Re: [RFC ABI V1 4/8] RDMA/core: Add support for custom types

[Jason Gunthorpe <jgunthorpe-ePGOBjL8dl3ta4EC/59zMFaTQe2KTcn/@public.gmane.org>]

On Wed, Jul 20, 2016 at 11:01:18AM +0300, Matan Barak wrote:

> But how would you know if rdma-cm is loaded and available? Don't we want to
> have a decent query interface rather than iterating through sysfs?
> I agree that it could be lower priority than the rest.

1) 'loaded available' I think you will agree we have a fairly bad user
   experience when it comes to autoloading modules. If we have one fd,
   and the core of that FD is part of the ib_core module that is
   pulled in on driver load then the a request for an API segment (eg uverbs,
   rdma_cm) can now actually trigger a module load, just like net does
   for PF's. So we actually get to a better, saner, place.

2) Yes, an in-band query interface makes much more sense than
   schlepping around sysfs. That could be done in the ioctl, or maybe
   via netlink, but as you say, lower priority.

> I meant moving to /dev/rdma_uapi, but when you fopen the device and get a
> fd, you could either have fd->NULL or fd->single_rdma_dev mapping.
> It's a little bit simpler, however, not that much from just doing 1:N
> mapping.

A major point of the unified interface is to sanely support rdma_cm
when working with multiple devices, so I'd expect 1:N is the only
reasonable option for that interface..

Jason
--
To unsubscribe from this list: send the line "unsubscribe linux-rdma" in
the body of a message to majordomo-u79uwXL29TY76Z2rM5mHXA@public.gmane.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html