11 Jul
2018
11 Jul
'18
10:55 a.m.
On Wed, Jul 11, 2018 at 10:04:52AM +0000, David Laight wrote:
I also suspect that 'write starvation' is also common - after all the purpose of the store buffer is to do reads in preference to writes in order to reduce the cpu stalls waiting for the memory bus (probably the cpu to cache interface).
I think your example is just: *(volatile int *)xxx = 1; while (!*(volatile int *)yyy) continue; running on two cpu with xxx and yyy swapped?
Yep. And Linux has been relying on that working for (afaict) basically forever.
You need a stronger bus cycle in there somewhere.
Since all spin-wait loops _should_ have cpu_relax() that is the natural place to put it.