[PATCH V3 0/2] sd fixes and cleanup

[PATCH V3 0/2] sd fixes and cleanup

[Damien Le Moal]

A couple of patches in this small series.

The first patch addresses a problem with large write same (or zeroout) block
layer calls being split into commands that are not aligned on the device
physical block size. This will work on regular disks but is not optimal.
The commands will however fail with host-managed zoned block device if the
commands target sequential write required zones. The patch simply makes sure
that the maximum write same sectors is aligned to physical block sizes.
Note that this fix can be implemented either in sd_config_write_same() or in
the block layer (__blkdev_issue_write_same() and __blkdev_issue_write_zeroes()).

The second patch does not introduce functional changes.

Changes from V2:
* Use round_down() in first patch to align maximum write same sectors to
  physical sector size.

Changes from V1:
* Removed useless addition of blank line an indentaion change in first patch.

*** BLURB HERE ***

Damien Le Moal (2):
  scsi: sd: Fix sd_config_write_same()
  scsi: sd: Use sectors_to_logical()

 drivers/scsi/sd.c | 36 ++++++++++++++++++++++++++++--------
 1 file changed, 28 insertions(+), 8 deletions(-)

-- 
2.13.5

[PATCH V3 1/2] scsi: sd: Fix sd_config_write_same()

[Damien Le Moal]

Reporting a maximum number of blocks that is not aligned on the device
physical size would cause a large write same request to be split into
physically unaligned chunks by __blkdev_issue_write_zeroes() and
__blkdev_issue_write_same(), even if the caller of these functions took
care to align its request to physical sectors.

So make sure the maximum reported is aligned to the device physical
block size. This is only an optional optimization for regular disks,
but this is mandatory to avoid failure of large write same requests
directed at sequential write required zones of host-managed ZBC disks.

Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
---
 drivers/scsi/sd.c | 20 ++++++++++++++++++++
 1 file changed, 20 insertions(+)

diff --git a/drivers/scsi/sd.c b/drivers/scsi/sd.c
index e2647f2d4430..0a824e9f4d63 100644
--- a/drivers/scsi/sd.c
+++ b/drivers/scsi/sd.c
@@ -898,6 +898,26 @@ static void sd_config_write_same(struct scsi_disk *sdkp)
 	else
 		sdkp->zeroing_mode = SD_ZERO_WRITE;
 
+	if (sdkp->max_ws_blocks &&
+	    sdkp->physical_block_size > logical_block_size) {
+		/*
+		 * Reporting a maximum number of blocks that is not aligned
+		 * on the device physical size would cause a large write same
+		 * request to be split into physically unaligned chunks by
+		 * __blkdev_issue_write_zeroes() and __blkdev_issue_write_same()
+		 * even if the caller of these functions took care to align the
+		 * large request. So make sure the maximum reported is aligned
+		 * to the device physical block size. This is only an optional
+		 * optimization for regular disks, but this is mandatory to
+		 * avoid failure of large write same requests directed at
+		 * sequential write required zones of host-managed ZBC disks.
+		 */
+		sdkp->max_ws_blocks =
+			round_down(sdkp->max_ws_blocks,
+				   bytes_to_logical(sdkp->device,
+						    sdkp->physical_block_size));
+	}
+
 out:
 	blk_queue_max_write_same_sectors(q, sdkp->max_ws_blocks *
 					 (logical_block_size >> 9));
-- 
2.13.5

[PATCH V3 2/2] scsi: sd: Use sectors_to_logical()

[Damien Le Moal]

Replace open coded conversions.

Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com>
---
 drivers/scsi/sd.c | 16 ++++++++--------
 1 file changed, 8 insertions(+), 8 deletions(-)

diff --git a/drivers/scsi/sd.c b/drivers/scsi/sd.c
index 0a824e9f4d63..c5d05b1c58a5 100644
--- a/drivers/scsi/sd.c
+++ b/drivers/scsi/sd.c
@@ -738,8 +738,8 @@ static int sd_setup_unmap_cmnd(struct scsi_cmnd *cmd)
 {
 	struct scsi_device *sdp = cmd->device;
 	struct request *rq = cmd->request;
-	u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
-	u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
+	u64 sector = sectors_to_logical(sdp, blk_rq_pos(rq));
+	u32 nr_sectors = sectors_to_logical(sdp, blk_rq_sectors(rq));
 	unsigned int data_len = 24;
 	char *buf;
 
@@ -772,8 +772,8 @@ static int sd_setup_write_same16_cmnd(struct scsi_cmnd *cmd, bool unmap)
 {
 	struct scsi_device *sdp = cmd->device;
 	struct request *rq = cmd->request;
-	u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
-	u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
+	u64 sector = sectors_to_logical(sdp, blk_rq_pos(rq));
+	u32 nr_sectors = sectors_to_logical(sdp, blk_rq_sectors(rq));
 	u32 data_len = sdp->sector_size;
 
 	rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
@@ -802,8 +802,8 @@ static int sd_setup_write_same10_cmnd(struct scsi_cmnd *cmd, bool unmap)
 {
 	struct scsi_device *sdp = cmd->device;
 	struct request *rq = cmd->request;
-	u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
-	u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
+	u64 sector = sectors_to_logical(sdp, blk_rq_pos(rq));
+	u32 nr_sectors = sectors_to_logical(sdp, blk_rq_sectors(rq));
 	u32 data_len = sdp->sector_size;
 
 	rq->special_vec.bv_page = alloc_page(GFP_ATOMIC | __GFP_ZERO);
@@ -833,8 +833,8 @@ static int sd_setup_write_zeroes_cmnd(struct scsi_cmnd *cmd)
 	struct request *rq = cmd->request;
 	struct scsi_device *sdp = cmd->device;
 	struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
-	u64 sector = blk_rq_pos(rq) >> (ilog2(sdp->sector_size) - 9);
-	u32 nr_sectors = blk_rq_sectors(rq) >> (ilog2(sdp->sector_size) - 9);
+	u64 sector = sectors_to_logical(sdp, blk_rq_pos(rq));
+	u32 nr_sectors = sectors_to_logical(sdp, blk_rq_sectors(rq));
 	int ret;
 
 	if (!(rq->cmd_flags & REQ_NOUNMAP)) {
-- 
2.13.5

Re: [PATCH V3 1/2] scsi: sd: Fix sd_config_write_same()

[Bart Van Assche]

On Tue, 2017-09-05 at 20:55 +0900, Damien Le Moal wrote:
> Reporting a maximum number of blocks that is not aligned on the device
> physical size would cause a large write same request to be split into
> physically unaligned chunks by __blkdev_issue_write_zeroes() and
> __blkdev_issue_write_same(), even if the caller of these functions took
> care to align its request to physical sectors.
> 
> So make sure the maximum reported is aligned to the device physical
> block size. This is only an optional optimization for regular disks,
> but this is mandatory to avoid failure of large write same requests
> directed at sequential write required zones of host-managed ZBC disks.

Reviewed-by: Bart Van Assche <bart.vanassche@wdc.com>

Re: [PATCH V3 2/2] scsi: sd: Use sectors_to_logical()

[Bart Van Assche]

On Tue, 2017-09-05 at 20:55 +0900, Damien Le Moal wrote:
> Replace open coded conversions.

Reviewed-by: Bart Van Assche <bart.vanassche@wdc.com>

Re: [PATCH V3 1/2] scsi: sd: Fix sd_config_write_same()

[Martin K. Petersen]

Damien,

> Reporting a maximum number of blocks that is not aligned on the device
> physical size would cause a large write same request to be split into
> physically unaligned chunks by __blkdev_issue_write_zeroes() and
> __blkdev_issue_write_same(), even if the caller of these functions
> took care to align its request to physical sectors.

Applied to 4.15/scsi-queue. Thanks!

-- 
Martin K. Petersen	Oracle Linux Engineering