[REPOST][PATCH 2/2] driver core: platform: allow platform drivers to bind to any device

[REPOST][PATCH 2/2] driver core: platform: allow platform drivers to bind to any device

[Kim Phillips]

Platform drivers such as the vfio-platform "meta-" driver [1]
should be allowed to specify that they can bind to any device,
much like PCI drivers can with PCI_ANY_ID.

Currently, binding platform drivers to devices depends on:

- a string match in the device node's compatible entry (OF)
- a string match in the ACPI id list (ACPI)
- a string match in the id_table (platform data)
- a string match on the driver name (fall-back)

none of which allow for the notion of "match any."

This patch adds the notion by adding a "match any device" boolean to
struct platform_driver, for drivers to be able to set and thus not cause
platform_match() to fail when a bind is requested.

[1] http://www.spinics.net/lists/kvm/msg96701.html

Signed-off-by: Kim Phillips <kim.phillips@linaro.org>
---
rebased onto 3.13-rc2, and reposted from first submission which
received no comments:

https://lkml.org/lkml/2013/10/11/48

 drivers/base/platform.c         | 4 ++++
 include/linux/platform_device.h | 1 +
 2 files changed, 5 insertions(+)

diff --git a/drivers/base/platform.c b/drivers/base/platform.c
index 3a94b79..78a5b62 100644
--- a/drivers/base/platform.c
+++ b/drivers/base/platform.c
@@ -736,6 +736,10 @@ static int platform_match(struct device *dev, struct device_driver *drv)
 	struct platform_device *pdev = to_platform_device(dev);
 	struct platform_driver *pdrv = to_platform_driver(drv);
 
+	/* the driver matches any device */
+	if (pdrv->match_any_dev)
+		return 1;
+
 	/* Attempt an OF style match first */
 	if (of_driver_match_device(dev, drv))
 		return 1;
diff --git a/include/linux/platform_device.h b/include/linux/platform_device.h
index 16f6654..55e76b1 100644
--- a/include/linux/platform_device.h
+++ b/include/linux/platform_device.h
@@ -179,6 +179,7 @@ struct platform_driver {
 	struct device_driver driver;
 	const struct platform_device_id *id_table;
 	bool prevent_deferred_probe;
+	bool match_any_dev;
 };
 
 #define to_platform_driver(drv)	(container_of((drv), struct platform_driver, \
-- 
1.8.5

Re: [REPOST][PATCH 2/2] driver core: platform: allow platform drivers to bind to any device

[Greg Kroah-Hartman]

On Tue, Dec 03, 2013 at 12:34:54PM +0000, Kim Phillips wrote:
> Platform drivers such as the vfio-platform "meta-" driver [1]
> should be allowed to specify that they can bind to any device,
> much like PCI drivers can with PCI_ANY_ID.
> 
> Currently, binding platform drivers to devices depends on:
> 
> - a string match in the device node's compatible entry (OF)
> - a string match in the ACPI id list (ACPI)
> - a string match in the id_table (platform data)
> - a string match on the driver name (fall-back)
> 
> none of which allow for the notion of "match any."
> 
> This patch adds the notion by adding a "match any device" boolean to
> struct platform_driver, for drivers to be able to set and thus not cause
> platform_match() to fail when a bind is requested.
> 
> [1] http://www.spinics.net/lists/kvm/msg96701.html
> 
> Signed-off-by: Kim Phillips <kim.phillips@linaro.org>
> ---
> rebased onto 3.13-rc2, and reposted from first submission which
> received no comments:
> 
> https://lkml.org/lkml/2013/10/11/48
> 
>  drivers/base/platform.c         | 4 ++++
>  include/linux/platform_device.h | 1 +
>  2 files changed, 5 insertions(+)
> 
> diff --git a/drivers/base/platform.c b/drivers/base/platform.c
> index 3a94b79..78a5b62 100644
> --- a/drivers/base/platform.c
> +++ b/drivers/base/platform.c
> @@ -736,6 +736,10 @@ static int platform_match(struct device *dev, struct device_driver *drv)
>  	struct platform_device *pdev = to_platform_device(dev);
>  	struct platform_driver *pdrv = to_platform_driver(drv);
>  
> +	/* the driver matches any device */
> +	if (pdrv->match_any_dev)
> +		return 1;

This breaks userspace in that it will never know to load the module that
can "bind to anything".

