x86, pat: Fix set_memory_wc related corruption
[linux-2.6/mini2440.git] / drivers / md / raid0.c
blob335f490dcad631ddaca2ad73509c900b2a505ad8
1 /*
2 raid0.c : Multiple Devices driver for Linux
3 Copyright (C) 1994-96 Marc ZYNGIER
4 <zyngier@ufr-info-p7.ibp.fr> or
5 <maz@gloups.fdn.fr>
6 Copyright (C) 1999, 2000 Ingo Molnar, Red Hat
9 RAID-0 management functions.
11 This program is free software; you can redistribute it and/or modify
12 it under the terms of the GNU General Public License as published by
13 the Free Software Foundation; either version 2, or (at your option)
14 any later version.
16 You should have received a copy of the GNU General Public License
17 (for example /usr/src/linux/COPYING); if not, write to the Free
18 Software Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
21 #include <linux/blkdev.h>
22 #include <linux/seq_file.h>
23 #include "md.h"
24 #include "raid0.h"
26 static void raid0_unplug(struct request_queue *q)
28 mddev_t *mddev = q->queuedata;
29 raid0_conf_t *conf = mddev->private;
30 mdk_rdev_t **devlist = conf->devlist;
31 int i;
33 for (i=0; i<mddev->raid_disks; i++) {
34 struct request_queue *r_queue = bdev_get_queue(devlist[i]->bdev);
36 blk_unplug(r_queue);
40 static int raid0_congested(void *data, int bits)
42 mddev_t *mddev = data;
43 raid0_conf_t *conf = mddev->private;
44 mdk_rdev_t **devlist = conf->devlist;
45 int i, ret = 0;
47 for (i = 0; i < mddev->raid_disks && !ret ; i++) {
48 struct request_queue *q = bdev_get_queue(devlist[i]->bdev);
50 ret |= bdi_congested(&q->backing_dev_info, bits);
52 return ret;
56 * inform the user of the raid configuration
58 static void dump_zones(mddev_t *mddev)
60 int j, k, h;
61 sector_t zone_size = 0;
62 sector_t zone_start = 0;
63 char b[BDEVNAME_SIZE];
64 raid0_conf_t *conf = mddev->private;
65 printk(KERN_INFO "******* %s configuration *********\n",
66 mdname(mddev));
67 h = 0;
68 for (j = 0; j < conf->nr_strip_zones; j++) {
69 printk(KERN_INFO "zone%d=[", j);
70 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
71 printk("%s/",
72 bdevname(conf->devlist[j*mddev->raid_disks
73 + k]->bdev, b));
74 printk("]\n");
76 zone_size = conf->strip_zone[j].zone_end - zone_start;
77 printk(KERN_INFO " zone offset=%llukb "
78 "device offset=%llukb size=%llukb\n",
79 (unsigned long long)zone_start>>1,
80 (unsigned long long)conf->strip_zone[j].dev_start>>1,
81 (unsigned long long)zone_size>>1);
82 zone_start = conf->strip_zone[j].zone_end;
84 printk(KERN_INFO "**********************************\n\n");
87 static int create_strip_zones(mddev_t *mddev)
89 int i, c, j, err;
90 sector_t curr_zone_end, sectors;
91 mdk_rdev_t *smallest, *rdev1, *rdev2, *rdev, **dev;
92 struct strip_zone *zone;
93 int cnt;
94 char b[BDEVNAME_SIZE];
95 raid0_conf_t *conf = kzalloc(sizeof(*conf), GFP_KERNEL);
97 if (!conf)
98 return -ENOMEM;
99 list_for_each_entry(rdev1, &mddev->disks, same_set) {
100 printk(KERN_INFO "raid0: looking at %s\n",
101 bdevname(rdev1->bdev,b));
102 c = 0;
104 /* round size to chunk_size */
105 sectors = rdev1->sectors;
106 sector_div(sectors, mddev->chunk_sectors);
107 rdev1->sectors = sectors * mddev->chunk_sectors;
109 list_for_each_entry(rdev2, &mddev->disks, same_set) {
