Boehm-gc, linux-threads and SMP.... crash

[Bryce McKinlay <bryce at albatross dot co dot nz>]

I'm having problems with deadlocks and crashes in heavily-threaded
applications on my SMP system.  The attached example (from kaffe's
regression tests) crashes maybe 70% of the time on my dual 450Mhz ia32
box, usually with an "Aborted" but sometimes by deadlocking.

The same test runs 100% reliably if
  - I boot the machine with a UP kernel, or
  - I configure with --disable-java-gc

System is:

Redhat 6.1
2.2.14pre kernel
glibc 2.1.2-11

Unfortunatly, if I run the test under gdb it allways seems to deadlock
and so I havn't been able to get a backtrace of the problem. However, I
configured boehm-gc for thread debugging, and the output leading up to a
deadlock usually looks something like this:

Sending restart signal to 0x1406
Sending restart signal to 0x1c08
Sending restart signal to 0x2009
World started
In GC_restart_handler for 0x1406
Continuing 0x1406
In GC_restart_handler for 0x2009
Continuing 0x2009
In GC_restart_handler for 0x1c08
Continuing 0x1c08
Sending suspend signal to 0x400
Suspending 0x400
SP = bffff7a8
TOS = c0000000

Is the collector SMP safe on linux?

regards

  [ bryce ]