You need a way to encode this in the platform device id that can be a
wildcard type of thing, so that userspace can know about this.

thanks,

greg k-h

Re: [REPOST][PATCH 2/2] driver core: platform: allow platform drivers to bind to any device

[Scott Wood]

On Wed, 2013-12-18 at 17:03 -0800, Greg Kroah-Hartman wrote:
> On Tue, Dec 03, 2013 at 12:34:54PM +0000, Kim Phillips wrote:
> > Platform drivers such as the vfio-platform "meta-" driver [1]
> > should be allowed to specify that they can bind to any device,
> > much like PCI drivers can with PCI_ANY_ID.
> > 
> > Currently, binding platform drivers to devices depends on:
> > 
> > - a string match in the device node's compatible entry (OF)
> > - a string match in the ACPI id list (ACPI)
> > - a string match in the id_table (platform data)
> > - a string match on the driver name (fall-back)
> > 
> > none of which allow for the notion of "match any."
> > 
> > This patch adds the notion by adding a "match any device" boolean to
> > struct platform_driver, for drivers to be able to set and thus not cause
> > platform_match() to fail when a bind is requested.
> > 
> > [1] http://www.spinics.net/lists/kvm/msg96701.html
> > 
> > Signed-off-by: Kim Phillips <kim.phillips@linaro.org>
> > ---
> > rebased onto 3.13-rc2, and reposted from first submission which
> > received no comments:
> > 
> > https://lkml.org/lkml/2013/10/11/48
> > 
> >  drivers/base/platform.c         | 4 ++++
> >  include/linux/platform_device.h | 1 +
> >  2 files changed, 5 insertions(+)
> > 
> > diff --git a/drivers/base/platform.c b/drivers/base/platform.c
> > index 3a94b79..78a5b62 100644
> > --- a/drivers/base/platform.c
> > +++ b/drivers/base/platform.c
> > @@ -736,6 +736,10 @@ static int platform_match(struct device *dev, struct device_driver *drv)
> >  	struct platform_device *pdev = to_platform_device(dev);
> >  	struct platform_driver *pdrv = to_platform_driver(drv);
> >  
> > +	/* the driver matches any device */
> > +	if (pdrv->match_any_dev)
> > +		return 1;
> 
> This breaks userspace in that it will never know to load the module that
> can "bind to anything".
> 
> You need a way to encode this in the platform device id that can be a
> wildcard type of thing, so that userspace can know about this.

How is userspace broken?  vfio platform is a new thing; there's nothing
existing to break.  I don't see how automatic module loading makes sense
for it.  Whether the module needs to be loaded depends on the user's
intentions, not based on what hardware you have.

-Scott

Re: [REPOST][PATCH 2/2] driver core: platform: allow platform drivers to bind to any device

[Greg Kroah-Hartman]

