[lkml]   [2019]   [Dec]   [20]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
SubjectRe: [PATCH 1/2] include: trace: Add SCMI header with trace events

> Thank you for sharing your experiments and thoughts. I have probably
> similar setup for stressing the communication channel, but I do also
> some wired things in the firmware. That's why I need to measure these
> delays. I am happy that it is useful for you also.
> I don't know if your firmware supports 'fast channel', but please keep
> in mind that it is not tracked in this 'transfer_id'.
> This transfer_id in v2 version does not show the real transfers
> to the firmware since there is another path called 'fast channel' or
> 'fast_switch' in the CPUfreq. It is in drivers/firmware/arm_scmi/perf.c
> and the atomic variable is not incremented in that path. Adding it also
> there just for atomic_inc() probably would add delays in the fast_switch
> and also brings little value.
> For the normal channel, where we have spinlocks and other stuff, this
> atomic_inc() could stay in my opinion.
> Regards,
> Lukasz
Hi Lukasz,

We currently do not use "fast channels" - although it is possible we might someday.
I find the transfer_id useful per your v2 even if it doesn't cover FC.  Thanks for
submitting and discussing this!

Jim Quinlan

 \ /
  Last update: 2019-12-20 17:24    [W:0.042 / U:5.948 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site