* [PATCH] gpss: core: no waiters left behind on deregister
@ 2019-06-24 8:33 Oliver Neukum
2019-06-25 7:04 ` Johan Hovold
2019-06-25 10:47 ` Sergei Shtylyov
0 siblings, 2 replies; 6+ messages in thread
From: Oliver Neukum @ 2019-06-24 8:33 UTC (permalink / raw)
To: johan, linux-usb; +Cc: Oliver Neukum
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);
gdev->ops->close(gdev);
}
up_write(&gdev->rwsem);
--
2.16.4
^ permalink raw reply related [flat|nested] 6+ messages in thread
* Re: [PATCH] gpss: core: no waiters left behind on deregister
2019-06-24 8:33 [PATCH] gpss: core: no waiters left behind on deregister Oliver Neukum
@ 2019-06-25 7:04 ` Johan Hovold
2019-06-26 11:04 ` Oliver Neukum
2019-06-25 10:47 ` Sergei Shtylyov
1 sibling, 1 reply; 6+ messages in thread
From: Johan Hovold @ 2019-06-25 7:04 UTC (permalink / raw)
To: Oliver Neukum; +Cc: johan, linux-usb
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.
> gdev->ops->close(gdev);
> }
> up_write(&gdev->rwsem);
Thanks,
Johan
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] gpss: core: no waiters left behind on deregister
2019-06-24 8:33 [PATCH] gpss: core: no waiters left behind on deregister Oliver Neukum
2019-06-25 7:04 ` Johan Hovold
@ 2019-06-25 10:47 ` Sergei Shtylyov
1 sibling, 0 replies; 6+ messages in thread
From: Sergei Shtylyov @ 2019-06-25 10:47 UTC (permalink / raw)
To: Oliver Neukum, johan, linux-usb
Hello!
The subject has gpss ISO gnss. You hardly meant the General Purpose
System Simulation. :-)
On 24.06.2019 11:33, 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);
> gdev->ops->close(gdev);
> }
> up_write(&gdev->rwsem);
MBR, Sergei
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] gpss: core: no waiters left behind on deregister
2019-06-25 7:04 ` Johan Hovold
@ 2019-06-26 11:04 ` Oliver Neukum
2019-06-26 11:41 ` Johan Hovold
0 siblings, 1 reply; 6+ messages in thread
From: Oliver Neukum @ 2019-06-26 11:04 UTC (permalink / raw)
To: Johan Hovold; +Cc: linux-usb
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:
mutex_lock(&gdev->read_mutex);
while (kfifo_is_empty(&gdev->read_fifo)) {
mutex_unlock(&gdev->read_mutex);
if (gdev->disconnected)
return 0;
if (file->f_flags & O_NONBLOCK)
return -EAGAIN;
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. 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.
Regards
Oliver
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] gpss: core: no waiters left behind on deregister
2019-06-26 11:04 ` Oliver Neukum
@ 2019-06-26 11:41 ` Johan Hovold
2019-07-01 10:00 ` Oliver Neukum
0 siblings, 1 reply; 6+ messages in thread
From: Johan Hovold @ 2019-06-26 11:41 UTC (permalink / raw)
To: Oliver Neukum; +Cc: Johan Hovold, linux-usb
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.
Johan
^ permalink raw reply [flat|nested] 6+ messages in thread
* Re: [PATCH] gpss: core: no waiters left behind on deregister
2019-06-26 11:41 ` Johan Hovold
@ 2019-07-01 10:00 ` Oliver Neukum
0 siblings, 0 replies; 6+ messages in thread
From: Oliver Neukum @ 2019-07-01 10:00 UTC (permalink / raw)
To: Johan Hovold; +Cc: linux-usb
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
^ permalink raw reply [flat|nested] 6+ messages in thread
end of thread, other threads:[~2019-07-01 10:14 UTC | newest]
Thread overview: 6+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2019-06-24 8:33 [PATCH] gpss: core: no waiters left behind on deregister Oliver Neukum
2019-06-25 7:04 ` Johan Hovold
2019-06-26 11:04 ` Oliver Neukum
2019-06-26 11:41 ` Johan Hovold
2019-07-01 10:00 ` Oliver Neukum
2019-06-25 10:47 ` Sergei Shtylyov
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).