RP2040 tweaks

[liamfraser]

RP2040 tweaks:

• Compiler warning fixes
• allow pico-sdk LWIP to be used if TinyUSB one does not exist
• Use custom memcpy in RP2040 code
• Don't clear registers when not needed as hardware reset already does it

[hathach]

look good, thank you

[earlephilhower]

I know this has been merged, but replacing memcpy with a local implementation makes no sense to me.

memcpy, by definition, does not have any alignment restrictions (but of course you need memmove if there's potential overlap). I would suggests that any memcpy that crashes on a non-4-byte-aligned input or output is busted or should not be called memcpy because lots of other things are going to break, not just USB.

None of the man pages (ISO C spec is $$$) I've seen show an alignment requirement.

While the implementation here of unaligned_memcpy looks like it works (but doesn't return the almost always unused pointer that ISO memcpy does), it feels like it's going to be slower than whatever GCC or the ROM would implement...

edit There are also lots of memcpy calls in the class drivers as well. So, only a couple memcpy calls out of a whole bunch are being replaced w/this specific commit...

[hathach]

@earlephilhower yeah, thank you for your feedback. You are rigtht, this is my thought as well initially. However, since it copy to/from a dual port USB RAM, it may have some extra hw limit, we have seen this in other ports e.g #1024, though that is a bit different since MCU is M4 and able to do unaligned access (hence memcpy is implemented in such as way), however the usb ram is limited to byte/aligned access only which could cause issue.

Update: It is confirmed that this does actually fixed an actual hw issue under certain testing condition.

[earlephilhower]

Gotcha. Thanks to all for looking into it.

[hathach]

Gotcha. Thanks to all for looking into it.

no problems, thank you for providing feedback on the PR.