2006-11-01 15:38:27

by John Clark

[permalink] [raw]
Subject: sky2 driver causes kernel crash as of 2.6.18.1

I have been compiling kernels from 2.6.16.16 on to see if there is any
improvement
in the Sky2 driver. The most recent official kernel version, 2.6.18.1,
as of 10/31/06
seems to still have problems.

The crash debug splat indicates that the transmit routine was being
executed when
the final crash occured. But before the crash there were a series of
diagnostics from
the driver:

NETDEV WATCHDOG: eth4: transmit time out
sky2 eth4: tx timeout
sky2 hardware hung? flushing

messages.

Is there any better driver in a 'unstable' kernel that someone has
tested sufficiently?

Thanks
John Clark


2006-11-01 15:58:57

by Oleg Verych

[permalink] [raw]
Subject: Re: sky2 driver causes kernel crash as of 2.6.18.1

On 2006-11-01, John Clark wrote:
> I have been compiling kernels from 2.6.16.16 on to see if there is any
> improvement
> in the Sky2 driver. The most recent official kernel version, 2.6.18.1,
> as of 10/31/06
> seems to still have problems.
>
> The crash debug splat indicates that the transmit routine was being
> executed when
> the final crash occured. But before the crash there were a series of
> diagnostics from
> the driver:
>
> NETDEV WATCHDOG: eth4: transmit time out
> sky2 eth4: tx timeout
> sky2 hardware hung? flushing
>
> messages.
>
> Is there any better driver in a 'unstable' kernel that someone has
> tested sufficiently?

Well, try "unstable" 2.6.19-rc4. I'm not sure, but bug may be related to
7bd656d12119708b37414bf909ab2995473da818

Although, there are some sky2 updates in future 2.6.18.2, maybe they
are related too.
____

2006-11-01 16:11:54

by Daniel J Blueman

[permalink] [raw]
Subject: Re: sky2 driver causes kernel crash as of 2.6.18.1

John Clark wrote:
> I have been compiling kernels from 2.6.16.16 on to see if there is any
> improvement
> in the Sky2 driver. The most recent official kernel version, 2.6.18.1,
> as of 10/31/06
> seems to still have problems.
>
> The crash debug splat indicates that the transmit routine was being
> executed when
> the final crash occured. But before the crash there were a series of
> diagnostics from
> the driver:
>
> NETDEV WATCHDOG: eth4: transmit time out
> sky2 eth4: tx timeout
> sky2 hardware hung? flushing
>
> messages.
>
> Is there any better driver in a 'unstable' kernel that someone has
> tested sufficiently?
>
> Thanks
> John Clark

I've been using sky2 v1.9 (in eg linux-2.6.19-rc3) on two sky2
platforms, running 100s of gigabytes via NFSv4 and CIFS without
problem. In earlier versions, I'd hit a race after a few GBs at high
load; Stephen Hemminger has done a great job.
--
Daniel J Blueman