Skip to contentSkip to conversation

Subject: [001/001 ] raid0 remove hashing, variable chunk size

PATCH raz ben yehuda 3 participants 3 messages May 13, 2009 – May 14, 2009 MD RAID

lore
Share & tools
lore
#0 lore raw

1. Beautify raid0 code
2. Remove hashing
3. Chunk size of 4K*n instead of 4K*2^n ( must apply a patch to mdadm ). patch was tested with backward compatibility to 2.6.18. i did not test 32bit, i do have a 32bit compiler for 2.6.30-rc2.

 md.c    |   24 ++-
 raid0.c |  428 ++++++++++++++++++++++++++++++----------------------------------
 raid0.h |    3 
 3 files changed, 221 insertions(+), 234 deletions(-)

---

drivers/md/md.c +17 -7
index ed5727c..068e18d 100644
@@ -444,8 +444,10 @@ static sector_t calc_num_sectors(mdk_rdev_t *rdev, unsigned chunk_size)
 {
 	sector_t num_sectors = rdev->sb_start;
 
-	if (chunk_size)
-		num_sectors &= ~((sector_t)chunk_size/512 - 1);
+	if (chunk_size) {
+		int chunk_sects = chunk_size>>9;
+		num_sectors = (num_sectors/chunk_sects)*chunk_sects;
+	}
 	return num_sectors;
 }
 
@@ -3512,7 +3514,7 @@ min_sync_store(mddev_t *mddev, const char *buf, size_t len)
 
 	/* Must be a multiple of chunk_size */
 	if (mddev->chunk_size) {
-		if (min & (sector_t)((mddev->chunk_size>>9)-1))
+		if (min % (sector_t)(mddev->chunk_size>>9))
 			return -EINVAL;
 	}
 	mddev->resync_min = min;
@@ -3549,7 +3551,7 @@ max_sync_store(mddev_t *mddev, const char *buf, size_t len)
 
 		/* Must be a multiple of chunk_size */
 		if (mddev->chunk_size) {
-			if (max & (sector_t)((mddev->chunk_size>>9)-1))
+			if (max % (sector_t)(mddev->chunk_size>>9))
 				return -EINVAL;
 		}
 		mddev->resync_max = max;
@@ -3992,12 +3994,20 @@ static int do_md_run(mddev_t * mddev)
 		}
 		/*
 		 * chunk-size has to be a power of 2
-		 */
-		if ( (1 << ffz(~chunk_size)) != chunk_size) {
+		*/
+		if ((1 << ffz(~chunk_size)) != chunk_size
+			&& mddev->level != 0) {
 			printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
 			return -EINVAL;
 		}
-
+		/*
+		* raid0 chunk size has to divide by a page
+		*/
+		if (mddev->level == 0 && (chunk_size % 4096)) {
+			printk(KERN_ERR "chunk_size of %d not valid\n",
+					chunk_size);
+			return -EINVAL;
+		}
 		/* devices must have minimum size of one chunk */
 		list_for_each_entry(rdev, &mddev->disks, same_set) {
 			if (test_bit(Faulty, &rdev->flags))
drivers/md/raid0.c +204 -224

583 lines not on this page · raw message

drivers/md/raid0.h +0 -3
index 824b12e..2a73f94 100644
@@ -16,9 +16,6 @@ struct raid0_private_data
 	struct strip_zone *strip_zone;
 	mdk_rdev_t **devlist; /* lists of rdevs, pointed to by strip_zone->dev */
 	int nr_strip_zones;
-
-	sector_t spacing;
-	int sector_shift; /* shift this before divide by spacing */
 };
 
 typedef struct raid0_private_data raid0_conf_t;
#1 lore raw
Quoted 11 lines · raz ben yehuda

1. Beautify raid0 code
2. Remove hashing
3. Chunk size of 4K*n instead of 4K*2^n ( must apply a patch to mdadm ). patch was tested with backward compatibility to 2.6.18. i did not test 32bit, i do have
a 32bit compiler for 2.6.30-rc2.

 md.c    |   24 ++-
 raid0.c |  428
++++++++++++++++++++++++++++++----------------------------------
 raid0.h |    3
 3 files changed, 221 insertions(+), 234 deletions(-)

That is a very small description for a very big patch.

As it does 3 different things, can you make it into three different patches? That tends to make it a lot easier to review.

