I also saw this with 2.5.36-bk3, patched with Robert's preempt-fix,
under similar circumstances (dbench 40 and parallel make).
Without doing a kernel compile at the same time, the text box was able
to run up to 128 dbench clients successfully with 2.5.37 and PREEMPT.
The 2.5.37 kernel was patched with this fix for PREEMPT:
and this fix for eepro100:
Both of those kernels were configured with SMP and PREEMPT.
The test box is dual PIII, 1GB, scsi, ext3 fs. The dbench runs and
kernel compiles were both run from the same disk.
I have been unable so far to repeat this failure with plain 2.5.37 and
no PREEMPT. Plain vanilla 2.5.37 no preempt has run up to 112 dbench
clients while doing make -j3 bzImage. At one point, the box appeared to
have frozen, with vmstat -n 1 3600 output stopped printing for a while,
but the system is still responsive to SYSRQ-H, P, T, just slow with
dbench 128 running.
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to firstname.lastname@example.org
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/