2010-12-08 15:24:02

by Takashi Iwai

[permalink] [raw]
Subject: [PATCH] mmc: Fix re-probing after hibernation

The commit 4c2ef25fe0b847d2ae818f74758ddb0be1c27d8e
mmc: fix all hangs related to mmc/sd card insert/removal during
suspend/resume
introduced a bug where the device probing no longer works after
hibernation. This was because the pm notifier expects
PM_POST_HIBERNATION call while the system sends PM_POST_RESTORE
instead, thus disable_rescan is kept as 1.

Cc: <[email protected]>
Signed-off-by: Takashi Iwai <[email protected]>
---
drivers/mmc/core/core.c | 2 +-
1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/drivers/mmc/core/core.c b/drivers/mmc/core/core.c
index 31ae07a..30094f6 100644
--- a/drivers/mmc/core/core.c
+++ b/drivers/mmc/core/core.c
@@ -1772,7 +1772,7 @@ int mmc_pm_notify(struct notifier_block *notify_block,
break;

case PM_POST_SUSPEND:
- case PM_POST_HIBERNATION:
+ case PM_POST_RESTORE:

spin_lock_irqsave(&host->lock, flags);
host->rescan_disable = 0;
--
1.7.3.1


2010-12-08 22:05:00

by Maxim Levitsky

[permalink] [raw]
Subject: Re: [PATCH] mmc: Fix re-probing after hibernation

On Wed, 2010-12-08 at 16:23 +0100, Takashi Iwai wrote:
> The commit 4c2ef25fe0b847d2ae818f74758ddb0be1c27d8e
> mmc: fix all hangs related to mmc/sd card insert/removal during
> suspend/resume
> introduced a bug where the device probing no longer works after
> hibernation. This was because the pm notifier expects
> PM_POST_HIBERNATION call while the system sends PM_POST_RESTORE
> instead, thus disable_rescan is kept as 1.
>
> Cc: <[email protected]>
> Signed-off-by: Takashi Iwai <[email protected]>
> ---
> drivers/mmc/core/core.c | 2 +-
> 1 files changed, 1 insertions(+), 1 deletions(-)
>
> diff --git a/drivers/mmc/core/core.c b/drivers/mmc/core/core.c
> index 31ae07a..30094f6 100644
> --- a/drivers/mmc/core/core.c
> +++ b/drivers/mmc/core/core.c
> @@ -1772,7 +1772,7 @@ int mmc_pm_notify(struct notifier_block *notify_block,
> break;
>
> case PM_POST_SUSPEND:
> - case PM_POST_HIBERNATION:
> + case PM_POST_RESTORE:
>
> spin_lock_irqsave(&host->lock, flags);
> host->rescan_disable = 0;

So that was the case....
Takasi, thank you very much.

I always use the 'unsafe resume' thing so I wasn't affected.

Best regards,
Maxim Levitsky

2010-12-09 01:53:08

by Chris Ball

[permalink] [raw]
Subject: Re: [PATCH] mmc: Fix re-probing after hibernation

Hi Takashi,

On Wed, Dec 08, 2010 at 04:23:59PM +0100, Takashi Iwai wrote:
> The commit 4c2ef25fe0b847d2ae818f74758ddb0be1c27d8e
> mmc: fix all hangs related to mmc/sd card insert/removal during
> suspend/resume
> introduced a bug where the device probing no longer works after
> hibernation. This was because the pm notifier expects
> PM_POST_HIBERNATION call while the system sends PM_POST_RESTORE
> instead, thus disable_rescan is kept as 1.
>
> Cc: <[email protected]>
> Signed-off-by: Takashi Iwai <[email protected]>
> ---
> drivers/mmc/core/core.c | 2 +-
> 1 files changed, 1 insertions(+), 1 deletions(-)
>
> diff --git a/drivers/mmc/core/core.c b/drivers/mmc/core/core.c
> index 31ae07a..30094f6 100644
> --- a/drivers/mmc/core/core.c
> +++ b/drivers/mmc/core/core.c
> @@ -1772,7 +1772,7 @@ int mmc_pm_notify(struct notifier_block *notify_block,
> break;
>
> case PM_POST_SUSPEND:
> - case PM_POST_HIBERNATION:
> + case PM_POST_RESTORE:
>
> spin_lock_irqsave(&host->lock, flags);
> host->rescan_disable = 0;