Quoted 1 line

You are meant to put your own email address here, not mine.

NeilBrown

#2 lore raw
In reply to raz ben yehuda
Quoted 430 lines · raz ben yehuda

1. Beautify raid0 code
2. Remove hashing
3. Chunk size of 4K*n instead of 4K*2^n ( must apply a patch to mdadm ). patch was tested with backward compatibility to 2.6.18. i did not test 32bit, i do have a 32bit compiler for 2.6.30-rc2.
 md.c    |   24 ++-
 raid0.c |  428 ++++++++++++++++++++++++++++++---------------------------------- raid0.h |    3

 3 files changed, 221 insertions(+), 234 deletions(-)

---

drivers/md/md.c +17 -7
index ed5727c..068e18d 100644
@@ -444,8 +444,10 @@ static sector_t calc_num_sectors(mdk_rdev_t *rdev, unsigned chunk_size)
  {
        sector_t num_sectors = rdev->sb_start;
 
-       if (chunk_size)
-               num_sectors &= ~((sector_t)chunk_size/512 - 1);
+       if (chunk_size) {
+               int chunk_sects = chunk_size>>9;
+               num_sectors = (num_sectors/chunk_sects)*chunk_sects;
+       }
        return num_sectors;
  }
 
@@ -3512,7 +3514,7 @@ min_sync_store(mddev_t *mddev, const char *buf, size_t len)
 
        /* Must be a multiple of chunk_size */
        if (mddev->chunk_size) {
-               if (min & (sector_t)((mddev->chunk_size>>9)-1))
+               if (min % (sector_t)(mddev->chunk_size>>9))
                        return -EINVAL;
        }
        mddev->resync_min = min;
@@ -3549,7 +3551,7 @@ max_sync_store(mddev_t *mddev, const char *buf, size_t len)
 
                /* Must be a multiple of chunk_size */
                if (mddev->chunk_size) {
-                       if (max & (sector_t)((mddev->chunk_size>>9)-1))
+                       if (max % (sector_t)(mddev->chunk_size>>9))
                                return -EINVAL;
                }
                mddev->resync_max = max;
@@ -3992,12 +3994,20 @@ static int do_md_run(mddev_t * mddev)
                }
                /*
                 * chunk-size has to be a power of 2
-                */
-               if ( (1 << ffz(~chunk_size)) != chunk_size) {
+               */
+               if ((1 << ffz(~chunk_size)) != chunk_size
+                       && mddev->level != 0) {
                        printk(KERN_ERR "chunk_size of %d not valid\n", chunk_size);
                        return -EINVAL;
                }
-
+               /*
+               * raid0 chunk size has to divide by a page
+               */
+               if (mddev->level == 0 && (chunk_size % 4096)) {
+                       printk(KERN_ERR "chunk_size of %d not valid\n",
+                                       chunk_size);
+                       return -EINVAL;
+               }
                /* devices must have minimum size of one chunk */
                list_for_each_entry(rdev, &mddev->disks, same_set) {
                        if (test_bit(Faulty, &rdev->flags))
drivers/md/raid0.c +46 -32
index c08d755..990ec6c 100644
@@ -53,32 +53,49 @@ static int raid0_congested(void *data, int bits)
  }
 
