[PATCH] mmc: sdhci-pxav3: fix race between runtime pm and irq

[PATCH] mmc: sdhci-pxav3: fix race between runtime pm and irq

[Jisheng Zhang]

This patch is to fix a race condition that may cause an unhandled irq,
which results in big sdhci interrupt numbers and endless "mmc1: got irq
while runtime suspended" msgs before v3.15.

Consider following scenario:

      CPU0                            CPU1
                              sdhci_pxav3_runtime_suspend()
                               spin_lock_irqsave(&host->lock, flags);
 sdhci_irq()
  spining on the &host->lock
                               host->runtime_suspended = true;
                               spin_unlock_irqrestore(&host->lock, flags);
  get the &host->lock
  runtime_suspended is true now
  return IRQ_NONE;

Fix this race by using the core sdhci.c supplied sdhci_runtime_suspend_host()
in runtime suspend hook which will disable card interrupts. We also use the
sdhci_runtime_resume_host() in the runtime resume hook accordingly.

Signed-off-by: Jisheng Zhang <jszhang@marvell.com>
Cc: <stable@vger.kernel.org> # v3.9+
---
 drivers/mmc/host/sdhci-pxav3.c | 15 ++++-----------
 1 file changed, 4 insertions(+), 11 deletions(-)

diff --git a/drivers/mmc/host/sdhci-pxav3.c b/drivers/mmc/host/sdhci-pxav3.c
index 4de39fb..6975c51 100644
--- a/drivers/mmc/host/sdhci-pxav3.c
+++ b/drivers/mmc/host/sdhci-pxav3.c
@@ -460,17 +460,15 @@ static int sdhci_pxav3_runtime_suspend(struct device *dev)
 	struct sdhci_host *host = dev_get_drvdata(dev);
 	struct sdhci_pltfm_host *pltfm_host = sdhci_priv(host);
 	struct sdhci_pxa *pxa = pltfm_host->priv;
-	unsigned long flags;
+	int ret;
 
-	spin_lock_irqsave(&host->lock, flags);
-	host->runtime_suspended = true;
-	spin_unlock_irqrestore(&host->lock, flags);
+	ret = sdhci_runtime_suspend_host(host);
 
 	clk_disable_unprepare(pxa->clk_io);
 	if (!IS_ERR(pxa->clk_core))
 		clk_disable_unprepare(pxa->clk_core);
 
-	return 0;
+	return ret;
 }
 
 static int sdhci_pxav3_runtime_resume(struct device *dev)
@@ -478,17 +476,12 @@ static int sdhci_pxav3_runtime_resume(struct device *dev)
 	struct sdhci_host *host = dev_get_drvdata(dev);
 	struct sdhci_pltfm_host *pltfm_host = sdhci_priv(host);
 	struct sdhci_pxa *pxa = pltfm_host->priv;
-	unsigned long flags;
 
 	clk_prepare_enable(pxa->clk_io);
 	if (!IS_ERR(pxa->clk_core))
 		clk_prepare_enable(pxa->clk_core);
 
-	spin_lock_irqsave(&host->lock, flags);
-	host->runtime_suspended = false;
-	spin_unlock_irqrestore(&host->lock, flags);
-
-	return 0;
+	return sdhci_runtime_resume_host(host);
 }
 #endif
 
-- 
2.1.4

Re: [PATCH] mmc: sdhci-pxav3: fix race between runtime pm and irq

[Ulf Hansson]

On 21 January 2015 at 13:45, Jisheng Zhang <jszhang@marvell.com> wrote:
> This patch is to fix a race condition that may cause an unhandled irq,
> which results in big sdhci interrupt numbers and endless "mmc1: got irq
> while runtime suspended" msgs before v3.15.
>
> Consider following scenario:
>
>       CPU0                            CPU1
>                               sdhci_pxav3_runtime_suspend()
>                                spin_lock_irqsave(&host->lock, flags);
>  sdhci_irq()
>   spining on the &host->lock
>                                host->runtime_suspended = true;
>                                spin_unlock_irqrestore(&host->lock, flags);
>   get the &host->lock
>   runtime_suspended is true now
>   return IRQ_NONE;
>
> Fix this race by using the core sdhci.c supplied sdhci_runtime_suspend_host()
> in runtime suspend hook which will disable card interrupts. We also use the
> sdhci_runtime_resume_host() in the runtime resume hook accordingly.
>
> Signed-off-by: Jisheng Zhang <jszhang@marvell.com>
> Cc: <stable@vger.kernel.org> # v3.9+
> ---
>  drivers/mmc/host/sdhci-pxav3.c | 15 ++++-----------
>  1 file changed, 4 insertions(+), 11 deletions(-)
>
> diff --git a/drivers/mmc/host/sdhci-pxav3.c b/drivers/mmc/host/sdhci-pxav3.c
> index 4de39fb..6975c51 100644
> --- a/drivers/mmc/host/sdhci-pxav3.c
> +++ b/drivers/mmc/host/sdhci-pxav3.c
> @@ -460,17 +460,15 @@ static int sdhci_pxav3_runtime_suspend(struct device *dev)
>         struct sdhci_host *host = dev_get_drvdata(dev);
>         struct sdhci_pltfm_host *pltfm_host = sdhci_priv(host);
>         struct sdhci_pxa *pxa = pltfm_host->priv;
> -       unsigned long flags;
> +       int ret;
>
> -       spin_lock_irqsave(&host->lock, flags);
> -       host->runtime_suspended = true;
> -       spin_unlock_irqrestore(&host->lock, flags);
> +       ret = sdhci_runtime_suspend_host(host);

If you get an error at this point, you shouldn't continue but instead
just return an error. Maybe even return -EBUSY.

