Trying to understand instruction cache (for repeated Jit compilation reusing memory pages)

The answer is simple, I can just use the builtinclearcache intrinsic (available in both gcc and clang).
For C compilers that don't have this intrinsic, I can use the clearcache libgcc function (tested with tcc on the raspberry pi 5).

Built-in Function: void builtinclearcache (void *begin, void *end)

This function is used to flush the processor’s instruction cache for the region of memory between begin inclusive and end exclusive. [...]

If the target does not require instruction cache flushes, builtinclearcache has no effect. Otherwise either instructions are emitted in-line to clear the instruction cache or a call to the clearcache function in libgcc is made.

With this I can set OFFSET2 as low as sizeof(instr_ret42) and it just works fine.