Ian Campbell wrote:
> (dropping the e1000 guys, it seems unnecessary to keep spamming them
> with this issue when it's unlikely to be anything to do with them. I've
> left their list on for now so they know. I'd suggest dropping it from
> any replies.)
>
> On Sat, 2008-09-13 at 09:57 +0100, Ian Campbell wrote:
>
>> On Fri, 2008-09-12 at 18:15 -0500, Tom Tucker wrote:
>>
>>> Iain sadly yes. There's a thread stuck holding the BUSY bit or a thread
>>> failed to clear the bit properly
>>> (maybe an error path). Data continues to arrives, but the transport
>>> never gets put back on the queue
>>> because it's BUSY bit is set. In other words, this is a different error
>>> than the one we've been chasing.
>>>
>>> If I sent you a patch, could you rebuild the kernel and give it a whirl?
>>> Also, can you give me a
>>> kernel.org relative commit-id or tag for the kernel that you're using?
>>>
>> I sure could. I'm using the Debian kernel which is currently at 2.6.26.3
>> (pkg version 2.6.26-4) although I have an update to 2.6.26.4 (via pkg
>> 2.6.26-5) pending.
>>
>> If I'm going to build my own I'll start with current git
>> (a551b98d5f6fce5897d497abd8bfb262efb33d2a) and repro there before trying
>> your patch.
>>
>
> FYI I've repro'd with
> commit a551b98d5f6fce5897d497abd8bfb262efb33d2a
> Merge: d1c6d2e... 50bed2e...
> Author: Linus Torvalds <[email protected]>
> Date: Thu Sep 11 11:50:15 2008 -0700
>
> Merge branch 'for-linus' of git://git.kernel.dk/linux-2.6-block
>
> * 'for-linus' of git://git.kernel.dk/linux-2.6-block:
> sg: disable interrupts inside sg_copy_buffer
> which was the latest git a few days back.
>
> I'm going to start bisecting between v2.6.25 and v2.6.26. There's 173
> commits in fs/nfs* net/sunrpc in that interval so with a day per test I
> should have something next week...
>
>
Iain:
I'm assuming you'll do this in advance of any patch from me? I was
simply going to add printk to the various shutdown paths and see if we
could get some finer grained debug output since the generic transport
debug output was too verbose.
Let me know and I'll help if I can,
Tom
> Ian.
>
On Tue, 2008-09-16 at 06:38 -0500, Tom Tucker wrote:
> Iain:
I'm actually Ian. (I don't actually mind that much, the other way round
is probably much more annoying for people called Iain).
> I'm assuming you'll do this in advance of any patch from me?
I was going to grind through it unless/until you sent a patch.
> I was simply going to add printk to the various shutdown paths and see if we
> could get some finer grained debug output since the generic transport
> debug output was too verbose.
>
> Let me know and I'll help if I can,
You mean logging on the client or server side? I'm currently bisecting
client. I'm happy to add some extra prints there if you point me in the
direction of the code paths of interest.
Ian.
--
Ian Campbell
Current Noise: Anathema - Leave No Trace
I haven't lost my mind; I know exactly where I left it.