[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-Id: <20180811.120324.88920967254643809.davem@davemloft.net>
Date: Sat, 11 Aug 2018 12:03:24 -0700 (PDT)
From: David Miller <davem@...emloft.net>
To: caleb.raitto@...il.com
Cc: herbert@...dor.apana.org.au, mst@...hat.com,
arei.gonglei@...wei.com, jasowang@...hat.com,
netdev@...r.kernel.org, linux-crypto@...r.kernel.org,
caraitto@...gle.com
Subject: Re: [PATCH v2 net-next 0/2] virtio_net: Expand affinity to
arbitrary numbers of cpu and vq
From: Caleb Raitto <caleb.raitto@...il.com>
Date: Thu, 9 Aug 2018 18:18:27 -0700
> Virtio-net tries to pin each virtual queue rx and tx interrupt to a cpu if
> there are as many queues as cpus.
>
> Expand this heuristic to configure a reasonable affinity setting also
> when the number of cpus != the number of virtual queues.
>
> Patch 1 allows vqs to take an affinity mask with more than 1 cpu.
> Patch 2 generalizes the algorithm in virtnet_set_affinity beyond
> the case where #cpus == #vqs.
>
> v2 changes:
> Renamed "virtio_net: Make vp_set_vq_affinity() take a mask." to
> "virtio: Make vp_set_vq_affinity() take a mask."
>
> Tested:
...
Yes, this is so much better than the NAPI module parameter thing!
I'll apply this, and we can adjust/tweak the logic if necessary
on top of these changes.
Thanks!
Powered by blists - more mailing lists