Re: [linus:master] [sched/fair] 829c1651e9: hackbench.throughput -7.9% regression

[Chen Yu]

On 2023-03-09 at 11:46:50 +0800, kernel test robot wrote:
> Greeting,
> 
> FYI, we noticed a -7.9% regression of hackbench.throughput due to commit:
> 
> commit: 829c1651e9c4a6f78398d3e67651cef9bb6b42cc ("sched/fair: sanitize vruntime of entity being placed")
> https://git.kernel.org/cgit/linux/kernel/git/torvalds/linux.git master
> 
> in testcase: hackbench
> on test machine: 104 threads 2 sockets (Skylake) with 192G memory
> with following parameters:
> 
> 	nr_threads: 100%
> 	iterations: 4
> 	mode: process
> 	ipc: socket
> 	cpufreq_governor: performance
>
In theory the patch deals with a corner case that the task sleeps for more than 1 minute,
but this is not the case for hackbench, and it did increase the preemption:

 1.08e+08 ±  2%    +648.7%  8.084e+08 ±  5%  hackbench.time.involuntary_context_switches 

It might be an issue described and fixed here:
https://lore.kernel.org/lkml/20230306132418.50389-1-zhangqiao22@xxxxxxxxxx/
https://lore.kernel.org/lkml/ZAiFxWLSb9HDazSI@vingu-book/
I'll check if above patch helps.

thanks,
Chenyu