diff mbox series

[1/6] PM: runtime: enable wake irq after runtime_suspend hook called

Message ID 1617874514-12282-1-git-send-email-chunfeng.yun@mediatek.com (mailing list archive)
State New
Headers show
Series [1/6] PM: runtime: enable wake irq after runtime_suspend hook called | expand

Commit Message

Chunfeng Yun April 8, 2021, 9:35 a.m. UTC
When the dedicated wake irq is level trigger, enable it before
calling runtime_suspend, will trigger an interrupt.

e.g.
for a low level trigger type, it's low level at running time (0),
and becomes high level when enters suspend (runtime_suspend (1) is
called), a wakeup signal at (2) make it become low level, wake irq
will be triggered.

                ------------------
               |           ^     ^|
----------------           |     | --------------
 |<---(0)--->|<--(1)--|   (3)   (2)    (4)

if we enable the wake irq before calling runtime_suspend during (0),
an interrupt will arise, it causes resume immediately;
enable wake irq after calling runtime_suspend, e.g. at (3) or (4),
will works.

This patch seems no side effect on edge trigger wake irq.

Signed-off-by: Chunfeng Yun <chunfeng.yun@mediatek.com>
---
 drivers/base/power/runtime.c | 5 ++---
 1 file changed, 2 insertions(+), 3 deletions(-)

Comments

Rafael J. Wysocki April 8, 2021, 5:41 p.m. UTC | #1
On Thu, Apr 8, 2021 at 11:35 AM Chunfeng Yun <chunfeng.yun@mediatek.com> wrote:
>
> When the dedicated wake irq is level trigger, enable it before
> calling runtime_suspend, will trigger an interrupt.
>
> e.g.
> for a low level trigger type, it's low level at running time (0),
> and becomes high level when enters suspend (runtime_suspend (1) is
> called), a wakeup signal at (2) make it become low level, wake irq
> will be triggered.
>
>                 ------------------
>                |           ^     ^|
> ----------------           |     | --------------
>  |<---(0)--->|<--(1)--|   (3)   (2)    (4)
>
> if we enable the wake irq before calling runtime_suspend during (0),
> an interrupt will arise, it causes resume immediately;

But that's necessary to avoid missing a wakeup interrupt, isn't it?

> enable wake irq after calling runtime_suspend, e.g. at (3) or (4),
> will works.
>
> This patch seems no side effect on edge trigger wake irq.
>
> Signed-off-by: Chunfeng Yun <chunfeng.yun@mediatek.com>
> ---
>  drivers/base/power/runtime.c | 5 ++---
>  1 file changed, 2 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/base/power/runtime.c b/drivers/base/power/runtime.c
> index a46a7e30881b..796739a015a5 100644
> --- a/drivers/base/power/runtime.c
> +++ b/drivers/base/power/runtime.c
> @@ -619,12 +619,12 @@ static int rpm_suspend(struct device *dev, int rpmflags)
>         __update_runtime_status(dev, RPM_SUSPENDING);
>
>         callback = RPM_GET_CALLBACK(dev, runtime_suspend);
> -
> -       dev_pm_enable_wake_irq_check(dev, true);
>         retval = rpm_callback(callback, dev);
>         if (retval)
>                 goto fail;
>
> +       dev_pm_enable_wake_irq_check(dev, true);
> +
>   no_callback:
>         __update_runtime_status(dev, RPM_SUSPENDED);
>         pm_runtime_deactivate_timer(dev);
> @@ -659,7 +659,6 @@ static int rpm_suspend(struct device *dev, int rpmflags)
>         return retval;
>
>   fail:
> -       dev_pm_disable_wake_irq_check(dev);
>         __update_runtime_status(dev, RPM_ACTIVE);
>         dev->power.deferred_resume = false;
>         wake_up_all(&dev->power.wait_queue);
> --
> 2.18.0
>
Chunfeng Yun April 9, 2021, 1:53 a.m. UTC | #2
On Thu, 2021-04-08 at 19:41 +0200, Rafael J. Wysocki wrote:
> On Thu, Apr 8, 2021 at 11:35 AM Chunfeng Yun <chunfeng.yun@mediatek.com> wrote:
> >
> > When the dedicated wake irq is level trigger, enable it before
> > calling runtime_suspend, will trigger an interrupt.
> >
> > e.g.
> > for a low level trigger type, it's low level at running time (0),
> > and becomes high level when enters suspend (runtime_suspend (1) is
> > called), a wakeup signal at (2) make it become low level, wake irq
> > will be triggered.
> >
> >                 ------------------
> >                |           ^     ^|
> > ----------------           |     | --------------
> >  |<---(0)--->|<--(1)--|   (3)   (2)    (4)
> >
> > if we enable the wake irq before calling runtime_suspend during (0),
> > an interrupt will arise, it causes resume immediately;
> 
> But that's necessary to avoid missing a wakeup interrupt, isn't it?
That's also what I worry about.
It may happen if the trigger level only keeps a very short time, and the
interrupt controller can't process it timely, but I don't think it
follow the level trigger mechanism, the HW should latch it until the ISR
is called. right?

