Re: [PATCH v3] sched/fair: bring back select_idle_smt, but differently

[Peter Zijlstra]

On Wed, Apr 07, 2021 at 11:47:17AM +0100, Mel Gorman wrote:

> Ok, cpusets do split domains. I can't imagine the logic of splitting SMT
> siblings across cpusets but if it's possible, it has to be checked and
> protecting that with cpusets_enabled() would be a little overkill and
> possibly miss some other corner case :(

Quite. The logic is that some people can't be arsed to look at how the
topology is enumerated and simply split logical CPUs by a random number.

Imagine we have 4 cores, with 2 threads each, for 8 logical CPUs.

Suppose you want a partition of 6 'cpus' and 2 spares. Hoping for a 3:1
core split.

If things are enumerated like: core0{smt0, smt1}, core1{smt0, smt1} ...
then 0-5 will get you 3 whole cores.

If OTOH things are enumerated like: smt0{core0, core1, core2, core3},
smt1{...} then 0-5 will get you the loonie side of the moon.

Funny thing is that x86 can iterate either way depending on how the BIOS
monkey filled out the tables.