i2c: i2c-omap: fix interrupt flood during resume

Commit Message

Kalle Jokiniemi Oct. 5, 2012, 8:55 a.m. UTC

The resume_noirq enables interrupts one-by-one starting from
first one. Now if the wake up event for suspend came from i2c
device, the i2c bus irq gets enabled before the threaded
i2c device irq, causing a flood of i2c bus interrupts as the
threaded irq that should clear the event is not enabled yet.

Fixed the issue by adding suspend_late and resume_early
functions that keep i2c bus interrupts disabled until
resume_noirq has run completely.

Issue was detected doing a wake up from autosleep with
twl4030 power key on N9. Patch tested on N9.

Signed-off-by: Kalle Jokiniemi <[e-mail]>
---
 drivers/i2c/busses/i2c-omap.c |   37 +++++++++++++++++++++++++++++++++++++
 1 files changed, 37 insertions(+), 0 deletions(-)

Comments

Kevin Hilman Oct. 5, 2012, 10:07 p.m. UTC | #1

+Grygorii (who's been working on various I2C related suspend/resume
           issues also)

Hi Kalle,

Kalle Jokiniemi <[e-mail]> writes:

> The resume_noirq enables interrupts one-by-one starting from
> first one. Now if the wake up event for suspend came from i2c
> device, the i2c bus irq gets enabled before the threaded
> i2c device irq, causing a flood of i2c bus interrupts as the
> threaded irq that should clear the event is not enabled yet.

Ugh, another reason we need some sort of driver dependency tracking in the
driver model.

> Fixed the issue by adding suspend_late and resume_early
> functions that keep i2c bus interrupts disabled until
> resume_noirq has run completely.

Hmm, any reason we couldn't put the disable in the .suspend_noirq
callback?  The reason being is that other drivers might use I2C in their
suspend or late_suspend callbacks.  I'm not aware of any using I2C in
their late_suspend callbacks in mainline, but in theory I2C should work
all the way through the late callbacks.

I know it might look strange to have a disable_irq() in the noirq
callback, since IRQs are already disabled at that point, but a comment
stating that it's just there to balance the (re)enable in the
early_resume callback should be clear.

Some other minor comments below...

> Issue was detected doing a wake up from autosleep with
> twl4030 power key on N9. Patch tested on N9.
>
> Signed-off-by: Kalle Jokiniemi <[e-mail]>
> ---
>  drivers/i2c/busses/i2c-omap.c |   37 +++++++++++++++++++++++++++++++++++++
>  1 files changed, 37 insertions(+), 0 deletions(-)
>
> diff --git a/drivers/i2c/busses/i2c-omap.c b/drivers/i2c/busses/i2c-omap.c
> index 801df60..b77b0c2 100644
> --- a/drivers/i2c/busses/i2c-omap.c
> +++ b/drivers/i2c/busses/i2c-omap.c
> @@ -1158,6 +1158,35 @@ omap_i2c_remove(struct platform_device *pdev)
>  	return 0;
>  }
>  
> +#ifdef CONFIG_SUSPEND

This should be CONFIG_PM_SLEEP in order to cover hibernation also.

> +static int omap_i2c_suspend_late(struct device *dev)
> +{
> +
> +	struct platform_device *pdev = to_platform_device(dev);
> +	struct omap_i2c_dev *_dev = platform_get_drvdata(pdev);
> +
> +	/*
> +	 * The noirq_resume enables the interrupts one by one,
> +	 * this causes a interrupt flood if the SW irq actually reading
> +	 * event from i2c device is enabled only after i2c bus irq, as the
> +	 * irq that should clear the event is still disabled. We have to
> +	 * disable the bus irq until all other irqs have been enabled.
> +	 */

This comment probably belongs in the resume handler.

