[08/11] mmc: core: Fixup processing of SDIO IRQs during system suspend/resume

Message ID 20190903142207.5825-9-ulf.hansson@linaro.org
State Superseded
Headers show
Series
  • mmc: core: PM fixes/improvements for SDIO IRQs
Related show

Commit Message

Ulf Hansson Sept. 3, 2019, 2:22 p.m.
System suspend/resume of SDIO cards, with SDIO IRQs enabled and when using
MMC_CAP2_SDIO_IRQ_NOTHREAD is unfortunate still suffering from a fragile
behaviour. Some problems have been taken care of so far, but more issues
remains.

For example, calling the ->ack_sdio_irq() callback to let host drivers
re-enable the SDIO IRQs is a bad idea, unless the IRQ have been consumed,
which may not be the case during system suspend/resume. This may lead to
that a host driver re-signals the same SDIO IRQ over and over again,
causing a storm of IRQs and gives a ping-pong effect towards the
sdio_irq_work().

Moreover, calling the ->enable_sdio_irq() callback at system resume to
re-enable already enabled SDIO IRQs for the host, causes the runtime PM
count for some host drivers to become in-balanced. This then leads to the
host to remain runtime resumed, no matter if it's needed or not.

To fix these problems, let's check if process_sdio_pending_irqs() actually
consumed the SDIO IRQ, before we continue to ack the IRQ by invoking the
->ack_sdio_irq() callback.

Additionally, there should be no need to re-enable SDIO IRQs as the host
driver already knows if they were enabled at system suspend, thus also
whether it needs to re-enable them at system resume. For this reason, drop
the call to ->enable_sdio_irq() during system resume.

In regards to these changes there is yet another issue, which is when there
is an SDIO IRQ being signaled by the host driver, but after the SDIO card
has been system suspended. Currently these IRQs are just thrown away, while
we should at least make sure to try to consume them when the SDIO card has
been system resumed. Fix this by calling sdio_signal_irq() after system
resumed the SDIO card.

Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>

---
 drivers/mmc/core/sdio.c     | 2 +-
 drivers/mmc/core/sdio_irq.c | 3 ++-
 2 files changed, 3 insertions(+), 2 deletions(-)

-- 
2.17.1

Comments

Matthias Kaehlcke Sept. 5, 2019, 6:43 p.m. | #1
On Tue, Sep 03, 2019 at 04:22:04PM +0200, Ulf Hansson wrote:
> System suspend/resume of SDIO cards, with SDIO IRQs enabled and when using

> MMC_CAP2_SDIO_IRQ_NOTHREAD is unfortunate still suffering from a fragile

> behaviour. Some problems have been taken care of so far, but more issues

> remains.

> 

> For example, calling the ->ack_sdio_irq() callback to let host drivers

> re-enable the SDIO IRQs is a bad idea, unless the IRQ have been consumed,

> which may not be the case during system suspend/resume. This may lead to

> that a host driver re-signals the same SDIO IRQ over and over again,

> causing a storm of IRQs and gives a ping-pong effect towards the

> sdio_irq_work().

> 

> Moreover, calling the ->enable_sdio_irq() callback at system resume to

> re-enable already enabled SDIO IRQs for the host, causes the runtime PM

> count for some host drivers to become in-balanced. This then leads to the

> host to remain runtime resumed, no matter if it's needed or not.

> 

> To fix these problems, let's check if process_sdio_pending_irqs() actually

> consumed the SDIO IRQ, before we continue to ack the IRQ by invoking the

> ->ack_sdio_irq() callback.

> 

> Additionally, there should be no need to re-enable SDIO IRQs as the host

> driver already knows if they were enabled at system suspend, thus also

> whether it needs to re-enable them at system resume. For this reason, drop

> the call to ->enable_sdio_irq() during system resume.

> 

> In regards to these changes there is yet another issue, which is when there

> is an SDIO IRQ being signaled by the host driver, but after the SDIO card

> has been system suspended. Currently these IRQs are just thrown away, while

> we should at least make sure to try to consume them when the SDIO card has

> been system resumed. Fix this by calling sdio_signal_irq() after system

> resumed the SDIO card.

> 

> Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>

> ---

>  drivers/mmc/core/sdio.c     | 2 +-

>  drivers/mmc/core/sdio_irq.c | 3 ++-

>  2 files changed, 3 insertions(+), 2 deletions(-)

> 

> diff --git a/drivers/mmc/core/sdio.c b/drivers/mmc/core/sdio.c

> index c557f1519b77..3114d496495a 100644

> --- a/drivers/mmc/core/sdio.c

> +++ b/drivers/mmc/core/sdio.c

> @@ -1015,7 +1015,7 @@ static int mmc_sdio_resume(struct mmc_host *host)

>  		if (!(host->caps2 & MMC_CAP2_SDIO_IRQ_NOTHREAD))

>  			wake_up_process(host->sdio_irq_thread);

>  		else if (host->caps & MMC_CAP_SDIO_IRQ)

> -			host->ops->enable_sdio_irq(host, 1);

> +			sdio_signal_irq(host);


You could possibly limit this to cards that remain powered during
suspend, but doing it always should do no harm.

