Add all SPI examples for the ESP32-H2

[jessebraham]

Adds all SPI-related examples for the ESP32-H2. This involved updating the GDMA and SPI peripheral drivers.

This was verified on hardware by @SergioGasquez and @JurajSadel, however I committed some git crimes since then so probably wouldn't hurt just to triple-check the examples.

[bjoernQ]

Maybe it would make sense to change the pins - at least on my boards GPIO7 doesn't seem to be available on the pin headers

[SergioGasquez]

Maybe it would make sense to change the pins - at least on my boards GPIO7 doesn't seem to be available on the pin headers

Agree, here is the diff my pin changes that I've used for testing: pins.zip

[SergioGasquez]

Btw, just tested the examples again and all of them seem to be working!

[jessebraham]

Maybe it would make sense to change the pins - at least on my boards GPIO7 doesn't seem to be available on the pin headers

Everybody seems to have different dev kits, I already have to change the pins for a bunch of examples in the repo so I guess I don't really care what they're set to. If you want them changed please tell me which pins to change them to.

[bjoernQ]

Ah yes GPIO 6 is also not available.

Also the clock seems to be slightly off:

The spi_loopback.rs example requests 100u32.kHz()

Also I guess we could add qspi_flash.rs (I have one of those flash chips so I can verify that)

[bjoernQ]

Yep qspi_flash.rs works - I used these pins

    let sclk = io.pins.gpio1;
    let miso = io.pins.gpio2;
    let mosi = io.pins.gpio3;
    let sio2 = io.pins.gpio4;
    let sio3 = io.pins.gpio5;
    let cs = io.pins.gpio11;

[jessebraham]

Okay thanks, I'll update each example to use those pins and add the missing example.

[SergioGasquez]

I've been investigating the frequency issue, here is what I've found:

I was able to solve it by using the XTAL clock (changing to 0 in https://github.com/esp-rs/esp-hal/pull/549/files#diff-13231643cd8aa34e91c614e3f5280962f1e3f2c53a8fa3012786d95865eaf3e6R2149):

Looking at esp-idf source, H2 and C6 default clock is the PLL and, the select method is the same for H2 and C6

The thing is, according to the TRM if we want to use the PLL clock, 1 should be correct, although it's 48M in C2 and 80M in C6.

@bjoernQ tested (my C6 devkit is not working for some reason) if the frequency issue was also present in C6, since it should be the same, but C6 works fine.

[jessebraham]

I've rebased and updated the SPI clock configuration as per the above comment. @SergioGasquez would you mind please verifying that things are now working correctly?

[SergioGasquez]

I've rebased and updated the SPI clock configuration as per the above comment. @SergioGasquez would you mind please verifying that things are now working correctly?

Just tried the spi_lookback example with the recent changes and freq seems to be working fine:

[bjoernQ]

but using XTAL_CLK would mean we limit the max SPI speed 🤔

[bjoernQ]

XTAL_CLOCK works because there is this

let apb_clk_freq: HertzU32 = HertzU32::Hz(clocks.apb_clock.to_Hz());

And we have

           desired_rates: RawClocks {
                cpu_clock: HertzU32::MHz(96),
                apb_clock: HertzU32::MHz(32),
                xtal_clock: HertzU32::MHz(32),
                i2c_clock: HertzU32::MHz(32),
            },

So, the APB_CLOCK of 32MHz matches the XTAL_CLK

[jessebraham]

I've fixed the comment and rebased this branch again.

Given that the SPI data transfer seems to be working (as indicated by the success of the various SPI examples) and that we have a lot of clock-related issues across the board for the H2, I think this is probably okay to merge as-is unless anybody has any major objections.

[bjoernQ]

Probably doesn't make sense to further block this but we should probably create an issue to remind us to re-think the choose of the clock source here?

[JurajSadel]

I'm doing some testing based on internal discussion, please don't merge this yet

[JurajSadel]

When CI is green this should be ready for final review. I tested all examples with LA but QSPI, @bjoernQ could you please try that?

[bjoernQ]

qspi_flash example works. Also using pll_48m_clock is the right fix for this 👍 This would also be needed to fix TIMG

[bjoernQ]

LGTM

[bjoernQ]

Frequency seems fine

[SergioGasquez]

Changes LGTM, although I couldn't verify the freq as my LA decided not to work