[PATCH v20 02/12] Add infrastructure for copy offload in block and request layer.
Damien Le Moal
dlemoal at kernel.org
Tue Jun 25 14:18:18 PDT 2024
On 6/26/24 03:18, Bart Van Assche wrote:
> On 6/24/24 2:55 PM, Damien Le Moal wrote:
>> I am still a little confused as to why we need 2 BIOs, one for src and one for
>> dst... Is it because of the overly complex scsi extended copy support ?
>>
>> Given that the main use case is copy offload for data within the same device,
>> using a single BIO which somehow can carry a list of LBA sources and a single
>> destination LBA would be far simpler and perfectly matching nvme simple copy and
>> ATA write gathered. And I think that this would also match the simplest case for
>> scsi extended copy as well.
>
> Hi Damien,
>
> What are the implications for the device mapper code if the copy source
> and destination LBAs are encoded in the bio payload instead of in
> bio->bi_sector?
DM can deal with "abnormal" BIOs on its own. There is code for that.
See is_abnormal_io() and __process_abnormal_io(). Sure, that will need more code
compared to a bio sector+size based simple split, but I do not think it is a big
deal given the potential benefits of the offloading.
--
Damien Le Moal
Western Digital Research
More information about the Linux-nvme
mailing list