Re: [PATCH v6 1/4] Runtime Interpreted Power Sequences

On Thursday 13 September 2012 15:50:37 Sascha Hauer wrote:

On Thu, Sep 13, 2012 at 09:29:20AM +0200, Thierry Reding wrote:

quoted

On Thu, Sep 13, 2012 at 10:03:27AM +0300, Tomi Valkeinen wrote:

quoted

On Thu, 2012-09-13 at 09:00 +0200, Sascha Hauer wrote:

quoted

On Thu, Sep 13, 2012 at 09:54:09AM +0300, Tomi Valkeinen wrote:

quoted

On Thu, 2012-09-13 at 15:36 +0900, Alex Courbot wrote:

quoted

On Thursday 13 September 2012 14:22:57 Tomi Valkeinen wrote:

quoted

However, I fear these board specific things may be quite a bit
anything,
so it may well be pwm, gpios and regulators are not enough for
them. For
example, there could be an FPGA on the board which requires some
configuration to accomplish the task at hand. It could be rather
difficult to handle it with a generic power sequence.

Right. Note that this framework is supposed to be extended - I
would like to at least add regulator voltage setting, and maybe
even support for clocks and pinmux (but that might be out of
place).

Yes, that's one concern of mine... I already can imagine someone
suggesting adding conditionals to the power sequence data. Perhaps
also
direct memory read/writes so you can twiddle registers directly. And
so
on. Where's the limit what it should contain? Can we soon write full
drivers with the DT data? =)

I have this concern aswell, that's why I'm sceptical about this patch
set. But what are the alternatives? Adding power code to the drivers
and
thus adding board specific code to them is backwards.

As was pointed out in earlier posts in this thread, these are almost
always device specific, not board specific.

Do you have examples of board specific power sequences or such?

It is true that most (perhaps all) power sequences can be associated
with a specific device, but if we go and implement drivers for these
kinds of devices we will probably end up with loads of variations of
the same scheme.

Lets take display panels as an example. One of the devices that we build
has gone through two generations so far and both are slightly different
in how they control the panel backlight: one has an external backlight
controller, the other has the display controller built into the panel.
However, from the board's perspective the control of the backlight
doesn't change, because both devices get the same inputs (an enable pin
and a PWM) that map to the same pins on the SoC.

This may not be a very good example because the timing isn't relevant,
but the basic point is still valid: if we provide a driver for both
panel devices, the code will be exactly the same. So we end up having to
refactor to avoid code duplication and use the same driver for a number
of backlight/panel combinations. Which in itself isn't very bad, but it
also means that we'll probably get to see a large number of "generic"
drivers which aren't very generic after all.

Another problem, which also applies to the case of power-sequences, is
that often the panel and backlight are not the same device.

Maybe that is the problem that needs to be addressed? They *are* not the
same device, still they are handled in a single platform callback (or
now power sequence). Maybe the amount of combinations dastrically go
down if we really make them two devices.

Most of our panels have:

- A regulator (or gpio) for turning them on

And the backlights have:

- A regulator (or gpio) for turning them on
- A PWM for controlling brightness.

The power sequence for the above is clear: Turn on the panel the panel,
wait until it stabilized and afterwards turn on the backlight.

Actually the sequence I submitted in this patchset only takes care of the 
backlight device (the panel - or LCD - should have its own). The regulator 
controls the power supply, the PWM the intensity, and on top of that it also 
has an enable GPIO. These 3 resources are exclusively for the LED - the LCD 
uses other ones. So as of now it seems that the LCD/backlight separation is 
effective and the resources needed are not so uniform across backlights (not 
even mentioning the delays).

The LCD's power sequence is even weirder - VDD must take at least 0.5ms for 
going from 10% to 90% of its power, you must wait 400ms after switching it off 
before switching it on again, and you should also transmit data for 200ms 
before switching the backlight's LED on (using its own sequence). That last 
point is interesting since it somehow makes the LCD and LED dependent on each 
other - on an unrelated note, this might be something to consider in Laurent's 
proposal for a panel framework.

Alex.