Re: [PATCH v12 3/3] Documentation/filesystems/proc.txt: add AVX512_elapsed_ms

["Li, Aubrey" <aubrey.li@linux.intel.com>]

On 2019/2/24 2:16, Thomas Gleixner wrote:
> On Thu, 21 Feb 2019, Aubrey Li wrote:
>> @@ -45,6 +45,7 @@ Table of Contents
>>    3.9   /proc/<pid>/map_files - Information about memory mapped files
>>    3.10  /proc/<pid>/timerslack_ns - Task timerslack value
>>    3.11	/proc/<pid>/patch_state - Livepatch patch operation state
>> +  3.12	/proc/<pid>/AVX512_elapsed_ms - time elapsed since last AVX512 use
> 
> So is this a separate file now?
>   
>> +3.12	/proc/<pid>/AVX512_elapsed_ms - time elapsed since last AVX512 use
>> +--------------------------------------------------------------------------
>> +If AVX512 is supported on the machine, this file displays time elapsed since
> 
> This is not a file and this documentation wants to be where the status file
> is described.
> 
>> +last AVX512 usage of the task in millisecond.
> 
> Since last usage is misleading. What you want to say is:
> 
>   The entry shows the milliseconds elapsed since the last time AVX512 usage
>   was recorded.
> 
>> +The per-task AVX512 usage tracking mechanism is added during context switch.
>> +When the task is scheduled out, the AVX512 timestamp of the task is tagged
>> +by jiffies if AVX512 usage is detected.
>> +
>> +When this interface is queried, AVX512_elapsed_ms is calculated as follows:
>> +
>> +	delta = (long)(jiffies_now - AVX512_timestamp);
>> +	AVX512_elpased_ms = jiffies_to_msecs(delta);
> 
> This information is not really helpful for someone who wants to use that
> field.
> 
>> +
>> +Because this tracking mechanism depends on context switch, the number of
>> +AVX512_elapsed_ms could be inaccurate if the AVX512 using task runs alone on
>> +a CPU and not scheduled out for a long time. An extreme experiment shows a
>> +task is spinning on the AVX512 ops on an isolated CPU, but the longest elapsed
>> +time is close to 4 seconds(HZ = 250).
>> +
>> +So 5s or even longer is an appropriate threshold for the job scheduler to poll
>> +and decide if the task should be classifed as an AVX512 task and migrated
>> +away from the core on which a Non-AVX512 task is running.
> 
> 5 seconds or long is appropriate? No. It really depends on the workload and
> the scheduling scenarios. What the documentation has to provide is the
> information that this value is a crystal ball estimate and what the reasons
> are why its inaccurate.
> 
> Something like this instead of this conglomorate of useful, irrelevant and
> misleading information:
> 
>   The AVX512_elapsed_ms entry shows the milliseconds elapsed since the last
>   time AVX512 usage was recorded. The recording happens on a best effort
>   basis when a task is scheduled out. This means that the value depends on
>   two factors:
> 
>     1) The time which the task spent on the CPU without being scheduled
>        out. With CPU isolation and a single runnable task this can take
>        several seconds.
> 
>     2) The time since the task was scheduled out last. Depending on the
>        reason for being scheduled out (time slice exhausted, syscall ...)
>        this can be arbitrary long time.
> 
>   As a consequence the value cannot be considered precise and authoritive
>   information. The application which uses this information has to be aware
>   of the overall scenario on the system in order to determine whether a
>   task is a real AVX512 user or not.
> 
> See? No jiffies, no code snippets, no absolute numbers and no magic
> recommendation which might be correct for your test scenario, but
> completely bogus for some other scenario.
> 
> Instead it contains the things which a application programmer who wants to
> use that value needs to know. He then has to map it to his scenario and
> build the crystal ball logic which makes it perhaps useful.

Thanks a lot, I'll try to refine it again.

Regards,
-Aubrey