[PATCH v3 5/5] pwm: airoha: Add support for EN7581 SoC

[Uwe Kleine-König]

Hello Benjamin,

On Thu, Sep 05, 2024 at 02:18:41PM +0200, Benjamin Larsson wrote:
> On 2024-09-05 11:30, Uwe Kleine-König wrote:
> > 1 second long pulses with a period size of 1 second, so a constant high
> > signal?
> 
> Hi, I think I was unclear. The SoC documentation is not that detailed. But I
> think I understand how it works now.
> 
> One register contains the minimum duration (d_min). And then there is one 8
> bit register for the signal low period (lp) and then another 8bit register
> for the high period (hp). Per my understanding a change of polarity is then
> just a swap of lp and hp.

that doesn't sound right.

A "normal" waveform with period = 10 ns and duty_cycle = 2 ns looks as
follows:

   _         _         _
  / \_______/ \_______/ \_______/ 
  ^         ^         ^         ^

assuming a monospace font that's 1 char per ns, the ^ marking the period
start.

Ignoring scaling, your hardware needs to have hp = 2 and lp = 8. If you
switch that (assuming you mean switching in the same way as I do) to hp
= 8 and lp = 2, you get:

   _______   _______   _______
  /       \_/       \_/       \_/
  ^         ^         ^         ^

which is still a "normal" polarity output as a period starts with the
active part.

I admit that's a bit artificial, because the waveform for

	period = 10 ns
	duty_cycle = 2 ns
	polarity = inversed

looks as follows:

     _______   _______   _______
  \_/       \_/       \_/       \_/
  ^         ^         ^         ^

which isn't any different from the 2nd waveform above if you ignore the
period start markers (which are not observable apart from the moments
where you reconfigure the output).

However it matters if you have a chip with >1 output that are not
independent.

> The period is d_min * (lp + hp) and duty_cycle (on time) is then d_min*hp
> (per my understanding of the linux api). This means that there can be
> different settings that result in the same pwm signal (if you double d_min
> and halving lp and hp the signal should be the same).

Sounds correct.

> This means that when requesting a period and duty cycle you need to search
> through the configuration space to find the optimal value.

Or restrict yourself consistently to something simpler than a exhaustive
search through the complete configuration space.

> MvH

(BTW, I had to research the meaning of MvH. In case someone else doesn't
know it: It's the usual abbreviation for "Med vänliga hälsningar" in
Sweden or "Med vennlig hilsen" in Norway; both meaning "With friendly
greetings".)

Best regards
Uwe
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 488 bytes
Desc: not available
URL: <http://lists.infradead.org/pipermail/linux-mediatek/attachments/20240905/b9446535/attachment.sig>