Re: [Patch v4 1/2] freezer: check OOM kill while being frozen

[Tejun Heo <tj@kernel.org>]

Hello,

On Tue, Sep 09, 2014 at 06:06:25PM +0200, Michal Hocko wrote:
> > Even for userland tasks, try_to_freeze() can currently be anywhere in
> > the kernel.  The frequently used ones are few but there are some odd
> 
> I always thought that user space tasks can be in the fridge only on the
> way out from the kernel (get_signal_to_deliver). I have quickly greped

It *can* be anywhere.  We used to have some deep in nfs.  They got
removed later due to deadlocks but in theory they still can be
anywhere.

> the code and the only place I can see seems to be run_guest but that
> one bails out quickly when there are signals pending so it should be
> safe in this context.
> cifs is doing something suspicious (cifs_reconnect) but I didn't check
> more closely all the contexts it is called from.

Prolly something similar with what nfs was doing?

> > ones out, and, again, there's nothing enforcing any structure on
> > try_to_freeze() usage. 
> 
> Would it make sense to have try_to_freeze_user_task or similar and check
> for kernel thread in try_to_freeze and complain loudly if called from
> user task context? I mean does it even make sense to call try_to_freeze
> in the middle of kernel operation for a user task?

I personally think the whole try_to_freeze() was a mistake at least
for userland tasks.  We should have collected them in a (mostly)
single place like a jobctl stop.  I'm not sure whether distinguishing
the two interfaces would buy us much tho.

> > The other thing is that we may do quite a bit during exiting including
> > allocating memory. 
> 
> yes, we can allocate memory and even page fault on the exit path. But
> TIF_MEMDIE will make sure that the allocation will be successful if
> there is some memory left.

TIF_MEMDIE ensures forward progress so that the task can exit;
however, I'm not sure whether all the things that a task does during
exit are safe for PM freezes.

Thanks.

-- 
tejun