110 printk(KERN_INFO "raid0: comparing %s(%llu)",
111 bdevname(rdev1->bdev,b),
112 (unsigned long long)rdev1->sectors);
113 printk(KERN_INFO " with %s(%llu)\n",
114 bdevname(rdev2->bdev,b),
115 (unsigned long long)rdev2->sectors);
116 if (rdev2 == rdev1) {
117 printk(KERN_INFO "raid0: END\n");
118 break;
120 if (rdev2->sectors == rdev1->sectors) {
122 * Not unique, don't count it as a new
123 * group
125 printk(KERN_INFO "raid0: EQUAL\n");
126 c = 1;
127 break;
129 printk(KERN_INFO "raid0: NOT EQUAL\n");
131 if (!c) {
132 printk(KERN_INFO "raid0: ==> UNIQUE\n");
133 conf->nr_strip_zones++;
134 printk(KERN_INFO "raid0: %d zones\n",
135 conf->nr_strip_zones);
138 printk(KERN_INFO "raid0: FINAL %d zones\n", conf->nr_strip_zones);
139 err = -ENOMEM;
140 conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
141 conf->nr_strip_zones, GFP_KERNEL);
142 if (!conf->strip_zone)
143 goto abort;
144 conf->devlist = kzalloc(sizeof(mdk_rdev_t*)*
145 conf->nr_strip_zones*mddev->raid_disks,
146 GFP_KERNEL);
147 if (!conf->devlist)
148 goto abort;
150 /* The first zone must contain all devices, so here we check that
151 * there is a proper alignment of slots to devices and find them all
153 zone = &conf->strip_zone[0];
154 cnt = 0;
155 smallest = NULL;
156 dev = conf->devlist;
157 err = -EINVAL;
158 list_for_each_entry(rdev1, &mddev->disks, same_set) {
159 int j = rdev1->raid_disk;
161 if (j < 0 || j >= mddev->raid_disks) {
162 printk(KERN_ERR "raid0: bad disk number %d - "
163 "aborting!\n", j);
164 goto abort;
166 if (dev[j]) {
167 printk(KERN_ERR "raid0: multiple devices for %d - "
168 "aborting!\n", j);
169 goto abort;
171 dev[j] = rdev1;
173 disk_stack_limits(mddev->gendisk, rdev1->bdev,
174 rdev1->data_offset << 9);
175 /* as we don't honour merge_bvec_fn, we must never risk
176 * violating it, so limit ->max_sector to one PAGE, as
177 * a one page request is never in violation.
180 if (rdev1->bdev->bd_disk->queue->merge_bvec_fn &&
181 queue_max_sectors(mddev->queue) > (PAGE_SIZE>>9))
182 blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
184 if (!smallest || (rdev1->sectors < smallest->sectors))
185 smallest = rdev1;
186 cnt++;
188 if (cnt != mddev->raid_disks) {
189 printk(KERN_ERR "raid0: too few disks (%d of %d) - "
190 "aborting!\n", cnt, mddev->raid_disks);
191 goto abort;
193 zone->nb_dev = cnt;
194 zone->zone_end = smallest->sectors * cnt;
196 curr_zone_end = zone->zone_end;
198 /* now do the other zones */
199 for (i = 1; i < conf->nr_strip_zones; i++)
201 zone = conf->strip_zone + i;
202 dev = conf->devlist + i * mddev->raid_disks;
204 printk(KERN_INFO "raid0: zone %d\n", i);
205 zone->dev_start = smallest->sectors;
206 smallest = NULL;
207 c = 0;
209 for (j=0; j<cnt; j++) {
210 char b[BDEVNAME_SIZE];
211 rdev = conf->devlist[j];
212 printk(KERN_INFO "raid0: checking %s ...",
213 bdevname(rdev->bdev, b));
214 if (rdev->sectors <= zone->dev_start) {
215 printk(KERN_INFO " nope.\n");
216 continue;
218 printk(KERN_INFO " contained as device %d\n", c);
219 dev[c] = rdev;
220 c++;
221 if (!smallest || rdev->sectors < smallest->sectors) {
222 smallest = rdev;
223 printk(KERN_INFO " (%llu) is smallest!.\n",
224 (unsigned long long)rdev->sectors);
