On Thu, 21 May 2015, Thierry Reding wrote:

On Thu, May 21, 2015 at 09:40:01AM +0100, Lee Jones wrote:

quoted

On Wed, 20 May 2015, Thierry Reding wrote:

quoted

On Wed, May 20, 2015 at 07:35:51AM +0100, Lee Jones wrote:

quoted

On Tue, 19 May 2015, Andrew Bresticker wrote:

quoted

On Thu, May 14, 2015 at 10:38 AM, Andrew Bresticker
[off-list ref] wrote:

quoted

On Thu, May 14, 2015 at 12:40 AM, Lee Jones [off-list ref] wrote:

quoted

On Thu, 14 May 2015, Jon Hunter wrote:

quoted

On 13/05/15 15:39, Lee Jones wrote:

quoted

On Mon, 04 May 2015, Andrew Bresticker wrote:

quoted

Add a binding document for the XUSB host complex on NVIDIA Tegra124
and later SoCs.  The XUSB host complex includes a mailbox for
communication with the XUSB micro-controller and an xHCI host-controller.

Signed-off-by: Andrew Bresticker <redacted>
Cc: Rob Herring <robh+dt@kernel.org>
Cc: Pawel Moll <redacted>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Ian Campbell <redacted>
Cc: Kumar Gala <redacted>
Cc: Samuel Ortiz <redacted>
Cc: Lee Jones <redacted>
---
Changes from v7:
 - Move non-shared resources into child nodes.
New for v7.
---
 .../bindings/mfd/nvidia,tegra124-xusb.txt          | 37 ++++++++++++++++++++++
 1 file changed, 37 insertions(+)
 create mode 100644 Documentation/devicetree/bindings/mfd/nvidia,tegra124-xusb.txt

diff --git a/Documentation/devicetree/bindings/mfd/nvidia,tegra124-xusb.txt b/Documentation/devicetree/bindings/mfd/nvidia,tegra124-xusb.txt
new file mode 100644
index 0000000..bc50110
--- /dev/null
+++ b/Documentation/devicetree/bindings/mfd/nvidia,tegra124-xusb.txt

@@ -0,0 +1,37 @@
+NVIDIA Tegra XUSB host copmlex
+==============================
+
+The XUSB host complex on Tegra124 and later SoCs contains an xHCI host
+controller and a mailbox for communication with the XUSB micro-controller.
+
+Required properties:
+--------------------
+ - compatible: For Tegra124, must contain "nvidia,tegra124-xusb".
+   Otherwise, must contain '"nvidia,<chip>-xusb", "nvidia,tegra124-xusb"'
+   where <chip> is tegra132.
+ - reg: Must contain the base and length of the XUSB FPCI registers.
+ - ranges: Bus address mapping for the XUSB block.  Can be empty since the
+   mapping is 1:1.
+ - #address-cells: Must be 2.
+ - #size-cells: Must be 2.
+
+Example:
+--------
+  usb at 0,70098000 {
+          compatible = "nvidia,tegra124-xusb";
+          reg = <0x0 0x70098000 0x0 0x1000>;
+          ranges;
+
+          #address-cells = <2>;
+          #size-cells = <2>;
+
+          usb-host at 0,70090000 {
+                  compatible = "nvidia,tegra124-xhci";
+                  ...
+          };
+
+          mailbox {
+                  compatible = "nvidia,tegra124-xusb-mbox";
+                  ...
+          };

This doesn't appear to be a proper MFD.  I would have the USB and
Mailbox devices probe seperately and use a phandle to point the USB
device to its Mailbox.

usb at xyz {
    mboxes = <&xusb-mailbox, [chan]>;
};

I am assuming that Andrew had laid it out like this to reflect the hw
structure. The mailbox and xhci controller are part of the xusb
sub-system and hence appear as child nodes. My understanding is that for
device-tree we want the device-tree structure to reflect the actual hw.
Is this not the case?

