From: Benjamin LaHaise <bcrl@kvack.org> Date: 2011-02-24 19:15:39
Hi folks,
I've recently noticed that one of the embedded systems I'm using (the
PCEngines ALIXes) is becoming CPU bound under heavy network traffic.
Upon investigation, it looks like the VT6105M isn't actually using the
hardware checksum offloading support of the hardware. Are there any
known reasons why this isn't enabled (hardware bugs?)? I'll test enabling
it in the driver, but I figured it would be worth asking if this path
has been explored already. Cheers,
-ben
From: David Miller <davem@davemloft.net> Date: 2011-02-24 20:23:18
From: Benjamin LaHaise <bcrl@kvack.org>
Date: Thu, 24 Feb 2011 13:58:05 -0500
I've recently noticed that one of the embedded systems I'm using (the
PCEngines ALIXes) is becoming CPU bound under heavy network traffic.
Upon investigation, it looks like the VT6105M isn't actually using the
hardware checksum offloading support of the hardware. Are there any
known reasons why this isn't enabled (hardware bugs?)? I'll test enabling
it in the driver, but I figured it would be worth asking if this path
has been explored already. Cheers,
As far as I can tell it was never attempted. So it should work.
If you do that, while you're here, you can make rhine_rx() take
a "napi" arg and make this driver use netif_gso_receive() too.
Don't forget to set NETIF_F_GRO or similar in netdev->flags during
probe, and also hookup the necessary ethtool hooks.
From: Roger Luethi <hidden> Date: 2011-02-25 08:00:11
On Thu, 24 Feb 2011 12:23:54 -0800, David Miller wrote:
From: Benjamin LaHaise <bcrl@kvack.org>
Date: Thu, 24 Feb 2011 13:58:05 -0500
quoted
Upon investigation, it looks like the VT6105M isn't actually using the
hardware checksum offloading support of the hardware. Are there any
known reasons why this isn't enabled (hardware bugs?)? I'll test enabling
As far as I can tell it was never attempted. So it should work.
If you do that, while you're here, you can make rhine_rx() take
a "napi" arg and make this driver use netif_gso_receive() too.
Don't forget to set NETIF_F_GRO or similar in netdev->flags during
probe, and also hookup the necessary ethtool hooks.
I have a patch to enable hw checksumming (the ethtool hooks are done, but I
somehow missed the NETIF_F_GRO bit). Care to give it a whirl?
Roger
From: Jan Ceuleers <hidden> Date: 2011-02-25 18:37:34
On 25/02/11 08:53, Roger Luethi wrote:
I have a patch to enable hw checksumming (the ethtool hooks are done, but I
somehow missed the NETIF_F_GRO bit). Care to give it a whirl?
Can you post that, preferably rebased to net-next? Even if Benjamin
doesn't get 'round to implementing all of the improvements Dave proposes
perhaps Dave will be clement enough to apply it as-is if it proves to be
a net positive?
Thanks, Jan