Thanks very much for tracking this down! The code suggests that
POST_RESTORE is only emitted on a *failed* hibernation restore --
from include/linux/notifier.h:

/* Hibernation and suspend events */
#define PM_POST_HIBERNATION 0x0002 /* Hibernation finished */
#define PM_POST_SUSPEND 0x0004 /* Suspend finished */
#define PM_POST_RESTORE 0x0006 /* Restore failed */

So, this all suggests that we want to add the POST_RESTORE case but
also want to keep the POST_HIBERNATION case. Do you agree? Was the
case you saw of failed probe after hibernation a failed restore from
hibernation image?

By the way, it looks like there are some other drivers which also
test for POST_HIBERNATION but not POST_RESTORE:

drivers/media/video/tlg2300/pd-main.c
drivers/s390/char/vmwatchdog.c
drivers/s390/cio/css.c
drivers/staging/brcm80211/brcmfmac/dhd_linux.c

Thanks,

- Chris.
--
Chris Ball <[email protected]> <http://printf.net/>
One Laptop Per Child

2010-12-09 20:38:24

by Rafael J. Wysocki

[permalink] [raw]
Subject: Re: [PATCH] mmc: Fix re-probing after hibernation

On Thursday, December 09, 2010, Takashi Iwai wrote:
> At Thu, 9 Dec 2010 01:53:03 +0000,
> Chris Ball wrote:
> >
> > Hi Takashi,
> >
> > On Wed, Dec 08, 2010 at 04:23:59PM +0100, Takashi Iwai wrote:
> > > The commit 4c2ef25fe0b847d2ae818f74758ddb0be1c27d8e
> > > mmc: fix all hangs related to mmc/sd card insert/removal during
> > > suspend/resume
> > > introduced a bug where the device probing no longer works after
> > > hibernation. This was because the pm notifier expects
> > > PM_POST_HIBERNATION call while the system sends PM_POST_RESTORE
> > > instead, thus disable_rescan is kept as 1.
> > >
> > > Cc: <[email protected]>
> > > Signed-off-by: Takashi Iwai <[email protected]>
> > > ---
> > > drivers/mmc/core/core.c | 2 +-
> > > 1 files changed, 1 insertions(+), 1 deletions(-)
> > >
> > > diff --git a/drivers/mmc/core/core.c b/drivers/mmc/core/core.c
> > > index 31ae07a..30094f6 100644
> > > --- a/drivers/mmc/core/core.c
> > > +++ b/drivers/mmc/core/core.c
> > > @@ -1772,7 +1772,7 @@ int mmc_pm_notify(struct notifier_block *notify_block,
> > > break;
> > >
> > > case PM_POST_SUSPEND:
> > > - case PM_POST_HIBERNATION:
> > > + case PM_POST_RESTORE:
> > >
> > > spin_lock_irqsave(&host->lock, flags);
> > > host->rescan_disable = 0;
> >
> > Thanks very much for tracking this down! The code suggests that
> > POST_RESTORE is only emitted on a *failed* hibernation restore --
> > from include/linux/notifier.h:
> >
> > /* Hibernation and suspend events */
> > #define PM_POST_HIBERNATION 0x0002 /* Hibernation finished */
> > #define PM_POST_SUSPEND 0x0004 /* Suspend finished */
> > #define PM_POST_RESTORE 0x0006 /* Restore failed */
> >
> > So, this all suggests that we want to add the POST_RESTORE case but
> > also want to keep the POST_HIBERNATION case. Do you agree? Was the
> > case you saw of failed probe after hibernation a failed restore from
> > hibernation image?
>
> Right, I took a look at Documentation/pm/notifiers.txt, and it's
> clearly written that PM_POST_RESTORE is only for the error case. But
> S4 resume succeeded actually on my machine, so it took time to spot
> out the culprit.
>
> Looking at the code in kernel/power/, this might be the case of
> user-space suspend. hibernate.c seems returning PM_POST_HIBERNATION
> while user.c returns PM_POST_RESTORE after image restoration, and I
> guess it's a bug of user.c.
>
> The untested patch is below (sorry I have no corresponding machine
> at my home.) Rafael, Pavel, could you check this?
>
>
> thanks,
>
> Takashi
>
> ===
> From 915fda609e02b02e4fc6b944e4432ea1d964adc5 Mon Sep 17 00:00:00 2001
> From: Takashi Iwai <[email protected]>
> Date: Thu, 9 Dec 2010 08:09:21 +0100
> Subject: [PATCH] PM: Fix PM_POST_* notification with user-space suspend
>
> The user-space hibernation sends a wrong notification after the image
> restoration because of thinko for the file flag check. RDONLY
> corresponds to hibernation and WRONLY to restoration, confusingly.

