| Message ID | 20191027140549.26272-3-damien.lemoal@wdc.com (mailing list archive) |
|---|---|
| State | Not Applicable |
| Headers | show |
| Series | Zone management commands support | expand |
The reason code for REQ_OP_RESET_ALL is kept in a different function so we can clearly differentiate between REQ_OP_RESET and REQ_OP_RESET_ALL when we add new tracepoints with blktrace framework. But if that is acceptable, then, Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com> On 10/27/19 7:06 AM, Damien Le Moal wrote: > There is no need for the function __blkdev_reset_all_zones() as > REQ_OP_ZONE_RESET_ALL can be handled directly in blkdev_reset_zones() > bio loop with an early break from the loop. This patch removes this > function and modifies blkdev_reset_zones(), simplifying the code. > > Signed-off-by: Damien Le Moal<damien.lemoal@wdc.com>
On 2019/10/28 8:49, Chaitanya Kulkarni wrote: > The reason code for REQ_OP_RESET_ALL is kept in a different function so > we can clearly differentiate between REQ_OP_RESET and REQ_OP_RESET_ALL > when we add new tracepoints with blktrace framework. Isn't the trace point under submit_bio() in generic_make_request_checks() ? So removing the function is not a problem for tracing as far as I can tell. Am I missing something ? > > But if that is acceptable, then, > > Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com> > > On 10/27/19 7:06 AM, Damien Le Moal wrote: >> There is no need for the function __blkdev_reset_all_zones() as >> REQ_OP_ZONE_RESET_ALL can be handled directly in blkdev_reset_zones() >> bio loop with an early break from the loop. This patch removes this >> function and modifies blkdev_reset_zones(), simplifying the code. >> >> Signed-off-by: Damien Le Moal<damien.lemoal@wdc.com> > >
On 10/28/19 12:54 AM, Damien Le Moal wrote: > Isn't the trace point under submit_bio() in > generic_make_request_checks() ? So removing the function is not a > problem for tracing as far as I can tell. Am I missing something ? Yes you are right, I completely missed that. Sorry I've created confusion with block_dump and tracepoint. Block trace code is fine. The block_dump code under the submit_bio() is only dumps the bios with the data:- 1171 if (unlikely(block_dump)) { 1172 char b[BDEVNAME_SIZE]; 1173 printk(KERN_DEBUG "%s(%d): %s block %Lu on %s (%u sectors)\n", 1174 current->comm, task_pid_nr(current), 1175 op_is_write(bio_op(bio)) ? "WRITE" : "READ", 1176 (unsigned long long)bio->bi_iter.bi_sector, 1177 bio_devname(bio, b), count); 1178 } I've posted patch-series [1] in the past to move that code out but it didn't go anywhere in anticipation of more data less requests. Since it is taking longer to have blktrace extensions RFC to move forward and [1] didn't go anywhere I wanted to use block_dump parameter in the blk-zoned.c (not an ideal situation) so that we can have atleast minimal debug support for the new REQ_OP_ZONE_XXX operations until we get block trace extensions in the kernel. Nonetheless, I'll just a send a patch on the top of this which will make discussion much easier. [1] :- https://lore.kernel.org/linux-block/DM6PR04MB57546ECC4CFDDB5535E3382586FB0@DM6PR04MB5754.namprd04.prod.outlook.com/T/
Looks good,
Reviewed-by: Christoph Hellwig <hch@lst.de>
diff --git a/block/blk-zoned.c b/block/blk-zoned.c index 7fe376eede86..14785011e798 100644 --- a/block/blk-zoned.c +++ b/block/blk-zoned.c @@ -202,32 +202,14 @@ int blkdev_report_zones(struct block_device *bdev, sector_t sector, } EXPORT_SYMBOL_GPL(blkdev_report_zones); -/* - * Special case of zone reset operation to reset all zones in one command, - * useful for applications like mkfs. - */ -static int __blkdev_reset_all_zones(struct block_device *bdev, gfp_t gfp_mask) -{ - struct bio *bio = bio_alloc(gfp_mask, 0); - int ret; - - /* across the zones operations, don't need any sectors */ - bio_set_dev(bio, bdev); - bio_set_op_attrs(bio, REQ_OP_ZONE_RESET_ALL, 0); - - ret = submit_bio_wait(bio); - bio_put(bio); - - return ret; -} - static inline bool blkdev_allow_reset_all_zones(struct block_device *bdev, + sector_t sector, sector_t nr_sectors) { if (!blk_queue_zone_resetall(bdev_get_queue(bdev))) return false; - if (nr_sectors != part_nr_sects_read(bdev->bd_part)) + if (sector || nr_sectors != part_nr_sects_read(bdev->bd_part)) return false; /* * REQ_OP_ZONE_RESET_ALL can be executed only if the block device is @@ -270,9 +252,6 @@ int blkdev_reset_zones(struct block_device *bdev, /* Out of range */ return -EINVAL; - if (blkdev_allow_reset_all_zones(bdev, nr_sectors)) - return __blkdev_reset_all_zones(bdev, gfp_mask); - /* Check alignment (handle eventual smaller last zone) */ zone_sectors = blk_queue_zone_sectors(q); if (sector & (zone_sectors - 1)) @@ -283,17 +262,24 @@ int blkdev_reset_zones(struct block_device *bdev, return -EINVAL; while (sector < end_sector) { - bio = blk_next_bio(bio, 0, gfp_mask); - bio->bi_iter.bi_sector = sector; bio_set_dev(bio, bdev); - bio_set_op_attrs(bio, REQ_OP_ZONE_RESET, 0); + /* + * Special case for the zone reset operation that reset all + * zones, this is useful for applications like mkfs. + */ + if (blkdev_allow_reset_all_zones(bdev, sector, nr_sectors)) { + bio->bi_opf = REQ_OP_ZONE_RESET_ALL; + break; + } + + bio->bi_opf = REQ_OP_ZONE_RESET; + bio->bi_iter.bi_sector = sector; sector += zone_sectors; /* This may take a while, so be nice to others */ cond_resched(); - } ret = submit_bio_wait(bio);
There is no need for the function __blkdev_reset_all_zones() as REQ_OP_ZONE_RESET_ALL can be handled directly in blkdev_reset_zones() bio loop with an early break from the loop. This patch removes this function and modifies blkdev_reset_zones(), simplifying the code. Signed-off-by: Damien Le Moal <damien.lemoal@wdc.com> --- block/blk-zoned.c | 40 +++++++++++++--------------------------- 1 file changed, 13 insertions(+), 27 deletions(-)