On 4/11/2019 8:40 PM, Christoph Hellwig wrote:
On Thu, Mar 07, 2019 at 08:02:30PM +0200, Max Gurtovoy wrote:
From: Israel Rukshin <israelr@xxxxxxxxxxxx>
Instead of adding many arguments to RDMA mapping functions with PI,
create a new data structure 'ib_scatterlist' which contains the needed
information on a mapped scatterlist. In the future, this will also be
used by ULPs that define these attributes explicitly.
I still don't like the idea of another magic structure at all. Even
if we absolutely need something like this for the signature MR case
(which I'm not sure about) we should not require any of that for the
normal case.
We basically add 4 lines of boilerplate code to each ib_map_mr_sg
caller, and probably have worse code generation as well.
Also the name for the structure is extremely misleading - it is
not a scatterlist structure, but really the arguments structure
for ib_map_mr_sg.
we'll remove this structure from V4 and will go with more arguments
passed to the function.
Jason suggested to used iterator passing instead of sg, nents and
offset. We'll check this idea in the future since this patchset is big
enough without this refactoring and still changes to scatterlist
functionality should be made as well.