>  	}

>  

>  out:

> diff --git a/drivers/mmc/core/sdio_irq.c b/drivers/mmc/core/sdio_irq.c

> index d7965b53a6d2..900871073bd7 100644

> --- a/drivers/mmc/core/sdio_irq.c

> +++ b/drivers/mmc/core/sdio_irq.c

> @@ -115,7 +115,8 @@ static void sdio_run_irqs(struct mmc_host *host)

>  	mmc_claim_host(host);

>  	if (host->sdio_irqs) {

>  		process_sdio_pending_irqs(host);

> -		host->ops->ack_sdio_irq(host);

> +		if (!host->sdio_irq_pending)

> +			host->ops->ack_sdio_irq(host);

>  	}

>  	mmc_release_host(host);

>  }


I'm by no means a SDIO expert, but as far as I can tell this looks
good. I verified that this patch fixes a problem with SDIO interrupts
that are ignored while suspending.

Reviewed-by: Matthias Kaehlcke <mka@chromium.org>
Ulf Hansson Sept. 6, 2019, 9:42 a.m. | #2
On Fri, 6 Sep 2019 at 01:48, Doug Anderson <dianders@chromium.org> wrote:
>

> Hi,

>

> On Tue, Sep 3, 2019 at 7:22 AM Ulf Hansson <ulf.hansson@linaro.org> wrote:

> >

> > System suspend/resume of SDIO cards, with SDIO IRQs enabled and when using

> > MMC_CAP2_SDIO_IRQ_NOTHREAD is unfortunate still suffering from a fragile

> > behaviour. Some problems have been taken care of so far, but more issues

> > remains.

> >

> > For example, calling the ->ack_sdio_irq() callback to let host drivers

> > re-enable the SDIO IRQs is a bad idea, unless the IRQ have been consumed,

> > which may not be the case during system suspend/resume. This may lead to

> > that a host driver re-signals the same SDIO IRQ over and over again,

> > causing a storm of IRQs and gives a ping-pong effect towards the

> > sdio_irq_work().

> >

> > Moreover, calling the ->enable_sdio_irq() callback at system resume to

> > re-enable already enabled SDIO IRQs for the host, causes the runtime PM

> > count for some host drivers to become in-balanced. This then leads to the

> > host to remain runtime resumed, no matter if it's needed or not.

> >

> > To fix these problems, let's check if process_sdio_pending_irqs() actually

> > consumed the SDIO IRQ, before we continue to ack the IRQ by invoking the

> > ->ack_sdio_irq() callback.

> >

> > Additionally, there should be no need to re-enable SDIO IRQs as the host

> > driver already knows if they were enabled at system suspend, thus also

> > whether it needs to re-enable them at system resume. For this reason, drop

> > the call to ->enable_sdio_irq() during system resume.

> >

> > In regards to these changes there is yet another issue, which is when there

> > is an SDIO IRQ being signaled by the host driver, but after the SDIO card

> > has been system suspended. Currently these IRQs are just thrown away, while

> > we should at least make sure to try to consume them when the SDIO card has

> > been system resumed. Fix this by calling sdio_signal_irq() after system

> > resumed the SDIO card.

> >

> > Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>

> > ---

> >  drivers/mmc/core/sdio.c     | 2 +-

> >  drivers/mmc/core/sdio_irq.c | 3 ++-

> >  2 files changed, 3 insertions(+), 2 deletions(-)

> >

> > diff --git a/drivers/mmc/core/sdio.c b/drivers/mmc/core/sdio.c

> > index c557f1519b77..3114d496495a 100644

> > --- a/drivers/mmc/core/sdio.c

> > +++ b/drivers/mmc/core/sdio.c

> > @@ -1015,7 +1015,7 @@ static int mmc_sdio_resume(struct mmc_host *host)

> >                 if (!(host->caps2 & MMC_CAP2_SDIO_IRQ_NOTHREAD))

> >                         wake_up_process(host->sdio_irq_thread);

> >                 else if (host->caps & MMC_CAP_SDIO_IRQ)

> > -                       host->ops->enable_sdio_irq(host, 1);

> > +                       sdio_signal_irq(host);

>

> Is this always safe?  On 1-function cards you won't poll CCCR_INTx so

> you'll always signal an interrupt at resume time, won't you?


Good point!

What we really want to do is to just schedule the work and not include
to set the sdio_irq_pending flag. Actually, the flag may have been
set, in case a host driver have called sdio_signal_irq() when the SDIO
card was suspended - but that is the intention.

Thanks for pointing out the issue, I will re-spin and fix it!

Kind regards
Uffe

Patch

diff --git a/drivers/mmc/core/sdio.c b/drivers/mmc/core/sdio.c
index c557f1519b77..3114d496495a 100644
--- a/drivers/mmc/core/sdio.c
+++ b/drivers/mmc/core/sdio.c
@@ -1015,7 +1015,7 @@  static int mmc_sdio_resume(struct mmc_host *host)
 		if (!(host->caps2 & MMC_CAP2_SDIO_IRQ_NOTHREAD))
 			wake_up_process(host->sdio_irq_thread);
 		else if (host->caps & MMC_CAP_SDIO_IRQ)
-			host->ops->enable_sdio_irq(host, 1);
+			sdio_signal_irq(host);
 	}
 
 out:
diff --git a/drivers/mmc/core/sdio_irq.c b/drivers/mmc/core/sdio_irq.c
index d7965b53a6d2..900871073bd7 100644
--- a/drivers/mmc/core/sdio_irq.c
+++ b/drivers/mmc/core/sdio_irq.c
@@ -115,7 +115,8 @@  static void sdio_run_irqs(struct mmc_host *host)
 	mmc_claim_host(host);
 	if (host->sdio_irqs) {
 		process_sdio_pending_irqs(host);
-		host->ops->ack_sdio_irq(host);
+		if (!host->sdio_irq_pending)
+			host->ops->ack_sdio_irq(host);
 	}
 	mmc_release_host(host);
 }