> 
> > enable wake irq after calling runtime_suspend, e.g. at (3) or (4),
> > will works.
> >
> > This patch seems no side effect on edge trigger wake irq.
> >
> > Signed-off-by: Chunfeng Yun <chunfeng.yun@mediatek.com>
> > ---
> >  drivers/base/power/runtime.c | 5 ++---
> >  1 file changed, 2 insertions(+), 3 deletions(-)
> >
> > diff --git a/drivers/base/power/runtime.c b/drivers/base/power/runtime.c
> > index a46a7e30881b..796739a015a5 100644
> > --- a/drivers/base/power/runtime.c
> > +++ b/drivers/base/power/runtime.c
> > @@ -619,12 +619,12 @@ static int rpm_suspend(struct device *dev, int rpmflags)
> >         __update_runtime_status(dev, RPM_SUSPENDING);
> >
> >         callback = RPM_GET_CALLBACK(dev, runtime_suspend);
> > -
> > -       dev_pm_enable_wake_irq_check(dev, true);
> >         retval = rpm_callback(callback, dev);
> >         if (retval)
> >                 goto fail;
> >
> > +       dev_pm_enable_wake_irq_check(dev, true);
> > +
> >   no_callback:
> >         __update_runtime_status(dev, RPM_SUSPENDED);
> >         pm_runtime_deactivate_timer(dev);
> > @@ -659,7 +659,6 @@ static int rpm_suspend(struct device *dev, int rpmflags)
> >         return retval;
> >
> >   fail:
> > -       dev_pm_disable_wake_irq_check(dev);
> >         __update_runtime_status(dev, RPM_ACTIVE);
> >         dev->power.deferred_resume = false;
> >         wake_up_all(&dev->power.wait_queue);
> > --
> > 2.18.0
> >
Ikjoon Jang April 9, 2021, 5:32 a.m. UTC | #3
Hi Chunfeng,

On Thu, Apr 8, 2021 at 5:35 PM Chunfeng Yun <chunfeng.yun@mediatek.com> wrote:
>
> When the dedicated wake irq is level trigger, enable it before
> calling runtime_suspend, will trigger an interrupt.
>
> e.g.
> for a low level trigger type, it's low level at running time (0),
> and becomes high level when enters suspend (runtime_suspend (1) is
> called), a wakeup signal at (2) make it become low level, wake irq
> will be triggered.
>
>                 ------------------
>                |           ^     ^|
> ----------------           |     | --------------
>  |<---(0)--->|<--(1)--|   (3)   (2)    (4)
>

Can't we just use a falling edge type for this irq line?

