Re: [PATCH 2/2] ACPI / platform: Initialize ACPI handles of platform devices in advance

[Greg Kroah-Hartman <gregkh@linuxfoundation.org>]

On Mon, Nov 19, 2012 at 09:44:21PM +0100, Rafael J. Wysocki wrote:
> On Monday, November 19, 2012 06:45:22 PM Rafael J. Wysocki wrote:
> > On Monday, November 19, 2012 06:32:06 PM Rafael J. Wysocki wrote:
> > > On Monday, November 19, 2012 08:23:34 AM Greg Kroah-Hartman wrote:
> > > > On Sun, Nov 18, 2012 at 10:13:59PM +0100, Rafael J. Wysocki wrote:
> > > > > From: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
> > > > > 
> > > > > The current platform device creation and registration code in
> > > > > acpi_create_platform_device() is quite convoluted.  This function
> > > > > takes an ACPI device node as an argument and eventually calls
> > > > > platform_device_register_resndata() to create and register a
> > > > > platform device object on the basis of the information contained
> > > > > in that code.  However, it doesn't associate the new platform
> > > > > device with the ACPI node directly, but instead it relies on
> > > > > acpi_platform_notify(), called from within device_add(), to find
> > > > > that ACPI node again with the help of acpi_platform_find_device()
> > > > > and acpi_platform_match() and then attach the new platform device
> > > > > to it.  This causes an additional ACPI namespace walk to happen and
> > > > > is clearly suboptimal.
> > > > > 
> > > > > Use the observation that it is now possible to initialize the ACPI
> > > > > handle of a device before calling device_add() for it to make this
> > > > > code more straightforward.  Namely, add a new field to struct
> > > > > platform_device_info allowing us to pass the ACPI handle of interest
> > > > > to platform_device_register_full(), which will then use it to
> > > > > initialize the new device's ACPI handle before registering it.
> > > > > This will cause acpi_platform_notify() to use the ACPI handle from
> > > > > the device structure directly instead of using the .find_device()
> > > > > routine provided by the device's bus type.  In consequence,
> > > > > acpi_platform_bus, acpi_platform_find_device(), and
> > > > > acpi_platform_match() are not necessary any more, so remove them.
> > > > 
> > > > Why can't you use the platform_data * that is already in struct device
> > > > for this, instead of adding an acpi-specific field to the
> > > > platform_device structure?
> > > 
> > > Hmm, I kind of don't understand the question. :-)
> > > 
> > > Yes, we have acpi_handle in struct device (it actually is being added by a
> > > patch you've acked) and we use it.  The whole point here is to streamline
> > > of the initalization of that field.
> > > 
> > > > If not that, surely there is another field in struct device that you
> > > > could use that is free for this type of device?
> > > 
> > > Yes, there is one and as I said above. :-)
> > > 
> > > I'd be happy to use the struct device's field directly, but
> > > platform_device_register_full() allocates memory for the struct device in
> > > question, so that field actually doesn't exist yet when it is called.
> > > 
> > > > >  struct platform_device_info {
> > > > >  		struct device *parent;
> > > > > +		void *acpi_handle;
> > > > 
> > > > Oh, and if I do accept this, I want a "real" structure pointer here
> > > > please, not a void * "handle".  That way is a slippery slope to the
> > > > Windows kernel programming style :)
> > > 
> > > This is (void *), because the field being initialized is (void *).  That field,
> > > in turn, is (void *), because ACPICA defines it that way.  I thought about
> > > wrapping that in some more meaningless data type, but I did't find a way
> > 
> > s/meaningless/meaningful/
> 
> Well, perhaps I'll describe the problem to you, maybe you can help. :-)
> 
> So, we want to have acpi_handle (or acpi_node) in addition to of_node in struct
> device (to be used in the analogous way plus for the execution of AML methods),
> but we don't want all users of device.h to have to include ACPI headers
> where the acpi_handle data type is defined.  For this reason, we're using
> (void *) as its data type now, which let's say I'm not really happy with.
> 
> I've been thinking about that for quite a while, though, and I'm not really
> sure what to do about that.  Perhaps we could define something like
> 
> struct acpi_dev_node {
> #ifdef CONFIG_ACPI
> 	void *handle;
> #endif
> };
> 
> in device.h and use that as "struct acpi_dev_node acpi_node;" in struct device.
> Then, we could add the following macro
> 
> #ifdef CONFIG_ACPI
> #define ACPI_HANDLE(dev)	((dev)->acpi_node.handle)
> #else
> #define ACPI_HANDLE(dev)	(NULL)
> #endif
> 
> and redefine DEVICE_ACPI_HANDLE(dev) as ((acpi_handle)ACPI_HANDLE(dev)).
> 
> Then, the $subject patch would add "struct acpi_dev_node acpi_node;" to
> struct platform_device_info and use ACPI_HANDLE(dev) instead of accessing
> the struct device's field directly.
> 
> I wonder what you think?

I like the hack of using an empty structure here, that's fine with me,
and makes me feel a little bit better about the whole "void *" stuff.
If you respin the patch with this, I'll ack it.

thanks,

greg k-h