Re: Bug in disk event polling

[Tejun Heo <tj@kernel.org>]

Hello, Alan.

On Sun, Feb 12, 2012 at 12:48:22PM -0500, Alan Stern wrote:
> > I think it would be better to use wq for most kthreads.  A lot of them
> > aren't strictly correct in the way they deal with
> > kthread_should_stop() and freezing.  kthread in general simply seems
> > way too difficult to use correctly.
> 
> Maybe so, but getting rid of it at this point would be a big change.  

Ooh, yeah, if we do it, it would be a gradual transition.

> Also, kthreads were originally considered more suitable for tasks that
> would need to run for a long time; is this no longer true?

wq no longer has problem w/ long running work items.  The only
limiting parameter is max_active, which is primarily there to protect
against berserk cases.

> > > kthread_run() can be adjusted on a case-by-case basis, by inserting
> > > calls to set_freezable() and try_to_freeze() at the appropriate places.  
> > > But what about async_schedule()?
> > 
> > Given the stuff async is used for, maybe just make all async execution
> > freezable?
> 
> That probably won't work.  What if a driver relies on async thread
> execution to carry out its I/O?
>
> As another example, sd_probe() calls async_schedule(sd_probe_async,...)
> to handle the long-running parts of probing a SCSI disk.  In turn,
> sd_remove() calls async_synchronize_full() to insure that probing is
> over before the device is unbound from sd.
> 
> What happens if a hot-unpluggable disk drive is unplugged while the
> system is asleep?  It's entirely possible that the bus subsystem's
> resume routine would see the device was gone and would try to 
> unregister it.  Then sd_remove would wait for the async thread 
> to finish, which would never happen because the thread would be frozen 
> and wouldn't be thawed until all the resume routines had finished.
> 
> In this case, the proper solution is to have the SCSI prepare method 
> call async_synchronize_full().  Other cases will require other 
> solutions.

Hmmm.... I don't know then. :)

Thanks.

-- 
tejun