questions about comments on https://gitlab.com/Xenomai/xenomai-hacker-space/-/issues/33

questions about comments on https://gitlab.com/Xenomai/xenomai-hacker-space/-/issues/33

[Chen, Hongzhan]

Hi Jan

I have some questions about comments on  https://gitlab.com/Xenomai/xenomai-hacker-space/-/issues/33 and need your help to comment more.

1. I do not quite understand the requirement "visualize task switches correctly in Kernelshark (color in CPU bar is not updating on cobalt_switch_context)".

When I click the cobalt_switch_context event listed in the event table on the second half in the kernelshark, actually color in cpu bar is updating to 
green because there is long green vertical  line to mark focused event like attached picture.  In addition , In spite of focused thing,  colors mainly 
change depending on what task is running for cpu plots in kernelshark. when cobalt_switch_context is traced, it still be in the prev context and its 
color is same with task prev in kernelshark. Do we want to update color of  cobalt_switch_context  to specific color different from all tasks even if there is no 
focused thing?  What result do we need to achieve when there is cobalt_switch_context, Could you be more specific?

2. Regarding  the requirement " visualize RT/Non-RT (out-of-band/in-band) state in CPU bar)",
    2.1. Do we need to differentiate state  running a Linux task out-of-band and running whatever code from the out-of-band execution stage as described in 
             " What you really need to know at this point" of  https://evlproject.org/dovetail/altsched/?
    2.2 What state do we want to show  in CPU bar? Different color covering the size of  RT/Non-RT state other than all tasks and CPUs?  If so, we may ask for a plugin
          in kernelshark , which would raise similar question as you mentioned in comments.

For the other two aspects, I still need more time to do research. 
  

Regards

Hongzhan Chen

-------------- next part --------------
A non-text attachment was scrubbed...
Name: IMG_4941.JPG
Type: image/jpeg
Size: 5243392 bytes
Desc: IMG_4941.JPG
URL: <http://xenomai.org/pipermail/xenomai/attachments/20211207/4291b117/attachment.jpe>

RE: questions about comments on https://gitlab.com/Xenomai/xenomai-hacker-space/-/issues/33

[Chen, Hongzhan]

>
>-----Original Message-----
>From: Chen, Hongzhan
>Sent: Tuesday, December 7, 2021 11:11 AM
>To: xenomai@xenomai.org<mailto:xenomai@xenomai.org>; Jan Kiszka jan.kiszka@siemens.com<mailto:jan.kiszka@siemens.com>
>Subject: questions about comments on https://gitlab.com/Xenomai/xenomai-hacker-space/-/issues/33
>
>Hi Jan
>
>I have some questions about comments on  https://gitlab.com/Xenomai/xenomai-hacker-space/-/issues/33 and need your help to comment more.
>
>1. I do not quite understand the requirement "visualize task switches correctly in Kernelshark (color in CPU bar is not updating on cobalt_switch_context)".
>
>When I click the cobalt_switch_context event listed in the event table on the second half in the kernelshark, actually color in cpu bar is updating to
>green because there is long green vertical  line to mark focused event like attached picture.  In addition , In spite of focused thing,  colors mainly
>change depending on what task is running for cpu plots in kernelshark. when cobalt_switch_context is traced, it still be in the prev context and its
>color is same with task prev in kernelshark. Do we want to update color of  cobalt_switch_context  to specific color different from all tasks even if there is no
>focused thing?  What result do we need to achieve when there is cobalt_switch_context, Could you be more specific?

After I understand what does means about sched_switch in kernelshark, I guess that we need to use cobalt_switch_context as dividing line for RT/non-RT tasks similar with sched_switch after we use xenomai.