228 zone->nb_dev = c;
229 sectors = (smallest->sectors - zone->dev_start) * c;
230 printk(KERN_INFO "raid0: zone->nb_dev: %d, sectors: %llu\n",
231 zone->nb_dev, (unsigned long long)sectors);
233 curr_zone_end += sectors;
234 zone->zone_end = curr_zone_end;
236 printk(KERN_INFO "raid0: current zone start: %llu\n",
237 (unsigned long long)smallest->sectors);
239 mddev->queue->unplug_fn = raid0_unplug;
240 mddev->queue->backing_dev_info.congested_fn = raid0_congested;
241 mddev->queue->backing_dev_info.congested_data = mddev;
244 * now since we have the hard sector sizes, we can make sure
245 * chunk size is a multiple of that sector size
247 if ((mddev->chunk_sectors << 9) % queue_logical_block_size(mddev->queue)) {
248 printk(KERN_ERR "%s chunk_size of %d not valid\n",
249 mdname(mddev),
250 mddev->chunk_sectors << 9);
251 goto abort;
254 blk_queue_io_min(mddev->queue, mddev->chunk_sectors << 9);
255 blk_queue_io_opt(mddev->queue,
256 (mddev->chunk_sectors << 9) * mddev->raid_disks);
258 printk(KERN_INFO "raid0: done.\n");
259 mddev->private = conf;
260 return 0;
261 abort:
262 kfree(conf->strip_zone);
263 kfree(conf->devlist);
264 kfree(conf);
265 mddev->private = NULL;
266 return err;
270 * raid0_mergeable_bvec -- tell bio layer if a two requests can be merged
271 * @q: request queue
272 * @bvm: properties of new bio
273 * @biovec: the request that could be merged to it.
275 * Return amount of bytes we can accept at this offset
277 static int raid0_mergeable_bvec(struct request_queue *q,
278 struct bvec_merge_data *bvm,
279 struct bio_vec *biovec)
281 mddev_t *mddev = q->queuedata;
282 sector_t sector = bvm->bi_sector + get_start_sect(bvm->bi_bdev);
283 int max;
284 unsigned int chunk_sectors = mddev->chunk_sectors;
285 unsigned int bio_sectors = bvm->bi_size >> 9;
287 if (is_power_of_2(chunk_sectors))
288 max = (chunk_sectors - ((sector & (chunk_sectors-1))
289 + bio_sectors)) << 9;
290 else
291 max = (chunk_sectors - (sector_div(sector, chunk_sectors)
292 + bio_sectors)) << 9;
293 if (max < 0) max = 0; /* bio_add cannot handle a negative return */
294 if (max <= biovec->bv_len && bio_sectors == 0)
295 return biovec->bv_len;
296 else
297 return max;
300 static sector_t raid0_size(mddev_t *mddev, sector_t sectors, int raid_disks)
302 sector_t array_sectors = 0;
303 mdk_rdev_t *rdev;
305 WARN_ONCE(sectors || raid_disks,
306 "%s does not support generic reshape\n", __func__);
308 list_for_each_entry(rdev, &mddev->disks, same_set)
309 array_sectors += rdev->sectors;
311 return array_sectors;
314 static int raid0_run(mddev_t *mddev)
316 int ret;
318 if (mddev->chunk_sectors == 0) {
319 printk(KERN_ERR "md/raid0: chunk size must be set.\n");
320 return -EINVAL;
322 if (md_check_no_bitmap(mddev))
323 return -EINVAL;
324 blk_queue_max_sectors(mddev->queue, mddev->chunk_sectors);
325 mddev->queue->queue_lock = &mddev->queue->__queue_lock;
327 ret = create_strip_zones(mddev);
328 if (ret < 0)
329 return ret;
331 /* calculate array device size */
332 md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));
334 printk(KERN_INFO "raid0 : md_size is %llu sectors.\n",
335 (unsigned long long)mddev->array_sectors);
336 /* calculate the max read-ahead size.
