Point-to-point bus in device tree

Point-to-point bus in device tree

[David Brown]

Some MSM SoCs have a small serial-type "bus" that is used to
communicate with the PMIC devices.  This interface is always
point-to-point.  I'm doing a device-tree conversion of the driver that
Ken Heitke posted last year <https://lkml.org/lkml/2011/2/17/503>.

A naive conversion to device tree, would result in something like
this:

	qcom,ssbi@500000 {
		compatible = "qcom,ssbi";
		reg = <0x500000 0x1000>;
		qcom,controller-type = "ssbi";

		qcom,pmic8058@0 {
			reg = <0x0 0x01>;
			...
		}
	}

There would end up being an extraneous register for the device on the
other end (there are no addresses), and there would need to be code in
the ssbi driver to traverse this small tree to find these nodes.

Another approach would be to make the pmic driver the actual device,
and just have it use the ssbi code as a library.  Something like:

	qcom,pmic8058@500000 {
		compatible = "qcom,pmic8058";
		reg = <0x500000 0x1000>;
		qcom,ssbi-type = "ssbi";
		...
	}

This is a lot cleaner, takes less code, but doesn't technically
represent how the hardware is really structured.

Does anyone have strong (or weak) feelings about this?

Thanks,
David

-- 
Sent by an employee of the Qualcomm Innovation Center, Inc.
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum.

Re: Point-to-point bus in device tree

[Stephen Warren]

On 04/05/2012 12:15 PM, David Brown wrote:
> Some MSM SoCs have a small serial-type "bus" that is used to
> communicate with the PMIC devices.  This interface is always
> point-to-point.  I'm doing a device-tree conversion of the driver that
> Ken Heitke posted last year <https://lkml.org/lkml/2011/2/17/503>.
> 
> A naive conversion to device tree, would result in something like
> this:
> 
> 	qcom,ssbi@500000 {
> 		compatible = "qcom,ssbi";
> 		reg = <0x500000 0x1000>;
> 		qcom,controller-type = "ssbi";
> 
> 		qcom,pmic8058@0 {
> 			reg = <0x0 0x01>;
> 			...
> 		}
> 	}
> 
> There would end up being an extraneous register for the device on the
> other end (there are no addresses), and there would need to be code in
> the ssbi driver to traverse this small tree to find these nodes.

Isn't that extra code simply:

of_platform_populate(pdev->dev.of_node, NULL, NULL, &pdev->dev);

That seems like pretty low overhead.

Re: Point-to-point bus in device tree

[David Brown]

On Thu, Apr 05, 2012 at 01:23:46PM -0600, Stephen Warren wrote:
> On 04/05/2012 12:15 PM, David Brown wrote:
> > Some MSM SoCs have a small serial-type "bus" that is used to
> > communicate with the PMIC devices.  This interface is always
> > point-to-point.  I'm doing a device-tree conversion of the driver that
> > Ken Heitke posted last year <https://lkml.org/lkml/2011/2/17/503>.
> > 
> > A naive conversion to device tree, would result in something like
> > this:
> > 
> > 	qcom,ssbi@500000 {
> > 		compatible = "qcom,ssbi";
> > 		reg = <0x500000 0x1000>;
> > 		qcom,controller-type = "ssbi";
> > 
> > 		qcom,pmic8058@0 {
> > 			reg = <0x0 0x01>;
> > 			...
> > 		}
> > 	}
> > 
> > There would end up being an extraneous register for the device on the
> > other end (there are no addresses), and there would need to be code in
> > the ssbi driver to traverse this small tree to find these nodes.
> 
> Isn't that extra code simply:
> 
> of_platform_populate(pdev->dev.of_node, NULL, NULL, &pdev->dev);
> 
> That seems like pretty low overhead.

True, but it still bothers me to have to have a bogus register.

David

-- 
Sent by an employee of the Qualcomm Innovation Center, Inc.
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum.

