Firefox 3.6.X and Thunderbird 3.0.X crashing with Radeon graphics on FBSD 8.0-STABLE SMP/sm64 box

[O. Hartmann]

Since the introduction of Thunderbird 3.0 and Firefox 3.6 I see 
spontanous crashes/coredumps of both thunderbird and firefox. Interingly 
Firefox 3.5.X works well on he same platform.

The platform is a FreeBSD 8.0-STABLE/amd64 (r205536: Tue Mar 23 22:19:04 
CET 2010), SMP box with 8GB of RAM, QuadCore Intel Q6600 on a P35-based 
motherboard.

Thunderbird 3 crashes rarely compared to Firefox 3. The longer the 
application thunderbird runs, the higher the likelyhood the app crashes 
and vanishes. Sometimes this happens immediately after starting 
thunderbird, sometimes it takes its few minutes or half an hour.

Firefox 3 is sensitive to its pull-down menus or requester showing up in 
some situations. I can provoke a crash by clicking onto a pull-down-menu 
in firefox 3, it immediately dumps a core.

Well, I wouldn't write s cross-posting if I would be sure this behaviour 
is due to some oddities in my installation, but since this odd behaviour 
occured both on Firefox 3.6 and Thunderbird 3 in several situations and 
with several tries to get a workaround, I feel desperately lost.

What I did so far:

- Recompiling EVERY port on my box (four times in a row to make sure 
everything is right, its a pain with nearly 950 ports).
- Deinstalling both Firefox 3 and Thunderbird 3 and installing the 
binary packages from FreeBSD.ORG


I have a private UP box, running the same OS FreeBSD 8.0-STABLE/amd64 on 
a single core Athlon 3500+ with only 2GB of RAM. There I use both 
Thunderbird 3.0.3 and Firefox 3.6.2 without any problem.

I suspect the X11 server or some part of the accelerator stuff 
triggering the crashes. On the both machines,

WITHOUT_NOUVEAU = YES

is defined. On the UP box at home, I utilise a HD4830 graphics 
accelerator with DR enabled. The lab's box have had
both HD4670 and now HD4770 accelerators, both do not work properly with 
the state-of-the-art drivers supported by the official pots collection. 
HD4670 never got to work since the new RadeonHD driver 1.3 was 
introduced (prior to that it worked), the new HD4770 works with 
explicitely disabling DRI and crashes whenever I leave a session (quit 
windowmaker). Well, this is another story, but I suspect the Radeon 
driver infrastructure causing the problems - but I'm not sure.

I use perl-threaded 5.10, but I guess this is not a problem since the 
problems occur whether I have threaded perl or not.

As I said, I feel like a dead man in the water ...

Regards,
Oliver