-----"Jason Gunthorpe" <jgg@xxxxxxxx> wrote: ----- >To: "zhengbin" <zhengbin13@xxxxxxxxxx> >From: "Jason Gunthorpe" <jgg@xxxxxxxx> >Date: 01/04/2020 12:15AM >Cc: bmt@xxxxxxxxxxxxxx, dledford@xxxxxxxxxx, >linux-rdma@xxxxxxxxxxxxxxx >Subject: [EXTERNAL] Re: [PATCH 0/5] RDMA: use true,false for bool >variable > >On Tue, Dec 24, 2019 at 04:40:07PM +0800, zhengbin wrote: >> zhengbin (5): >> RDMA/siw: use true,false for bool variable >> IB/hfi1: use true,false for bool variable >> IB/iser: use true,false for bool variable >> RDMA/mlx4: use true,false for bool variable >> RDMA/mlx5: use true,false for bool variable > >Applied to for-next, but Leon seems right about the funny >relaxed_ird_negotiation that is never set, is that debugging or >something? > Sorry for not coming back earlier. I had a quite severe bicycle accident which has put me completely off for the last two and a half weeks. This flag was introduced with a change to the MPA setup sequence. While siw originally insisted in correct peer behavior (not sending data in RDMA mode until handshake completed), this led to issues with another iWarp implementation, which under certain circumstances brakes the correct setup sequence. So we allowed that peer behavior, while it brakes the spec. If we agree that this 'permissive' behavior to be the only behavior we want to support, we can simply remove that flag and related code, maybe just leaving a comment at the right place stating we are permissive? Thanks Bernard.
