| Message ID | 20191027140549.26272-1-damien.lemoal@wdc.com (mailing list archive) |
|---|---|
| Headers | show |
| Series | Zone management commands support | expand |
On 10/27/19 8:05 AM, Damien Le Moal wrote: > This series implements a few improvements and cleanups to zone block > device zone reset operations with the first three patches. > > The remaining of the series patches introduce zone open, close and > finish support, allowing users of zoned block devices to explicitly > control the condition (state) of zones. > > While these operations are not stricktly necessary for the correct > operation of zoned block devices, the open and close operations can > improve performance for some device implementations of the ZBC and ZAC > standards under write workloads. The finish zone operation, which > transition a zone to the full state, can also be useful to protect a > zone data by preventing further zone writes. > > These operations are implemented by introducing the new > REQ_OP_ZONE_OPEN, REQ_OP_ZONE_CLOSE and REQ_OP_ZONE_FINISH request codes > and the function blkdev_zone_mgmt() to issue these requests. This new > function also replaces the former blkdev_reset_zones() function to reset > zones write pointer. > > The new ioctls BLKOPENZONE, BLKCLOSEZONE and BLKFINISHZONE are also > defined to allow applications to issue these new requests without > resorting to a device passthrough interface (e.g. SG_IO). > > Support for these operations is added to the SCSI sd driver, to the dm > infrastructure (dm-linear and dm-flakey targets) and to the null_blk > driver. Can patch 3 go in separately, doesn't look like we need it in this series? Also need the DM folks to review/sign off on patch 7. Mike?
On 2019/11/02 4:01, Jens Axboe wrote: > On 10/27/19 8:05 AM, Damien Le Moal wrote: >> This series implements a few improvements and cleanups to zone block >> device zone reset operations with the first three patches. >> >> The remaining of the series patches introduce zone open, close and >> finish support, allowing users of zoned block devices to explicitly >> control the condition (state) of zones. >> >> While these operations are not stricktly necessary for the correct >> operation of zoned block devices, the open and close operations can >> improve performance for some device implementations of the ZBC and ZAC >> standards under write workloads. The finish zone operation, which >> transition a zone to the full state, can also be useful to protect a >> zone data by preventing further zone writes. >> >> These operations are implemented by introducing the new >> REQ_OP_ZONE_OPEN, REQ_OP_ZONE_CLOSE and REQ_OP_ZONE_FINISH request codes >> and the function blkdev_zone_mgmt() to issue these requests. This new >> function also replaces the former blkdev_reset_zones() function to reset >> zones write pointer. >> >> The new ioctls BLKOPENZONE, BLKCLOSEZONE and BLKFINISHZONE are also >> defined to allow applications to issue these new requests without >> resorting to a device passthrough interface (e.g. SG_IO). >> >> Support for these operations is added to the SCSI sd driver, to the dm >> infrastructure (dm-linear and dm-flakey targets) and to the null_blk >> driver. > > Can patch 3 go in separately, doesn't look like we need it in this > series? Yes, I think it can go in now in 5.4-rc if Martin is willing to take it. That will create a small conflict in your tree for patch 6 though. Martin, Can you take patch 3 now ? > > Also need the DM folks to review/sign off on patch 7. Mike? >
Damien,
> Can you take patch 3 now ?
Yep. Applied to 5.4/scsi-fixes.
On 2019/11/05 13:53, Martin K. Petersen wrote: > > Damien, > >> Can you take patch 3 now ? > > Yep. Applied to 5.4/scsi-fixes. Thanks !
On 10/27/19 8:05 AM, Damien Le Moal wrote: > This series implements a few improvements and cleanups to zone block > device zone reset operations with the first three patches. > > The remaining of the series patches introduce zone open, close and > finish support, allowing users of zoned block devices to explicitly > control the condition (state) of zones. > > While these operations are not stricktly necessary for the correct > operation of zoned block devices, the open and close operations can > improve performance for some device implementations of the ZBC and ZAC > standards under write workloads. The finish zone operation, which > transition a zone to the full state, can also be useful to protect a > zone data by preventing further zone writes. > > These operations are implemented by introducing the new > REQ_OP_ZONE_OPEN, REQ_OP_ZONE_CLOSE and REQ_OP_ZONE_FINISH request codes > and the function blkdev_zone_mgmt() to issue these requests. This new > function also replaces the former blkdev_reset_zones() function to reset > zones write pointer. > > The new ioctls BLKOPENZONE, BLKCLOSEZONE and BLKFINISHZONE are also > defined to allow applications to issue these new requests without > resorting to a device passthrough interface (e.g. SG_IO). > > Support for these operations is added to the SCSI sd driver, to the dm > infrastructure (dm-linear and dm-flakey targets) and to the null_blk > driver. Applied for 5.5, thanks. I've got the last sd patch pending, the conflict is rather ugly. I'll setup a post branch for drivers with this in, once the dependent fix has landed in Linus's tree.