rp2: native C modules cause crash

[peterhinch]

To replicate compile the features0.c demo for armv7m. The resultant features0.mpy imports OK, but any attempt to run it e.g.

import features0
features0.factorial(3)

results in a crash.

[kadamski]

Makes sense. RP Pico is Cortex-M0+ which is armv6m, not armv7m (cortex-m3). If you do a quick hack and modify py/dynruntime.mk to use -mcpu=cortex-m0 instead of -mcpu=cortex-m3 in case of armv7m, it works without a crash.

[kadamski]

I was looking into fixing this problem properly but it is quite complicated, from what I've seen. But at least we know the root cause.

[pigrew]

An example diff is posted on the RPi forums. It works for me.

[jimmo]

See also #7616 (comment) for a related issue.

Fortunately this one is easier to work around!

[pigrew]

Ok... less great news. With the patch I found, I'm getting linker errors when I try to call mp_get_buffer_raise(data_obj, &bufinfo, MP_BUFFER_READ) from my module:

GEN build/fasti2c.config.h
CC fasti2c.c
LINK build/fasti2c.o
LinkError: build/fasti2c.o: undefined symbol: memset
make: *** [../micropython/py/dynruntime.mk:147: build/fasti2c.native.mpy] Error 1

I'm using the ARM "GNU Arm Embedded Toolchain 10 2021.07" on Windows through msys2/mingw.

[dpgeorge]

Since c1b9d22, features0 and features1 natmod examples can build for armv6m and run on the rp2 port.