> if we enable the wake irq before calling runtime_suspend during (0),
> an interrupt will arise, it causes resume immediately;
> enable wake irq after calling runtime_suspend, e.g. at (3) or (4),
> will works.
>
> This patch seems no side effect on edge trigger wake irq.
>
> Signed-off-by: Chunfeng Yun <chunfeng.yun@mediatek.com>
> ---
>  drivers/base/power/runtime.c | 5 ++---
>  1 file changed, 2 insertions(+), 3 deletions(-)
>
> diff --git a/drivers/base/power/runtime.c b/drivers/base/power/runtime.c
> index a46a7e30881b..796739a015a5 100644
> --- a/drivers/base/power/runtime.c
> +++ b/drivers/base/power/runtime.c
> @@ -619,12 +619,12 @@ static int rpm_suspend(struct device *dev, int rpmflags)
>         __update_runtime_status(dev, RPM_SUSPENDING);
>
>         callback = RPM_GET_CALLBACK(dev, runtime_suspend);
> -
> -       dev_pm_enable_wake_irq_check(dev, true);
>         retval = rpm_callback(callback, dev);
>         if (retval)
>                 goto fail;
>
> +       dev_pm_enable_wake_irq_check(dev, true);
> +
>   no_callback:
>         __update_runtime_status(dev, RPM_SUSPENDED);
>         pm_runtime_deactivate_timer(dev);
> @@ -659,7 +659,6 @@ static int rpm_suspend(struct device *dev, int rpmflags)
>         return retval;
>
>   fail:
> -       dev_pm_disable_wake_irq_check(dev);
>         __update_runtime_status(dev, RPM_ACTIVE);
>         dev->power.deferred_resume = false;
>         wake_up_all(&dev->power.wait_queue);
> --
> 2.18.0
>
Tony Lindgren April 9, 2021, 5:39 a.m. UTC | #4
* Chunfeng Yun <chunfeng.yun@mediatek.com> [210409 01:54]:
> On Thu, 2021-04-08 at 19:41 +0200, Rafael J. Wysocki wrote:
> > On Thu, Apr 8, 2021 at 11:35 AM Chunfeng Yun <chunfeng.yun@mediatek.com> wrote:
> > >
> > > When the dedicated wake irq is level trigger, enable it before
> > > calling runtime_suspend, will trigger an interrupt.
> > >
> > > e.g.
> > > for a low level trigger type, it's low level at running time (0),
> > > and becomes high level when enters suspend (runtime_suspend (1) is
> > > called), a wakeup signal at (2) make it become low level, wake irq
> > > will be triggered.
> > >
> > >                 ------------------
> > >                |           ^     ^|
> > > ----------------           |     | --------------
> > >  |<---(0)--->|<--(1)--|   (3)   (2)    (4)
> > >
> > > if we enable the wake irq before calling runtime_suspend during (0),
> > > an interrupt will arise, it causes resume immediately;
> > 
> > But that's necessary to avoid missing a wakeup interrupt, isn't it?
> That's also what I worry about.

Yeah sounds like this patch will lead into missed wakeirqs.

Regards,

Tony
Tony Lindgren April 9, 2021, 5:40 a.m. UTC | #5
* Ikjoon Jang <ikjn@chromium.org> [210409 05:33]:
> Hi Chunfeng,
> 
> On Thu, Apr 8, 2021 at 5:35 PM Chunfeng Yun <chunfeng.yun@mediatek.com> wrote:
> >
> > When the dedicated wake irq is level trigger, enable it before
> > calling runtime_suspend, will trigger an interrupt.
> >
> > e.g.
> > for a low level trigger type, it's low level at running time (0),
> > and becomes high level when enters suspend (runtime_suspend (1) is
> > called), a wakeup signal at (2) make it become low level, wake irq
> > will be triggered.
> >
> >                 ------------------
> >                |           ^     ^|
> > ----------------           |     | --------------
> >  |<---(0)--->|<--(1)--|   (3)   (2)    (4)
> >
> 
> Can't we just use a falling edge type for this irq line?

Sounds reasonable to me :)

Tony
Chunfeng Yun April 9, 2021, 8:36 a.m. UTC | #6
On Fri, 2021-04-09 at 08:39 +0300, Tony Lindgren wrote:
> * Chunfeng Yun <chunfeng.yun@mediatek.com> [210409 01:54]:
> > On Thu, 2021-04-08 at 19:41 +0200, Rafael J. Wysocki wrote:
> > > On Thu, Apr 8, 2021 at 11:35 AM Chunfeng Yun <chunfeng.yun@mediatek.com> wrote:
> > > >
> > > > When the dedicated wake irq is level trigger, enable it before
> > > > calling runtime_suspend, will trigger an interrupt.
> > > >
> > > > e.g.
> > > > for a low level trigger type, it's low level at running time (0),
> > > > and becomes high level when enters suspend (runtime_suspend (1) is
> > > > called), a wakeup signal at (2) make it become low level, wake irq
> > > > will be triggered.
> > > >
> > > >                 ------------------
> > > >                |           ^     ^|
> > > > ----------------           |     | --------------
> > > >  |<---(0)--->|<--(1)--|   (3)   (2)    (4)
> > > >
> > > > if we enable the wake irq before calling runtime_suspend during (0),
> > > > an interrupt will arise, it causes resume immediately;
> > > 
> > > But that's necessary to avoid missing a wakeup interrupt, isn't it?
> > That's also what I worry about.
> 
> Yeah sounds like this patch will lead into missed wakeirqs.
If miss level trigger wakeirqs, that means HW doesn't latch it? is it HW
limitation?
> 
> Regards,
> 
> Tony
Chunfeng Yun April 9, 2021, 8:43 a.m. UTC | #7
On Fri, 2021-04-09 at 13:32 +0800, Ikjoon Jang wrote:
> Hi Chunfeng,
> 
> On Thu, Apr 8, 2021 at 5:35 PM Chunfeng Yun <chunfeng.yun@mediatek.com> wrote:
> >
> > When the dedicated wake irq is level trigger, enable it before
> > calling runtime_suspend, will trigger an interrupt.
> >
> > e.g.
> > for a low level trigger type, it's low level at running time (0),
> > and becomes high level when enters suspend (runtime_suspend (1) is
> > called), a wakeup signal at (2) make it become low level, wake irq
> > will be triggered.
> >
> >                 ------------------
> >                |           ^     ^|
> > ----------------           |     | --------------
> >  |<---(0)--->|<--(1)--|   (3)   (2)    (4)
> >
> 
> Can't we just use a falling edge type for this irq line?
I'll try it, but the original code still doesn't process above mentioned
case.

