On Thu, May 3, 2018 at 3:31 PM, Alex Rosenbaum <rosenbaumalex@xxxxxxxxx> wrote: > On Thu, May 3, 2018 at 12:12 AM, Jason Gunthorpe <jgg@xxxxxxxx> wrote: >> On Wed, May 02, 2018 at 09:15:09AM -0500, Rohit Zambre wrote: >>> On Tue, May 1, 2018 at 3:46 PM, Jason Gunthorpe <jgg@xxxxxxxxxxxx> wrote: >>> > On Tue, May 01, 2018 at 08:13:07PM +0000, Rohit Zambre wrote: >>> Could you point me to this new CQ poller you are referring to here? >> >> ibv_create_cq_ex() > > > I also agree that passing the td is the correct approach here, > probably via the pd. > > Working on these patches you've noticed that the ex(tended) cq already > has an IBV_CREATE_CQ_ATTR_SINGLE_THREADED create attr. This removes > the CQ lock in mlx5 provider code. Yes, I did notice the extended CQ takes the SINGLE_THREADED attribute. > Using the ibv_td will allow even better locking scheme as it allows > remove the lock for the cq with all it's connected qp. handling mlx5 > CQE's requires reading the QP->WRE sturtcs under a lock. So, pass a thread domain as a part of the init_attr? -- To unsubscribe from this list: send the line "unsubscribe linux-rdma" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