The patch is correct. I'm taking it to my tree.

Thanks,
Rafael


> Signed-off-by: Takashi Iwai <[email protected]>
> ---
> kernel/power/user.c | 2 +-
> 1 files changed, 1 insertions(+), 1 deletions(-)
>
> diff --git a/kernel/power/user.c b/kernel/power/user.c
> index e819e17..11d92b2 100644
> --- a/kernel/power/user.c
> +++ b/kernel/power/user.c
> @@ -137,7 +137,7 @@ static int snapshot_release(struct inode *inode, struct file *filp)
> free_all_swap_pages(data->swap);
> if (data->frozen)
> thaw_processes();
> - pm_notifier_call_chain(data->mode == O_WRONLY ?
> + pm_notifier_call_chain(data->mode == O_RDONLY ?
> PM_POST_HIBERNATION : PM_POST_RESTORE);
> atomic_inc(&snapshot_device_available);
>
>

2010-12-10 07:40:37

by Takashi Iwai

[permalink] [raw]
Subject: Re: [PATCH] mmc: Fix re-probing after hibernation

At Thu, 9 Dec 2010 21:37:24 +0100,
Rafael J. Wysocki wrote:
>
> On Thursday, December 09, 2010, Takashi Iwai wrote:
> > At Thu, 9 Dec 2010 01:53:03 +0000,
> > Chris Ball wrote:
> > >
> > > Hi Takashi,
> > >
> > > On Wed, Dec 08, 2010 at 04:23:59PM +0100, Takashi Iwai wrote:
> > > > The commit 4c2ef25fe0b847d2ae818f74758ddb0be1c27d8e
> > > > mmc: fix all hangs related to mmc/sd card insert/removal during
> > > > suspend/resume
> > > > introduced a bug where the device probing no longer works after
> > > > hibernation. This was because the pm notifier expects
> > > > PM_POST_HIBERNATION call while the system sends PM_POST_RESTORE
> > > > instead, thus disable_rescan is kept as 1.
> > > >
> > > > Cc: <[email protected]>
> > > > Signed-off-by: Takashi Iwai <[email protected]>
> > > > ---
> > > > drivers/mmc/core/core.c | 2 +-
> > > > 1 files changed, 1 insertions(+), 1 deletions(-)
> > > >
> > > > diff --git a/drivers/mmc/core/core.c b/drivers/mmc/core/core.c
> > > > index 31ae07a..30094f6 100644
> > > > --- a/drivers/mmc/core/core.c
> > > > +++ b/drivers/mmc/core/core.c
> > > > @@ -1772,7 +1772,7 @@ int mmc_pm_notify(struct notifier_block *notify_block,
> > > > break;
> > > >
> > > > case PM_POST_SUSPEND:
> > > > - case PM_POST_HIBERNATION:
> > > > + case PM_POST_RESTORE:
> > > >
> > > > spin_lock_irqsave(&host->lock, flags);
> > > > host->rescan_disable = 0;
> > >
> > > Thanks very much for tracking this down! The code suggests that
> > > POST_RESTORE is only emitted on a *failed* hibernation restore --
> > > from include/linux/notifier.h:
> > >
> > > /* Hibernation and suspend events */
> > > #define PM_POST_HIBERNATION 0x0002 /* Hibernation finished */
> > > #define PM_POST_SUSPEND 0x0004 /* Suspend finished */
> > > #define PM_POST_RESTORE 0x0006 /* Restore failed */
> > >
> > > So, this all suggests that we want to add the POST_RESTORE case but
> > > also want to keep the POST_HIBERNATION case. Do you agree? Was the
> > > case you saw of failed probe after hibernation a failed restore from
> > > hibernation image?
> >
> > Right, I took a look at Documentation/pm/notifiers.txt, and it's
> > clearly written that PM_POST_RESTORE is only for the error case. But
> > S4 resume succeeded actually on my machine, so it took time to spot
> > out the culprit.
> >
> > Looking at the code in kernel/power/, this might be the case of
> > user-space suspend. hibernate.c seems returning PM_POST_HIBERNATION
> > while user.c returns PM_POST_RESTORE after image restoration, and I
> > guess it's a bug of user.c.
> >
> > The untested patch is below (sorry I have no corresponding machine
> > at my home.) Rafael, Pavel, could you check this?
> >
> >
> > thanks,
> >
> > Takashi
> >
> > ===
> > From 915fda609e02b02e4fc6b944e4432ea1d964adc5 Mon Sep 17 00:00:00 2001
> > From: Takashi Iwai <[email protected]>
> > Date: Thu, 9 Dec 2010 08:09:21 +0100
> > Subject: [PATCH] PM: Fix PM_POST_* notification with user-space suspend
> >
> > The user-space hibernation sends a wrong notification after the image
> > restoration because of thinko for the file flag check. RDONLY
> > corresponds to hibernation and WRONLY to restoration, confusingly.
>
> The patch is correct. I'm taking it to my tree.

