Re: [PATCH v4 4/4] pwm: meson: make full use of common clock framework

[Heiner Kallweit <hkallweit1@gmail.com>]

On 14.04.2023 21:39, Martin Blumenstingl wrote:
> Hello Heiner,
> 
> On Thu, Apr 13, 2023 at 7:55 AM Heiner Kallweit <hkallweit1@gmail.com> wrote:
> [...]
>> Tested-by: Martin Blumenstingl <martin.blumenstingl@googlemail.com>
> Unfortunately I have some bad news and I need to take back my Tested-by :-(
> Previously my test was: cycle through all available CPU frequencies
> while stressing the CPU.
> My assumption was: if the system doesn't lock up everything's fine
> because we have a high enough voltage.
> 
> This evening however I got a memory corruption error while trying to
> log in via UART - which I thought was strange.
> So I connected my logic analyzer to my Odroid-C1 and did some experiments:
> 
> period = 30518, duty cycle = 15259 (typically used for the 32kHz
> output to the SDIO wifi chip)
> before your patches / after applying your patches:
> PWM: duty cycle: 50.000000% / 50.000000%
> PWM: period: 30.6 µs / 30.5 µs
> Timing: Time: 15.292 µs (65.395 kHz) / 15.250 µs (65.574 kHz)
> Timing: Average: 15.296 µs (65.377 kHz) / 15.264 µs (65.513 kHz)
> driver debug messages with your patches applied:
> fin_freq: 850000000 Hz
> period=30518 cnt=25940
> duty=15259 duty_cnt=12970
> 
> Then I tried period = 12218, duty cycle = 0 (typically used for the
> highest CPU voltage):
> before your patches / after applying your patches:
> PWM: duty cycle: 0.338983% / n/a (constant low output)
> PWM: period: 12.3 µs / n/a
> Timing: Time: 12.250 µs (81.633 kHz) / n/a
> Timing: Average: 6.148 µs (162.668 kHz) / n/a
> driver debug messages with your patches applied:
> fin_freq: 850000000 Hz
> period=12218 cnt=10385
> 
With a 850MHz input clock we should see a 0.01% duty cycle with 1.2ns
clock pulses. Can we rule out an issue with the measuring equipment?
Is your logic analyzer able to display such short clock pulses?

> Finally I tried period = 12218, duty cycle = 12218 (typically used for
> the lowest CPU voltage):
> before your patches / after applying your patches:
> PWM: duty cycle: 99.661017% / n/a (constant low output)
> PWM: period: 12.3 µs / n/a
> Timing: Time: 12.250 µs (81.633 kHz) / n/a
> Timing: Average: 6.148 µs (162.668 kHz) / n/a
> driver debug messages with your patches applied:
> fin_freq: 850000000 Hz
> period=12218 cnt=10385
> 
Here I have no idea yet.

> After seeing the constant low output with period 12218 I realized that
> my previous test was no good: the CPU was fed the highest possible
> voltage all the time.
> It's not clear to me why period 12218 would give no PWM output at all
> while period 30518 works fine.
> I did an experiment by removing CLK_SET_RATE_PARENT from the divider's
> init.flags -> now XTAL (24MHz) is the only possible clock (it's the
> hardware default). It does indeed bring back the exact same results as
> before (where the XTAL clock was also used; with the changes from this
> series FCLK_DIV3 is now chosen, which runs at 850MHz).
> 
> Do you have any idea what could cause this?
> The FCLK_DIV3 input seems to work as otherwise period 30518 would also not work.
> The calculated values also look sane, so it's not that we have some
> 32-bit overflow (as I'm testing on a 32-bit Meson8b SoC).
> 
At first I'd like to verify that the registers have the expected values.
Can you provide the values of PWM_A/B (depending on which channel is used in your
case) and PWM_MISC_AB at the end of meson_pwm_enable()? Thanks!

> 
> Best regards,
> Martin

Heiner