Cortex-M0 vs. M3 : LLVM and LD

One of the issues that you run into using Clang/LLVM as your compiler for bare-metal ARM Cortex cores is you have to directly use arm-none-eabi-ld to do your linking.

Directly using ld can be a bit nerve wrecking at times to get the options correct (and the order of options does matter) as normally you are just let gcc use collect2 and have it internally execute ld to perform your linking.

One of the areas using it directly that can bite you is not linking to the proper libgcc.a for the Cortex-M that you are targeting. Looking into your arm-none-eabi/lib/gcc/arm-none-eabi/X.X.X tool-chain directory and you will find multiple directories. One for each ARM architecture; armv6-m, armv7-ar, armv7-m, thumb, thumb2, etc…

Add a library include for architecture directory that matches the core that you compiled against and everything will be fine:

Cortex M3 example:

arm-none-eabi-ld -Map bin/main.axf.map -T src/cortex_M3.ld --library-path /Users/administrator/Code/llvm_superproject/install/arm-none-eabi/newlib-syscalls/arm-none-eabi/lib/thumb/thumb2 --library-path /Users/administrator/Code/llvm_superproject/install/arm-none-eabi/newlib-syscalls/arm-none-eabi/lib/thumb  --library-path /Users/administrator/Code/llvm_superproject/install/arm-none-eabi/newlib-syscalls/arm-none-eabi/lib  --library-path /Users/administrator/Code/llvm_superproject/install/arm-none-eabi/lib/gcc/arm-none-eabi/4.8.3/armv7-m -g   obj/printf_with_malloc.o obj/startup.o --start-group -lgcc -lc --end-group -o bin/main.axf

Cortex M0+ example:

arm-none-eabi-ld -Map bin/main.axf.map -T src/cortex_M0.ld --library-path /Users/administrator/Code/llvm_superproject/install/arm-none-eabi/newlib-syscalls/arm-none-eabi/lib/thumb/thumb2 --library-path /Users/administrator/Code/llvm_superproject/install/arm-none-eabi/lib/gcc/arm-none-eabi/4.8.3/armv6-m  --gc-sections --print-gc-sections  obj/printf_with_malloc.o obj/startup.o --start-group -lgcc -lc --end-group -o bin/main.axf

ARM Cortex-M instruction sets

ARM Cortex-M	Thumb	Thumb-2	Hardware multiply	Hardware divide	Saturated math	DSP extensions	Floating-point	ARM architecture	Core architecture
Cortex-M0^[1]	Most	Subset	1 or 32 cycle	No	No	No	No	ARMv6-M^[6]	Von Neumann
Cortex-M0+^[2]	Most	Subset	1 or 32 cycle	No	No	No	No	ARMv6-M^[6]	Von Neumann
Cortex-M1^[3]	Most	Subset	3 or 33 cycle	No	No	No	No	ARMv6-M^[6]	Von Neumann
Cortex-M3^[4]	Entire	Entire	1 cycle	Yes	Yes	No	No	ARMv7-M^[7]	Harvard
Cortex-M4^[5]	Entire	Entire	1 cycle	Yes	Yes	Yes	Optional	ARMv7E-M^[7]	Harvard

SushiHangover

PowerShell, Learn it or Perish ;-)

Cortex-M0 vs. M3 : LLVM and LD

Comments