[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <20230224112207.055165bd@kernel.org>
Date: Fri, 24 Feb 2023 11:22:07 -0800
From: Jakub Kicinski <kuba@...nel.org>
To: "Nambiar, Amritha" <amritha.nambiar@...el.com>
Cc: Alexander H Duyck <alexander.duyck@...il.com>,
<netdev@...r.kernel.org>, <davem@...emloft.net>,
<edumazet@...gle.com>, <pabeni@...hat.com>,
"Saeed Mahameed" <saeed@...nel.org>,
"Samudrala, Sridhar" <sridhar.samudrala@...el.com>
Subject: Re: Kernel interface to configure queue-group parameters
On Fri, 24 Feb 2023 01:14:15 -0800 Nambiar, Amritha wrote:
> On 2/16/2023 9:32 AM, Jakub Kicinski wrote:
> > On Thu, 16 Feb 2023 02:35:35 -0800 Nambiar, Amritha wrote:
> >> Right. Something like /sys/class/net/<iface>/napis/napi<0-N>
> >> Maybe, initially there would be as many napis as queues due to 1:1
> >> association, but as the queues bitmap is tuned for the napi, only those
> >> napis that have queue[s] associated with it would be exposed.
> >
> > Forget about using sysfs, please. We've been talking about making
> > "queues first class citizen", mapping to pollers is part of that
> > problem space. And it's complex enough to be better suited for netlink.
>
> Okay. Can ethtool netlink be an option for this? For example,
>
> ethtool --show-napis
> Lists all the napi instances and associated queue[s] list for each napi
> for the specified network device.
>
> ethtool --set-napi
> Configure the attributes (say, queue[s] list) for each napi
>
> napi <napi_id>
> The napi instance to configure
>
> queues <q_id1, q_id2, ...>
> The queue[s] that are to be serviced by the napi instance.
The netdev-genl family is a better target.
But the work is doing the refactoring within the kernel to abstract
all this stuff away from the drivers, so that the kernel has
a stronger model of queues. If we just expose the calls to the drivers
directly we'll end up with a lot of code duplication and not-so-subtle
differences between vendors :(
Powered by blists - more mailing lists