lkml.org 
[lkml]   [2020]   [Jan]   [9]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH 2/2] PM / devfreq: Add devfreq_transitions debugfs file
From
Date
On 1/9/20 12:44 AM, Lukasz Luba wrote:
> Hi,
>
> On 1/8/20 2:20 PM, Dmitry Osipenko wrote:
>> 08.01.2020 00:48, Bjorn Andersson пишет:
>>> On Tue 07 Jan 01:05 PST 2020, Chanwoo Choi wrote:
>>>
>>>> Add new devfreq_transitions debugfs file to track the frequency transitions
>>>> of all devfreq devices for the simple profiling as following:
>>>> - /sys/kernel/debug/devfreq/devfreq_transitions
>>>>
>>>> And the user can decide the storage size (CONFIG_NR_DEVFREQ_TRANSITIONS)
>>>> in Kconfig in order to save the transition history.
>>>>
>>>> [Detailed description of each field of 'devfreq_transitions' debugfs file]
>>>> - time_ms    : Change time of frequency transition. (unit: millisecond)
>>>> - dev_name    : Device name of h/w.
>>>> - dev        : Device name made by devfreq core.
>>>> - parent_dev    : If devfreq device uses the passive governor,
>>>>           show parent devfreq device name.
>>>> - load_%    : If devfreq device uses the simple_ondemand governor,
>>>>           load is used by governor whene deciding the new frequency.
>>>>           (unit: percentage)
>>>> - old_freq_hz    : Frequency before changing. (unit: hz)
>>>> - new_freq_hz    : Frequency after changed. (unit: hz)
>>>>
>>>> [For example on Exynos5422-based Odroid-XU3 board]
>>>> $ cat /sys/kernel/debug/devfreq/devfreq_transitions
>>>> time_ms    dev_name                       dev        parent_dev load_% old_freq_hz  new_freq_hz
>>>> ---------- ------------------------------ ---------- ---------- ---------- ------------ ------------
>>>> 14600      soc:bus_noc                    devfreq2   devfreq1   0      100000000    67000000
>>>> 14600      soc:bus_fsys_apb               devfreq3   devfreq1   0      200000000    100000000
>>>> 14600      soc:bus_fsys                   devfreq4   devfreq1   0      200000000    100000000
>>>> 14600      soc:bus_fsys2                  devfreq5   devfreq1   0      150000000    75000000
>>>> 14602      soc:bus_mfc                    devfreq6   devfreq1   0      222000000    96000000
>>>> 14602      soc:bus_gen                    devfreq7   devfreq1   0      267000000    89000000
>>>> 14602      soc:bus_g2d                    devfreq9   devfreq1   0      300000000    84000000
>>>> 14602      soc:bus_g2d_acp                devfreq10  devfreq1   0      267000000    67000000
>>>> 14602      soc:bus_jpeg                   devfreq11  devfreq1   0      300000000    75000000
>>>> 14602      soc:bus_jpeg_apb               devfreq12  devfreq1   0      167000000    84000000
>>>> 14603      soc:bus_disp1_fimd             devfreq13  devfreq1   0      200000000    120000000
>>>> 14603      soc:bus_disp1                  devfreq14  devfreq1   0      300000000    120000000
>>>> 14606      soc:bus_gscl_scaler            devfreq15  devfreq1   0      300000000    150000000
>>>> 14606      soc:bus_mscl                   devfreq16  devfreq1   0      333000000    84000000
>>>> 14608      soc:bus_wcore                  devfreq1              9      333000000    84000000
>>>> 14783      10c20000.memory-controller     devfreq0              35     825000000    633000000
>>>> 15873      soc:bus_wcore                  devfreq1              41     84000000     400000000
>>>> 15873      soc:bus_noc                    devfreq2   devfreq1   0      67000000     100000000
>>>> [snip]
>>>>
>>>
>>> Wouldn't it make more sense to expose this through the tracing
>>> framework - like many other subsystems does?
>>
>> I think devfreq core already has some tracing support and indeed it
>> should be better to extend it rather than duplicate.
>>
>
> In my opinion this debugfs interface should be considered as a helpful
> validation entry point. We had some issues with wrong bootloader
> configurations in clock tree, where some frequencies could not be set
> in the kernel. Similar useful description can be find in clock subsystem
> where there is clock tree summary file.
>
> It is much cheaper to poke a few files in debug dir by some automated
> test than starting tracing, provoking desired code flow in the
> devfreq for every device, paring the results... A simple boot test
> which reads only these new files can be enough to rise the flag.
> Secondly the tracing is not always compiled.
>
> It could capture old/wrong bootloaders which pinned devices
> improperly to PLLs or wrong DT values in OPP table.
> (a workaround for Odroid xu4 patchset:
> https://protect2.fireeye.com/url?k=364d2fbb-6b9993d3-364ca4f4-0cc47a3356b2-98db7e7cf023414c&u=https://lkml.org/lkml/2019/7/15/276
> )
>
> Chanwoo what do think about some sanity check summary?
> It could be presented in a 3rd file: 'devfreq_sanity', which
> could report if the devices could set their registered OPPs
> and got the same values, i.e. set 166MHz --> set to 150MHz
> in reality. If a config option i.e. DEVFREQ_SANITY is set
> then during the registration of a new device it checks OPPs
> if they are possible to set. It could be done before assigning
> the governor for the device and results present in of of your files.

Firstly, I'm welcoming to add new debugging feature.

As you suggested, it is required for the OPP control.
But, I'm not sure that it have to be verified in either OPP or devfreq.
Or it have be verified when adding the OPP table into devicetree file.

If we add this debugging feature, I think 'resource_sanity' as following path.
/sys/kernel/debug/devfreq/devfreqX/resource_sanity (not fixed name)
- show the sanity result of regulator voltage and frequency

--
Best Regards,
Chanwoo Choi
Samsung Electronics

\
 
 \ /
  Last update: 2020-01-09 11:39    [W:0.125 / U:7.268 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site