Re: [PATCH] gpss: core: no waiters left behind on deregister

[Oliver Neukum <oneukum@suse.com>]

Am Mittwoch, den 26.06.2019, 13:41 +0200 schrieb Johan Hovold:
> On Wed, Jun 26, 2019 at 01:04:07PM +0200, Oliver Neukum wrote:
> > Am Dienstag, den 25.06.2019, 09:04 +0200 schrieb Johan Hovold:
> > > On Mon, Jun 24, 2019 at 10:33:23AM +0200, Oliver Neukum wrote:
> > > > If you deregister a device you need to wake up all waiters
> > > > as there will be no further wakeups.
> > > > 
> > > > Signed-off-by: Oliver Neukum <oneukum@suse.com>
> > > > ---
> > > >  drivers/gnss/core.c | 2 +-
> > > >  1 file changed, 1 insertion(+), 1 deletion(-)
> > > > 
> > > > diff --git a/drivers/gnss/core.c b/drivers/gnss/core.c
> > > > index e6f94501cb28..0d13bd2cefd5 100644
> > > > --- a/drivers/gnss/core.c
> > > > +++ b/drivers/gnss/core.c
> > > > @@ -303,7 +303,7 @@ void gnss_deregister_device(struct gnss_device *gdev)
> > > >  	down_write(&gdev->rwsem);
> > > >  	gdev->disconnected = true;
> > > >  	if (gdev->count) {
> > > > -		wake_up_interruptible(&gdev->read_queue);
> > > > +		wake_up_interruptible_all(&gdev->read_queue);
> > > 
> > > GNSS core doesn't have any exclusive waiters, so no need to use use the
> > > exclusive wake-up (all) interface.
> > 
> > Well, yes, but that is the problem. In gnss_read() you drop the lock:
> > That means that an arbitrary number of tasks can get here.
> > 
> >                 ret = wait_event_interruptible(gdev->read_queue,
> >                                 gdev->disconnected ||
> >                                 !kfifo_is_empty(&gdev->read_fifo));
> > 
> > Meaning that an arbitrary number can be sleeping here.
> 
> I understand wait you're getting at, but I think your mistaken regarding
> exclusive wait. Note that wait_event_interruptible() uses nonexclusive
> wait.
> 
> > Yet in gnss_deregister_device() you use a simple wake_up:
> > 
> > void gnss_deregister_device(struct gnss_device *gdev)
> > 
> > {
> > 
> >         down_write(&gdev->rwsem);
> >         gdev->disconnected = true;
> >         if (gdev->count) {
> >                 wake_up_interruptible(&gdev->read_queue);
> > 
> > 
> > wake_up_interruptible() will wake up one waiting task. But after that
> > the device is gone. There will be no further events. The other tasks
> > will sleep forever.
> 
> No, wake_up_interruptible() will wake up all nonexclusive waiters,
> which is all we care about here.

You are right and tracing this is hard.

	Regards & Sorry
		Oliver