337 * For read-ahead of large files to be effective, we need to
338 * readahead at least twice a whole stripe. i.e. number of devices
339 * multiplied by chunk size times 2.
340 * If an individual device has an ra_pages greater than the
341 * chunk size, then we will not drive that device as hard as it
342 * wants. We consider this a configuration error: a larger
343 * chunksize should be used in that case.
346 int stripe = mddev->raid_disks *
347 (mddev->chunk_sectors << 9) / PAGE_SIZE;
348 if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
349 mddev->queue->backing_dev_info.ra_pages = 2* stripe;
352 blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
353 dump_zones(mddev);
354 return 0;
357 static int raid0_stop(mddev_t *mddev)
359 raid0_conf_t *conf = mddev->private;
361 blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
362 kfree(conf->strip_zone);
363 kfree(conf->devlist);
364 kfree(conf);
365 mddev->private = NULL;
366 return 0;
369 /* Find the zone which holds a particular offset
370 * Update *sectorp to be an offset in that zone
372 static struct strip_zone *find_zone(struct raid0_private_data *conf,
373 sector_t *sectorp)
375 int i;
376 struct strip_zone *z = conf->strip_zone;
377 sector_t sector = *sectorp;
379 for (i = 0; i < conf->nr_strip_zones; i++)
380 if (sector < z[i].zone_end) {
381 if (i)
382 *sectorp = sector - z[i-1].zone_end;
383 return z + i;
385 BUG();
389 * remaps the bio to the target device. we separate two flows.
390 * power 2 flow and a general flow for the sake of perfromance
392 static mdk_rdev_t *map_sector(mddev_t *mddev, struct strip_zone *zone,
393 sector_t sector, sector_t *sector_offset)
395 unsigned int sect_in_chunk;
396 sector_t chunk;
397 raid0_conf_t *conf = mddev->private;
398 unsigned int chunk_sects = mddev->chunk_sectors;
400 if (is_power_of_2(chunk_sects)) {
401 int chunksect_bits = ffz(~chunk_sects);
402 /* find the sector offset inside the chunk */
403 sect_in_chunk = sector & (chunk_sects - 1);
404 sector >>= chunksect_bits;
405 /* chunk in zone */
406 chunk = *sector_offset;
407 /* quotient is the chunk in real device*/
408 sector_div(chunk, zone->nb_dev << chunksect_bits);
409 } else{
410 sect_in_chunk = sector_div(sector, chunk_sects);
411 chunk = *sector_offset;
412 sector_div(chunk, chunk_sects * zone->nb_dev);
415 * position the bio over the real device
416 * real sector = chunk in device + starting of zone
417 * + the position in the chunk
419 *sector_offset = (chunk * chunk_sects) + sect_in_chunk;
420 return conf->devlist[(zone - conf->strip_zone)*mddev->raid_disks
421 + sector_div(sector, zone->nb_dev)];
425 * Is io distribute over 1 or more chunks ?
427 static inline int is_io_in_chunk_boundary(mddev_t *mddev,
428 unsigned int chunk_sects, struct bio *bio)
430 if (likely(is_power_of_2(chunk_sects))) {
431 return chunk_sects >= ((bio->bi_sector & (chunk_sects-1))
432 + (bio->bi_size >> 9));
433 } else{
434 sector_t sector = bio->bi_sector;
435 return chunk_sects >= (sector_div(sector, chunk_sects)
436 + (bio->bi_size >> 9));
440 static int raid0_make_request(struct request_queue *q, struct bio *bio)
442 mddev_t *mddev = q->queuedata;
443 unsigned int chunk_sects;
444 sector_t sector_offset;
445 struct strip_zone *zone;
446 mdk_rdev_t *tmp_dev;
447 const int rw = bio_data_dir(bio);
448 int cpu;
450 if (unlikely(bio_barrier(bio))) {
451 bio_endio(bio, -EOPNOTSUPP);
452 return 0;
455 cpu = part_stat_lock();
456 part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
457 part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
458 bio_sectors(bio));
459 part_stat_unlock();
461 chunk_sects = mddev->chunk_sectors;
462 if (unlikely(!is_io_in_chunk_boundary(mddev, chunk_sects, bio))) {
463 sector_t sector = bio->bi_sector;
464 struct bio_pair *bp;
465 /* Sanity check -- queue functions should prevent this happening */
466 if (bio->bi_vcnt != 1 ||
467 bio->bi_idx != 0)
468 goto bad_map;
469 /* This is a one page bio that upper layers
470 * refuse to split for us, so we need to split it.