> +	disable_irq(_dev->irq);
> +	return 0;
> +}
> +
> +static int omap_i2c_resume_early(struct device *dev)
> +{
> +
> +	struct platform_device *pdev = to_platform_device(dev);
> +	struct omap_i2c_dev *_dev = platform_get_drvdata(pdev);
> +
> +	enable_irq(_dev->irq);
> +
> +	return 0;
> +}
> +#endif
>  #ifdef CONFIG_PM_RUNTIME
>  static int omap_i2c_runtime_suspend(struct device *dev)
>  {
> @@ -1178,10 +1207,18 @@ static int omap_i2c_runtime_resume(struct device *dev)
>  
>  	return 0;
>  }
> +#endif
>  
> +#if defined(CONFIG_SUSPEND) || defined(CONFIG_PM_RUNTIME)

#ifdef CONFIG_PM will cover this

>  static struct dev_pm_ops omap_i2c_pm_ops = {
> +#ifdef CONFIG_SUSPEND

again, CONFIG_PM_SLEEP here

> +	.suspend_late = omap_i2c_suspend_late,
> +	.resume_early = omap_i2c_resume_early,
> +#endif
> +#ifdef CONFIG_PM_RUNTIME
>  	.runtime_suspend = omap_i2c_runtime_suspend,
>  	.runtime_resume = omap_i2c_runtime_resume,
> +#endif
>  };
>  #define OMAP_I2C_PM_OPS (&omap_i2c_pm_ops)
>  #else

Kevin
--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Kalle Jokiniemi Oct. 8, 2012, 6:21 a.m. UTC | #2

pe, 2012-10-05 kello 15:07 -0700, Kevin Hilman kirjoitti:
> +Grygorii (who's been working on various I2C related suspend/resume
>            issues also)
> 
> Hi Kalle,
> 
> Kalle Jokiniemi <[e-mail]> writes:
> 
> > The resume_noirq enables interrupts one-by-one starting from
> > first one. Now if the wake up event for suspend came from i2c
> > device, the i2c bus irq gets enabled before the threaded
> > i2c device irq, causing a flood of i2c bus interrupts as the
> > threaded irq that should clear the event is not enabled yet.
> 
> Ugh, another reason we need some sort of driver dependency tracking in the
> driver model.
> 
> > Fixed the issue by adding suspend_late and resume_early
> > functions that keep i2c bus interrupts disabled until
> > resume_noirq has run completely.
> 
> Hmm, any reason we couldn't put the disable in the .suspend_noirq
> callback?

I just liked the symmetry of late/early.

>   The reason being is that other drivers might use I2C in their
> suspend or late_suspend callbacks.  I'm not aware of any using I2C in
> their late_suspend callbacks in mainline, but in theory I2C should work
> all the way through the late callbacks.

Makes sense, I'll change that.

> 
> I know it might look strange to have a disable_irq() in the noirq
> callback, since IRQs are already disabled at that point, but a comment
> stating that it's just there to balance the (re)enable in the
> early_resume callback should be clear.
> 
> Some other minor comments below...
> 
> > Issue was detected doing a wake up from autosleep with
> > twl4030 power key on N9. Patch tested on N9.
> >
> > Signed-off-by: Kalle Jokiniemi <[e-mail]>
> > ---
> >  drivers/i2c/busses/i2c-omap.c |   37 +++++++++++++++++++++++++++++++++++++
> >  1 files changed, 37 insertions(+), 0 deletions(-)
> >
> > diff --git a/drivers/i2c/busses/i2c-omap.c b/drivers/i2c/busses/i2c-omap.c
> > index 801df60..b77b0c2 100644
> > --- a/drivers/i2c/busses/i2c-omap.c
> > +++ b/drivers/i2c/busses/i2c-omap.c
> > @@ -1158,6 +1158,35 @@ omap_i2c_remove(struct platform_device *pdev)
> >  	return 0;
> >  }
> >  
> > +#ifdef CONFIG_SUSPEND
> 
> This should be CONFIG_PM_SLEEP in order to cover hibernation also.

Ok, I'll fix these, thanks.


> 
> > +static int omap_i2c_suspend_late(struct device *dev)
> > +{
> > +
> > +	struct platform_device *pdev = to_platform_device(dev);
> > +	struct omap_i2c_dev *_dev = platform_get_drvdata(pdev);
> > +
> > +	/*
> > +	 * The noirq_resume enables the interrupts one by one,
> > +	 * this causes a interrupt flood if the SW irq actually reading
> > +	 * event from i2c device is enabled only after i2c bus irq, as the
> > +	 * irq that should clear the event is still disabled. We have to
> > +	 * disable the bus irq until all other irqs have been enabled.
> > +	 */
> 
> This comment probably belongs in the resume handler.

