[Regression, bisected] Receive failure in et131x.

[Regression, bisected] Receive failure in et131x.

[Nick Bowler]

Continuing the theme from the last release cycle -- my PCI-express
et131x card (which works in 2.6.32) fails in the latest Linus git.  The
card works at first (DHCP and NFS mounts during bootup are OK) but
quickly fails once any reasonable traffic is put through it.  Transmit
seems to continue to work after a failure, but receive is mostly dead
forever -- the occasional frame appears to make it through.  The kernel
log contains many messages of the following form:

et131x 0000:02:00.0: NICRxPkts PSR Entry 1546 indicates length of 12920 and/or bad bi(614)
et131x 0000:02:00.0: NICRxPkts PSR Entry 1547 indicates length of 27745 and/or bad bi(92)
et131x 0000:02:00.0: NICRxPkts PSR Entry 1548 indicates length of 30064 and/or bad bi(116)
...

Bisection reveals the following.

99fd99f618daecae638550275cb132ab1ffe464c is the first bad commit
commit 99fd99f618daecae638550275cb132ab1ffe464c
Author: Alan Cox <alan@linux.intel.com>
Date:   Wed Nov 18 14:07:58 2009 +0000

    Staging: et131x: clean up the avail fields in the rx registers

    These have a wrap bit but again need little work to clean out. There are a
    couple of uglies left that want addressing in later clean up. Notably we should
    probably keep the local psr copy and wrap as two values.

    Signed-off-by: Alan Cox <alan@linux.intel.com>
    Signed-off-by: Greg Kroah-Hartman <gregkh@suse.de>

:040000 040000 8f0034c85fab183134fa19c7d58cee687991b6e7 a98b1db6fa82d8107fc5699fa2ae0851f38a14b0 M      drivers

-- 
Nick Bowler, Elliptic Technologies (http://www.elliptictech.com/)

Re: [Regression, bisected] Receive failure in et131x.

[Alan Cox]

On Tue, 29 Dec 2009 13:00:05 -0500
Nick Bowler <nbowler@elliptictech.com> wrote:

> Continuing the theme from the last release cycle -- my PCI-express
> et131x card (which works in 2.6.32) fails in the latest Linus git.  The
> card works at first (DHCP and NFS mounts during bootup are OK) but
> quickly fails once any reasonable traffic is put through it.  Transmit
> seems to continue to work after a failure, but receive is mostly dead
> forever -- the occasional frame appears to make it through.  The kernel
> log contains many messages of the following form:
> 
> et131x 0000:02:00.0: NICRxPkts PSR Entry 1546 indicates length of 12920 and/or bad bi(614)
> et131x 0000:02:00.0: NICRxPkts PSR Entry 1547 indicates length of 27745 and/or bad bi(92)
> et131x 0000:02:00.0: NICRxPkts PSR Entry 1548 indicates length of 30064 and/or bad bi(116)

Thanks - I'll beat that one over the head Jan 2nd and get it back into
order. What setup are you using and at what speed ?

Alan

Re: [Regression, bisected] Receive failure in et131x.

[Nick Bowler]

On 21:08 Tue 29 Dec     , Alan Cox wrote:
> Thanks - I'll beat that one over the head Jan 2nd and get it back into
> order.

Great, thanks a lot.

> What setup are you using and at what speed ?

I assume by "setup" you are referring to the network configuration.  I'd
copy the ifconfig output for you, but I'm not at the machine right now.
However, it's pretty plain (defaults everywhere -- 1500 byte mtu, one
ipv4 address).  I am using gigabit.

-- 
Nick Bowler, Elliptic Technologies (http://www.elliptictech.com/)