Re: [RFC][PATCH] Power domains for platform bus type

[Rafael J. Wysocki]

On Monday, January 31, 2011, Kevin Hilman wrote:
> Grant Likely <grant.likely@xxxxxxxxxxxx> writes:
> 
> > On Mon, Jan 31, 2011 at 11:16:51PM +0100, Rafael J. Wysocki wrote:
> >> On Monday, January 31, 2011, Alan Stern wrote:
> >> > On Mon, 31 Jan 2011, Rafael J. Wysocki wrote:
> >> > 
> >> > > On Monday, January 31, 2011, Alan Stern wrote:
> >> > > > On Sun, 30 Jan 2011, Rafael J. Wysocki wrote:
> >> > > > 
> >> > > > > > One thing about this implementation is slightly questionable.  The new
> >> > > > > > power_domain callbacks were added to the __weak platform PM routines,
> >> > > > > > which means they will have to be included in every overriding routine
> >> > > > > > provided by a platform imiplementation.
> >> > > > > > 
> >> > > > > > Would it be better to separate these things?  Have the power_domain 
> >> > > > > > callbacks occur in a static outer function which then calls a public 
> >> > > > > > __weak inner function that can be overridden?
> >> > > > > 
> >> > > > > That certainly is a good idea, but I wasn't sure how to do that.  It looks
> >> > > > > like I could keep the __weak functions as they are and modify
> >> > > > > platform_dev_pm_ops instead to point to a new set of function that in turn
> >> > > > > would call the __weak ones.  For example, the .suspend pointer in
> >> > > > > platform_dev_pm_ops might point to a new function, say
> >> > > > > platform_pm_full_suspend() that would call the power domain functions and
> >> > > > > the "original" platform_pm_suspend().  Is that what you mean?
> >> > > > 
> >> > > > Yes.  But what about the platform_bus_set_pm_ops() interface?  Should
> >> > > > platform-specific replacements for the pm_ops functions also include
> >> > > > the power_domain callbacks?
> >> > > 
> >> > > Well, whoever uses platform_bus_set_pm_ops(), he can simply prevent power
> >> > > domains from being used by not defining them in the first place. :-)
> >> > 
> >> > But what about the case where the user _does_ want to have power
> >> > domains?
> >> 
> >> Ah, OK.  The caller of platform_bus_set_pm_ops() will replace the original
> >> platform_dev_pm_ops with his own set of operations, so he will not see the
> >> power domains.
> >> 
> >> > Do you want to make the replacement routines responsible for
> >> > invoking the power-domain callbacks, or should the platform core handle
> >> > this automatically?
> >> 
> >> Well, if someone replaces the entire platform_dev_pm_ops object, this means
> >> that on his platform power management is substantially different from the
> >> generic one.  In that case, IMO, he should be responsible for handling all
> >> of the subsystem-level aspects of power management, including power domains.
> >
> > Part of point of doing something like power_domain is to *get rid* of
> > platform_bus_set_pm_ops().  It is a horrid, stop-gap interface that
> > doesn't scale.  I don't think much consideration needs to be made for
> > users of platform_bus_set_pm_ops() in this regard.
> 
> As the author of platform_bus_set_pm_ops(), I humbly agree.  
> 
> Also, the __weak functions here were obsoleted by
> platform_bus_set_pm_ops().  Once Magnus moves to
> platform_bus_set_pm_ops() (or this new interface) the __weak attributes
> should be removed (c.f. commit log below[1] where
> platform_bus_set_pm_ops() was added.)
> 
> Kevin
> 
> commit c64a0926710153b9d44c979d2942f4a8648fd74e
> Author: Kevin Hilman <khilman@xxxxxx>
> Date:   Wed Aug 25 12:50:00 2010 -0700
> 
>     driver core: platform_bus: allow runtime override of dev_pm_ops
>     
>     Currently, the platform_bus allows customization of several of the
>     busses dev_pm_ops methods by using weak symbols so that platform code
>     can override them.  The weak-symbol approach is not scalable when
>     wanting to support multiple platforms in a single kernel binary.
>     
>     Instead, provide __init methods for platform code to customize the
>     dev_pm_ops methods at runtime.
>     
>     NOTE: after these dynamic methods are merged, the weak symbols should
>           be removed from drivers/base/platform.c.  AFAIK, this will only
>           affect SH and sh-mobile which should be converted to use this
>           runtime approach instead of the weak symbols.  After SH &
>           sh-mobile are converted, the weak symobols could be removed.
> --

So, it seems there are two possibilities, either (1) keep the $subject patch
as is in the hope that Magnus will use power domains instead of overriding
the __weak callbacks and remove the _weak attribute when that happens, or (2)
modify it along the lines suggested by Alan (ie. so that the _weak callbacks
stay as they are, but they won't be pointed to by platform_dev_pm_ops directly).
I'm pretty much fine with each of them, so I'd prefer to do whichever is
generally more useful.  Magnus?

Rafael
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/