Re: [PATCHv10 0/9] write hints with nvme fdp, scsi streams

[Nitesh Shetty <nj.shetty@samsung.com>]

[-- Attachment #1: Type: text/plain, Size: 1148 bytes --]

On 09/12/24 09:20PM, Martin K. Petersen wrote:
>
>Bart,
>
>> Does "cookie" refer to the SCSI ROD token? Storing the ROD token in
>> the REQ_OP_COPY_DST bio implies that the REQ_OP_COPY_DST bio is only
>> submitted after the REQ_OP_COPY_SRC bio has completed.
>
>Obviously. You can't issue a WRITE USING TOKEN until you have the token.
>
>> NVMe users may prefer that REQ_OP_COPY_SRC and REQ_OP_COPY_DST bios
>> are submitted simultaneously.
>
>What would be the benefit of submitting these operations concurrently?
>As I have explained, it adds substantial complexity and object lifetime
>issues throughout the stack. To what end?
>
>-- 
Bart,
We did implement payload based approach in the past[1] which aligns
with this. Since we wait till the REQ_OP_COPY_SRC completes, there won't
be issue with async type of dm IOs.
Since this would be an internal kernel plumbing, we can optimize/change
the approach moving forward.
If you are okay with the approach, I can give a respin to that version.

Thanks,
Nitesh Shetty

[1]
https://lore.kernel.org/linux-block/20230605121732.28468-1-nj.shetty@samsung.com/T/#mecd04c060cd4285a4b036ca79cc58713308771fe

[-- Attachment #2: Type: text/plain, Size: 0 bytes --]