Ok. I'll move it.

- Kalle


> 
> > +	disable_irq(_dev->irq);
> > +	return 0;
> > +}
> > +
> > +static int omap_i2c_resume_early(struct device *dev)
> > +{
> > +
> > +	struct platform_device *pdev = to_platform_device(dev);
> > +	struct omap_i2c_dev *_dev = platform_get_drvdata(pdev);
> > +
> > +	enable_irq(_dev->irq);
> > +
> > +	return 0;
> > +}
> > +#endif
> >  #ifdef CONFIG_PM_RUNTIME
> >  static int omap_i2c_runtime_suspend(struct device *dev)
> >  {
> > @@ -1178,10 +1207,18 @@ static int omap_i2c_runtime_resume(struct device *dev)
> >  
> >  	return 0;
> >  }
> > +#endif
> >  
> > +#if defined(CONFIG_SUSPEND) || defined(CONFIG_PM_RUNTIME)
> 
> #ifdef CONFIG_PM will cover this
> 
> >  static struct dev_pm_ops omap_i2c_pm_ops = {
> > +#ifdef CONFIG_SUSPEND
> 
> again, CONFIG_PM_SLEEP here
> 
> > +	.suspend_late = omap_i2c_suspend_late,
> > +	.resume_early = omap_i2c_resume_early,
> > +#endif
> > +#ifdef CONFIG_PM_RUNTIME
> >  	.runtime_suspend = omap_i2c_runtime_suspend,
> >  	.runtime_resume = omap_i2c_runtime_resume,
> > +#endif
> >  };
> >  #define OMAP_I2C_PM_OPS (&omap_i2c_pm_ops)
> >  #else
> 
> Kevin


--
To unsubscribe from this list: send the line "unsubscribe linux-omap" in
the body of a message to [e-mail]
More majordomo info at  http://vger.kernel.org/majordomo-info.html

Patch

diff --git a/drivers/i2c/busses/i2c-omap.c b/drivers/i2c/busses/i2c-omap.c
index 801df60..b77b0c2 100644
--- a/drivers/i2c/busses/i2c-omap.c
+++ b/drivers/i2c/busses/i2c-omap.c
@@ -1158,6 +1158,35 @@  omap_i2c_remove(struct platform_device *pdev)
 	return 0;
 }
 
+#ifdef CONFIG_SUSPEND
+static int omap_i2c_suspend_late(struct device *dev)
+{
+
+	struct platform_device *pdev = to_platform_device(dev);
+	struct omap_i2c_dev *_dev = platform_get_drvdata(pdev);
+
+	/*
+	 * The noirq_resume enables the interrupts one by one,
+	 * this causes a interrupt flood if the SW irq actually reading
+	 * event from i2c device is enabled only after i2c bus irq, as the
+	 * irq that should clear the event is still disabled. We have to
+	 * disable the bus irq until all other irqs have been enabled.
+	 */
+	disable_irq(_dev->irq);
+	return 0;
+}
+
+static int omap_i2c_resume_early(struct device *dev)
+{
+
+	struct platform_device *pdev = to_platform_device(dev);
+	struct omap_i2c_dev *_dev = platform_get_drvdata(pdev);
+
+	enable_irq(_dev->irq);
+
+	return 0;
+}
+#endif
 #ifdef CONFIG_PM_RUNTIME
 static int omap_i2c_runtime_suspend(struct device *dev)
 {
@@ -1178,10 +1207,18 @@  static int omap_i2c_runtime_resume(struct device *dev)
 
 	return 0;
 }
+#endif
 
+#if defined(CONFIG_SUSPEND) || defined(CONFIG_PM_RUNTIME)
 static struct dev_pm_ops omap_i2c_pm_ops = {
+#ifdef CONFIG_SUSPEND
+	.suspend_late = omap_i2c_suspend_late,
+	.resume_early = omap_i2c_resume_early,
+#endif
+#ifdef CONFIG_PM_RUNTIME
 	.runtime_suspend = omap_i2c_runtime_suspend,
 	.runtime_resume = omap_i2c_runtime_resume,
+#endif
 };
 #define OMAP_I2C_PM_OPS (&omap_i2c_pm_ops)
 #else