sleeping function called from invalid context

["Hareesh Nagarajan" <hareesh.nagarajan@xxxxxxxxx>]

I've made the following changes to kernel/exit.c:

fastcall NORET_TYPE void do_exit(long code)
{
	struct task_struct *tsk = current;
	int group_dead;

	if(tsk->pk)
		__do_abort(tsk->pk, 0);
	profile_task_exit(tsk);
		...
		...
	exit_mm(tsk);
		...
}

Now, __do_abort is a function that I've written that internally calls
might_sleep() and then calls kmem_cache_alloc. When __do_abort gets
called (tsk->pk != NULL; ofcourse), I see the following error in
dmesg:

[ 5266.639222] note: test2[2414] exited with preempt_count 2
[ 5266.639293] Debug: sleeping function called from invalid context at
include/linux/rwsem.h:43
[ 5266.639836] in_atomic():1, irqs_disabled():0
[ 5266.640037]  [<c011d431>] __might_sleep+0xa1/0xc0
[ 5266.640612]  [<c0124928>] exit_mm+0x38/0x2b0
[ 5266.640734]  [<c01257a7>] do_exit+0xe7/0x840

This pretty much means, we called a function that sleeps, while the
kernel was locked (CONFIG_PREMPT=y). The code path, for this reads as
follows:

exit_mm() -> calls down_read(...) -> which in turn calls __might_sleep
and there the following check is made:

if ((in_atomic() || irqs_disabled()) && ...) {
   ...
}

and in our case in_atomic is 1.

So my question is why is this detected at exit_mm()? Why is this not
detected in __do_abort() where I have a might_sleep() check? Earlier
when I had Profiling enabled, the same error was detected at
profile_task_exit(tsk); which also does a down_read(...).

Any hints will be appreciated,

Thanks,

./hareesh
PS: when I remove kmem_cache_alloc from the body of __do_abort() the
errors go away.

--
Kernelnewbies: Help each other learn about the Linux kernel.
Archive:       http://mail.nl.linux.org/kernelnewbies/
FAQ:           http://kernelnewbies.org/faq/