Now, since sdhci_runtime_suspend_host() always returns 0 (should it be
converted to void? ), perhaps you could ignore the error completely
and return 0, as before?

>
>         clk_disable_unprepare(pxa->clk_io);
>         if (!IS_ERR(pxa->clk_core))
>                 clk_disable_unprepare(pxa->clk_core);
>
> -       return 0;
> +       return ret;
>  }
>
>  static int sdhci_pxav3_runtime_resume(struct device *dev)
> @@ -478,17 +476,12 @@ static int sdhci_pxav3_runtime_resume(struct device *dev)
>         struct sdhci_host *host = dev_get_drvdata(dev);
>         struct sdhci_pltfm_host *pltfm_host = sdhci_priv(host);
>         struct sdhci_pxa *pxa = pltfm_host->priv;
> -       unsigned long flags;
>
>         clk_prepare_enable(pxa->clk_io);
>         if (!IS_ERR(pxa->clk_core))
>                 clk_prepare_enable(pxa->clk_core);
>
> -       spin_lock_irqsave(&host->lock, flags);
> -       host->runtime_suspended = false;
> -       spin_unlock_irqrestore(&host->lock, flags);
> -
> -       return 0;
> +       return sdhci_runtime_resume_host(host);
>  }
>  #endif
>
> --
> 2.1.4
>

Kind regards
Uffe

Re: [PATCH] mmc: sdhci-pxav3: fix race between runtime pm and irq

[Jisheng Zhang]

Dear Ulf,

On Fri, 23 Jan 2015 00:23:29 -0800
Ulf Hansson <ulf.hansson@linaro.org> wrote:

> On 21 January 2015 at 13:45, Jisheng Zhang <jszhang@marvell.com> wrote:
> > This patch is to fix a race condition that may cause an unhandled irq,
> > which results in big sdhci interrupt numbers and endless "mmc1: got irq
> > while runtime suspended" msgs before v3.15.
> >
> > Consider following scenario:
> >
> >       CPU0                            CPU1
> >                               sdhci_pxav3_runtime_suspend()
> >                                spin_lock_irqsave(&host->lock, flags);
> >  sdhci_irq()
> >   spining on the &host->lock
> >                                host->runtime_suspended = true;
> >                                spin_unlock_irqrestore(&host->lock, flags);
> >   get the &host->lock
> >   runtime_suspended is true now
> >   return IRQ_NONE;
> >
> > Fix this race by using the core sdhci.c supplied
> > sdhci_runtime_suspend_host() in runtime suspend hook which will disable
> > card interrupts. We also use the sdhci_runtime_resume_host() in the
> > runtime resume hook accordingly.
> >
> > Signed-off-by: Jisheng Zhang <jszhang@marvell.com>
> > Cc: <stable@vger.kernel.org> # v3.9+
> > ---
> >  drivers/mmc/host/sdhci-pxav3.c | 15 ++++-----------
> >  1 file changed, 4 insertions(+), 11 deletions(-)
> >
> > diff --git a/drivers/mmc/host/sdhci-pxav3.c
> > b/drivers/mmc/host/sdhci-pxav3.c index 4de39fb..6975c51 100644
> > --- a/drivers/mmc/host/sdhci-pxav3.c
> > +++ b/drivers/mmc/host/sdhci-pxav3.c
> > @@ -460,17 +460,15 @@ static int sdhci_pxav3_runtime_suspend(struct
> > device *dev) struct sdhci_host *host = dev_get_drvdata(dev);
> >         struct sdhci_pltfm_host *pltfm_host = sdhci_priv(host);
> >         struct sdhci_pxa *pxa = pltfm_host->priv;
> > -       unsigned long flags;
> > +       int ret;
> >
> > -       spin_lock_irqsave(&host->lock, flags);
> > -       host->runtime_suspended = true;
> > -       spin_unlock_irqrestore(&host->lock, flags);
> > +       ret = sdhci_runtime_suspend_host(host);
> 
> If you get an error at this point, you shouldn't continue but instead
> just return an error. Maybe even return -EBUSY.

Thanks for reviewing. What about just return "ret" if get an error here?
Even if sdhci_runtime_suspend_host() return any error in the future, we
are still safe. I'm cooking a patch to behave like this.

> 
> Now, since sdhci_runtime_suspend_host() always returns 0 (should it be
> converted to void? ), perhaps you could ignore the error completely
> and return 0, as before?

Thanks,
Jisheng