>>>>> "Jens" == Jens Axboe writes: Jens> Great, the two different values and needing to sync them was Jens> horrible. What kind of testing did you do? Have to be a little Jens> extra careful at this point. Yeah, we should probably let it soak a bit in -next just to make sure. There really aren't many devices from this millennium that don't support clustering. Which I guess is why we haven't seen any problems. I ended up disabling clustering in one of the FC drivers to test with a real workload. Threw in a BUG_ON(nsegs > queue_max_segments(q)) for good measure. I also tested mixing and matching clustered and non-clustered bottom devices while stacking with DM. New version below, fixing the things you and Matthew pointed out... block: Deprecate QUEUE_FLAG_CLUSTER and use queue_limits instead When stacking devices, a request_queue is not always available. This forced us to have a no_cluster flag in the queue_limits that could be used as a carrier until the request_queue had been set up for a metadevice. There were several problems with that approach. First of all it was up to the stacking device to remember to set queue flag after stacking had completed. Also, the queue flag and the queue limits had to be kept in sync at all times. We got that wrong, which could lead to us issuing commands that went beyond the max scatterlist limit set by the driver. The proper fix is to avoid having two flags for tracking the same thing. We deprecate QUEUE_FLAG_CLUSTER and use the queue limit directly in the block layer merging functions. The queue_limit 'no_cluster' is turned into 'cluster' to avoid double negatives and to ease stacking. Clustering defaults to being enabled as before. The queue flag logic is removed from the stacking function, and explicitly setting the cluster flag is no longer necessary in DM and MD. Reported-by: Ed Lin Signed-off-by: Martin K. Petersen diff --git a/block/blk-merge.c b/block/blk-merge.c index 77b7c26..74bc4a7 100644 --- a/block/blk-merge.c +++ b/block/blk-merge.c @@ -21,7 +21,7 @@ static unsigned int __blk_recalc_rq_segments(struct request_queue *q, return 0; fbio = bio; - cluster = test_bit(QUEUE_FLAG_CLUSTER, &q->queue_flags); + cluster = blk_queue_cluster(q); seg_size = 0; nr_phys_segs = 0; for_each_bio(bio) { @@ -87,7 +87,7 @@ EXPORT_SYMBOL(blk_recount_segments); static int blk_phys_contig_segment(struct request_queue *q, struct bio *bio, struct bio *nxt) { - if (!test_bit(QUEUE_FLAG_CLUSTER, &q->queue_flags)) + if (!blk_queue_cluster(q)) return 0; if (bio->bi_seg_back_size + nxt->bi_seg_front_size > @@ -123,7 +123,7 @@ int blk_rq_map_sg(struct request_queue *q, struct request *rq, int nsegs, cluster; nsegs = 0; - cluster = test_bit(QUEUE_FLAG_CLUSTER, &q->queue_flags); + cluster = blk_queue_cluster(q); /* * for each bio in rq diff --git a/block/blk-settings.c b/block/blk-settings.c index 701859f..e55f5fc 100644 --- a/block/blk-settings.c +++ b/block/blk-settings.c @@ -126,7 +126,7 @@ void blk_set_default_limits(struct queue_limits *lim) lim->alignment_offset = 0; lim->io_opt = 0; lim->misaligned = 0; - lim->no_cluster = 0; + lim->cluster = 1; } EXPORT_SYMBOL(blk_set_default_limits); @@ -464,15 +464,6 @@ EXPORT_SYMBOL(blk_queue_io_opt); void blk_queue_stack_limits(struct request_queue *t, struct request_queue *b) { blk_stack_limits(&t->limits, &b->limits, 0); - - if (!t->queue_lock) - WARN_ON_ONCE(1); - else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) { - unsigned long flags; - spin_lock_irqsave(t->queue_lock, flags); - queue_flag_clear(QUEUE_FLAG_CLUSTER, t); - spin_unlock_irqrestore(t->queue_lock, flags); - } } EXPORT_SYMBOL(blk_queue_stack_limits); @@ -545,7 +536,7 @@ int blk_stack_limits(struct queue_limits *t, struct queue_limits *b, t->io_min = max(t->io_min, b->io_min); t->io_opt = lcm(t->io_opt, b->io_opt); - t->no_cluster |= b->no_cluster; + t->cluster &= b->cluster; t->discard_zeroes_data &= b->discard_zeroes_data; /* Physical block size a multiple of the logical block size? */ @@ -641,7 +632,6 @@ void disk_stack_limits(struct gendisk *disk, struct