Xenomai Community Call Minutes - March 10, 2021

Xenomai Community Call Minutes - March 10, 2021

[Wang, Rick Y]

Attendees:
Philippe Gerum, Jan Kiszka, Fino Meng, Florian Bezdeka, Song Chen, 
Antoine Hoarau, Hongzhan Chen, Dao Zhang, Jianbo Cao, Yanfeng Pu,
Anderson Huang, Florent Pirou, Xiaoyan Gong, Bryan Cheng, Rick Wang

Agenda:
*Community update - Jan, Philippe
*Open

*Community update
- x32 ABI support is removed from CXP.
- ARM64 Dovetail enabling is working in progress.
- Regression issue found with current patches merged.
- x86-32 app crash on the latest build.
- The gdb issue is fixed by Hongzhan. Patch submitted pending on merge.
- Y2038 prolem patches to rebase to Philippe's tree.

Next steps:
1. Extensive test on v3.2 dev tree
2. Fix x32 is not working issue
3. Merge gdb issue fix patches
4. Enable contributor access to giblab bug tracking
5. EVL porting from Apr.


Thanks,
Rick

Re: Xenomai Community Call Minutes - March 10, 2021

[Philippe Gerum]

Hi,

I will not be able to attend the next meeting planned tomorrow. Here is
the update from my side since the last time we met:

- The Xenomai/arm64 port to Dovetail is complete, available from my
  private tree. Tested on kvm/aarch64, imx8mm, rpi4 and Odroid C4 so
  far.

- All pending patches now merged into my 3.2 tree.

- A regression with Xenomai/I-pipe on this tree was fixed a couple of
  days ago, causing switchtest to freeze the machine. You may want to
  rebase on it.

Thanks,

-- 
Philippe.

Re: Xenomai Community Call Minutes - March 10, 2021

[Philippe Gerum]

Philippe Gerum <rpm@xenomai.org> writes:

> Hi,
>
> I will not be able to attend the next meeting planned tomorrow. Here is
> the update from my side since the last time we met:
>
> - The Xenomai/arm64 port to Dovetail is complete, available from my
>   private tree. Tested on kvm/aarch64, imx8mm, rpi4 and Odroid C4 so
>   far.
>
> - All pending patches now merged into my 3.2 tree.
>
> - A regression with Xenomai/I-pipe on this tree was fixed a couple of
>   days ago, causing switchtest to freeze the machine. You may want to
>   rebase on it.
>
> Thanks,

- I have also addressed the COW-breaking issue for Dovetail
  (dovetail/v5.10 and dovetail/master), leveraging the existing work
  already done in copy_page_present() for supporting the pin_user_page
  API. Tested extensively here by tracing the behavior upon mlocking and
  fork with real-time tasks, but since the issue with the I-pipe was not
  trivial to reproduce with the I-pipe it seems, any help appreciated
  validating this.  As a result, force_commit_memory() disappeared from
  the Dovetail API.

TIA,


-- 
Philippe.

Re: Xenomai Community Call Minutes - March 10, 2021

[Jan Kiszka]

On 23.03.21 11:32, Philippe Gerum wrote:
> 
> Philippe Gerum <rpm@xenomai.org> writes:
> 
>> Hi,
>>
>> I will not be able to attend the next meeting planned tomorrow. Here is
>> the update from my side since the last time we met:
>>
>> - The Xenomai/arm64 port to Dovetail is complete, available from my
>>   private tree. Tested on kvm/aarch64, imx8mm, rpi4 and Odroid C4 so
>>   far.
>>
>> - All pending patches now merged into my 3.2 tree.
>>
>> - A regression with Xenomai/I-pipe on this tree was fixed a couple of
>>   days ago, causing switchtest to freeze the machine. You may want to
>>   rebase on it.
>>
>> Thanks,
> 
> - I have also addressed the COW-breaking issue for Dovetail
>   (dovetail/v5.10 and dovetail/master), leveraging the existing work
>   already done in copy_page_present() for supporting the pin_user_page
>   API. Tested extensively here by tracing the behavior upon mlocking and
>   fork with real-time tasks, but since the issue with the I-pipe was not
>   trivial to reproduce with the I-pipe it seems, any help appreciated
>   validating this.  As a result, force_commit_memory() disappeared from
>   the Dovetail API.

Running smokey (full set) over Dovetail triggers it easily. Like it did
over I-pipe.

Jan

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

Re: Xenomai Community Call Minutes - March 10, 2021

[Philippe Gerum]

Jan Kiszka <jan.kiszka@siemens.com> writes:

> On 23.03.21 11:32, Philippe Gerum wrote:
>> 
>> Philippe Gerum <rpm@xenomai.org> writes:
>> 
>>> Hi,
>>>
>>> I will not be able to attend the next meeting planned tomorrow. Here is
>>> the update from my side since the last time we met:
>>>
>>> - The Xenomai/arm64 port to Dovetail is complete, available from my
>>>   private tree. Tested on kvm/aarch64, imx8mm, rpi4 and Odroid C4 so
>>>   far.
>>>
>>> - All pending patches now merged into my 3.2 tree.
>>>
>>> - A regression with Xenomai/I-pipe on this tree was fixed a couple of
>>>   days ago, causing switchtest to freeze the machine. You may want to
>>>   rebase on it.
>>>
>>> Thanks,
>> 
>> - I have also addressed the COW-breaking issue for Dovetail
>>   (dovetail/v5.10 and dovetail/master), leveraging the existing work
>>   already done in copy_page_present() for supporting the pin_user_page
>>   API. Tested extensively here by tracing the behavior upon mlocking and
>>   fork with real-time tasks, but since the issue with the I-pipe was not
>>   trivial to reproduce with the I-pipe it seems, any help appreciated
>>   validating this.  As a result, force_commit_memory() disappeared from
>>   the Dovetail API.
>
> Running smokey (full set) over Dovetail triggers it easily. Like it did
> over I-pipe.
>

Which test in particular?

