Re: [PATCH RESEND V4 5/9] of: Add NVIDIA Tegra xHCI controller binding

[Andrew Bresticker]

On Fri, Oct 31, 2014 at 4:32 AM, Thierry Reding
<thierry.reding@xxxxxxxxx> wrote:
> On Thu, Oct 30, 2014 at 10:26:47AM -0700, Andrew Bresticker wrote:
>> On Thu, Oct 30, 2014 at 10:24 AM, Thierry Reding
>> <thierry.reding@xxxxxxxxx> wrote:
>> > On Thu, Oct 30, 2014 at 10:19:21AM -0700, Andrew Bresticker wrote:
>> >> On Thu, Oct 30, 2014 at 6:55 AM, Thierry Reding
>> >> <thierry.reding@xxxxxxxxx> wrote:
>> >> > On Wed, Oct 29, 2014 at 09:37:14AM -0700, Andrew Bresticker wrote:
>> >> >> On Wed, Oct 29, 2014 at 2:43 AM, Thierry Reding
>> >> >> <thierry.reding@xxxxxxxxx> wrote:
>> >> >> > On Tue, Oct 28, 2014 at 03:27:50PM -0700, Andrew Bresticker wrote:
>> >> >> > [...]
>> >> >> >> diff --git a/Documentation/devicetree/bindings/pinctrl/nvidia,tegra124-xusb-padctl.txt b/Documentation/devicetree/bindings/pinctrl/nvidia,tegra124-xusb-padctl.txt
>> >> >> > [...]
>> >> >> >> +Optional properties:
>> >> >> >> +-------------------
>> >> >> >> +- vbus-{0,1,2}-supply: VBUS regulator for the corresponding UTMI pad.
>> >> >> >> +- vddio-hsic-supply: VDDIO regulator for the HSIC pads.
>> >> >> >> +- nvidia,usb3-port-{0,1}-lane: PCIe/SATA lane to which the corresponding USB3
>> >> >> >> +  port is mapped.  See <dt-bindings/pinctrl/pinctrl-tegra-xusb.h> for the list
>> >> >> >> +  of valid values.
>> >> >> >
>> >> >> > I dislike how we now need to provide a list of all pins in the header
>> >> >> > file, where previously we used strings for this. This could become very
>> >> >> > ugly if the set of pins changes in future generations of this IP block.
>> >> >> >
>> >> >> > Could we instead derive this from the pinmux nodes? For example you have
>> >> >> > this in the example below:
>> >> >> >
>> >> >> >         usb3p0 {
>> >> >> >                 nvidia,lanes = "pcie-0";
>> >> >> >                 ...
>> >> >> >         };
>> >> >> >
>> >> >> > Perhaps what we need is to either key off the node name or add another
>> >> >> > property, such as:
>> >> >> >
>> >> >> >         nvidia,usb3-port = <0>;
>> >> >> >
>> >> >> > This would match the nvidia,usb2-port property that you've added below.
>> >> >>
>> >> >> That is actually how I described the USB3 port to SS lane mapping
>> >> >> originally, but in review of an earlier version of this series,
>> >> >> Stephen suggested that I make it a separate, not pinconfig property
>> >> >> since it wasn't a value written directly to the hardware.  I'm fine
>> >> >> with changing it back as the pinconfig property makes more sense to me
>> >> >> as well.
>> >> >
>> >> > Hmm... I had considered it a mux option of the specific lane. If the
>> >> > function is usb3, it'd still need to be muxed to one of the ports. So
>> >> > it's additional information associated with the usb3 function.
>> >> >
>> >> > I did look through the driver changes and can't really make out which
>> >> > part of the code actually performs this assignment. Can you point me to
>> >> > it?
>> >>
>> >> There's not really an assignment.  The property is used to map between
>> >> a lane (e.g. PCIe-0 or SATA) and the USB3.0 port it's mapped to.  For
>> >> an example of where it's used, take a look at usb3_phy_power_on().
>> >> There are certain per-lane registers which need to be programmed in
>> >> addition to the per-USB3.0 port pad registers.  This mapping is used
>> >> to determine which lane needs to be programmed.
>> >
>> > Are you saying the mapping of lane to USB port is fixed? That is, PCIe-0
>> > lane is always used for USB port X and SATA always for USB port Y?
>>
>> No, sorry if that was unclear, it's not fixed - it's a board specific
>> property.
>
> Okay, but there's no register that contains the mapping of the port to a
> lane, similar to what's done for the functions, right?

Correct.

> I mean the driver only uses the lane to find out which register to write.
> Doesn't that imply that two lanes (or more) could be mapped to the same
> USB 3.0 port?

I guess?  Not sure how that would work in hardware to have two
SuperSpeed lanes wired up to a single port.

> I'm not sure I'm being clear here, so let me try another way. In order
> to establish a mapping between USB port and lane, I would've expected
> one of the following to happen:
>
>         - A value derived from the lane number is written to a register
>           belonging to a given port.
>
>         - A value derived from the port number is written to a register
>           belonging to a given lane.
>
> I can't see the code do either of the above, which to me implies that
> there's a fixed mapping between lanes and ports. What am I missing?

It's fixed in that it's not a software-modifiable property.  It's
describing an electrical connection between lane and port.  For
example, it's possible that one board has PCIe lane 1 wired up to
SuperSpeed port 1 and on another board for the SATA lane to be wired
up to SuperSpeed port 1.  Does that make sense?
--
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/