Regards

Hongzhan Chen

>
>2. Regarding  the requirement " visualize RT/Non-RT (out-of-band/in-band) state in CPU bar)",
>    2.1. Do we need to differentiate state  running a Linux task out-of-band and running whatever code from the out-of-band execution stage as described in
>             " What you really need to know at this point" of  https://evlproject.org/dovetail/altsched/?
>    2.2 What state do we want to show  in CPU bar? Different color covering the size of  RT/Non-RT state other than all tasks and CPUs?  If so, we may ask for a plugin
>          in kernelshark , which would raise similar question as you mentioned in comments.
>
>For the other two aspects, I still need more time to do research.
>
>
>Regards
>
>Hongzhan Chen

Re: questions about comments on https://gitlab.com/Xenomai/xenomai-hacker-space/-/issues/33

[Jan Kiszka]

On 07.12.21 04:11, Chen, Hongzhan wrote:
> Hi Jan
> 
> I have some questions about comments on  https://gitlab.com/Xenomai/xenomai-hacker-space/-/issues/33 and need your help to comment more.
> 
> 1. I do not quite understand the requirement "visualize task switches correctly in Kernelshark (color in CPU bar is not updating on cobalt_switch_context)".
> 
> When I click the cobalt_switch_context event listed in the event table on the second half in the kernelshark, actually color in cpu bar is updating to 
> green because there is long green vertical  line to mark focused event like attached picture.  In addition , In spite of focused thing,  colors mainly 
> change depending on what task is running for cpu plots in kernelshark. when cobalt_switch_context is traced, it still be in the prev context and its 
> color is same with task prev in kernelshark. Do we want to update color of  cobalt_switch_context  to specific color different from all tasks even if there is no 
> focused thing?  What result do we need to achieve when there is cobalt_switch_context, Could you be more specific?
> 

I assume this one is resolved now, at least regarding what the goal
would be.

> 2. Regarding  the requirement " visualize RT/Non-RT (out-of-band/in-band) state in CPU bar)",
>     2.1. Do we need to differentiate state  running a Linux task out-of-band and running whatever code from the out-of-band execution stage as described in 
>              " What you really need to know at this point" of  https://evlproject.org/dovetail/altsched/?
>     2.2 What state do we want to show  in CPU bar? Different color covering the size of  RT/Non-RT state other than all tasks and CPUs?  If so, we may ask for a plugin
>           in kernelshark , which would raise similar question as you mentioned in comments.

I suspect it would be best to decouple to stage state (in-band, oob)
from the current task, maybe by adding another bar underneath the task
bar for each CPU that shows this state.

In fact, visualizing the dovetail state could be seen as another use
case for visualizing things like interrupt masks of preemption locks
that Linux upstream has. Maybe there are already ideas or even bits to
make that possible, and would could re-used them. Or we could sell our
demand via enabling those other scenarios as well...

Jan

-- 
Siemens AG, T RDA IOT
Corporate Competence Center Embedded Linux

RE: questions about comments on https://gitlab.com/Xenomai/xenomai-hacker-space/-/issues/33

[Chen, Hongzhan]

>-----Original Message-----
>From: Jan Kiszka <jan.kiszka@siemens.com> 
>Sent: Thursday, December 9, 2021 5:35 PM
>To: Chen, Hongzhan <hongzhan.chen@intel.com>; xenomai@xenomai.org
>Subject: Re: questions about comments on https://gitlab.com/Xenomai/xenomai-hacker-space/-/issues/33
>
>On 07.12.21 04:11, Chen, Hongzhan wrote:
>> Hi Jan
>> 
>> I have some questions about comments on  https://gitlab.com/Xenomai/xenomai-hacker-space/-/issues/33 and need your help to comment more.
>> 
>> 1. I do not quite understand the requirement "visualize task switches correctly in Kernelshark (color in CPU bar is not updating on cobalt_switch_context)".
>> 
>> When I click the cobalt_switch_context event listed in the event table on the second half in the kernelshark, actually color in cpu bar is updating to 
>> green because there is long green vertical  line to mark focused event like attached picture.  In addition , In spite of focused thing,  colors mainly 
>> change depending on what task is running for cpu plots in kernelshark. when cobalt_switch_context is traced, it still be in the prev context and its 
>> color is same with task prev in kernelshark. Do we want to update color of  cobalt_switch_context  to specific color different from all tasks even if there is no 
>> focused thing?  What result do we need to achieve when there is cobalt_switch_context, Could you be more specific?
>> 
>
>I assume this one is resolved now, at least regarding what the goal
>would be.
>
>> 2. Regarding  the requirement " visualize RT/Non-RT (out-of-band/in-band) state in CPU bar)",
>>     2.1. Do we need to differentiate state  running a Linux task out-of-band and running whatever code from the out-of-band execution stage as described in 
>>              " What you really need to know at this point" of  https://evlproject.org/dovetail/altsched/?
>>     2.2 What state do we want to show  in CPU bar? Different color covering the size of  RT/Non-RT state other than all tasks and CPUs?  If so, we may ask for a plugin
>>           in kernelshark , which would raise similar question as you mentioned in comments.
>
>I suspect it would be best to decouple to stage state (in-band, oob)
>from the current task, maybe by adding another bar underneath the task
>bar for each CPU that shows this state.

