Re: [PATCH] usb: musb: fix pm_runtime mismatch

From: Felipe Balbi
Date: Thu Dec 15 2011 - 20:38:52 EST


Hi,

On Fri, Dec 16, 2011 at 03:13:13AM +0200, Felipe Contreras wrote:
> On Fri, Dec 16, 2011 at 1:50 AM, Felipe Balbi <balbi@xxxxxx> wrote:
> > On Fri, Dec 16, 2011 at 01:31:02AM +0200, Felipe Contreras wrote:
> >> On Fri, Dec 16, 2011 at 1:01 AM, Felipe Balbi <balbi@xxxxxx> wrote:
> >> > On Fri, Dec 16, 2011 at 12:42:14AM +0200, Felipe Contreras wrote:
> >> >> --- a/drivers/usb/musb/musb_core.c
> >> >> +++ b/drivers/usb/musb/musb_core.c
> >> >> @@ -2012,8 +2012,6 @@ musb_init_controller(struct device *dev, int nIrq, void __iomem *ctrl)
> >> >>       if (status < 0)
> >> >>               goto fail3;
> >> >>
> >> >> -     pm_runtime_put(musb->controller);
> >> >
> >> > To me the real fix would be add the missing pm_runtime_get_sync(). On
> >> > probe() we're actually accessing MUSB's address space which needs it's
> >> > clocks turned on. I guess it's only working now by chance, probably
> >> > because glue layer calls pm_runtime_get_sync() to access it's own
> >> > address space and that uses the same clocks.
> >>
> >> Are you sure it's "musb-hdrc", and not "musb-omap2430" the one
> >> accessing the relevant address-space? From the runtime_pm
> >> documentation it looks like only the probe function should deal with
> >> this.
> >>
> >> If "musb-hdrc" was truly accessing these registers, then I would get
> >> the same failure because the clocks are turned off, but I don't...
> >
> > see musb_core_init(); You don't see any problems when accessing those
> > addresses because musb_platform_init() will fall into
> > omap2430_musb_init() which calls pm_runtime_get_sync(), and the same
> > clock actually enables both address spaces (musb-omap2430 and
> > musb-hdrc).
>
> That's true, but how would I go test this theory? Call
> pm_runtime_put_sync() at the end of omap2430_musb_init()?

sounds like a plan... Not sure if it will work always though. If I
remember correctly, pm_runtime_put_sync() will only be synchronous to
the current device, but will go up the parent tree asynchronously. Which
means that dev->parent will be see a scheduled runtime_put

> Also, I think most pm_runtime_disable() calls shouldn't be there...
> That would tell PM to activate power for the device, which is not what
> we want. The core drivers code will take care of truly disabling the
> pm_runtime stuff _without_ activating the device.

true, sounds correct.

--
balbi

Attachment: signature.asc
Description: Digital signature