feat(embedded-hal-bus): Spi - Add support for cs to clock delay

[vpochapuis]

What

Add support for cs to clock delay in embedded-hal-bus spi's device implementation.

Resolves https://github.com/rust-embedded/embedded-hal/issues/539

Why

This feature is defined at https://docs.rs/embedded-hal/latest/embedded_hal/spi/index.html#cs-to-clock-delays and issued at https://github.com/rust-embedded/embedded-hal/issues/539

Some chips that communicates through SPI necessitate this feature in order to correctly communicate.

How

Added a field in the device structs in order to store the cs_to_clock_delay, with its adequate constructor parameter. The delay is applied in the shared transaction implementation, just after the CS is toggled.

Notes

• I might have misunderstood where the clock is being controlled, I would need help if that's the case
• This change is breaking as the API to create the device is different, should this be avoided?
• I didn't see the usage of core::time::Duration in the code, I would like to know if its a bad practice in embedded system implementation or not and why

[Dirbaio]

Thanks for the PR! This is a welcome addition, just a few notes:

• Some SPI chips also need a delay after the data, between the last clock and deasserting CS. Could you add that too? I think it could be a separate setting so you can set them to different values.
• Perhaps the delay could default to 0 and be set with a .set_cs_to_clock_delay_ns() setter instead. This has two advantages: it keeps the new() signature simple for the simple case when you don't need delays, and is backwards-compatible.

[vpochapuis]

Thanks for the PR! This is a welcome addition, just a few notes:

* Some SPI chips also need a delay after the data, between the last clock and deasserting CS. Could you add that too? I think it could be a separate setting so you can set them to different values.

* Perhaps the delay could default to 0 and be set with a `.set_cs_to_clock_delay_ns()` setter instead. This has two advantages: it keeps the `new()` signature simple for the simple case when you don't need delays, and is backwards-compatible.

Thank you for the feedback! I will do my best to apply those.

Note: I have some issue testing this on STM32F439ZI, unrelated issue to this, about the SPI communication CLK pin behavior

If you ever encountered something similar and could give a quick tip

with the clock pin first bit floating for a while before starting normally under embassy_stm32:

Maybe related to https://github.com/embassy-rs/embassy/issues/1094 but I actually have no idea (timer prescaler config issue?), maybe you saw this in the past and could have quick tip?

I will continue by testing on ESP32C3 awaiting I fix my issue on the STM32.

[Dirbaio]

it might be a bug yes. please file an issue or join #embassy, we can discuss there.

[vpochapuis]

it might be a bug yes. please file an issue or join #embassy, we can discuss there.

Issue created and solve at https://github.com/embassy-rs/embassy/issues/3039

[SoAsEr]

I hate to revive this, but the documentation has now been asking to do something impossible for two years (https://github.com/rust-embedded/embedded-hal/commit/ff1f4695be29d7b3b80a79b6b5d41d493661299a). Is this in a state to be merged anytime?