Support eMMC devices via SDMMC (STM32)

[antonok-edm]

I've added support for eMMC memory devices in embassy-stm32 using the SDMMC peripheral.

I've confirmed that this code works correctly for basic initialization, reads, and writes on a device that I've previously been using under stm32f4xx-hal, using 1, 4, and 8 data lanes, each with several different operating frequencies up to 24MHz.

Note that I don't have any appropriate hardware to test:

• gpio_v1 config
• sdmmc_v2 config
• SD card regressions

[Dirbaio]

Thanks for the PR!

Could you not add a new trait + generic param for the mode, track it at runtime instead (with e.g. an enum?). Reasons are:

• There's code duplication in init_emmc/init_card and read_sd_status/read_ext_csd. With an enum you can easily "if emmc do this, if sd do that" in the few things that do change without having to copypaste the entire methods.
• Less generics makes the API simpler for the user.

[antonok-edm]

@Dirbaio sure, I'll try that. Poor timing on my part - right after I submitted this PR, I already had to rebase it because of #4005 doing the same thing for DMA 😅

[antonok-edm]

@Dirbaio I just pushed a new enum-based implementation in case you'd like to generally check over it. There are a couple of things that I think are slightly less clean than with traits but it's not too bad overall. I didn't attempt to address the code duplication you mentioned yet; will get to that tomorrow hopefully.

(previous implementation is still in antonok-edm/embassy#emmc-trait if needed)

[antonok-edm]

@Dirbaio ok, should be ready for review now - I've deduplicated the init methods and pulled common code out from all the block read/write functions into complete_datapath_transfer.

[Dirbaio]

thank you!

[Dirbaio]

Seems this breaks the sdcard test

bender run

[Dirbaio]

hm I think the "task arena is full" error is a red herring. Task arena is gone since #4020. Try rebasing on latest main, you'll likely get the same error in PR CI as in the merge queue CI, which is "waiting for card: NoCard".

[antonok-edm]

@Dirbaio SD card CI is now fixed, but there is still a usart_rx_ringbuffered error which seems unrelated, even after rebasing on main. Not sure if you have any ideas there.

[Dirbaio]

yeah it's just flaky, i've removed it in #4054

bender run

[chemicstry]

Hey, thank you for implementing this right when I need it!

I, however, am having problems and would appreciate any help. I'm using STM32F429ZIT6 with EMMC16G-MW28-03022 flash on 8 bit bus and it just freezes on init_emmc(). Seems like an infinite loop as no other async tasks are being executed. What is strange, is that when I set breakpoints anywhere in the code it manages to initialize the eMMC and reports correct capacity. Any read after that fails with panicked at embassy-stm32\src\dma\dma_bdma.rs:289:21: DMA: error on DMA@40026400 channel 6. I tried using 1, 4 and 8 bit buses with no luck.

Have you experienced anything similar and could nudge me where to look?

EDIT: Nevermind, got it to work. It was some weird interraction with my logging buffer getting full and fatfs crashing due to empty emmc.