On 3/18/2020 12:50 AM, Bart Van Assche wrote:
On 2020-03-17 08:41, Max Gurtovoy wrote:
On 3/17/2020 5:19 PM, Chuck Lever wrote:
If you want to guarantee that there is an SRQ for each comp_vector and a
comp_vector for each SRQ, stick with a CQ allocation API that enables
explicit selection of the comp_vector value, and cache that value in the
caller, not in the core data structures.
I'm Ok with that as well. This is exactly what we do in the nvmf/rdma
but I wanted to stick also with the SRP target approach.
Any objection to remove the call for ib_alloc_cq_any() from ib_srpt and
use ib_alloc_cq() ?
Hi Max,
Wasn't that call introduced by Chuck (see also commit 20cf4e026730
("rdma: Enable ib_alloc_cq to spread work over a device's comp_vectors")
# v5.4)? Anyway, I'm fine with the proposed change.
Yes this was introduced by Chuck, but no contract is broken :)
I'll update srpt in V2.
Thanks,
Bart.
