On Fri, Aug 18, 2017 at 03:16:12PM +0530, Selvin Xavier wrote: > Removing the "modify_port" hook is causing some issues with ib_cm module. > While adding a mad agent for the new device, ib_cm invokes > ib_modify_port which fails > with return value -ENOSYS. So the mad agent gets unregistered. This > break the connection > establishment. It is broken for bnxt_re also. > > The following patch helps. But it is like a workaround to solve the problem. > > diff --git a/drivers/infiniband/core/cm.c b/drivers/infiniband/core/cm.c > index d5ca101..59911dd 100644 > --- a/drivers/infiniband/core/cm.c > +++ b/drivers/infiniband/core/cm.c > @@ -4189,7 +4189,7 @@ static void cm_add_one(struct ib_device *ib_device) > goto error2; > > ret = ib_modify_port(ib_device, i, 0, &port_modify); > - if (ret) > + if (ret && ret != -ENOSYS) > goto error3; > > count++; > > Or should we have the modify_port hook with some basic checks? I think that your proposal is right thing to do. The driver should properly return the status of its callbacks (-ENOSYS) and it is responsibility of the caller to decide what to do in such case. The dummy function (return 0) hides the fact that modify_port do nothing and it can be wrong for some callers. Thanks
Attachment:
signature.asc
Description: PGP signature