> 
> > if we enable the wake irq before calling runtime_suspend during (0),
> > an interrupt will arise, it causes resume immediately;
> > enable wake irq after calling runtime_suspend, e.g. at (3) or (4),
> > will works.
> >
> > This patch seems no side effect on edge trigger wake irq.
> >
> > Signed-off-by: Chunfeng Yun <chunfeng.yun@mediatek.com>
> > ---
> >  drivers/base/power/runtime.c | 5 ++---
> >  1 file changed, 2 insertions(+), 3 deletions(-)
> >
> > diff --git a/drivers/base/power/runtime.c b/drivers/base/power/runtime.c
> > index a46a7e30881b..796739a015a5 100644
> > --- a/drivers/base/power/runtime.c
> > +++ b/drivers/base/power/runtime.c
> > @@ -619,12 +619,12 @@ static int rpm_suspend(struct device *dev, int rpmflags)
> >         __update_runtime_status(dev, RPM_SUSPENDING);
> >
> >         callback = RPM_GET_CALLBACK(dev, runtime_suspend);
> > -
> > -       dev_pm_enable_wake_irq_check(dev, true);
> >         retval = rpm_callback(callback, dev);
> >         if (retval)
> >                 goto fail;
> >
> > +       dev_pm_enable_wake_irq_check(dev, true);
> > +
> >   no_callback:
> >         __update_runtime_status(dev, RPM_SUSPENDED);
> >         pm_runtime_deactivate_timer(dev);
> > @@ -659,7 +659,6 @@ static int rpm_suspend(struct device *dev, int rpmflags)
> >         return retval;
> >
> >   fail:
> > -       dev_pm_disable_wake_irq_check(dev);
> >         __update_runtime_status(dev, RPM_ACTIVE);
> >         dev->power.deferred_resume = false;
> >         wake_up_all(&dev->power.wait_queue);
> > --
> > 2.18.0
> >
Rafael J. Wysocki April 9, 2021, 11:14 a.m. UTC | #8
On Fri, Apr 9, 2021 at 10:36 AM Chunfeng Yun <chunfeng.yun@mediatek.com> wrote:
>
> On Fri, 2021-04-09 at 08:39 +0300, Tony Lindgren wrote:
> > * Chunfeng Yun <chunfeng.yun@mediatek.com> [210409 01:54]:
> > > On Thu, 2021-04-08 at 19:41 +0200, Rafael J. Wysocki wrote:
> > > > On Thu, Apr 8, 2021 at 11:35 AM Chunfeng Yun <chunfeng.yun@mediatek.com> wrote:
> > > > >
> > > > > When the dedicated wake irq is level trigger, enable it before
> > > > > calling runtime_suspend, will trigger an interrupt.
> > > > >
> > > > > e.g.
> > > > > for a low level trigger type, it's low level at running time (0),
> > > > > and becomes high level when enters suspend (runtime_suspend (1) is
> > > > > called), a wakeup signal at (2) make it become low level, wake irq
> > > > > will be triggered.
> > > > >
> > > > >                 ------------------
> > > > >                |           ^     ^|
> > > > > ----------------           |     | --------------
> > > > >  |<---(0)--->|<--(1)--|   (3)   (2)    (4)
> > > > >
> > > > > if we enable the wake irq before calling runtime_suspend during (0),
> > > > > an interrupt will arise, it causes resume immediately;
> > > >
> > > > But that's necessary to avoid missing a wakeup interrupt, isn't it?
> > > That's also what I worry about.
> >
> > Yeah sounds like this patch will lead into missed wakeirqs.
> If miss level trigger wakeirqs, that means HW doesn't latch it? is it HW
> limitation?

