I see the GCC ARM port issue regarding thread unsafe function returns is still biting people. I find the use of -fomit-frame-pointer interferes with some debuggers' abilities to unwind the stack. Some time ago I suggested a FreeRTOS port fix for it. I have been using the fix successfully for some time now. It seems to take care of the issue. I think it should be in the official ARM port so that the problem doesn't keep coming back.

Here it is again (without explanation):
In the portSAVE_CONTEXT() macro of portmacro.h add the commented line in the position shown below

____/* Set R0 to point to the task stack pointer. */ \
________"STMDBSP,{SP}^ \n\t" \
________"NOP \n\t" \
________"SUBSP, SP, #4 \n\t" \
________"LDMIASP!,{R0} \n\t" \
\
/*gb - step over possible GCC stacked items (GCC bug)*/ \
________"SUBR0, R0, #8 \n\t" \

No response to this? Looks like a good idea.

Does this mean it uses an extra 8 bytes of RAM for each context? Should not be a problem I don't think.

Yes, eight bytes of stack space are wasted for each task, except the running task (and except for those tasks that were in the non thread-safe state when their context was saved). Also, the cost of a context switch goes up by the execution of one register-to-register instruction.