How do I configure newlib printf?

[Peter]

Im using GCC + Eclipse with the S110 SoftDevice and the HRS example code. Everything is working well except for printf (which I've added a call to via app_trace). I've set the ENABLE_DEBUG_LOG_SUPPORT #define. I've also called app_trace_init().

I get a Hard_Fault when I call app_trace. Stepping through the assembly:

1. printf is entered.
2. printf attempts to dereference _impure_ptr (part of the thread support for errno in newlib), but _impure_ptr points to 0x8000222, which I believe is in the region of the SoftDevice and not readable.
3. At this point, a Hard_Fault occurs.

Does newlib need to be initialized (i.e. do I need to call _init or _fini somewhere)? Has anyone else used printf with GCC + newlib?

I'm using the gcc_nrf51_s110_xxac.ld linker script that comes with the SDK--so my memory and flash regions should be correct.

[pabigot]

BSPACM provides some information about its newlib interface. I use printf and related stdio functions on nRF51 hardware with this solution.

In short you need to provide implementations of the system functions newlib expects. (Be aware that the nRF51 SDK, like many other ARM Cortex-M vendor libraries, hides a reference to -lnosys in a GROUP directive in the linker script. You'll need to get rid of that.)

[Peter]

But doesn't the nosys library just provide stubs for functions that are undefined? i.e., since retarget.c defines _read and _write, those versions will be used instead of the stubs in nosys. I can test this by commenting out _write in retarget.c--when I print the symbols with nm, _write comes from somewhere else (nosys I assume) and _read comes from retarget.c. Can't I let nosys define stubs for functions I don't care about like _exit?

[pabigot]

Yes. I was unaware of the retarget.c feature since I don't use anything but the CMSIS headers from the Nordic SDK.

I've also run into trouble where two things (e.g. retarget.o and -lnosys) define the same symbol, especially if they both come from libraries, as the order the libraries are read by the linker becomes important. In BSPACM I use weak symbols for the stubs to avoid some of these problems.