> > Split the FPU save area from the task struct. This allows easy 
> > migration of FPU context, and it's generally cleaner. It also allows 
> > the following two optimizations:

the atomicity bug is still there:

[ 1230.409772] BUG: sleeping function called from invalid context at mm/slab.c:>
[ 1230.409772] in_atomic():0, irqs_disabled():1
[ 1230.409772] Pid: 1187, comm: awk Not tainted 2.6.25-rc4-sched-devel.git-x86->[ 1230.409772]
[ 1230.409772] Call Trace:
[ 1230.409772]  [<ffffffff8022255b>] ? do_page_fault+0x3eb/0xa80
[ 1230.409772]  [<ffffffff8022c68e>] __might_sleep+0xce/0xf0
[ 1230.409772]  [<ffffffff8028c299>] kmem_cache_alloc+0xf9/0x120
[ 1230.409772]  [<ffffffff80214338>] init_fpu+0x98/0x100
[ 1230.409772]  [<ffffffff8020ec10>] math_state_restore+0x20/0x80
[ 1230.409772]  [<ffffffff809b8cc9>] error_exit+0x0/0x60

config and bootlog attached.

	Ingo