[Devel] [PATCH vz10 v4 1/3] sched: Clean up vCPU handling logic

[Konstantin Khorenko]

On 3/19/26 10:47, Dmitry Sepp wrote:
> The idea behind the change is to transition from the existing spatial
> vCPU handling approach that introduces costly modification to the
> scheduling logic to ensure the requested CPU count is obeyed (10%+
> performance drop in some tests) to temporal isolation that can be
> provided by the cgroup2 cpu.max.
> 
> Drop the legacy unneeded vCPU handling code. Remove the 'cpu.rate'
> control in favor of the internal calculation based on 'quota' and
> 'period' from 'cpu.max'. As 'cpu.max' is not implicitly used to set the
> rate, do not override nr_cpus when handling writes to 'cpu.max'.
> 
> https://virtuozzo.atlassian.net/browse/VSTOR-124385
> 
> Signed-off-by: Dmitry Sepp <dmitry.sepp at virtuozzo.com>
> ---
>   include/linux/sched.h          |   6 -
>   include/linux/sched/topology.h |   5 -
>   kernel/sched/core.c            |  98 +-------
>   kernel/sched/fair.c            | 408 ---------------------------------
>   kernel/sched/sched.h           |  10 -
>   5 files changed, 12 insertions(+), 515 deletions(-)
> 
...

>   
> -static int tg_set_cpu_limit(struct task_group *tg,
> -			    unsigned long cpu_rate, unsigned int nr_cpus)
> +static int tg_set_cpu_limit(struct task_group *tg, unsigned int nr_cpus)
>   {
>   	int ret;
>   	unsigned long rate;
> +	unsigned long cpu_rate = tg->cpu_rate;
>   	u64 quota = RUNTIME_INF;
>   	u64 burst = tg_get_cfs_burst(tg);
>   	u64 period = default_cfs_period();
> @@ -10090,21 +10041,6 @@ static int tg_set_cpu_limit(struct task_group *tg,
>   	return ret;
>   }
>   

static int tg_set_cpu_limit(struct task_group *tg, unsigned int nr_cpus)
{
         int ret;
         unsigned long rate;
         unsigned long cpu_rate = tg->cpu_rate;
         u64 quota = RUNTIME_INF;
         u64 burst = tg_get_cfs_burst(tg);
         u64 period = default_cfs_period();

         rate = (cpu_rate && nr_cpus) ?
                 min_t(unsigned long, cpu_rate, nr_cpus * MAX_CPU_RATE) :
                 max_t(unsigned long, cpu_rate, nr_cpus * MAX_CPU_RATE);
         if (rate) {
                 quota = div_u64(period * rate, MAX_CPU_RATE);
                 quota = max(quota, min_cfs_quota_period);
         }

         cpus_read_lock();
         mutex_lock(&cfs_constraints_mutex);
         ret = __tg_set_cfs_bandwidth(tg, period, quota, burst);
         if (!ret) {
                 tg->cpu_rate = cpu_rate;
// the line above is a no-op.
// i will apply the patch as is but in case this was not intended, please fix it with an incremental patch


                 tg->nr_cpus = nr_cpus;
         }
         mutex_unlock(&cfs_constraints_mutex);
         cpus_read_unlock();

         return ret;
}