STM32F103 command line development

For any number of excellent reasons, I am violently opposed to using the Arduino GUI. So these notes are about using vim and make to do code development.

First Demo (never worked)

• Blinking LED using linux tools
• STM32 Discovery using linux tools
• ARM Cortex M3 projects

As a starting point, I poked around in the ARM Cortex M3 projects link above and found an entry "LED Blink Example for STM32F103" and downloaded STM32F103VHB6_RevZ_Demo1_Lanchon_20080210.zip. I unzip this, cd inside, and type make. It launches the first command:

arm-none-eabi-gcc -mcpu=cortex-m3 -mthumb -Wall -g -O0 -I . -I lib/inc   -c -o main.o main.c

What do I do? Can these compilers both be installed? Can I just modify the makefile to use the compiler I have been using? Why are there two compilers for ARM anyway?

I actually try modifying the makefile. And it mostly works (at least the compiler doesn't blow up). It complains about a couple of odd constructions that I can easily clean up. Then it gets really upset at link time.

It is also worth noting that there is a lot of code here; over 16,000 lines of code just to blink an LED. Well it is a comprehensive library, not a simple demo. There are a lot of things that I may be glad to use in the future.

Pick your compiler

arm-none-eabi-gcc-cs.x86_64              1:5.2.0-4.fc24                 fedora  
gcc-arm-linux-gnu.x86_64                 6.1.1-2.fc24                   updates 

Compare however "none" and "linux" in the two names. In the gcc naming scheme "none" indicates that the compiler does not expect a C library and underlying operating system, where "linux" indicates the compiler is expecting to generate code to be linked in the usual way to run under the linux operating system. This is the party line anyway, although I have been using "gcc-arm-linux-gnu" for bare-metal embedded projects for quite some time. In other words, all that has been said so far in this paragraph is confusing rubbish.

The real trick is in the linker script. In the "linux" flavor compiler, the default path will expect to find main() and will try to get crt0, crt1.o (and similar things) from libc. Embedded projects will have none of this. An embedded project will have a special linker script, some assembler code to get things going after reset, and often will be wise to avoid having main() just to avoid confusion.

All this being said, I don't see any assembler code in the demo file I downloaded. There must be magic hidden in the code sourcery compiler.

Can we install them both safely? Who knows? Let's cross our fingers and give it a try:

dnf install arm-none-eabi-gcc-cs
(1/2): arm-none-eabi-binutils-cs-2.25-3.fc24.x86_64.rpm                                                                                                                                                                                                                                       840 kB/s | 2.0 MB     00:02    
(2/2): arm-none-eabi-gcc-cs-5.2.0-4.fc24.x86_64.rpm       

arm-none-eabi-gcc -Wl,--gc-sections,-Map=main.elf.map,-cref,-u,Reset_Handler -I . -I lib/inc -L lib -T stm32.ld main.o stm32f10x_it.o lib/libstm32.a --output main.elf
/usr/lib/gcc/arm-none-eabi/5.2.0/../../../../arm-none-eabi/bin/ld: cannot find crt0.o: No such file or directory
/usr/lib/gcc/arm-none-eabi/5.2.0/../../../../arm-none-eabi/bin/ld: cannot find -lc
collect2: error: ld returned 1 exit status
Makefile:36: recipe for target 'main.elf' failed
make: *** [main.elf] Error 1

Second demo (sort of)