Re: [PATCH-cgroup 1/2] cgroup/cpuset: Account for boot time isolated CPUs

[Waiman Long <longman@redhat.com>]

On 8/27/24 04:01, Michal Koutný wrote:
> Hi.
>
> On Tue, Aug 20, 2024 at 03:55:35PM GMT, Waiman Long <longman@redhat.com> wrote:
>> The prstate_housekeeping_conflict() function does check the
>> HK_TYPE_DOMAIN housekeeping cpumask to make sure that CPUs outside of it
>> can only be used in isolated partition.
>> Given the fact that we are going to make housekeeping cpumasks
>> dynamic, the current check may not be right anymore. Save the boot
>> time HK_TYPE_DOMAIN cpumask and check against it instead of the
>> upcoming dynamic HK_TYPE_DOMAIN housekeeping cpumask.
> Why is (will be) checking against the stored HK_TYPE_DOMAIN mask correct
> when this mask becomes dynamic?

In term of isolated CPUs, there are 2 categories - static and dynamic. 
Statically isolated CPUs are those that are designed as isolated at boot 
time by "isolcpus". Other isolated CPUs created by the cpuset isolated 
partitions are considered dynamic in the sense that its state can change 
at run time. The degree of CPU isolation of dynamically isolated CPUs 
isn't as good as that of the statically isolated ones. So I want to 
handle them separately which is what the prstate_housekeeping_conflict() 
is intended to do.

As it is my intention to make the housekeeping cpumasks dynamic, I need 
to keep a copy of the statically isolated CPUs and check against them. 
There is no point to check dynamically isolated CPUs as the test may 
produce a false positive result.

In the future when dynamic CPU isolation is as almost as good as the 
static ones, we can get rid of this distinction and treat all of them as 
dynamic.

Cheers,
Longman