[saagarjha]

God I wish people stopped sharing this thread because it’s full of things that just cause confusion. To respond to your point though it’s a per-thread flag that the kernel can flip for you if Apple had wanted it to. It’s probably better to just run the ARM code with full TSO, though; the performance penalty is like 10-15% for most code, which is probably less than making a system call into the kernel to be rescheduled with the right ordering whenever you wanted to switch. That doesn’t change the fact that the real challenge here is ABI, not ordering.