Re: [PATCH v18 02/12] Add infrastructure for copy offload in block and request layer.

[Nitesh Shetty]

>>On Thu, Dec 14, 2023 at 6:41 AM Keith Busch <kbusch@xxxxxxxxxx> wrote:
>>>
>>> On Wed, Dec 06, 2023 at 03:32:34PM +0530, Kanchan Joshi wrote:
>>> >  static inline bool bio_has_data(struct bio *bio)
>>> >  {
>>> > -     if (bio &&
>>> > -         bio->bi_iter.bi_size &&
>>> > -         bio_op(bio) != REQ_OP_DISCARD &&
>>> > -         bio_op(bio) != REQ_OP_SECURE_ERASE &&
>>> > -         bio_op(bio) != REQ_OP_WRITE_ZEROES)
>>> > +     if (bio && (bio_op(bio) == REQ_OP_READ || bio_op(bio) == REQ_OP_WRITE))
>>> >               return true;
>>>
>>> There are other ops besides READ and WRITE that have data, but this is
>>> might be fine by the fact that other ops with data currently don't call
>>> this function.
>>>
>>> > diff --git a/include/linux/blk_types.h b/include/linux/blk_types.h
>>> > index 7c2316c91cbd..bd821eaa7a02 100644
>>> > --- a/include/linux/blk_types.h
>>> > +++ b/include/linux/blk_types.h
>>> > @@ -393,6 +393,10 @@ enum req_op {
>>> >       /* reset all the zone present on the device */
>>> >       REQ_OP_ZONE_RESET_ALL   = (__force blk_opf_t)17,
>>> >
>>> > +     /* copy offload dst and src operation */
>>> > +     REQ_OP_COPY_SRC         = (__force blk_opf_t)19,
>>>
>>> Should this be an even numbered OP? The odd ones are for data
>>> WRITEs.

Our request opcode needs to be write based(even) so that while forming nvme-tcp
packets we send this as part of the nvme capsule.

But now I think this design can be simplified as you suggested, if we align
COPY_SRC to even and COPY_DST to odd. This requires us to change the design by
sending dst bio first hence forming a write based request, followed by
src bio's.
Will send a follow up series next week fixing this.

Thank you,
Nitesh Shetty