Hi Kees,

On 11/10/2021 1:21 AM, Kees Cook wrote:

quoted

I suppose these three CPUs have different cache policy. It also could be
related with netperf throughput testing.

Does moving the syscall implementation somewhere else change things?

I moved the syscall implementation to a stand alone file and put the file
to net directory for testing. The new patch is like:
https://zerobin.net/?a2b782afadf3c428#Me8l4AJuhiSCfaLVWVzydAVIK6ves0EVIVD76wLnVQo=


The test result is like following:
     - on Cascade Lake: -10.4%
    356365           -10.4%     319180        netperf.Throughput_total_tps
     22272           -10.4%      19948        netperf.Throughput_tps


     - on Cooper Lake: -4.0%
    345772 ±  4%      -4.0%     331814        netperf.Throughput_total_tps
     21610 ±  4%      -4.0%      20738        netperf.Throughput_tps


     - on Ice Lake: -1.1%
    509824            -1.1%     504434        netperf.Throughput_total_tps
     31864            -1.1%      31527        netperf.Throughput_tps


Regards
Yin, Fengwei