block_device *bdev, sector_t offset) { struct request_queue *t = disk->queue; - struct request_queue *b = bdev_get_queue(bdev); if (bdev_stack_limits(&t->limits, bdev, offset >> 9) < 0) { char top[BDEVNAME_SIZE], bottom[BDEVNAME_SIZE]; @@ -652,17 +642,6 @@ void disk_stack_limits(struct gendisk *disk, struct block_device *bdev, printk(KERN_NOTICE "%s: Warning: Device %s is misaligned\n", top, bottom); } - - if (!t->queue_lock) - WARN_ON_ONCE(1); - else if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) { - unsigned long flags; - - spin_lock_irqsave(t->queue_lock, flags); - if (!test_bit(QUEUE_FLAG_CLUSTER, &b->queue_flags)) - queue_flag_clear(QUEUE_FLAG_CLUSTER, t); - spin_unlock_irqrestore(t->queue_lock, flags); - } } EXPORT_SYMBOL(disk_stack_limits); diff --git a/block/blk-sysfs.c b/block/blk-sysfs.c index 013457f..41fb691 100644 --- a/block/blk-sysfs.c +++ b/block/blk-sysfs.c @@ -119,7 +119,7 @@ static ssize_t queue_max_integrity_segments_show(struct request_queue *q, char * static ssize_t queue_max_segment_size_show(struct request_queue *q, char *page) { - if (test_bit(QUEUE_FLAG_CLUSTER, &q->queue_flags)) + if (blk_queue_cluster(q)) return queue_var_show(queue_max_segment_size(q), (page)); return queue_var_show(PAGE_CACHE_SIZE, (page)); diff --git a/drivers/md/dm-table.c b/drivers/md/dm-table.c index 90267f8..e2da191 100644 --- a/drivers/md/dm-table.c +++ b/drivers/md/dm-table.c @@ -1131,11 +1131,6 @@ void dm_table_set_restrictions(struct dm_table *t, struct request_queue *q, */ q->limits = *limits; - if (limits->no_cluster) - queue_flag_clear_unlocked(QUEUE_FLAG_CLUSTER, q); - else - queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, q); - if (!dm_table_supports_discards(t)) queue_flag_clear_unlocked(QUEUE_FLAG_DISCARD, q); else diff --git a/drivers/md/md.c b/drivers/md/md.c index 84c46a1..52694d2 100644 --- a/drivers/md/md.c +++ b/drivers/md/md.c @@ -4296,9 +4296,6 @@ static int md_alloc(dev_t dev, char *name) goto abort; mddev->queue->queuedata = mddev; - /* Can be unlocked because the queue is new: no concurrency */ - queue_flag_set_unlocked(QUEUE_FLAG_CLUSTER, mddev->queue); - blk_queue_make_request(mddev->queue, md_make_request); disk = alloc_disk(1 << shift); diff --git a/drivers/scsi/scsi_lib.c b/drivers/scsi/scsi_lib.c index b55b0ec..3852e51 100644 --- a/drivers/scsi/scsi_lib.c +++ b/drivers/scsi/scsi_lib.c @@ -1643,9 +1643,8 @@ struct request_queue *__scsi_alloc_queue(struct Scsi_Host *shost, blk_queue_max_segment_size(q, dma_get_max_seg_size(dev)); - /* New queue, no concurrency on queue_flags */ if (!shost->use_clustering) - queue_flag_clear_unlocked(QUEUE_FLAG_CLUSTER, q); + q->limits.cluster = 0; /* * set a reasonable default alignment on word boundaries: the diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h index aae86fd..95aeeeb 100644 --- a/include/linux/blkdev.h +++ b/include/linux/blkdev.h @@ -250,7 +250,7 @@ struct queue_limits { unsigned char misaligned; unsigned char discard_misaligned; - unsigned char no_cluster; + unsigned char cluster; signed char discard_zeroes_data; }; @@ -380,7 +380,6 @@ struct request_queue #endif }; -#define QUEUE_FLAG_CLUSTER 0 /* cluster several segments into 1 */ #define QUEUE_FLAG_QUEUED 1 /* uses generic tag queueing */ #define QUEUE_FLAG_STOPPED 2 /* queue is stopped */ #define QUEUE_FLAG_SYNCFULL 3 /* read queue has been filled */ @@ -403,7 +402,6 @@ struct request_queue #define QUEUE_FLAG_SECDISCARD 19 /* supports SECDISCARD */ #define QUEUE_FLAG_DEFAULT ((1 << QUEUE_FLAG_IO_STAT) | \ - (1 << QUEUE_FLAG_CLUSTER) | \ (1 << QUEUE_FLAG_STACKABLE) | \ (1 << QUEUE_FLAG_SAME_COMP) | \ (1 << QUEUE_FLAG_ADD_RANDOM)) @@ -510,6 +508,11 @@ static inline void queue_flag_clear(unsigned int flag, struct request_queue *q) #define rq_data_dir(rq) ((rq)->cmd_flags & 1) +static inline unsigned int blk_queue_cluster(struct request_queue *q) +{ + return q->limits.cluster; +} + /* * We regard a request as sync, if either a read or a sync write */ -- dm-devel mailing list dm-devel@redhat.com https://www.redhat.com/mailman/listinfo/dm-devel