What about painting a hollow blue box under the baseline of corresponding task 
to mark running in OOB state, starting from first cobalt_switch_context switching
task in  to next cobalt_switch_context switching it out like attached pic?

>
>In fact, visualizing the dovetail state could be seen as another use
>case for visualizing things like interrupt masks of preemption locks
>that Linux upstream has. Maybe there are already ideas or even bits to
>make that possible, and would could re-used them. Or we could sell our
>demand via enabling those other scenarios as well...
>
>Jan
>
>-- 
>Siemens AG, T RDA IOT
>Corporate Competence Center Embedded Linux
-------------- next part --------------
A non-text attachment was scrubbed...
Name: IMG_4960.JPG
Type: image/jpeg
Size: 5292938 bytes
Desc: IMG_4960.JPG
URL: <http://xenomai.org/pipermail/xenomai/attachments/20211213/4aa89416/attachment.jpe>

RE: questions about comments on https://gitlab.com/Xenomai/xenomai-hacker-space/-/issues/33

[Chen, Hongzhan]

 
>>-----Original Message-----
>>From: Jan Kiszka <jan.kiszka@siemens.com> 
>>Sent: Thursday, December 9, 2021 5:35 PM
>>To: Chen, Hongzhan <hongzhan.chen@intel.com>; xenomai@xenomai.org
>>Subject: Re: questions about comments on https://gitlab.com/Xenomai/xenomai-hacker-space/-/issues/33
>>
>>On 07.12.21 04:11, Chen, Hongzhan wrote:
>>> Hi Jan
>>> 
>>> I have some questions about comments on  https://gitlab.com/Xenomai/xenomai-hacker-space/-/issues/33 and need your help to comment more.
>>> 
>>> 1. I do not quite understand the requirement "visualize task switches correctly in Kernelshark (color in CPU bar is not updating on cobalt_switch_context)".
>>> 
>>> When I click the cobalt_switch_context event listed in the event table on the second half in the kernelshark, actually color in cpu bar is updating to 
>>> green because there is long green vertical  line to mark focused event like attached picture.  In addition , In spite of focused thing,  colors mainly 
>>> change depending on what task is running for cpu plots in kernelshark. when cobalt_switch_context is traced, it still be in the prev context and its 
>>> color is same with task prev in kernelshark. Do we want to update color of  cobalt_switch_context  to specific color different from all tasks even if there is no 
>>> focused thing?  What result do we need to achieve when there is cobalt_switch_context, Could you be more specific?
>>> 
>>
>>I assume this one is resolved now, at least regarding what the goal
>>would be.
>>
>>> 2. Regarding  the requirement " visualize RT/Non-RT (out-of-band/in-band) state in CPU bar)",
>>>     2.1. Do we need to differentiate state  running a Linux task out-of-band and running whatever code from the out-of-band execution stage as described in 
>>>              " What you really need to know at this point" of  https://evlproject.org/dovetail/altsched/?
>>>     2.2 What state do we want to show  in CPU bar? Different color covering the size of  RT/Non-RT state other than all tasks and CPUs?  If so, we may ask for a plugin
>>>           in kernelshark , which would raise similar question as you mentioned in comments.
>>
>>I suspect it would be best to decouple to stage state (in-band, oob)
>>from the current task, maybe by adding another bar underneath the task
 >bar for each CPU that shows this state.
>
>What about painting a hollow blue box under the baseline of corresponding task 
>to mark running in OOB state, starting from first cobalt_switch_context switching
>task in  to next cobalt_switch_context switching it out like attached pic?