Yes, the DT files should reflect h/w.  I have requested to see what
the memory map looks like, so I might provide a more appropriate
solution to accepting a pretty pointless MFD.

FWIW, the address map for XUSB looks like this:

XUSB_HOST: 0x70090000 - 0x7009a000
    xHCI registers: 0x70090000 - 0x70098000
    FPCI configuration registers: 0x70098000 - 0x70099000
    IPFS configuration registers: 0x70099000 - 0x7009a000

quoted

Two solutions spring to mind.  You can either call
of_platform_populate() from the USB driver, as some already do:

  drivers/usb/dwc3/dwc3-exynos.c:
    ret = of_platform_populate(node, NULL, NULL, dev);
  drivers/usb/dwc3/dwc3-keystone.c:
    error = of_platform_populate(node, NULL, NULL, dev);
  drivers/usb/dwc3/dwc3-omap.c:
    ret = of_platform_populate(node, NULL, NULL, dev);
  drivers/usb/dwc3/dwc3-qcom.c:
    ret = of_platform_populate(node, NULL, NULL, qdwc->dev);
  drivers/usb/dwc3/dwc3-st.c:
    ret = of_platform_populate(node, NULL, NULL, dev);
  drivers/usb/musb/musb_am335x.c:
    ret = of_platform_populate(pdev->dev.of_node, NULL, NULL, &pdev->dev);

This still requires a small, separate driver to setup the regmap and
do of_platform_populate().  The only difference is it lives in
drivers/usb/ instead of drivers/mfd/.

quoted

Or use the "simple-mfd", which is currently in -next:

  git show next/master:Documentation/devicetree/bindings/mfd/mfd.txt

I'm not too opposed to this, but Thierry was when I brought this up
before.  The issue here is that if we ever have to do something
besides setting up a regmap in the MFD, we'd have to change the
binding and break DT backwards-compatibility.

Any thoughts on this?  A minimal MFD seems to be the best way to
future-proof this binding/driver should it need to be extended in the
future.  If this is a firm NAK from you however, I'll need to let
Jassi now so that he can un-queue the mailbox patches for 4.2....

I was waiting to hear Thierry's thoughts.  However, I am unconvinced
that you need an MFD driver for this and refuse to take a shell (read
"pointless") one on an "if we ever ..." clause.

Will you break backwards capability though?  I'm not sure you will.
Old DTBs will still use 'simple-mfd' and probe the devices in the
normal way.  *If* you introduce an MFD driver at a later date then the
old DTB will miss out the *new* functionality, which is expected and
accepted.

I'm a little confused by the simple-mfd approach. The only code I see in
linux-next for this is a single line that adds the "simple-mfd" string
to the OF device ID table in drivers/of/platform.c. As far as I can tell
this will merely cause child devices to be created. There won't be a
shared regmap and resources won't be set up properly either. Having a
proper MFD driver seems to be the only way to achieve what we need.

The reason why every other simple-mfd users seems to get away with this
is because they also match on syscon and that sets up a regmap of its
own and the child device drivers use the syscon API to get at it. So I
don't see how we can make use of simple-mfd to achieve what we need,
unless we essentially copy what syscon does (but do proper resource
management while at it).

If you have shared resources and your device isn't classed as a syscon
device then yes, simple-mfd probably isn't suitable for this use-case.
You might need to go into more detail with regards to "proper resource
management", as I'm not entirely sure I agree.

Still, this doesn't change the fact that, from what I've seen, I still
don't think you need a dedicated MFD driver.

What do you think of:

usb-host at 0,70090000 {
	compatible = "nvidia,tegra124-xhci";
	reg = <0x0 0x70090000 0x0 0x80CF>,
	      <0x0 0x70098800 0x0 0x0800>,
	      <0x0 0x70099000 0x0 0x1000>;

	/* Something from the datasheet */
	reg-names = "xhci-before-mbox", "xhci-after-mbox", "ipfs";

	interrupts = <GIC_SPI 39 IRQ_TYPE_LEVEL_HIGH>;
	ranges;

	xusb_mbox: mailbox {
		compatible = "nvidia,tegra124-xusb-mbox";
		reg = <0x0 0x700980e0 0x0 0x13>,
		      <0x0 0x70098428 0x0 0x03>;

		/* Something from the datasheet */
		reg-names = "mbox-one", "mbox-two";
		interrupts = <GIC_SPI 40 IRQ_TYPE_LEVEL_HIGH>;
	};
};

Then hvae the XHCI driver call of_platform_populate() as I proposed
above?