-- 
Philippe.

Re: Xenomai Community Call Minutes - March 10, 2021

[Philippe Gerum]

Philippe Gerum <rpm@xenomai.org> writes:

> Jan Kiszka <jan.kiszka@siemens.com> writes:
>
>> On 23.03.21 11:32, Philippe Gerum wrote:
>>> 
>>> Philippe Gerum <rpm@xenomai.org> writes:
>>> 
>>>> Hi,
>>>>
>>>> I will not be able to attend the next meeting planned tomorrow. Here is
>>>> the update from my side since the last time we met:
>>>>
>>>> - The Xenomai/arm64 port to Dovetail is complete, available from my
>>>>   private tree. Tested on kvm/aarch64, imx8mm, rpi4 and Odroid C4 so
>>>>   far.
>>>>
>>>> - All pending patches now merged into my 3.2 tree.
>>>>
>>>> - A regression with Xenomai/I-pipe on this tree was fixed a couple of
>>>>   days ago, causing switchtest to freeze the machine. You may want to
>>>>   rebase on it.
>>>>
>>>> Thanks,
>>> 
>>> - I have also addressed the COW-breaking issue for Dovetail
>>>   (dovetail/v5.10 and dovetail/master), leveraging the existing work
>>>   already done in copy_page_present() for supporting the pin_user_page
>>>   API. Tested extensively here by tracing the behavior upon mlocking and
>>>   fork with real-time tasks, but since the issue with the I-pipe was not
>>>   trivial to reproduce with the I-pipe it seems, any help appreciated
>>>   validating this.  As a result, force_commit_memory() disappeared from
>>>   the Dovetail API.
>>
>> Running smokey (full set) over Dovetail triggers it easily. Like it did
>> over I-pipe.
>>
>
> Which test in particular?


Reply to self: anything that forks under testsuite I guess.

-- 
Philippe.

Re: Xenomai Community Call Minutes - March 10, 2021

[Jan Kiszka]

On 23.03.21 14:56, Jan Kiszka wrote:
> On 23.03.21 11:32, Philippe Gerum wrote:
>>
>> Philippe Gerum <rpm@xenomai.org> writes:
>>
>>> Hi,
>>>
>>> I will not be able to attend the next meeting planned tomorrow. Here is
>>> the update from my side since the last time we met:
>>>
>>> - The Xenomai/arm64 port to Dovetail is complete, available from my
>>>   private tree. Tested on kvm/aarch64, imx8mm, rpi4 and Odroid C4 so
>>>   far.
>>>
>>> - All pending patches now merged into my 3.2 tree.
>>>
>>> - A regression with Xenomai/I-pipe on this tree was fixed a couple of
>>>   days ago, causing switchtest to freeze the machine. You may want to
>>>   rebase on it.
>>>
>>> Thanks,
>>
>> - I have also addressed the COW-breaking issue for Dovetail
>>   (dovetail/v5.10 and dovetail/master), leveraging the existing work
>>   already done in copy_page_present() for supporting the pin_user_page
>>   API. Tested extensively here by tracing the behavior upon mlocking and
>>   fork with real-time tasks, but since the issue with the I-pipe was not
>>   trivial to reproduce with the I-pipe it seems, any help appreciated
>>   validating this.  As a result, force_commit_memory() disappeared from
>>   the Dovetail API.
> 
> Running smokey (full set) over Dovetail triggers it easily. Like it did
> over I-pipe.

Correction: Not easily, only easily on /certain/ setups. One was
qemu-arm64 with xenomai-images.

Jan

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

Re: Xenomai Community Call Minutes - March 10, 2021

[Philippe Gerum]

Jan Kiszka <jan.kiszka@siemens.com> writes:

> On 23.03.21 14:56, Jan Kiszka wrote:
>> On 23.03.21 11:32, Philippe Gerum wrote:
>>>
>>> Philippe Gerum <rpm@xenomai.org> writes:
>>>
>>>> Hi,
>>>>
>>>> I will not be able to attend the next meeting planned tomorrow. Here is
>>>> the update from my side since the last time we met:
>>>>
>>>> - The Xenomai/arm64 port to Dovetail is complete, available from my
>>>>   private tree. Tested on kvm/aarch64, imx8mm, rpi4 and Odroid C4 so
>>>>   far.
>>>>
>>>> - All pending patches now merged into my 3.2 tree.
>>>>
>>>> - A regression with Xenomai/I-pipe on this tree was fixed a couple of
>>>>   days ago, causing switchtest to freeze the machine. You may want to
>>>>   rebase on it.
>>>>
>>>> Thanks,
>>>
>>> - I have also addressed the COW-breaking issue for Dovetail
>>>   (dovetail/v5.10 and dovetail/master), leveraging the existing work
>>>   already done in copy_page_present() for supporting the pin_user_page
>>>   API. Tested extensively here by tracing the behavior upon mlocking and
>>>   fork with real-time tasks, but since the issue with the I-pipe was not
>>>   trivial to reproduce with the I-pipe it seems, any help appreciated
>>>   validating this.  As a result, force_commit_memory() disappeared from
>>>   the Dovetail API.
>> 
>> Running smokey (full set) over Dovetail triggers it easily. Like it did
>> over I-pipe.
>
> Correction: Not easily, only easily on /certain/ setups. One was
> qemu-arm64 with xenomai-images.
>

This is what I thought, because I could not figure out a way to
reproduce on my x86, arm and arm64 setups when that issue surfaced. This
said, with hindsight looking at the copy_pte code, the bug ended up
being shallow.

-- 
Philippe.