Re: 2.6.23-rc2-mm1: hang, prop_norm_single involved

[Peter Zijlstra <peterz@infradead.org>]

On Thu, 2007-08-09 at 14:45 +0200, Peter Zijlstra wrote:
> On Thu, 2007-08-09 at 15:10 +0400, Alexey Dobriyan wrote:
> > LTP run reproducably hangs during rwtest01 test
> > rwtest -N rwtest01 -c -q -i 60s -f sync 10%25000:rs-sync=$$
> > Calltrace is always the same:
> > 

> [root@opteron ~]# PATH=/testcases/bin/:$PATH /testcases/bin/rwtest -N rwtest01 -c -q -i 60s -f sync 10%25000:rs-sync=$$
> rwtest01    1  PASS  :  Test passed
> [root@opteron ~]# PATH=/testcases/bin/:$PATH /testcases/bin/rwtest -N rwtest01 -c -q -i 60s -f sync 10%25000:rs-sync=$$
>    
> I can reproduce, but not always.
> 
> Also, since the task->dirties member is initialized in fork.c this
> should either _always_ happen or never. So this does point to some
> memory corruption, ->dirties is the very last member of the task struct.
> 
> /me goes try with slab_debug,...

to no avail, banging head against the wall what is happening here.

Andrew, could you:

# mm-dirty-balancing-for-tasks.patch

while I try to figure this one out?

That seems to make the unhappies I could reproduce here go away.