lkml.org 
[lkml]   [2012]   [Apr]   [2]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH V3 5/6] dt: Document Tegra20/30 pinctrl binding
On 04/02/2012 12:49 AM, Simon Glass wrote:
> Hi Stephen,
>
> On Thu, Mar 22, 2012 at 11:27 AM, Stephen Warren <swarren@wwwdotorg.org> wrote:
>> Define a new binding for the Tegra pin controller, which is capable of
>> defining all aspects of desired pin multiplexing and pin configuration.
>> This is all based on the new common pinctrl bindings.
>>
>> Add Tegra30 binding based on Tegra20 binding.
>>
>> Add some basic stuff that was missing before:
>> * How many and what reg property entries must be provided.
>> * An example.
>>
>> Signed-off-by: Stephen Warren <swarren@wwwdotorg.org>
>> ---
>> v3: Fix typo in Tegra20 binding example
...
>> +Example board file extract:
>> +
>> + pinctrl@70000000 {
>> + sdmmc4_default: pinmux {
>> + sdmmc4_clk_pcc4 {
>> + nvidia,pins = "sdmmc4_clk_pcc4",
>> + "sdmmc4_rst_n_pcc3";
>> + nvidia,function = "sdmmc4";
>> + nvidia,pull = <0>;
>> + nvidia,tristate = <0>;
>> + };
>> + sdmmc4_dat0_paa0 {
>> + nvidia,pins = "sdmmc4_dat0_paa0",
>> + "sdmmc4_dat1_paa1",
>> + "sdmmc4_dat2_paa2",
>> + "sdmmc4_dat3_paa3",
>> + "sdmmc4_dat4_paa4",
>> + "sdmmc4_dat5_paa5",
>> + "sdmmc4_dat6_paa6",
>> + "sdmmc4_dat7_paa7";
>
> I see that you have done with a hierarchical approach which I like a lot.
>
> However, the large number of strings here (using a string to name a
> function and a pin) is going to create quite a bit of overhead, not to
> mention FDT space.

IIRC, I profiled this in the middle of last year and while there was
measurable overhead using strings relative to integers, it was tiny; on
the order of perhaps a couple mS.

> Have you given up on the /define/ patch that you created?

Not entirely, but it's obvious it will take a /long/ time to get that or
equivalent functionality into dtc. I'd rather not block pinmux-in-dtc by
waiting for it, especially when the string alternative works fine with
what I consider reasonable overhead.

> If so, I wonder if we could at least provide an alternate binding
> using numbering.

I'd rather only have a single binding; alternatives are just going to
add a bunch of complexity.

Perhaps we can have a flag-day in the future and change the binding once
dtc has grown named-constants support or we've got cpp integrated into
the build flow for this.

> I have just figured out how to get the C preprocessor
> out of U-Boot's FDT path,

I don't understand exactly what that means.

> but if that is the only way, perhaps the
> kernel should use that to get numbered symbols?
>
> I would much prefer a parallel property which provides the names, that
> can be omitted.

If we did have multiple possible data representations, either-or seems
better; allowing both to co-exist just opens up the potential for them
to say different things.

> (Similarly for your 'nvidia,pull' property, it would be nice to have a
> symbolic name)
>
>> + nvidia,function = "sdmmc4";
>> + nvidia,pull = <2>;
>> + nvidia,tristate = <0>;
>> + };
>> + };
>> + };
>> +
>> + sdhci@78000400 {
>> + pinctrl-names = "default";
>> + pinctrl-0 = <&sdmmc4_default>;
>> + };


\
 
 \ /
  Last update: 2012-04-02 17:51    [from the cache]
©2003-2011 Jasper Spaans