Re: Point-to-point bus in device tree

[David Brown]

On Thu, Apr 05, 2012 at 01:08:29PM -0700, David Brown wrote:
> On Thu, Apr 05, 2012 at 01:23:46PM -0600, Stephen Warren wrote:
> > On 04/05/2012 12:15 PM, David Brown wrote:
> > > Some MSM SoCs have a small serial-type "bus" that is used to
> > > communicate with the PMIC devices.  This interface is always
> > > point-to-point.  I'm doing a device-tree conversion of the driver that
> > > Ken Heitke posted last year <https://lkml.org/lkml/2011/2/17/503>.
> > > 
> > > A naive conversion to device tree, would result in something like
> > > this:
> > > 
> > > 	qcom,ssbi@500000 {
> > > 		compatible = "qcom,ssbi";
> > > 		reg = <0x500000 0x1000>;
> > > 		qcom,controller-type = "ssbi";
> > > 
> > > 		qcom,pmic8058@0 {
> > > 			reg = <0x0 0x01>;
> > > 			...
> > > 		}
> > > 	}
> > > 
> > > There would end up being an extraneous register for the device on the
> > > other end (there are no addresses), and there would need to be code in
> > > the ssbi driver to traverse this small tree to find these nodes.
> > 
> > Isn't that extra code simply:
> > 
> > of_platform_populate(pdev->dev.of_node, NULL, NULL, &pdev->dev);
> > 
> > That seems like pretty low overhead.
> 
> True, but it still bothers me to have to have a bogus register.

Ok, not sure where I got that idea, but I don't appear to actually
need a register in the pmic8058 node for of_platform_populate to
create a platform device.  It only needs a compatible field.  So, it
look like that matches just what I need.

Thanks,
David

-- 
Sent by an employee of the Qualcomm Innovation Center, Inc.
The Qualcomm Innovation Center, Inc. is a member of the Code Aurora Forum.

Re: Point-to-point bus in device tree

[Grant Likely]

On Fri, 6 Apr 2012 08:16:10 -0700, David Brown <davidb@codeaurora.org> wrote:
> On Thu, Apr 05, 2012 at 01:08:29PM -0700, David Brown wrote:
> > On Thu, Apr 05, 2012 at 01:23:46PM -0600, Stephen Warren wrote:
> > > On 04/05/2012 12:15 PM, David Brown wrote:
> > > > Some MSM SoCs have a small serial-type "bus" that is used to
> > > > communicate with the PMIC devices.  This interface is always
> > > > point-to-point.  I'm doing a device-tree conversion of the driver that
> > > > Ken Heitke posted last year <https://lkml.org/lkml/2011/2/17/503>.
> > > > 
> > > > A naive conversion to device tree, would result in something like
> > > > this:
> > > > 
> > > > 	qcom,ssbi@500000 {
> > > > 		compatible = "qcom,ssbi";
> > > > 		reg = <0x500000 0x1000>;
> > > > 		qcom,controller-type = "ssbi";
> > > > 
> > > > 		qcom,pmic8058@0 {
> > > > 			reg = <0x0 0x01>;
> > > > 			...
> > > > 		}
> > > > 	}
> > > > 
> > > > There would end up being an extraneous register for the device on the
> > > > other end (there are no addresses), and there would need to be code in
> > > > the ssbi driver to traverse this small tree to find these nodes.
> > > 
> > > Isn't that extra code simply:
> > > 
> > > of_platform_populate(pdev->dev.of_node, NULL, NULL, &pdev->dev);
> > > 
> > > That seems like pretty low overhead.
> > 
> > True, but it still bothers me to have to have a bogus register.
> 
> Ok, not sure where I got that idea, but I don't appear to actually
> need a register in the pmic8058 node for of_platform_populate to
> create a platform device.  It only needs a compatible field.  So, it
> look like that matches just what I need.

Correct.  Don't use a reg and all will look good.

g.