ublk: remove segment count and size limits
authorUday Shankar <ushankar@purestorage.com>
Tue, 30 Apr 2024 21:16:24 +0000 (15:16 -0600)
committerJens Axboe <axboe@kernel.dk>
Tue, 30 Apr 2024 21:36:50 +0000 (15:36 -0600)
ublk_drv currently creates block devices with the default max_segments
and max_segment_size limits of BLK_MAX_SEGMENTS (128) and
BLK_MAX_SEGMENT_SIZE (65536) respectively. These defaults can
artificially constrain the I/O size seen by the ublk server - for
example, suppose that the ublk server has configured itself to accept
I/Os up to 1M and the application is also issuing 1M sized I/Os. If the
I/O buffer used by the application is backed by 4K pages, the buffer
could consist of up to 1M / 4K = 256 physically discontiguous segments
(even if the buffer is virtually contiguous). As such, the I/O could
exceed the default max_segments limit and get split. This can cause
unnecessary performance issues if the ublk server is optimized to handle
1M I/Os. The block layer's segment count/size limits exist to model
hardware constraints which don't exist in ublk_drv's case, so just
remove those limits for the block devices created by ublk_drv.

Signed-off-by: Uday Shankar <ushankar@purestorage.com>
Reviewed-by: Riley Thomasson <riley@purestorage.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20240430211623.2802036-1-ushankar@purestorage.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
drivers/block/ublk_drv.c

index bea3d5c..374e4ef 100644 (file)
@@ -2177,7 +2177,8 @@ static int ublk_ctrl_start_dev(struct ublk_device *ub, struct io_uring_cmd *cmd)
                .max_hw_sectors         = p->max_sectors,
                .chunk_sectors          = p->chunk_sectors,
                .virt_boundary_mask     = p->virt_boundary_mask,
-
+               .max_segments           = USHRT_MAX,
+               .max_segment_size       = UINT_MAX,
        };
        struct gendisk *disk;
        int ret = -EINVAL;