Re: [RFC] usb: host: u132-hcd: Remove deprecated create_singlethread_workqueue

[Michal Hocko]

On Mon 01-08-16 14:00:57, Alan Stern wrote:
> On Mon, 1 Aug 2016, Tejun Heo wrote:
> 
> > Hello,
> > 
> > On Mon, Aug 01, 2016 at 03:50:36PM +0200, Michal Hocko wrote:
> > > > All that would do is deferring the deadlock, right?  I'm not sure it
> > > > makes a lot of sense to protect an IO path against memory pressure
> > > > half-way.  It either can be depended during memory reclaim or it
> > > > can't. 
> > > 
> > > Completely agreed! If the rescuer thread can block on a memory
> > > allocation be it GFP_NOIO or others it is basically useless.
> > ...
> > > > Can MM people please chime in?  The question is about USB stoage
> > > > devices and memory reclaim.  USB doesn't guarantee forward progress
> > > > under memory pressure but tries a best-effort attempt with GFP_NOIO
> > > > and ATOMIC.  Is this the right thing to do?
> > > 
> > > If any real IO depends on those devices then this is not sufficient and
> > > they need some form of guarantee for progress (aka mempool).
> > 
> > Oliver, Alan, what do you think?  If USB itself can't operate without
> > allocating memory during transactions, whatever USB storage drivers
> > are doing isn't all that meaningful.  Can we proceed with the
> > workqueue patches?  Also, it could be that the only thing GFP_NOIO and
> > GFP_ATOMIC are doing is increasing the chance of IO failures under
> > memory pressure.  Maybe it'd be a good idea to reconsider the
> > approach?
> 
> I agree that USB's approach to memory allocation won't prevent failures 
> when there is severe pressure.

Or even worse, silent hangs for GFP_NOIO requests. If the allocation
size that is issued from that context is not large (basically < order-4)
then the allocation would be retried basically for ever without invoking
the OOM killer. Now, this is rather unlikely to become a real problem
unless there is a serious flood of these GFP_NOIO allocation requests.
But the main point remains. GFP_NOIO doesn't guanratee a forward
progress. Success of such an allocation depends on on a different
context with the full reclaim capabilities (including the OOM killer).

-- 
Michal Hocko
SUSE Labs