Thanks! So it's an over-3-years-old old bug ;)

Meanwhile, the fix below should be still added to mmc-core.
It's still needed for the error case.


Takashi

===
From: Takashi Iwai <[email protected]>
Subject: [PATCH] mmc: Fix re-probing with PM_POST_RESTORE notification

In the error-path where PM notifies PM_POST_RESTORE, the rescan-blockage
should be cleared as well. Otherwise it'll be never re-probed.

Also, as a bonus, this fixes a bug in S4 with user-mode suspend in the
current code, as it sends PM_POST_RESTORE instead of
PM_POST_HIBERNATION wrongly.

Cc: <[email protected]>
Signed-off-by: Takashi Iwai <[email protected]>
---
drivers/mmc/core/core.c | 1 +
1 files changed, 1 insertions(+), 0 deletions(-)

diff --git a/drivers/mmc/core/core.c b/drivers/mmc/core/core.c
index 31ae07a..57dcf8f 100644
--- a/drivers/mmc/core/core.c
+++ b/drivers/mmc/core/core.c
@@ -1773,6 +1773,7 @@ int mmc_pm_notify(struct notifier_block *notify_block,

case PM_POST_SUSPEND:
case PM_POST_HIBERNATION:
+ case PM_POST_RESTORE:

spin_lock_irqsave(&host->lock, flags);
host->rescan_disable = 0;
--
1.7.3.1

2010-12-11 21:56:09

by Chris Ball

[permalink] [raw]
Subject: Re: [PATCH] mmc: Fix re-probing after hibernation

Hi Takashi,

On Fri, Dec 10, 2010 at 08:40:31AM +0100, Takashi Iwai wrote:
> Subject: [PATCH] mmc: Fix re-probing with PM_POST_RESTORE notification
>
> In the error-path where PM notifies PM_POST_RESTORE, the rescan-blockage
> should be cleared as well. Otherwise it'll be never re-probed.
>
> Also, as a bonus, this fixes a bug in S4 with user-mode suspend in the
> current code, as it sends PM_POST_RESTORE instead of
> PM_POST_HIBERNATION wrongly.
>
> Cc: <[email protected]>
> Signed-off-by: Takashi Iwai <[email protected]>
> ---
> drivers/mmc/core/core.c | 1 +
> 1 files changed, 1 insertions(+), 0 deletions(-)
>
> diff --git a/drivers/mmc/core/core.c b/drivers/mmc/core/core.c
> index 31ae07a..57dcf8f 100644
> --- a/drivers/mmc/core/core.c
> +++ b/drivers/mmc/core/core.c
> @@ -1773,6 +1773,7 @@ int mmc_pm_notify(struct notifier_block *notify_block,
>
> case PM_POST_SUSPEND:
> case PM_POST_HIBERNATION:
> + case PM_POST_RESTORE:
>
> spin_lock_irqsave(&host->lock, flags);
> host->rescan_disable = 0;

Thanks, have pushed this to mmc-next and queued it for .37/stable.

--
Chris Ball <[email protected]> <http://printf.net/>
One Laptop Per Child