 
-static int create_strip_zones (mddev_t *mddev)
+static void raid0_dump_zones(mddev_t *mddev)
  {
-       int i, c, j;
-       sector_t current_start, curr_zone_start;
-       sector_t min_spacing;
+       int j, k, h;
+       char b[BDEVNAME_SIZE];
        raid0_conf_t *conf = mddev_to_conf(mddev);
-       mdk_rdev_t *smallest, *rdev1, *rdev2, *rdev;
-       struct strip_zone *zone;
-       int cnt;
+       printk(KERN_INFO "***** %s configuration ******",
+               mdname(mddev));
+       h = 0;
+       for (j = 0; j < conf->nr_strip_zones; j++) {
+               printk("\nzone%d", j);
+               printk("=[");
+               for (k = 0; k < conf->strip_zone[j].nb_dev; k++)
+                       printk("%s/", bdevname(
+                               conf->strip_zone[j].dev[k]->bdev, b));
+               printk("]\n\t zone offset=%llu device offset=%llu size=%llukb\n",
+                       (unsigned long long)conf->strip_zone[j].zone_start,
+                       (unsigned long long)conf->strip_zone[j].dev_start,
+                       (unsigned long long)conf->strip_zone[j].sectors>>1);
+       }
+       printk(KERN_INFO "**********************************\n\n");
+}
+
+
+static void raid0_count_zones(mddev_t *mddev, struct list_head *disks)
+{
+       int c = 0;
        char b[BDEVNAME_SIZE];
-
+       mdk_rdev_t  *rdev1, *rdev2;
+       raid0_conf_t *conf = mddev_to_conf(mddev);
        /*
         * The number of 'same size groups'
         */
        conf->nr_strip_zones = 0;
-
-       list_for_each_entry(rdev1, &mddev->disks, same_set) {
+       list_for_each_entry(rdev1, disks, same_set) {
                printk(KERN_INFO "raid0: looking at %s\n",
                        bdevname(rdev1->bdev,b));
                c = 0;
-               list_for_each_entry(rdev2, &mddev->disks, same_set) {
+               list_for_each_entry(rdev2, disks, same_set) {
                        printk(KERN_INFO "raid0:   comparing %s(%llu)",
-                              bdevname(rdev1->bdev,b),
+                              bdevname(rdev1->bdev, b),
                               (unsigned long long)rdev1->sectors);
-                       printk(KERN_INFO " with %s(%llu)\n",
-                              bdevname(rdev2->bdev,b),
+                       printk(KERN_INFO "with %s(%llu)\n",
+                              bdevname(rdev2->bdev, b),
                               (unsigned long long)rdev2->sectors);
                        if (rdev2 == rdev1) {
                                printk(KERN_INFO "raid0:   END\n");
@@ -103,81 +120,75 @@ static int create_strip_zones (mddev_t *mddev)
                }
        }
        printk(KERN_INFO "raid0: FINAL %d zones\n", conf->nr_strip_zones);
+}
 
-       conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
-                               conf->nr_strip_zones, GFP_KERNEL);
-       if (!conf->strip_zone)
-               return 1;
-       conf->devlist = kzalloc(sizeof(mdk_rdev_t*)*
-                               conf->nr_strip_zones*mddev->raid_disks,
-                               GFP_KERNEL);
-       if (!conf->devlist)
-               return 1;
 
-       /* The first zone must contain all devices, so here we check that
-        * there is a proper alignment of slots to devices and find them all
-        */
-       zone = &conf->strip_zone[0];
-       cnt = 0;
-       smallest = NULL;
-       zone->dev = conf->devlist;
-       list_for_each_entry(rdev1, &mddev->disks, same_set) {
-               int j = rdev1->raid_disk;
+/*
+ * The first zone must contain all devices, so here we check that
+ * there is a proper alignment of slots to devices and find them all
+ */
+static int raid0_create_first_zone(mddev_t *mddev, struct list_head *disks)
+{
+       mdk_rdev_t *smallest = NULL;
+       mdk_rdev_t  *rdev;
+       int cnt = 0;
+       raid0_conf_t *conf = mddev_to_conf(mddev);
+       struct strip_zone *zone0 = &conf->strip_zone[0];
 
+       zone0->dev = conf->devlist;
+       list_for_each_entry(rdev, disks, same_set) {
+               int j = rdev->raid_disk;
                if (j < 0 || j >= mddev->raid_disks) {

                       printk(KERN_ERR "raid0: bad disk number %d - " "aborting!\n", j);
-                       goto abort;
+                       return -1;
               }
-               if (zone->dev[j]) {
+               if (zone0->dev[j]) {
                       printk(KERN_ERR "raid0: multiple devices for %d - " "aborting!\n", j);
-                       goto abort;
+                       return -1;
               }
-               zone->dev[j] = rdev1;

drivers/md/raid0.c +3 -14
-
-               blk_queue_stack_limits(mddev->queue,
-                                      rdev1->bdev->bd_disk->queue);
-               /* as we don't honour merge_bvec_fn, we must never risk
-                * violating it, so limit ->max_sector to one PAGE, as
-                * a one page request is never in violation.
-                */
-
-               if (rdev1->bdev->bd_disk->queue->merge_bvec_fn &&
-                   mddev->queue->max_sectors > (PAGE_SIZE>>9))
-                       blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
-
-               if (!smallest || (rdev1->sectors < smallest->sectors))
-                       smallest = rdev1;
+               zone0->dev[j] = rdev;
+               if (!smallest || (rdev->sectors < smallest->sectors))
+                       smallest = rdev;

