Re: [PATCH] IB/ipoib: Enable pkey and device name decoupling

[Leon Romanovsky <leon-DgEjT+Ai2ygdnm+yROfE0A@public.gmane.org>]

[-- Attachment #1: Type: text/plain, Size: 1583 bytes --]

On Sat, Oct 21, 2017 at 01:42:15PM -0600, Jason Gunthorpe wrote:
> On Thu, Oct 19, 2017 at 09:28:31AM +0300, Alex Vesker wrote:
> > On 10/19/2017 9:14 AM, Leon Romanovsky wrote:
> > >On Thu, Oct 19, 2017 at 12:11:00AM -0600, Jason Gunthorpe wrote:
> > >>On Tue, Oct 17, 2017 at 01:21:21PM +0300, Leon Romanovsky wrote:
> > >>>>Yes, we should add a deprecation one shot printk to the kernel for the
> > >>>>sysfs interface to encourage people to use ip
> > >>>Please don't do that, it won't help for anyone, and especially for the people
> > >>>who didn't hear about "ip" in 2017.
> > >>This is the standard kernel way to encourage people to use the new
> > >>interfaces...
> > >>
> > >>>IPoIB netlink doesn't support enhanced IPoIB device because child
> > >>>device in netlink code was not allocated with rdma_alloc_netdev call
> > >>>as it was done for other flows.
> > >>I don't understand, isn't this just a (bad) bug to be fixed?
> > >Yes and we want to fix it BEFORE adding discouraging warnings.
>
> > We plan to block ipoib_netlink in the meanwhile until we fix it.
>
> That isn't a good idea, don't break APIs for performance reasons..

It is not "performance", but simple attempt to prevent kernel panic,
while calling that function, because I'm unsure if we success to provide
proper fix till merge window.

>
> > The fix is a little tricky since the netdev is allocated in
> > rtnl_newlink and it should be allocated with the new logic
> > introduced in enhanced IPoIB, setting rn_ops, allocating an
> > rdma_netdev if possible.
>
> Should get on it then .. :)
>
> Jason

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]