The hollow box is commonly used in task plot such as hollow green box for wake up latency
as described in page 15 of [1].

Regards

Hongzhan Chen


[1]: https://elinux.org/images/6/64/Elc2011_rostedt.pdf


>
>>
>>In fact, visualizing the dovetail state could be seen as another use
>>case for visualizing things like interrupt masks of preemption locks
>>that Linux upstream has. Maybe there are already ideas or even bits to
>>make that possible, and would could re-used them. Or we could sell our
>>demand via enabling those other scenarios as well...
>>
>>Jan
>>
>>-- 
>>Siemens AG, T RDA IOT
>>Corporate Competence Center Embedded Linux

Re: questions about comments on https://gitlab.com/Xenomai/xenomai-hacker-space/-/issues/33

[Jan Kiszka]

On 13.12.21 03:37, Chen, Hongzhan wrote:
> 
> 
>  
>>> -----Original Message-----
>>> From: Jan Kiszka <jan.kiszka@siemens.com> 
>>> Sent: Thursday, December 9, 2021 5:35 PM
>>> To: Chen, Hongzhan <hongzhan.chen@intel.com>; xenomai@xenomai.org
>>> Subject: Re: questions about comments on https://gitlab.com/Xenomai/xenomai-hacker-space/-/issues/33
>>>
>>> On 07.12.21 04:11, Chen, Hongzhan wrote:
>>>> Hi Jan
>>>>
>>>> I have some questions about comments on  https://gitlab.com/Xenomai/xenomai-hacker-space/-/issues/33 and need your help to comment more.
>>>>
>>>> 1. I do not quite understand the requirement "visualize task switches correctly in Kernelshark (color in CPU bar is not updating on cobalt_switch_context)".
>>>>
>>>> When I click the cobalt_switch_context event listed in the event table on the second half in the kernelshark, actually color in cpu bar is updating to 
>>>> green because there is long green vertical  line to mark focused event like attached picture.  In addition , In spite of focused thing,  colors mainly 
>>>> change depending on what task is running for cpu plots in kernelshark. when cobalt_switch_context is traced, it still be in the prev context and its 
>>>> color is same with task prev in kernelshark. Do we want to update color of  cobalt_switch_context  to specific color different from all tasks even if there is no 
>>>> focused thing?  What result do we need to achieve when there is cobalt_switch_context, Could you be more specific?
>>>>
>>>
>>> I assume this one is resolved now, at least regarding what the goal
>>> would be.
>>>
>>>> 2. Regarding  the requirement " visualize RT/Non-RT (out-of-band/in-band) state in CPU bar)",
>>>>     2.1. Do we need to differentiate state  running a Linux task out-of-band and running whatever code from the out-of-band execution stage as described in 
>>>>              " What you really need to know at this point" of  https://evlproject.org/dovetail/altsched/?
>>>>     2.2 What state do we want to show  in CPU bar? Different color covering the size of  RT/Non-RT state other than all tasks and CPUs?  If so, we may ask for a plugin
>>>>           in kernelshark , which would raise similar question as you mentioned in comments.
>>>
>>> I suspect it would be best to decouple to stage state (in-band, oob)
>> >from the current task, maybe by adding another bar underneath the task
>  >bar for each CPU that shows this state.
>>
>> What about painting a hollow blue box under the baseline of corresponding task 
>> to mark running in OOB state, starting from first cobalt_switch_context switching
>> task in  to next cobalt_switch_context switching it out like attached pic?
> 
> The hollow box is commonly used in task plot such as hollow green box for wake up latency
> as described in page 15 of [1].
> 

Then we only need to find a pattern that cannot be confused with
existing plots.

Can that plotting be provided by a custom plugin?

Jan

-- 
Siemens AG, T RDA IOT
Corporate Competence Center Embedded Linux