               cnt++;
       }
       if (cnt != mddev->raid_disks) {
               printk(KERN_ERR "raid0: too few disks (%d of %d) - " "aborting!\n", cnt, mddev->raid_disks);
-               goto abort;
+               return -1;
       }
-       zone->nb_dev = cnt;
-       zone->sectors = smallest->sectors * cnt;
-       zone->zone_start = 0;
+       zone0->nb_dev = cnt;
+       zone0->sectors = smallest->sectors * cnt;
+       zone0->zone_start = 0;
+       return 0;

drivers/md/raid0.c +13 -2
+}
+
 
-       current_start = smallest->sectors;
-       curr_zone_start = zone->sectors;
+
+static void raid0_set_higher_zones(mddev_t *mddev)
+{
+       int i, j, c;
+       mdk_rdev_t *rdev;
+       struct strip_zone *zone;
+       raid0_conf_t *conf = mddev_to_conf(mddev);
+       mdk_rdev_t *smallest;
+       sector_t current_start =
+               conf->strip_zone[0].sectors/conf->strip_zone[0].nb_dev;
+       sector_t curr_zone_start = conf->strip_zone[0].sectors;

       /* now do the other zones */
-       for (i = 1; i < conf->nr_strip_zones; i++)
-       {
+       for (i = 1; i < conf->nr_strip_zones; i++) {
               zone = conf->strip_zone + i;
               zone->dev = conf->strip_zone[i-1].dev + mddev->raid_disks; -
               printk(KERN_INFO "raid0: zone %d\n", i); zone->dev_start = current_start;
               smallest = NULL;
               c = 0;
-
-               for (j=0; j<cnt; j++) {
+               for (j = 0; j < conf->strip_zone[0].nb_dev; j++) { char b[BDEVNAME_SIZE];
                       rdev = conf->strip_zone[0].dev[j];
-                       printk(KERN_INFO "raid0: checking %s ...",
+                       printk("raid0: checking %s ...", bdevname(rdev->bdev, b));
                       if (rdev->sectors <= current_start) { printk(KERN_INFO " nope.\n");

drivers/md/raid0.c +20 -36
@@ -197,54 +208,38 @@ static int create_strip_zones (mddev_t *mddev)
                zone->sectors = (smallest->sectors - current_start) * c;
                printk(KERN_INFO "raid0: zone->nb_dev: %d, sectors: %llu\n",
                        zone->nb_dev, (unsigned long long)zone->sectors);
-
                zone->zone_start = curr_zone_start;
                curr_zone_start += zone->sectors;
-
                current_start = smallest->sectors;
                printk(KERN_INFO "raid0: current zone start: %llu\n",
                        (unsigned long long)current_start);
        }
+}
 
-       /* Now find appropriate hash spacing.
-        * We want a number which causes most hash entries to cover
-        * at most two strips, but the hash table must be at most
-        * 1 PAGE.  We choose the smallest strip, or contiguous collection
-        * of strips, that has big enough size.  We never consider the last
-        * strip though as it's size has no bearing on the efficacy of the hash
-        * table.
-        */
-       conf->spacing = curr_zone_start;
-       min_spacing = curr_zone_start;
-       sector_div(min_spacing, PAGE_SIZE/sizeof(struct strip_zone*));
-       for (i=0; i < conf->nr_strip_zones-1; i++) {
-               sector_t s = 0;
-               for (j = i; j < conf->nr_strip_zones - 1 &&
-                               s < min_spacing; j++)
-                       s += conf->strip_zone[j].sectors;
-               if (s >= min_spacing && s < conf->spacing)
-                       conf->spacing = s;
-       }
+static int raid0_create_strip_zones(mddev_t *mddev, struct list_head *disks)
+{
+       raid0_conf_t *conf = mddev_to_conf(mddev);
 
