Re: generating firewire traffic

On 6/10/09, Stefan Richter <[email protected]> wrote:
> Carl Karsten wrote:
>> I can cause a kernel panic:
>
> It's only of peripheral interest to kino-dev then. :-)
> ([email protected] and [email protected] cover kernel
> driver topics.)

yeah yeah... dvgrab is kinda in this pool, and I figured you or
someone here would have some clue what I was talking about.  thanks
for the tolerance. :)

>
>> http://bugzilla.kernel.org/show_bug.cgi?id=13075
>> by plugging in a dv cam and running the following for a few minutes:
>>
>> dvgrab - | nc 192.168.1.216 2007 | nc 192.168.1.216 2008 > /dev/null
>>
>> on 192.168.1.216 I am running my home made echo server:
>> while /bin/true;do  nc -l -p 2007 -c cat; done &
>> while /bin/true;do  nc -l -p 2008 -c cat; done &
>>
>> To make it easier to debug, I would like to remove the need for the
>> cam and the dvgrab command and just use something that stimulates the
>> firewire hardware without anything plugged into the port.
>
> You need the camera and a capture program; only then the controller
> generates interrupts and utilizes its DMA units.  You cannot achieve
> this locally.  Instead of the camera, there could be a second PC or the
> same PC with a second card and a streamer program.

same pc... hadn't thought of that.  excellent.

>
> Well, you actually /can/ achieve something similar without camera or
> second PC or second card --- by /transmitting/ video (which works even
> if nothing is connected).  But different DMA units of the controller and
> different DMA programming of the kernel drivers would be involved then,
> so you may get same or different results --- more likely different results.

worth a shot.   One of the k-devs I was chatting with didn't have a
firewire cable, so if this works, very handy.

>
>> Or if I need something plugged in, then not having to worry about what
>> is plugged in (cam, drive, 2nd firewire port...)
>>
>> and in case it is somehow dependent on the timing created by a dv cam,
>> something like cat /dev/raw1394 so that "install dvgrab" isn't needed.
>
> # cat /dev/dv1394/$n
> But that would utilize different driver code paths, i.e. ohci1394 +
> dv1394 instead of ohci1394 + ieee1394 + raw1394.
>
> /dev/raw1394 isn't cat-friendly.  But you could try testdv on it or
> something along the lines.

I am willing to try anything.

>
>> Or convince me that the panic isn't dependent on firewire at all,
>> which I have been told, but have not been able to reproduce.
>>
>> Or buy me a pony and I'll stop worrying about this :)
>
> Since the panic happens in the networking code, it is either a pure
> networking bug, or it is an ieee1394 bug which corrupts the kernel in a
> way that can only be seen by the networking driver (perhaps only the
> forcedeth driver in particular) while network traffic is going on.
>
> Can you capture to local disk without panic?

yes.

I can also generate the same network traffic for 8+ hours without
panic.  read from FW and panic in under 10 min, sometimes uner 1 min.

>
> One other experiment which could possibly give some further information
> is to repeat everything after replacing ohci1394 + ieee1394 + raw1394 by
> firewire-ohci + firewire-core.  Requires libraw1394 v2.  (Latest is best.)

If this next P is too tangled up, it will make more sense when I can
get back to the panic machine and see what's going on.  but that may
be 24 hours or so, so in the mean time:

I installed:
linux-image-2.6.30-999-generic_2.6.30-999.1244629645_amd64.deb
hooked my panic maker script into rc.local, rebooted a few times to
make sure it was working and ran out the door kinda late to an
appointment (car traffic was light, so I arrived 15 min early.)
network traffic on the other hand seems to have whacked something such
that I can't see what's going on, but I am inclined to say it hasn't
crashed.  Ben Hutchings (dvswitch dev) was saying something about a fw
related fix that he is planning to port to lenny.  I was under the
impression it was a few weeks old, but maybe the patch is that old but
it just got applied.   Is there a chance that something relevant
changed from .29 to .30?

-- 
Carl K

------------------------------------------------------------------------------
Crystal Reports - New Free Runtime and 30 Day Trial
Check out the new simplified licensing option that enables unlimited
royalty-free distribution of the report engine for externally facing 
server and web deployment.
http://p.sf.net/sfu/businessobjects