stm32_common: Add USB OTG FS/HS usbdev peripheral driver

[bergzand]

Contribution description

This PR adds an USBdev driver for the ST (synopsys). It is used with the STM32F2, STM32F4, STM32F7 and STM32H7 devices. The STM32F1 and STM32F3 devices unfortunately use a different device type.

The goal is to support the both the full speed and the high speed version in a single driver, the differences between the two are minimal and drivers from other project are able to combine the two instances too.

The peripheral itself has a 'Core ID' register containing a version number-like value. So far I've identified 2 different iterations for both the FS and the HS devices. They have slightly different register flags and behave slightly different in some cases. The crude method to detect the version is to check if certain register flag definitions are defined.
It is assumed that when both the FS and the HS peripheral is available on a single device, both are of the same version.

• CID 0x1x00: 4 FS endpoints, 6 HS endpoints (f401, f411…)
• CID 0x2x00: 6 FS endpoints, 9 HS endpoints (f446, f7xx…)

Testing procedure

• examples/usbus_minimal should work when adding xtimer.
• tests/usbus_cdc_ecm should work with cdcecm: only activate OUT endpoint after interface selection #12533 cherry-picked.
• tests/usbus_cdc_acm should work with cdcacm: Activate data out endpoint on DTE present signal #12536 cherry-picked.

Issues/PRs references

One of the remaining issues with the driver is that it requires xtimer (or a different solution) to sleep between one of the register modifications.

depends on #12534, #12533, #12536 and #12579

TODO:

• Add xtimer dependency when periph_usbdev is used.
• Add usbdev feature to non-nucleo boards
• Test with the HS peripheral in FS mode
• Add DMA support for the HS peripheral
• Add support for the non-nucleo boards
• Migrate DMA enable/disable to a #define
• Test isochronous transfers

Note

@chrysn, @basilfx: FYI The EFM32 devices seem to use an identical IP block from Synopsys, only the serialization phy seems to be different.

[bergzand]

Almost done here, I've tested this myself on the nucleo-f401re and the nucleo-f446re with a jumper wired USB cable. Both CID versions should work and both the FS and the HS peripheral work (in FS mode with the built-in phy). With the HS phy, both DMA and non-DMA is functional.

I have not yet tested isochronous transfers.

[bergzand]

Ready for review! I won't be force pushing anymore to this branch unless instructed otherwise.

I'm going to test isochronous transfers at some point, but I won't consider it blocking for this PR as it is not used within RIOT-provided USB functionality at the moment.

Still waiting for #12579 though. :)

[bergzand]

I'm carefully going to claim that isochronous transfers work. I have a small test setup where I can at least get the events from the transfer and receive events for the transfers. I also receive correct transfer lengths, I'm only haven't tested the content of the transfer yet.

[bergzand]

Rebased now that #12579 is merged, no longer waiting for other PRs

[bergzand]

Squashed after offline approval from @aabadie.

[aabadie]

One minor nit remaining.

Please squash! @dylad, do you still have comments ?

[aabadie]

Suggested change

	USEMODULE += xtimer
	USEMODULE += xtimer

Looks like there's a tab here

[bergzand]

😲 Fixed!

[dylad]

All good on my side.

[aabadie]

ACK! Now let's wait for Murdock (which is quite busy)

[aabadie]

All green, let's go !

[bergzand]

Thanks!

[dylad]

Good job with this PR @bergzand !