Re: [PATCH] usb: musb: fix context save over suspend.

[NeilBrown]

On Tue, 12 Feb 2013 13:03:36 -0800 Kevin Hilman <khilman@xxxxxxxxxx> wrote:

> NeilBrown <neilb@xxxxxxx> writes:
> 
> > On Mon, 21 Jan 2013 13:38:59 +0200 Igor Grinberg <grinberg@xxxxxxxxxxxxxx>
> > wrote:
> >
> >> -----BEGIN PGP SIGNED MESSAGE-----
> >> Hash: SHA1
> >> 
> >> Hi Neil,
> >> 
> >> On 01/21/13 11:28, NeilBrown wrote:
> >> > 
> >> > 
> >> > The standard suspend sequence involves runtime_resuming
> >> > devices before suspending the system.
> >> > So just saving context in runtime_suspend and restoring it
> >> > in runtime resume isn't enough.  We  must also save in "suspend"
> >> > and restore in "resume".
> >> > 
> >> > Without this patch, and OMAP3 system with off_mode enabled will find
> >> > the musb port non-functional after suspend/resume.  With the patch it
> >> > works perfectly.
> >> 
> >> Hmmm... Some time ago, this has been removed in
> >> 5d193ce8 (usb: musb: PM: fix context save/restore in suspend/resume path)
> >> 
> >> Am I missing something? Or things changed and now this patch is correct?
> >
> > Hi Igor,
> >  thanks for alerting me to that patch .... does anyone else get the feeling
> >  that power management to too complex to be understood by a mere human?
> 
> Yes.  ;)
> 
> >  That commit (5d193ce8) suggests that the musb-hdrc device is an
> >  'omap_device', or maybe has a PM domain set to something else.
> >  However it isn't/doesn't.  dev->pm_domain is NULL.  So no PM domain layer
> >  will ever call the musb_core musb_runtime_suspend/resume.
> >
> >  The parent device - musb-omap2430 - is an omap device, does have pm_domain
> >  set, and does have its omap2430_runtime_suspend/resume called for system
> >  suspend and so the context for that device is saved and restored.
> >  However that doesn't help the context for musb-hdrc.
> >
> >  Whether musb ever was an omap_device is beyond my archaeological skills to
> >  determine.
> >
> >  Kevin:  Was musb-hdrc ever a device with a pm_domain? or was it only ever
> >      the various possible parents that had domains?
> >      Are you able to defend your earlier patch in today's kernel?  It
> >      certainly causes my device not to work properly.
> 
> Sorry for the delay here, I'm back to a place where I can test this on
> real hardware.
> 
> My patch was fixing a real hang when musb was built-in (or loaded), in
> host-mode (mini-A cable attached) but no devices attached.  I just tried
> to reproduce this, and with your patch, the system hangs during suspend.
> 

Odd.  I plug in a mini-A cable, note that the 'mode' file holds
'a_idle' (sometimes just plugging in the cable isn't enough to trigger that,
but sometimes it is....) and suspend/resume work perfectly.  Though after
resume it is back to b_idle.

unplug/replug and it is back to  a_idle.  suspend/resume and back to b_idle.

> That being said, your description makes sense why this context
> save/restore is needed.  Perhaps your patch needs to add a check whether
> the device is runtime suspended (I gather this is what Ruslan's patch is
> doing.)

I'm not sure it is possible for the device to be runtime suspended at this
point.  Certainly my device never is, even if it was just before the suspend
sequence started. Something is waking it up...
(instruments the code).

Ahh - usb_suspend() calls choose_wakeup() which might call
pm_runtime_resume() if the could be a need to reprogram the wakeup setting.
As that is a 'might', the device might not be runtime-awake when 'suspend'
runs.

Can you see if this, on top of my previous patch, does any better on your
hardware?

Thanks,
NeilBrown


diff --git a/drivers/usb/musb/musb_core.c b/drivers/usb/musb/musb_core.c
index 956db0e..00deb94 100644
--- a/drivers/usb/musb/musb_core.c
+++ b/drivers/usb/musb/musb_core.c
@@ -2278,7 +2278,8 @@ static int musb_suspend(struct device *dev)
 	}
 
 	spin_unlock_irqrestore(&musb->lock, flags);
-	musb_save_context(musb);
+	if (!pm_runtime_status_suspended(dev))
+		musb_save_context(musb);
 	return 0;
 }
 
@@ -2289,7 +2290,8 @@ static int musb_resume_noirq(struct device *dev)
 	 * module got reset through the PSC (vs just being disabled).
 	 */
 	struct musb	*musb = dev_to_musb(dev);
-	musb_restore_context(musb);
+	if (!pm_runtime_status_suspended(dev))
+		musb_restore_context(musb);
 	return 0;
 }

Attachment: signature.asc
Description: PGP signature