If it's level-triggered, it won't be missed, but then it is just
pointless to suspend the device when wakeup is being signaled in the
first place.

I'm not sure if I understand the underlying problem correctly.  Is it
about addressing spurious wakeups?
Chunfeng Yun April 10, 2021, 1:44 a.m. UTC | #9
On Fri, 2021-04-09 at 13:14 +0200, Rafael J. Wysocki wrote:
> On Fri, Apr 9, 2021 at 10:36 AM Chunfeng Yun <chunfeng.yun@mediatek.com> wrote:
> >
> > On Fri, 2021-04-09 at 08:39 +0300, Tony Lindgren wrote:
> > > * Chunfeng Yun <chunfeng.yun@mediatek.com> [210409 01:54]:
> > > > On Thu, 2021-04-08 at 19:41 +0200, Rafael J. Wysocki wrote:
> > > > > On Thu, Apr 8, 2021 at 11:35 AM Chunfeng Yun <chunfeng.yun@mediatek.com> wrote:
> > > > > >
> > > > > > When the dedicated wake irq is level trigger, enable it before
> > > > > > calling runtime_suspend, will trigger an interrupt.
> > > > > >
> > > > > > e.g.
> > > > > > for a low level trigger type, it's low level at running time (0),
> > > > > > and becomes high level when enters suspend (runtime_suspend (1) is
> > > > > > called), a wakeup signal at (2) make it become low level, wake irq
> > > > > > will be triggered.
> > > > > >
> > > > > >                 ------------------
> > > > > >                |           ^     ^|
> > > > > > ----------------           |     | --------------
> > > > > >  |<---(0)--->|<--(1)--|   (3)   (2)    (4)
> > > > > >
> > > > > > if we enable the wake irq before calling runtime_suspend during (0),
> > > > > > an interrupt will arise, it causes resume immediately;
> > > > >
> > > > > But that's necessary to avoid missing a wakeup interrupt, isn't it?
> > > > That's also what I worry about.
> > >
> > > Yeah sounds like this patch will lead into missed wakeirqs.
> > If miss level trigger wakeirqs, that means HW doesn't latch it? is it HW
> > limitation?
> 
> If it's level-triggered, it won't be missed, but then it is just
> pointless to suspend the device when wakeup is being signaled in the
> first place.
Got it
> 
> I'm not sure if I understand the underlying problem correctly.  Is it
> about addressing spurious wakeups?
In fact, it's default value is the same as the wakeup signal, maybe the
above case, using level trigger, should be avoided, it is not clear and
causes confusion, as Ikjoon and Tony suggested, using falling edge type
is better.

Thanks a lot
diff mbox series

Patch

diff --git a/drivers/base/power/runtime.c b/drivers/base/power/runtime.c
index a46a7e30881b..796739a015a5 100644
--- a/drivers/base/power/runtime.c
+++ b/drivers/base/power/runtime.c
@@ -619,12 +619,12 @@  static int rpm_suspend(struct device *dev, int rpmflags)
 	__update_runtime_status(dev, RPM_SUSPENDING);
 
 	callback = RPM_GET_CALLBACK(dev, runtime_suspend);
-
-	dev_pm_enable_wake_irq_check(dev, true);
 	retval = rpm_callback(callback, dev);
 	if (retval)
 		goto fail;
 
+	dev_pm_enable_wake_irq_check(dev, true);
+
  no_callback:
 	__update_runtime_status(dev, RPM_SUSPENDED);
 	pm_runtime_deactivate_timer(dev);
@@ -659,7 +659,6 @@  static int rpm_suspend(struct device *dev, int rpmflags)
 	return retval;
 
  fail:
-	dev_pm_disable_wake_irq_check(dev);
 	__update_runtime_status(dev, RPM_ACTIVE);
 	dev->power.deferred_resume = false;
 	wake_up_all(&dev->power.wait_queue);