"cobalt/vfile: seq_file seek index must progress"

"cobalt/vfile: seq_file seek index must progress"

[Jan Kiszka]

Hi Philippe,

https://source.denx.de/Xenomai/xenomai/-/commit/dd9a5ee0ed6f734f40d3733f9fff23394c8647bd

Bug fix unrelated to dovetail? There is no description, so I cannot
easily asses this.

Jan

Re: "cobalt/vfile: seq_file seek index must progress"

[Jan Kiszka]

On 20.05.21 18:13, Jan Kiszka wrote:
> Hi Philippe,
> 
> https://source.denx.de/Xenomai/xenomai/-/commit/dd9a5ee0ed6f734f40d3733f9fff23394c8647bd
> 
> Bug fix unrelated to dovetail? There is no description, so I cannot
> easily asses this.
> 

Also, it seems to revert parts of
https://source.denx.de/Xenomai/xenomai/-/commit/4cc1e7727d2144098e2f512be71dfe7199d0f7cd
- granted, that was 11 years ago and had no description either.

I'd like to clarify this before tagging some 3.1.1.

Thanks,
Jan

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

Re: "cobalt/vfile: seq_file seek index must progress"

[Philippe Gerum]

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

> Hi Philippe,
>
> https://source.denx.de/Xenomai/xenomai/-/commit/dd9a5ee0ed6f734f40d3733f9fff23394c8647bd
>
> Bug fix unrelated to dovetail? There is no description, so I cannot
> easily asses this.
>

Yeah, not good. The idea is that the offset field we receive from the
kernel must progress in order for the loop to stop properly,
independently from our own tracking of the end-of-list condition.

AFAIR, this was very easily reproducible by running a test app creating
a few tenths of threads (10-20 would suffice) which would exit shortly
after, while running a cat loop on /proc/xenomai/sched/next. At some
point, the cat loop would cause the vfile handler to crash.

-- 
Philippe.

Re: "cobalt/vfile: seq_file seek index must progress"

[Jan Kiszka]

On 21.05.21 16:20, Philippe Gerum wrote:
> 
> Jan Kiszka <jan.kiszka@siemens.com> writes:
> 
>> Hi Philippe,
>>
>> https://source.denx.de/Xenomai/xenomai/-/commit/dd9a5ee0ed6f734f40d3733f9fff23394c8647bd
>>
>> Bug fix unrelated to dovetail? There is no description, so I cannot
>> easily asses this.
>>
> 
> Yeah, not good. The idea is that the offset field we receive from the
> kernel must progress in order for the loop to stop properly,
> independently from our own tracking of the end-of-list condition.
> 
> AFAIR, this was very easily reproducible by running a test app creating
> a few tenths of threads (10-20 would suffice) which would exit shortly
> after, while running a cat loop on /proc/xenomai/sched/next. At some
> point, the cat loop would cause the vfile handler to crash.
> 

OK, so this is an important independent bug fix.

It's not yet in next. Maybe you could sent it along with the reasoning
as patch to the list. Would then also pick it into stable.

Thanks,
Jan

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