+       raid0_count_zones(mddev, disks);
+       conf->strip_zone = kzalloc(sizeof(struct strip_zone)*
+                               conf->nr_strip_zones, GFP_KERNEL);
+       if (!conf->strip_zone)
+               return 1;
+       conf->devlist = kzalloc(sizeof(mdk_rdev_t *)*
+                               conf->nr_strip_zones*mddev->raid_disks,
+                               GFP_KERNEL);
+       if (!conf->devlist)
+               return 1;
+       if (raid0_create_first_zone(mddev, disks))
+               return 1;
+       raid0_set_higher_zones(mddev);
        mddev->queue->unplug_fn = raid0_unplug;
-
        mddev->queue->backing_dev_info.congested_fn = raid0_congested;
        mddev->queue->backing_dev_info.congested_data = mddev;
-
        printk(KERN_INFO "raid0: done.\n");
        return 0;
- abort:
-       return 1;
  }
 
-/**
- *     raid0_mergeable_bvec -- tell bio layer if a two requests can be merged
- *     @q: request queue
- *     @bvm: properties of new bio
- *     @biovec: the request that could be merged to it.
- *
- *     Return amount of bytes we can accept at this offset
- */
  static int raid0_mergeable_bvec(struct request_queue *q,
                                struct bvec_merge_data *bvm,
                                struct bio_vec *biovec)
@@ -254,124 +249,119 @@ static int raid0_mergeable_bvec(struct request_queue *q,
        int max;
        unsigned int chunk_sectors = mddev->chunk_size >> 9;
        unsigned int bio_sectors = bvm->bi_size >> 9;
-
-       max =  (chunk_sectors - ((sector & (chunk_sectors - 1)) + bio_sectors)) << 9;
-       if (max < 0) max = 0; /* bio_add cannot handle a negative return */
+       max =  (chunk_sectors - ((sector % chunk_sectors) + bio_sectors)) << 9;
+       if (max < 0)
+               max = 0;/*  bio_add cannot handle a negative return */
        if (max <= biovec->bv_len && bio_sectors == 0)
                return biovec->bv_len;

       else
               return max;
+       return 0;
 }

+
 static sector_t raid0_size(mddev_t *mddev, sector_t sectors, int raid_disks) {
-       sector_t array_sectors = 0;
+       int i;
       mdk_rdev_t *rdev;
+       sector_t array_sectors = 0;
+       raid0_conf_t *conf = mddev_to_conf(mddev);
+       mdk_rdev_t **devlist = conf->strip_zone[0].dev;
+       for (i = 0; i < mddev->raid_disks; i++) {
+               rdev = devlist[i];
+               if (test_bit(In_sync, &rdev->flags))
+                       array_sectors += rdev->sectors;
+       }
+       return array_sectors;

drivers/md/raid0.c +39 -5
+}
 
-       WARN_ONCE(sectors || raid_disks,
-                 "%s does not support generic reshape\n", __func__);
+static void raid0_set_queue_limits(mddev_t *mddev)
+{
+       mdk_rdev_t  *rdev;
 
-       list_for_each_entry(rdev, &mddev->disks, same_set)
-               array_sectors += rdev->sectors;
+       list_for_each_entry(rdev, &mddev->disks, same_set) {
+               blk_queue_stack_limits(mddev->queue,
+                              rdev->bdev->bd_disk->queue);
+               /* as we don't honour merge_bvec_fn, we must never risk
+                * violating it, so limit ->max_sector to one PAGE, as
+                * a one page request is never in violation.
+                */
+               if (rdev->bdev->bd_disk->queue->merge_bvec_fn &&
+                   mddev->queue->max_sectors > (PAGE_SIZE>>9))
+                       blk_queue_max_sectors(mddev->queue, PAGE_SIZE>>9);
+       }
+       printk(KERN_INFO "raid0:%s queue limits: hardware"
+               " segments=%d physical segments=%d\n"
+               " max hardware sectors = %d max sectors %d\n",
+               mdname(mddev),
+               mddev->queue->max_hw_segments,
+               mddev->queue->max_phys_segments,
+               mddev->queue->max_hw_sectors,
+               mddev->queue->max_sectors);
+}
+
+/* calculate the max read-ahead size.
+ * For read-ahead of large files to be effective, we need to
+ * readahead at least twice a whole stripe. i.e. number of devices
+ * multiplied by chunk size times 2.
+ * If an individual device has an ra_pages greater than the
+ * chunk size, then we will not drive that device as hard as it
+ * wants.  We consider this a configuration error: a larger
+ * chunksize should be used in that case.
+ */
+static void raid0_set_max_ra(mddev_t *mddev)
+{
+       int stripe = mddev->raid_disks * mddev->chunk_size / PAGE_SIZE;
+       if (mddev->queue->backing_dev_info.ra_pages < 2*stripe)
+               mddev->queue->backing_dev_info.ra_pages = 2*stripe;
 
