[PATCH] PM / SDIO: Use empty system suspend/resume callbacks at the bus level (was: Re: Recent ...)

["Rafael J. Wysocki" <rjw@sisk.pl>]

On Sunday, March 25, 2012, NeilBrown wrote:
> On Sun, 25 Mar 2012 23:25:37 +0200 "Rafael J. Wysocki" <rjw@sisk.pl> wrote:
> 
> > On Sunday, March 25, 2012, Rafael J. Wysocki wrote:
> > > Hi,
> > > 
> > > On Sunday, March 25, 2012, NeilBrown wrote:
> > > > 
> > > > Hi Rafael,
> > > > 
> > > >  Your recent patch:
> > > >    commit 35cd133c
> > > >    PM: Run the driver callback directly if the subsystem one is not there
> > > > 
> > > >  breaks suspend for my libertas wifi and probably other SDIO devices.
> > > 
> > > Well, the patch is not recent.  The _commit_ is more than three months old
> > > and the patch has been around since the last November (at least).
> > > 
> > > >  SDIO (and possible MMC in general) has a protocol where the suspend
> > > >  method can return -ENOSYS and this means "There is no point in suspending,
> > > >  just turn me off".
> > > >
> > > >  The device itself "mmc1:0001" (I think) doesn't have any bus etc 'suspend'
> > > >  function so the new code call the device's suspend function which returns
> > > >  ENOSYS and the suspend fails.
> > > > 
> > > >  The previous code ignores the device as there is no bus suspend, and when it
> > > >  gets to suspend the ancestor - which for me is omap_hsmmc.1, it calls the
> > > >  device suspend function catches the ENOSYS, and turns it off.
> > > 
> > > Well, I can only call that a blatant abuse of the PM infrastructure.
> > > 
> > > >  I suspect just reverting it isn't the right long term solution, however I
> > > >  can confirm that it works for me for now.
> > > 
> > > It's not a solution at all, because there's code that depends on it already in
> > > the tree and the fact that it works for you doesn't mean it won't break other
> > > systems.  So no, it's not an option.
> > > 
> > > >  I'm happy to try any alternate fixes you would like to suggest (but I cannot
> > > >  promise how quickly I will get the testing done).
> > > > 
> > > >  (I'm testing with 3.3)
> > > 
> > > The only fix I can think of is to rework SDIO to stop abusing the PM callbacks.
> > > I'll have a look at that next week, although I can't promise anything any time
> > > soon, because I'm heading to San Francisco on Saturday.
> > 
> > Well, this is kind of a long shot, but I wonder if the patch below makes
> > any difference?
> 
> Hi Rafael,
>  thanks for looking into this so quickly.
> 
> I removed the revert and applied this patch instead and can confirm that
> suspend completes now (and resume works and the wifi works after resume).
> 
> Further, this patch fits with my (fairly shallow) understanding of the
> problem.
> 
> Thanks!

Thanks for the confirmation.

Below it goes again with a changelog and tags.

I don't really think that SDIO does the right thing here overall, but that's
all I can do to address the problem timely.

Thanks,
Rafael

---
From: Rafael J. Wysocki <rjw@sisk.pl>
Subject: PM / SDIO: Use empty system suspend/resume callbacks at the bus level

Neil Brown reports that commit 35cd133c

   PM: Run the driver callback directly if the subsystem one is not there

breaks suspend for his libertas wifi, because SDIO has a protocol
where the suspend method can return -ENOSYS and this means "There is
no point in suspending, just turn me off".  Moreover, the suspend
methods provided by SDIO drivers are not supposed to be called by
the PM core or bus-level suspend routines (which aren't presend for
SDIO).  Instead, when the SDIO core gets to suspend the device's
ancestor, it calls the device driver's suspend function, catches the
ENOSYS, and turns the device off.

The commit above breaks the SDIO core's assumption that the device
drivers' callbacks won't be executed if it doesn't provide any
bus-level callbacks.  If fact, however, this assumption has never
been really satisfied, because device class or device type suspend
might very well use the driver's callback even without that commit.

The simplest way to address this problem is to make the SDIO core
tell the PM core to ignore driver callbacks, for example by providing
no-operation suspend/resume callbacks at the bus level for it,
which is implemented by this change.

Reported-and-tested-by: Neil Brown <neilb@suse.de>
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
---
 drivers/mmc/core/sdio_bus.c |   12 +++++++++---
 1 file changed, 9 insertions(+), 3 deletions(-)

Index: linux/drivers/mmc/core/sdio_bus.c
===================================================================
--- linux.orig/drivers/mmc/core/sdio_bus.c
+++ linux/drivers/mmc/core/sdio_bus.c
@@ -192,9 +192,15 @@ static int sdio_bus_remove(struct device
 	return ret;
 }
 
-#ifdef CONFIG_PM_RUNTIME
+#ifdef CONFIG_PM
+
+static int pm_no_operation(struct device *dev)
+{
+	return 0;
+}
 
 static const struct dev_pm_ops sdio_bus_pm_ops = {
+	SET_SYSTEM_SLEEP_PM_OPS(pm_no_operation, pm_no_operation)
 	SET_RUNTIME_PM_OPS(
 		pm_generic_runtime_suspend,
 		pm_generic_runtime_resume,
@@ -204,11 +210,11 @@ static const struct dev_pm_ops sdio_bus_
 
 #define SDIO_PM_OPS_PTR	(&sdio_bus_pm_ops)
 
-#else /* !CONFIG_PM_RUNTIME */
+#else /* !CONFIG_PM */
 
 #define SDIO_PM_OPS_PTR	NULL
 
-#endif /* !CONFIG_PM_RUNTIME */
+#endif /* !CONFIG_PM */
 
 static struct bus_type sdio_bus_type = {
 	.name		= "sdio",