Firewire lockup on 2.6.22.3 in SMP but not UP, old drivers

Firewire lockup on 2.6.22.3 in SMP but not UP, old drivers

[Andrew Burgess]

I can pull video via firewire from a set top box using
test_mpeg2 for hours (97GB is my current record) if using a
UP kernel. The machine locks hard after about 500MB using an
SMP kernel. No messages, no altsysreq keys.  It seems to
lock faster if I make the machine busy (disk and cpu). One
time it locked up running firewire_tester and once when I
switched cables between cable tv set top boxes (I think
while test_mpeg2 was running)

kernel 2.6.22.3 x86_64. I enabled all the mutex debug stuff
that I could find and turned on the NMI softlock detection
stuff. Nada. I tried the driver debug stuff but that seems
more for development (tracing).

I did not try a serial console to catch any messages that
cannot make it through syslog.

The hardware is ASUS A8V Deluxe:

  00:07.0 FireWire (IEEE 1394): VIA Technologies, Inc. IEEE 1394 Host Controller (rev 80)

Google shows this to be common although there are two other
chip versions out there.

The PC is about two years old, runs alot of stuff (server,
compiles, etc) and runs fine SMP until I touch firewire.

I have ordered a PCI firewire card to eliminate that part of
the hardware. Its also a VIA chipset.

If that doesn't help, does that make it seem like a driver
issue? (The UP vs SMP seems an odd symptom for hardware) If
so, what's the best way to proceed? Debug old drivers or
switch to new ones?  I have decades of C experience but know
nothing of 1394.

I want to use firewire with mythtv. Libs used by myth:

  athlon:aab > ldd /usr/bin/myth*|grep 1394|sort|uniq
  libavc1394.so.0 => /usr/lib64/libavc1394.so.0 (0x0000003141200000)
  libraw1394.so.8 => /usr/lib64/libraw1394.so.8 (0x000000313f600000)
  librom1394.so.0 => /usr/lib64/librom1394.so.0 (0x0000003140600000)

How close are these libraries to working with a set top box
with the new drivers?

Any quick hacks come to mind? Wrap the existing drivers in
the big kernel lock just to see if that gets things to work?
(that probably makes it obvious that I'm a kernel newbie :-)

I'll cc lkml for any general debug hints.

Thanks for any help
Andrew

Re: Firewire lockup on 2.6.22.3 in SMP but not UP, old drivers

[Stefan Richter]

Andrew Burgess wrote:
> what's the best way to proceed? Debug old drivers or
> switch to new ones?

As I mentioned in the branch of the discussion which lost the Cc: LKML,
this kind of bug looks like requiring a good look at the interaction of
raw1394's and ohci1394's locking (and maybe ieee1394's, but this seems
less likely).

It's hard work, and I don't have much experience with the code in
question nor do I have much time at the moment nor is there anybody else
who fixes ieee1394 subsystem bugs on a regular basis.  Nevertheless,
this kind of bugs should still be fixed in the "old" drivers.  They
haven't reached their end of life yet.
-- 
Stefan Richter
-=====-=-=== =--- =-===
http://arcgraph.de/sr/