Although I am not entirely convinced this is necessary, as the SPI data stall issue only appeared with optimisation off (-O0). Perhaps re-visit this if Upstream needs more free CPU time later...