-       return array_sectors;

 }

drivers/md/raid0.c +1 -1
-static int raid0_run (mddev_t *mddev)
+static int raid0_is_power2_chunk(mddev_t *mddev)

 {
-       unsigned  cur=0, i=0, nb_zone;
-       s64 sectors;
-       raid0_conf_t *conf;
+       if ((1 << ffz(~mddev->chunk_size)) == mddev->chunk_size)
+               return 1;
+       return 0;

drivers/md/raid0.c +5 -0
+}
 
+static int raid0_run(mddev_t *mddev)
+{
+       raid0_conf_t *conf;
+       int segment_boundary;
Quoted 4 lines

       if (mddev->chunk_size == 0) {
               printk(KERN_ERR "md/raid0: non-zero chunk size required.\n"); return -EINVAL;
       }

Are we not handling this in mdadm itself, I think so ?
[09:03:40 sinhas]$ sudo mdadm -C /dev/md0 -n2 -l0 -c0 /dev/sda5 /dev/sda6 -f
mdadm: invalid chunk/rounding value: 0

I think we also make sure that chunk size > 4K. I tested it.

Quoted 9 lines

-       printk(KERN_INFO "%s: setting max_sectors to %d, segment boundary to %d\n",
+       blk_queue_max_sectors(mddev->queue, mddev->chunk_size>>9);
+       if (!raid0_is_power2_chunk(mddev)) {
+               printk(KERN_INFO "raid0: not power 2 chunk size\n");
+               segment_boundary = ~(ffz(~mddev->chunk_size))>>1;
+       } else{
+               printk(KERN_INFO "raid0: is power 2 chunk size\n");
+               segment_boundary = (mddev->chunk_size>>1)-1;
+       }

I hope this piece of code gets executed someday,

[09:03:22 sinhas]$ sudo mdadm -C /dev/md0 -n2 -l0 -c62 /dev/sda5 /dev/sda6 -f [sudo] password for sinhas:
mdadm: invalid chunk/rounding value: 62

Quoted 246 lines

+       blk_queue_segment_boundary(mddev->queue, segment_boundary);
+       printk(KERN_INFO "%s: setting max_sectors"
+                       " to %d, segment boundary to %d\n", mdname(mddev),
              mddev->chunk_size >> 9,
-              (mddev->chunk_size>>1)-1);
-       blk_queue_max_sectors(mddev->queue, mddev->chunk_size >> 9);
-       blk_queue_segment_boundary(mddev->queue, (mddev->chunk_size>>1) - 1);
+              segment_boundary);
       mddev->queue->queue_lock = &mddev->queue->__queue_lock;

-       conf = kmalloc(sizeof (raid0_conf_t), GFP_KERNEL);
+       conf = kmalloc(sizeof(raid0_conf_t), GFP_KERNEL); if (!conf)
               goto out;
       mddev->private = (void *)conf;
-
       conf->strip_zone = NULL;
       conf->devlist = NULL;
-       if (create_strip_zones (mddev))
+       if (raid0_create_strip_zones(mddev, &mddev->disks)) goto out_free_conf;
-
       /* calculate array device size */
-       md_set_array_sectors(mddev, raid0_size(mddev, 0, 0));

