Re: [lkp] [x86/irq] 82ba4faca1: [No primary change] [unixbench]

[Aaron Lu]

On 08/24/2016 11:04 AM, kernel test robot wrote:
> 
> FYI, we noticed interrupts.CAL:Function_call_interrupts +771.1% improvement due to commit:

This is expected.

CAL:Function_call_interrupts now contains TLB shootdowns while it
doesn't previously.

Thanks,
Aaron

> 
> commit 82ba4faca1bffad429f15c90c980ffd010366c25 ("x86/irq: Do not substract irq_tlb_count from irq_call_count")
> https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git master
> 
> in testcase: unixbench
> on test machine: 8 threads Nehalem with 6G memory
> with following parameters:
> 
> 	runtime: 300s
> 	nr_task: 1
> 	test: shell8
> 
> iN addition to that, the commit also has significant impact (interrupts.CAL:Function_call_interrupts +1822.5%)
> on the following tests:
> 
> +------------------+----------------------------------+
> | testcase         | unixbench                        |
> | test machine     | 8 threads Nehalem with 6G memory |
> | test parameters  | nr_task=100%                     |
> |                  | runtime=300s                     |
> |                  | test=execl                       |
> +------------------+----------------------------------+
> 
> 
> Disclaimer:
> Results have been estimated based on internal Intel analysis and are provided
> for informational purposes only. Any difference in system hardware or software
> design or configuration may affect actual performance.
> 
> Details are as below:
> -------------------------------------------------------------------------------------------------->
> 
> 
> To reproduce:
> 
>         git clone git://git.kernel.org/pub/scm/linux/kernel/git/wfg/lkp-tests.git
>         cd lkp-tests
>         bin/lkp install job.yaml  # job file is attached in this email
>         bin/lkp run     job.yaml
> 
> =========================================================================================
> compiler/kconfig/nr_task/rootfs/runtime/tbox_group/test/testcase:
>   gcc-6/x86_64-rhel/1/debian-x86_64-2015-02-07.cgz/300s/nhm-white/shell8/unixbench
> 
> commit: 
>   ace7fab7a6 ("x86/mm: Fix swap entry comment and macro")
>   82ba4faca1 ("x86/irq: Do not substract irq_tlb_count from irq_call_count")
> 
> ace7fab7a6cdd363 82ba4faca1bffad429f15c90c9 
> ---------------- -------------------------- 
>          %stddev     %change         %stddev
>              \          |                \  
>      48379 Â  0%    +771.1%     421448 Â  0%  interrupts.CAL:Function_call_interrupts
>      19934 Â  8%     -10.3%      17882 Â  4%  meminfo.AnonHugePages
>     177575 Â  8%     -11.9%     156428 Â  7%  sched_debug.cpu.avg_idle.min
>       0.36 Â 35%    +431.2%       1.93 Â 55%  sched_debug.cpu.cpu_load[0].min
>      25.84 Â  1%     +11.3%      28.77 Â  4%  sched_debug.cpu.cpu_load[1].min
>       5.38 Â  3%     -13.9%       4.63 Â  5%  sched_debug.cpu.cpu_load[1].stddev
> 
> 
> 
> 
>                        interrupts.CAL:Function_call_interrupts
> 
>   450000 ++-----------------------------------------------------------------+
>          O O OO O O OO O O OO O O OO O O O OO                               |
>   400000 ++                                                                 |
>   350000 ++                                                                 |
>          |                                                                  |
>   300000 ++                                                                 |
>   250000 ++                                                                 |
>          |                                                                  |
>   200000 ++                                                                 |
>   150000 ++                                                                 |
>          |                                                                  |
>   100000 ++                                                                 |
>    50000 *+ .**.*.*.*     .**.*.*. *.*.*.*.*   .*. *.*.*.**    .*   .*.* .*.*
>          | *         :.*.*        *         :.*   *        + .*  *.*    *   |
>        0 ++----------*----------------------*---------------*---------------+
> 
> 
> 	[*] bisect-good sample
> 	[O] bisect-bad  sample
> 
> ***************************************************************************************************
> nhm-white: 8 threads Nehalem with 6G memory
> =========================================================================================
> compiler/kconfig/nr_task/rootfs/runtime/tbox_group/test/testcase:
>   gcc-6/x86_64-rhel/100%/debian-x86_64-2015-02-07.cgz/300s/nhm-white/execl/unixbench
> 
> commit: 
>   ace7fab7a6 ("x86/mm: Fix swap entry comment and macro")
>   82ba4faca1 ("x86/irq: Do not substract irq_tlb_count from irq_call_count")
> 
> ace7fab7a6cdd363 82ba4faca1bffad429f15c90c9 
> ---------------- -------------------------- 
>        fail:runs  %reproduction    fail:runs
>            |             |             |    
>          %stddev     %change         %stddev
>              \          |                \  
>      25890 Â  1%   +1822.5%     497748 Â  1%  interrupts.CAL:Function_call_interrupts
>      15522 Â 80%    +354.6%      70567 Â146%  latency_stats.avg.max
>       0.28 Â 11%     +23.9%       0.34 Â  4%  sched_debug.cfs_rq:/.nr_running.stddev
>     535.29 Â  0%   +2071.2%      11622 Â 55%  sched_debug.cfs_rq:/.runnable_load_avg.max
>      18.12 Â 35%     -66.9%       6.00 Â 85%  sched_debug.cfs_rq:/.runnable_load_avg.min
>     159.21 Â  0%   +2305.2%       3829 Â 55%  sched_debug.cfs_rq:/.runnable_load_avg.stddev
>     109.46 Â  3%   +1268.4%       1497 Â 53%  sched_debug.cpu.cpu_load[0].avg
>     535.25 Â  0%   +2071.2%      11621 Â 55%  sched_debug.cpu.cpu_load[0].max
>     536.29 Â  0%   +2067.3%      11623 Â 55%  sched_debug.cpu.cpu_load[1].max
>     158.13 Â  0%   +2317.9%       3823 Â 55%  sched_debug.cpu.cpu_load[1].stddev
>     535.92 Â  0%   +2068.9%      11623 Â 55%  sched_debug.cpu.cpu_load[2].max
>     158.04 Â  0%   +2319.1%       3823 Â 55%  sched_debug.cpu.cpu_load[2].stddev
>     535.00 Â  0%   +2072.4%      11622 Â 55%  sched_debug.cpu.cpu_load[3].max
>     157.68 Â  0%   +2324.3%       3822 Â 55%  sched_debug.cpu.cpu_load[3].stddev
>     533.17 Â  0%   +2079.7%      11621 Â 55%  sched_debug.cpu.cpu_load[4].max
>     157.11 Â  0%   +2332.9%       3822 Â 55%  sched_debug.cpu.cpu_load[4].stddev
>     168510 Â  2%    +840.5%    1584778 Â 51%  sched_debug.cpu.load.avg
>     600190 Â  1%   +1904.9%   12033432 Â 54%  sched_debug.cpu.load.max
>     197074 Â  1%   +1909.8%    3960765 Â 54%  sched_debug.cpu.load.stddev
>      -1804 Â-12%     -28.0%      -1299 Â-13%  sched_debug.cpu.nr_uninterruptible.min
>       1327 Â 19%     -24.6%       1000 Â 19%  sched_debug.cpu.nr_uninterruptible.stddev
> 
> 
> 
> 
> 
> 
> 
> Thanks,
> Xiaolong
>