bvec_iter iter;
206	unsigned i;
207
208	if (force)
209		goto bounce;
210	bio_for_each_segment(from, *bio_orig, iter)
211		if (page_to_pfn(from.bv_page) > queue_bounce_pfn(q))
212			goto bounce;
213
214	return;
215bounce:
216	bio = bio_clone_bioset(*bio_orig, GFP_NOIO, fs_bio_set);
217
218	bio_for_each_segment_all(to, bio, i) {
219		struct page *page = to->bv_page;
220
221		if (page_to_pfn(page) <= queue_bounce_pfn(q) && !force)
222			continue;
223
224		to->bv_page = mempool_alloc(pool, q->bounce_gfp);
225		inc_zone_page_state(to->bv_page, NR_BOUNCE);
226
227		if (rw == WRITE) {
228			char *vto, *vfrom;
229
230			flush_dcache_page(page);
231
232			vto = page_address(to->bv_page) + to->bv_offset;
233			vfrom = kmap_atomic(page) + to->bv_offset;
234			memcpy(vto, vfrom, to->bv_len);
235			kunmap_atomic(vfrom);
236		}
237	}
238
239	trace_block_bio_bounce(q, *bio_orig);
240
241	bio->bi_flags |= (1 << BIO_BOUNCED);
242
243	if (pool == page_pool) {
244		bio->bi_end_io = bounce_end_io_write;
245		if (rw == READ)
246			bio->bi_end_io = bounce_end_io_read;
247	} else {
248		bio->bi_end_io = bounce_end_io_write_isa;
249		if (rw == READ)
250			bio->bi_end_io = bounce_end_io_read_isa;
251	}
252
253	bio->bi_private = *bio_orig;
254	*bio_orig = bio;
255}
256
257void blk_queue_bounce(struct request_queue *q, struct bio **bio_orig)
258{
259	int must_bounce;
260	mempool_t *pool;
261
262	/*
263	 * Data-less bio, nothing to bounce
264	 */
265	if (!bio_has_data(*bio_orig))
266		return;
267
268	must_bounce = must_snapshot_stable_pages(q, *bio_orig);
269
270	/*
271	 * for non-isa bounce case, just check if the bounce pfn is equal
272	 * to or bigger than the highest pfn in the system -- in that case,
273	 * don't waste time iterating over bio segments
274	 */
275	if (!(q->bounce_gfp & GFP_DMA)) {
276		if (queue_bounce_pfn(q) >= blk_max_pfn && !must_bounce)
277			return;
278		pool = page_pool;
279	} else {
280		BUG_ON(!isa_page_pool);
281		pool = isa_page_pool;
282	}
283
284	/*
285	 * slow path
286	 */
287	__blk_queue_bounce(q, bio_orig, pool, must_bounce);
288}
289
290EXPORT_SYMBOL(blk_queue_bounce);
291