[PATCH v20 02/12] Add infrastructure for copy offload in block and request layer.

Bart Van Assche bvanassche at acm.org
Fri Jun 28 06:53:32 PDT 2024


On 6/25/24 10:22 PM, Christoph Hellwig wrote:
> It's not just dm.  You also need it in the partition remapping code
> (mandatory), md (nice to have), etc.
> 
> And then we have the whole mess of what is in the payload for the I/O
> stack vs what is in the payload for the on the wire protocol, which
> will have different formatting and potentially also different sizes.

Drivers like dm-linear rely on bio splitting. If the COPY_SRC and
COPY_DST operations travel independently down a stacked block driver
hierarchy, a separate data structure is needed to keep track of which
operations have been split and to combine the split operations into
requests. Isn't this an argument in favor of storing the source and
destination parameters in a single bio?

Thanks,

Bart.



More information about the Linux-nvme mailing list