That's a little bonghits. It requires the drivers to jump through hoops
to properly manage register accesses (needs to differentiate on the base
depending on the register offset). So if you're going to NAK the MFD
approach I'd rather go a completely different route and keep only a top-
level node in DT here.

One of the problems that the MFD design tries to solve is that the XHCI
controller needs a reference to the mailbox and the pad controller for a
PHY. The pad controller at the same time requires a reference to the
mailbox, so we have a circular dependency that we can only resolve by
introducing two separate devices, instantiated by some top-level entity.
For that reason I don't think your proposal is going to work either. The
circular dependency can't be broken because the XHCI driver will not be
able to of_platform_populate() before getting a PHY, and the PHY will
never show up until of_platform_populate() is called.

So if this isn't going to be an MFD, then I think we should simply go
and instantiate platform devices from the XUSB driver directly. The
problem arising from that is that we have no place to put the top-level
driver. We could take it into drivers/soc/tegra, or perhaps even have it
in the XHCI driver.

If we instantiate platform devices we can either set up the resources
such that we don't have to jump through hoops (I think the resource tree
will allow that) or set up a shared regmap. The latter might be the
easier way out, though it'd also be copying much of what MFD does, but
so be it if that's the only way we can get the matter settled.

I understand the difficulties identified and empathise with your
situation.  I just can't bring myself to justify that a USB device
which has it's own Mailbox is an MFD.  If you take a look above, you
can see some examples of other USB drivers registering sub-devices.  I
think you can make this work well for your setup.

quoted

quoted

There is also the matter of clocks, resets, power supplies, etc. which
simple-mfd can't take into account in its current form. From a hardware
point of view, (some of) the clocks and resets are shared by the XHCI
and the mailbox blocks, so the device tree node would have to take that
into account. And a driver would also have to know which clocks, resets
and power supplies to probe and the order in which they need to be
enabled. simple-mfd doesn't provide any of that currently, so we'd
likely need to hack around that in all sorts of weird ways in the child
drivers. It makes much more sense for a top-level MFD driver to set up
the shared hardware resources and then instantiate the child devices and
let the drivers for those only care about the child-specific resources.

A catch-all driver will inevitably lead to implementing a midlayer with
potentially all sorts of quirks to make it work with the various devices
out there.

A much better implementation, in my opinion, would be to make simple-mfd
a subclassable object and then have drivers use a helper library to call
code that is common for simple-mfd kinds of devices. Something like this
for example:

	struct tegra_xusb {
		...
		struct mfd_simple mfd;
		...
	};

	static int tegra_xusb_probe(struct platform_device *pdev)
	{
		struct tegra_xusb *xusb;
		...
		err = mfd_simple_register(&xusb->mfd);
		...
	}

Now all these drivers reuse all the code provided by the mfd_simple
helper, which will instantiate the children, but it is also very easy to
tie in the platform-specific glue (clocks, resets, regulators, ...) via
the device-specific drivers.

I'm not keen on creating a not-so-simple-mfd driver.  Let's work with
what we've got for the time being.

What we currently have is not a driver at all, it's merely an alias for
simple-bus.

Right, which is exactly what it was designed to be.  Initially we were
using simple-bus, but some people (rightly) thought this an abuse 'cos
MFD isn't really a bus.

-- 
Lee Jones
Linaro STMicroelectronics Landing Team Lead
Linaro.org ? Open source software for ARM SoCs
Follow Linaro: Facebook | Twitter | Blog