drivers/md/raid0.c +3 -64
-
-       printk(KERN_INFO "raid0 : md_size is %llu sectors.\n",
-               (unsigned long long)mddev->array_sectors);
-       printk(KERN_INFO "raid0 : conf->spacing is %llu sectors.\n",
-               (unsigned long long)conf->spacing);
-       {
-               sector_t s = raid0_size(mddev, 0, 0);
-               sector_t space = conf->spacing;
-               int round;
-               conf->sector_shift = 0;
-               if (sizeof(sector_t) > sizeof(u32)) {
-                       /*shift down space and s so that sector_div will work */
-                       while (space > (sector_t) (~(u32)0)) {
-                               s >>= 1;
-                               space >>= 1;
-                               s += 1; /* force round-up */
-                               conf->sector_shift++;
-                       }
-               }
-               round = sector_div(s, (u32)space) ? 1 : 0;
-               nb_zone = s + round;
-       }
-       printk(KERN_INFO "raid0 : nb_zone is %d.\n", nb_zone);
-
-       printk(KERN_INFO "raid0 : Allocating %zu bytes for hash.\n",
-                               nb_zone*sizeof(struct strip_zone*));
-       conf->hash_table = kmalloc (sizeof (struct strip_zone *)*nb_zone, GFP_KERNEL);
-       if (!conf->hash_table)
-               goto out_free_conf;
-       sectors = conf->strip_zone[cur].sectors;
-
-       conf->hash_table[0] = conf->strip_zone + cur;
-       for (i=1; i< nb_zone; i++) {
-               while (sectors <= conf->spacing) {
-                       cur++;
-                       sectors += conf->strip_zone[cur].sectors;
-               }
-               sectors -= conf->spacing;
-               conf->hash_table[i] = conf->strip_zone + cur;
-       }
-       if (conf->sector_shift) {
-               conf->spacing >>= conf->sector_shift;
-               /* round spacing up so when we divide by it, we
-                * err on the side of too-low, which is safest
-                */
-               conf->spacing++;
-       }
-
-       /* calculate the max read-ahead size.
-        * For read-ahead of large files to be effective, we need to
-        * readahead at least twice a whole stripe. i.e. number of devices
-        * multiplied by chunk size times 2.
-        * If an individual device has an ra_pages greater than the
-        * chunk size, then we will not drive that device as hard as it
-        * wants.  We consider this a configuration error: a larger
-        * chunksize should be used in that case.
-        */
-       {
-               int stripe = mddev->raid_disks * mddev->chunk_size / PAGE_SIZE;
-               if (mddev->queue->backing_dev_info.ra_pages < 2* stripe)
-                       mddev->queue->backing_dev_info.ra_pages = 2* stripe;
-       }
-
-
+       md_set_array_sectors(mddev, raid0_size(mddev, 0, mddev->raid_disks));
+       raid0_set_queue_limits(mddev);
+       raid0_set_max_ra(mddev);

       blk_queue_merge_bvec(mddev->queue, raid0_mergeable_bvec);
+       raid0_dump_zones(mddev);
       return 0;

 out_free_conf:

drivers/md/raid0.c +0 -0
@@ -388,8 +378,6 @@ static int raid0_stop (mddev_t *mddev)
        raid0_conf_t *conf = mddev_to_conf(mddev);
 

       blk_sync_queue(mddev->queue); /* the unplug fn references 'conf'*/
-       kfree(conf->hash_table);
-       conf->hash_table = NULL;
       kfree(conf->strip_zone);
       conf->strip_zone = NULL;
       kfree(conf);

drivers/md/raid0.c +29 -12
@@ -398,15 +386,36 @@ static int raid0_stop (mddev_t *mddev)
        return 0;
  }
 
+
+static void raid0_position_bio(mddev_t *mddev, struct bio *bio, sector_t sector)
+{
+       sector_t sect_in_chunk;
+       mdk_rdev_t *tmp_dev;
+       sector_t chunk_in_dev;
+       sector_t rsect;
+       sector_t x;
+       raid0_conf_t *conf = mddev_to_conf(mddev);
+       sector_t chunk_sects = mddev->chunk_size >> 9;
+       struct strip_zone *zone = &conf->strip_zone[0];
+
+       while (sector >= zone->zone_start + zone->sectors)
+               zone++;
+       sect_in_chunk = sector % chunk_sects;
+       x = (sector - zone->zone_start) / chunk_sects;
+       sector_div(x, zone->nb_dev);
+       chunk_in_dev = x;
+       x = sector / chunk_sects;
+       tmp_dev = zone->dev[sector_div(x, zone->nb_dev)];
+       rsect = (chunk_in_dev * chunk_sects) + zone->dev_start + sect_in_chunk;
+       bio->bi_bdev = tmp_dev->bdev;
+       bio->bi_sector = rsect + tmp_dev->data_offset;
+}
+
+
  static int raid0_make_request (struct request_queue *q, struct bio *bio)
  {
        mddev_t *mddev = q->queuedata;
-       unsigned int sect_in_chunk, chunksect_bits, chunk_sects;
-       raid0_conf_t *conf = mddev_to_conf(mddev);
-       struct strip_zone *zone;
-       mdk_rdev_t *tmp_dev;
-       sector_t chunk;
-       sector_t sector, rsect;
+       unsigned int chunk_sects = mddev->chunk_size >> 9;
        const int rw = bio_data_dir(bio);
        int cpu;
 
