Re: [RFC PATCH v1] stop_machine: Add stop_housekeeping_cpuslocked()

[Frederic Weisbecker]

Le Mon, Jan 06, 2025 at 03:08:11PM +0200, Costa Shulyupin a écrit :
> On Wed, 18 Dec 2024 at 19:38, Frederic Weisbecker <frederic@xxxxxxxxxx> wrote:
> > And so given the cost of such analysis and resulting possible patches, here
> > is an important question: is it worth the effort? What is the usecase of
> > shutting down a CPU while other isolated CPUs run critical isolated stuff?
> 
> The goal is to implement dynamic CPU isolation for realtime tasks such as DPDK.
> 
> Thomas Gleixner in https://lore.kernel.org/lkml/87bjz2210r.ffs@tglx/ suggested:
> "CPU hotplug solves this problem without any hackery. Take a CPU offline,
> change the mask of that CPU and bring it online again. Repeat until all
> CPU changes are done."
> 
> Unfortunately, CPU HP interferes with realtime tests and is
> unsuitable for dynamic CPU isolation.
> 
> Meanwhile, the maximum number of hyperthreads is climbing to 1024,
> increasing the demand for CPU HP.

I must confess I don't understand well your constraints. Why would you
change the set of isolated CPUs while running realtime tests (btw. did you mean
"tasks"?).

Do I understand it correctly that your server may run different kinds of
workloads concurrently, some of them isolated and some of them not, and these
workloads may be added / removed concurrently _anytime_? And therefore
a newly added isolated workload (which then adds CPUs to the isolated set)
mustn't disturb unrelated already running isolated workloads?

When you refer to realtime tasks, do you mean isolated?

Thanks.

> 
> Thank you,
> Costa
>