On: Those Win9x Crashes on Fast Machines

[acqq]

> I've seen code that makes assumptions about how much real time must have passed based on PAUSE loops.

Note: here, the PAUSE instruction is not the problem at all, but the "code that makes assumptions."

Because the "seen" code is not named, I assume it's something internal for some company?

[jeffbee → acqq]

Yes, private code. Basically there was some mutex fairness thing that was written on a SKX and on a Zen CPU where PAUSE is 50x faster it didn't have good fairness, it was too tight.