Re: [BUG]ethtool: NIC blink

Re: [BUG]ethtool: NIC blink

[Ben Hutchings]

[-- Attachment #1: Type: text/plain, Size: 1111 bytes --]

[Re-sending with cc to netdev]

On Tue, 2014-05-20 at 12:53 +0000, Shawn Warren wrote:
> Hello Ben and NetDev,
> 
> 
> I am curious how exactly the ethtool –p flag works when blinking
> interface lights. I ran into an issue on a DELL R820 with an Intel
> proc that when I blinked an interface and put ethtool in the
> background, the load on the server began steadily increasing. I ran
> the following command to reproduce the issue:
> 
> 
> # ethtool –p eth2 3600 &
> 
> 
> The load increased until the ethtool process was halted. Is this a
> known bug/issue?

It's not a bug in the ethtool utility.

Perhaps you have run into the problem explained in
<https://git.kernel.org/linus/68f512f21a64c9b264df6c61a9333e7890faf74b>.
The load will increase as processes wait (uninterruptibly) to obtain the
RTNL lock.  They will appear as being in D state.

This was fixed in Linux 3.0, but older versions can't practically be
fixed as it required an incompatible change to the driver API.

Ben.

-- 
Ben Hutchings
Life would be so much easier if we could look at the source code.

[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 828 bytes --]

Re: [BUG]ethtool: NIC blink

[Shawn Warren]

Hello Ben,

Thank you for sharing this link. This sounds about right for what I saw
but I was not checking the network connections to the server looking for
RTNL. Once I have a test server up and confirm this I will respond back.

Thanks,

Shawn Warren
Rackspace hosting



On 5/20/14 9:39 AM, "Ben Hutchings" <ben@decadent.org.uk> wrote:

>[Re-sending with cc to netdev]
>
>On Tue, 2014-05-20 at 12:53 +0000, Shawn Warren wrote:
>> Hello Ben and NetDev,
>> 
>> 
>> I am curious how exactly the ethtool ­p flag works when blinking
>> interface lights. I ran into an issue on a DELL R820 with an Intel
>> proc that when I blinked an interface and put ethtool in the
>> background, the load on the server began steadily increasing. I ran
>> the following command to reproduce the issue:
>> 
>> 
>> # ethtool ­p eth2 3600 &
>> 
>> 
>> The load increased until the ethtool process was halted. Is this a
>> known bug/issue?
>
>It's not a bug in the ethtool utility.
>
>Perhaps you have run into the problem explained in
><https://git.kernel.org/linus/68f512f21a64c9b264df6c61a9333e7890faf74b>.
>The load will increase as processes wait (uninterruptibly) to obtain the
>RTNL lock.  They will appear as being in D state.
>
>This was fixed in Linux 3.0, but older versions can't practically be
>fixed as it required an incompatible change to the driver API.
>
>Ben.
>
>-- 
>Ben Hutchings
>Life would be so much easier if we could look at the source code.