On Tue, Dec 15, 2015 at 06:10:54PM +0100, Bart Van Assche wrote: > Maybe I should explain why I started looking into this. If LIO exports a > block device for which the discard_zeroes_data flag has been set then it > would be possible for LIO to set the LBPRZ bit in the READ CAPACITY(16) > response (something it does not yet do today). This sounds like a useful goal. > If the granularity of > discard requests of the block device is larger than the block size of the > SCSI device, which layer should then handle the start and end blocks of the > SCSI UNMAP command if the start and/or end are not aligned on a discard > boundary ? Should the caller of blkdev_issue_discard() implement this or > should this functionality be added in blkdev_issue_discard() itself ? There is two sides to it: interface and implementation. To implement UNMAP / WRITE SAME with LBPRZ you should use blkdev_issue_zeroout with the discard flag set to true. But as-is this will have the same issue as blkdev_issue_discard as it will end up using blkdev_issue_discard if the functionality is supported by the device. I'm not sure how we could communicate the split size up the stack, so I fear the answer might be that blk_bio_discard_split might have to create additional WRITE (ZERO) bios if it hits an alignment mismatch. -- To unsubscribe from this list: send the line "unsubscribe linux-block" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
