> On 11 Aug 2021, at 13:44, Jason Gunthorpe <jgg@xxxxxxxxxx> wrote:
>
> On Wed, Aug 11, 2021 at 08:16:50AM +0300, Prabhakar Kushwaha wrote:
>> Qedr code is tightly coupled with existing both INIT transitions.
>> Here, during first INIT transition all variables are reset and
>> the RESET state is checked in post_recv() before any posting.
>>
>> Commit dc70f7c3ed34 ("RDMA/cma: Remove unnecessary INIT->INIT
>> transition") exposed this bug.
>
>
> Since we are reverting this the patch still makse sense? Certainly
> having a driver depend on two init->init transitions seems wrong to me
If what I wrote about adhering to the IBTA spec and transition the QP to INIT during rdma_{connect,accept} makes sense, I think we need a two-phased approach.
1. Get the ULPs to adhere. That would not demand any changes in cma.
2. Once 1 has been proven working, we can fix cma and make the transition to INIT in rdma_{connect,accept}.
OK?
Thxs, Håkon
