linux-kernel - Re: [RFC PATCH 6/6] sched/dl: Try not to select a too fast core

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]

Message-ID: <20190507155732.7ravrnld54rb6k5a@queper01-lin>
Date:   Tue, 7 May 2019 16:57:34 +0100
From:   Quentin Perret <quentin.perret@....com>
To:     Luca Abeni <luca.abeni@...tannapisa.it>
Cc:     linux-kernel@...r.kernel.org,
        Greg Kroah-Hartman <gregkh@...uxfoundation.org>,
        "Rafael J . Wysocki" <rafael@...nel.org>,
        Ingo Molnar <mingo@...hat.com>,
        Peter Zijlstra <peterz@...radead.org>,
        Vincent Guittot <vincent.guittot@...aro.org>,
        "Paul E . McKenney" <paulmck@...ux.ibm.com>,
        Joel Fernandes <joel@...lfernandes.org>,
        Luc Van Oostenryck <luc.vanoostenryck@...il.com>,
        Morten Rasmussen <morten.rasmussen@....com>,
        Juri Lelli <juri.lelli@...hat.com>,
        Daniel Bristot de Oliveira <bristot@...hat.com>,
        Patrick Bellasi <patrick.bellasi@....com>,
        Tommaso Cucinotta <tommaso.cucinotta@...tannapisa.it>
Subject: Re: [RFC PATCH 6/6] sched/dl: Try not to select a too fast core

On Monday 06 May 2019 at 06:48:36 (+0200), Luca Abeni wrote:
> From: luca abeni <luca.abeni@...tannapisa.it>
> 
> When a task can fit on multiple CPU cores, try to select the slowest
> core that is able to properly serve the task. This avoids useless
> future migrations, leaving the "fast cores" idle for more heavyweight
> tasks.

But only if the _current_ capacity of big CPUs (at the current freq) is
higher than the current capacity of the littles, is that right ? So we
don't really have a guarantee to pack small tasks on little cores ...

What is the rationale for looking at the current freq in dl_task_fit() ?
Energy reasons ? If so, I'd argue you should look at the energy model to
break the tie between CPU candidates ... ;)

And in the mean time, you could just look at arch_scale_cpu_capacity()
to check if a task fits ?

> Signed-off-by: luca abeni <luca.abeni@...tannapisa.it>
> ---
>  kernel/sched/cpudeadline.c | 17 ++++++++++++-----
>  1 file changed, 12 insertions(+), 5 deletions(-)
> 
> diff --git a/kernel/sched/cpudeadline.c b/kernel/sched/cpudeadline.c
> index 2a4ac7b529b7..897ed71af515 100644
> --- a/kernel/sched/cpudeadline.c
> +++ b/kernel/sched/cpudeadline.c
> @@ -143,17 +143,24 @@ int cpudl_find(struct cpudl *cp, struct task_struct *p,
>  	       struct cpumask *later_mask)
>  {
>  	const struct sched_dl_entity *dl_se = &p->dl;
> +	struct cpumask tmp_mask;

Hmm, these can get pretty big, so not sure about having one on the stack ...

>  
>  	if (later_mask &&
> -	    cpumask_and(later_mask, cp->free_cpus, &p->cpus_allowed)) {
> +	    cpumask_and(&tmp_mask, cp->free_cpus, &p->cpus_allowed)) {
>  		int cpu, max_cpu = -1;
> -		u64 max_cap = 0;
> +		u64 max_cap = 0, min_cap = SCHED_CAPACITY_SCALE * SCHED_CAPACITY_SCALE;
>  
> -		for_each_cpu(cpu, later_mask) {
> +		cpumask_clear(later_mask);
> +		for_each_cpu(cpu, &tmp_mask) {
>  			u64 cap;
>  
> -			if (!dl_task_fit(&p->dl, cpu, &cap))
> -				cpumask_clear_cpu(cpu, later_mask);
> +			if (dl_task_fit(&p->dl, cpu, &cap) && (cap <= min_cap)) {
> +				if (cap < min_cap) {
> +					min_cap = cap;
> +					cpumask_clear(later_mask);
> +				}
> +				cpumask_set_cpu(cpu, later_mask);
> +			}
>  
>  			if (cap > max_cap) {
>  				max_cap = cap;
> -- 
> 2.20.1
> 

Thanks,
Quentin