@@ -414,18 +423,14 @@ static int raid0_make_request (struct request_queue *q, struct bio *bio)
                bio_endio(bio, -EOPNOTSUPP);
                return 0;
        }
-
        cpu = part_stat_lock();
        part_stat_inc(cpu, &mddev->gendisk->part0, ios[rw]);
        part_stat_add(cpu, &mddev->gendisk->part0, sectors[rw],
                      bio_sectors(bio));
        part_stat_unlock();
 
-       chunk_sects = mddev->chunk_size >> 9;
-       chunksect_bits = ffz(~chunk_sects);
-       sector = bio->bi_sector;
-
-       if (unlikely(chunk_sects < (bio->bi_sector & (chunk_sects - 1)) + (bio->bi_size >> 9))) {
+       if (unlikely(chunk_sects < ((bio->bi_sector % chunk_sects)
+                       + (bio->bi_size >> 9)))) {
                struct bio_pair *bp;
                /* Sanity check -- queue functions should prevent this happening */

               if (bio->bi_vcnt != 1 ||

drivers/md/raid0.c +4 -29
@@ -434,7 +439,8 @@ static int raid0_make_request (struct request_queue *q, struct bio *bio)
                /* This is a one page bio that upper layers
                 * refuse to split for us, so we need to split it.
                 */
-               bp = bio_split(bio, chunk_sects - (bio->bi_sector & (chunk_sects - 1)));
+               bp = bio_split(bio, chunk_sects -
+                                       (bio->bi_sector % chunk_sects));
                if (raid0_make_request(q, &bp->bio1))
                        generic_make_request(&bp->bio1);
                if (raid0_make_request(q, &bp->bio2))
@@ -443,34 +449,7 @@ static int raid0_make_request (struct request_queue *q, struct bio *bio)
                bio_pair_release(bp);
                return 0;
        }
-
-
-       {
-               sector_t x = sector >> conf->sector_shift;
-               sector_div(x, (u32)conf->spacing);
-               zone = conf->hash_table[x];
-       }
-
-       while (sector >= zone->zone_start + zone->sectors)
-               zone++;
-
-       sect_in_chunk = bio->bi_sector & (chunk_sects - 1);
-
-
-       {
-               sector_t x = (sector - zone->zone_start) >> chunksect_bits;
-
-               sector_div(x, zone->nb_dev);
-               chunk = x;
-
-               x = sector >> chunksect_bits;
-               tmp_dev = zone->dev[sector_div(x, zone->nb_dev)];
-       }
-       rsect = (chunk << chunksect_bits) + zone->dev_start + sect_in_chunk;
-
-       bio->bi_bdev = tmp_dev->bdev;
-       bio->bi_sector = rsect + tmp_dev->data_offset;
-
+       raid0_position_bio(mddev, bio, bio->bi_sector);
        /*
         * Let the main block layer submit the IO and resolve recursion:
         */
@@ -485,6 +464,7 @@ bad_map:
        return 0;
  }
 
+
  static void raid0_status (struct seq_file *seq, mddev_t *mddev)
  {
  #undef MD_DEBUG
drivers/md/raid0.h +0 -0
index 824b12e..2a73f94 100644
@@ -16,9 +16,6 @@ struct raid0_private_data
        struct strip_zone *strip_zone;

       mdk_rdev_t **devlist; /* lists of rdevs, pointed to by strip_zone->dev */ int nr_strip_zones;
-
-       sector_t spacing;
-       int sector_shift; /* shift this before divide by spacing */ };

 typedef struct raid0_private_data raid0_conf_t;

--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Signature · 13 lines
Regards,
Sandeep.





 	
“To learn is to change. Education is a process that changes the learner.”
--
To unsubscribe from this list: send the line "unsubscribe linux-raid" in
the body of a message to [email protected]
More majordomo info at  http://vger.kernel.org/majordomo-info.html