Re: [PATCH 1/4] block: implement compatible DISCARD support

["Martin K. Petersen" <mkp@mkp.net>]

>>>>> "Christoph" == Christoph Hellwig <hch@infradead.org> writes:
>> 5) It use allocated_page instead of ZERO_PAGE.

Christoph> That's incorrect - both the scsi WRITE SAME and ATA UNMAP
Christoph> implementations write to the payload.  I have some plans to
Christoph> change that an get rid of the payload entirely, but I need to
Christoph> get back to the discard work and look at it in more detail.

I've been away for a couple of weeks (got back yesterday).  Just a heads
up that I've been working on block layer support for WRITE SAME as well
as a discard revamp the last little while.  I'll try to post those
patches today or tomorrow.

Just to give you an idea: I distinguish between discarding, clearing,
and filling a block range.

blkdev_issue_clear() is used for explicitly zeroing a region.  If the
target supports RZAT/TPRZ, discard will be used instead of zero filling.

blkdev_issue_discard() is for trimming without caring about what's left
in the blocks.  And blkdev_issue_fill() can be used to do a write same
pass on a range (for bulk initializing RAID blocks, for instance).

-- 
Martin K. Petersen	Oracle Linux Engineering