[LTP] [PATCH v2] tst_test: Add min_runtime to control lower bound of scaled runtime

[Li Wang]

Hi All,

Li Wang <liwang@redhat.com> wrote:


> But the default 1024 min_samples I still have no idea how long
> the .min_runtime needs. Maybe we can estimate and set .min_runtime
> on a slow system manually.
>

I choose to run the fuzzy_sync tests on a Cortex-A55 CPU with
only one core assigned, using the taskset command to simulate
an extreme single-core execution scenario.

Tested on:
    Cortex-A55, 1.2GHz
    Linux 5.14, aarch64

For example:
    time taskset -c 0 ./cve-2016-7117 (hacked the test only do sampling)

This setup allows me to evaluate how long the sampling phase
takes under constrained conditions.

Based on the observed sampling duration, I apply the following policy
for setting .min_runtime:

  1. If the sampling time is very short (less than 1 second),
      I simply set .min_runtime = 2.

  2. If the sampling time is longer than 2 seconds but still less than
     .runtime, I set .min_runtime to twice the sampling time, while
     keeping .runtime unchanged.

  3. If the sampling time exceeds .runtime, I also double the sampling
      time for .min_runtime, but leave .runtime as is.


-- 
Regards,
Li Wang