472 if (likely(is_power_of_2(chunk_sects)))
473 bp = bio_split(bio, chunk_sects - (sector &
474 (chunk_sects-1)));
475 else
476 bp = bio_split(bio, chunk_sects -
477 sector_div(sector, chunk_sects));
478 if (raid0_make_request(q, &bp->bio1))
479 generic_make_request(&bp->bio1);
480 if (raid0_make_request(q, &bp->bio2))
481 generic_make_request(&bp->bio2);
483 bio_pair_release(bp);
484 return 0;
487 sector_offset = bio->bi_sector;
488 zone = find_zone(mddev->private, &sector_offset);
489 tmp_dev = map_sector(mddev, zone, bio->bi_sector,
490 &sector_offset);
491 bio->bi_bdev = tmp_dev->bdev;
492 bio->bi_sector = sector_offset + zone->dev_start +
493 tmp_dev->data_offset;
495 * Let the main block layer submit the IO and resolve recursion:
497 return 1;
499 bad_map:
500 printk("raid0_make_request bug: can't convert block across chunks"
501 " or bigger than %dk %llu %d\n", chunk_sects / 2,
502 (unsigned long long)bio->bi_sector, bio->bi_size >> 10);
504 bio_io_error(bio);
505 return 0;
508 static void raid0_status(struct seq_file *seq, mddev_t *mddev)
510 #undef MD_DEBUG
511 #ifdef MD_DEBUG
512 int j, k, h;
513 char b[BDEVNAME_SIZE];
514 raid0_conf_t *conf = mddev->private;
516 sector_t zone_size;
517 sector_t zone_start = 0;
518 h = 0;
520 for (j = 0; j < conf->nr_strip_zones; j++) {
521 seq_printf(seq, " z%d", j);
522 seq_printf(seq, "=[");
523 for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
524 seq_printf(seq, "%s/", bdevname(
525 conf->devlist[j*mddev->raid_disks + k]
526 ->bdev, b));
528 zone_size = conf->strip_zone[j].zone_end - zone_start;
529 seq_printf(seq, "] ze=%lld ds=%lld s=%lld\n",
530 (unsigned long long)zone_start>>1,
531 (unsigned long long)conf->strip_zone[j].dev_start>>1,
532 (unsigned long long)zone_size>>1);
533 zone_start = conf->strip_zone[j].zone_end;
535 #endif
536 seq_printf(seq, " %dk chunks", mddev->chunk_sectors / 2);
537 return;
540 static struct mdk_personality raid0_personality=
542 .name = "raid0",
543 .level = 0,
544 .owner = THIS_MODULE,
545 .make_request = raid0_make_request,
546 .run = raid0_run,
547 .stop = raid0_stop,
548 .status = raid0_status,
549 .size = raid0_size,
552 static int __init raid0_init (void)
554 return register_md_personality (&raid0_personality);
557 static void raid0_exit (void)
559 unregister_md_personality (&raid0_personality);
562 module_init(raid0_init);
563 module_exit(raid0_exit);
564 MODULE_LICENSE("GPL");
565 MODULE_ALIAS("md-personality-2"); /* RAID0 */
566 MODULE_ALIAS("md-raid0");
567 MODULE_ALIAS("md-level-0");