On Thu, Dec 19, 2013 at 02:29:21PM -0600, Scott Wood wrote:
> On Wed, 2013-12-18 at 17:03 -0800, Greg Kroah-Hartman wrote:
> > On Tue, Dec 03, 2013 at 12:34:54PM +0000, Kim Phillips wrote:
> > > Platform drivers such as the vfio-platform "meta-" driver [1]
> > > should be allowed to specify that they can bind to any device,
> > > much like PCI drivers can with PCI_ANY_ID.
> > > 
> > > Currently, binding platform drivers to devices depends on:
> > > 
> > > - a string match in the device node's compatible entry (OF)
> > > - a string match in the ACPI id list (ACPI)
> > > - a string match in the id_table (platform data)
> > > - a string match on the driver name (fall-back)
> > > 
> > > none of which allow for the notion of "match any."
> > > 
> > > This patch adds the notion by adding a "match any device" boolean to
> > > struct platform_driver, for drivers to be able to set and thus not cause
> > > platform_match() to fail when a bind is requested.
> > > 
> > > [1] http://www.spinics.net/lists/kvm/msg96701.html
> > > 
> > > Signed-off-by: Kim Phillips <kim.phillips@linaro.org>
> > > ---
> > > rebased onto 3.13-rc2, and reposted from first submission which
> > > received no comments:
> > > 
> > > https://lkml.org/lkml/2013/10/11/48
> > > 
> > >  drivers/base/platform.c         | 4 ++++
> > >  include/linux/platform_device.h | 1 +
> > >  2 files changed, 5 insertions(+)
> > > 
> > > diff --git a/drivers/base/platform.c b/drivers/base/platform.c
> > > index 3a94b79..78a5b62 100644
> > > --- a/drivers/base/platform.c
> > > +++ b/drivers/base/platform.c
> > > @@ -736,6 +736,10 @@ static int platform_match(struct device *dev, struct device_driver *drv)
> > >  	struct platform_device *pdev = to_platform_device(dev);
> > >  	struct platform_driver *pdrv = to_platform_driver(drv);
> > >  
> > > +	/* the driver matches any device */
> > > +	if (pdrv->match_any_dev)
> > > +		return 1;
> > 
> > This breaks userspace in that it will never know to load the module that
> > can "bind to anything".
> > 
> > You need a way to encode this in the platform device id that can be a
> > wildcard type of thing, so that userspace can know about this.
> 
> How is userspace broken?  vfio platform is a new thing; there's nothing
> existing to break.  I don't see how automatic module loading makes sense
> for it.  Whether the module needs to be loaded depends on the user's
> intentions, not based on what hardware you have.

As your driver is saying it can "match any device", you need to tell
userspace that, which you aren't doing here.  That's what the ids that
are exported to userspace do, please add some for platform devices to
allow this to work properly, like almost all other busses have.

greg k-h

Re: [REPOST][PATCH 2/2] driver core: platform: allow platform drivers to bind to any device

[Scott Wood]

On Thu, 2013-12-19 at 12:44 -0800, Greg Kroah-Hartman wrote:
> On Thu, Dec 19, 2013 at 02:29:21PM -0600, Scott Wood wrote:
> > On Wed, 2013-12-18 at 17:03 -0800, Greg Kroah-Hartman wrote:
> > > On Tue, Dec 03, 2013 at 12:34:54PM +0000, Kim Phillips wrote:
> > > > diff --git a/drivers/base/platform.c b/drivers/base/platform.c
> > > > index 3a94b79..78a5b62 100644
> > > > --- a/drivers/base/platform.c
> > > > +++ b/drivers/base/platform.c
> > > > @@ -736,6 +736,10 @@ static int platform_match(struct device *dev, struct device_driver *drv)
> > > >  	struct platform_device *pdev = to_platform_device(dev);
> > > >  	struct platform_driver *pdrv = to_platform_driver(drv);
> > > >  
> > > > +	/* the driver matches any device */
> > > > +	if (pdrv->match_any_dev)
> > > > +		return 1;
> > > 
> > > This breaks userspace in that it will never know to load the module that
> > > can "bind to anything".
> > > 
> > > You need a way to encode this in the platform device id that can be a
> > > wildcard type of thing, so that userspace can know about this.
> > 
> > How is userspace broken?  vfio platform is a new thing; there's nothing
> > existing to break.  I don't see how automatic module loading makes sense
> > for it.  Whether the module needs to be loaded depends on the user's
> > intentions, not based on what hardware you have.
> 
> As your driver is saying it can "match any device", you need to tell
> userspace that, which you aren't doing here.

Userspace either knows how to set up VFIO platform devices, in which
case it knows what to do, or it doesn't, in which case why would we want
the module loaded?

> That's what the ids that are exported to userspace do, please add some for platform devices to
> allow this to work properly, like almost all other busses have.

This isn't about platform devices versus "almost all other buses", it's
about VFIO being fundamentally different from normal drivers.

-Scott