Re: [PATCH rdma-rc v1] IB/IPoIB: Fix legacy IPoIB due to wrong number of queues

From: Leon Romanovsky <leon@kernel.org>
To: Jason Gunthorpe <jgg@nvidia.com>
Cc: Dragos Tatulea <dtatulea@nvidia.com>,
	linux-rdma@vger.kernel.org, Tariq Toukan <tariqt@nvidia.com>
Subject: Re: [PATCH rdma-rc v1] IB/IPoIB: Fix legacy IPoIB due to wrong number of queues
Date: Tue, 24 Jan 2023 08:27:09 +0200	[thread overview]
Message-ID: <Y896PRioqqz6nVCZ@unreal> (raw)
In-Reply-To: <Y87SpbbdYE3A+y46@nvidia.com>

On Mon, Jan 23, 2023 at 02:32:05PM -0400, Jason Gunthorpe wrote:
> On Sun, Jan 22, 2023 at 02:44:35PM +0200, Leon Romanovsky wrote:
> 
> > > And the return of a really big number from ops->get_num_rx_queues is
> > > pretty ugly too, ideally that would be fixed to pass in some function
> > > arguments and obtain the ppriv so it can return the actual maximum
> > > number of queues and we don't waste a bunch of memory..
> > 
> > .get_num_rx_queues() is declared as void, so it can't have any complex
> > logic except returns some global define.
> 
> Well, yes, you'd have to add some arguments..

Jason, please be realistic.

We already were in this place, where we wanted to change netdev stack
for our IPoIB deadlock. As you probably remember, that didn't went well.

I see a little value to change bunch of netdev drivers just to save some
bytes in legacy IPoIB.

Thanks

>  
> Jason