2022-04-05 01:36:45

by Guilherme G. Piccoli

[permalink] [raw]
Subject: Re: [RFC 0/1] Add sysctl entry for controlling crash_kexec_post_notifiers

On 01/04/2022 17:22, Alejandro Jimenez wrote:
> I noticed that in contrast to other kernel core parameters (e.g. kernel.panic,
> kernel.panic_on_warn, kernel.panic_print) crash_kexec_post_notifiers is not
> available as a sysctl tunable. I am aware that because it is a kernel core
> parameter, there is already an entry under:
>
> /sys/module/kernel/parameters/crash_kexec_post_notifiers
>
> and that allows us to read/modify it at runtime. However, I thought it should
> also be available via sysctl, since users that want to read/set this value at
> runtime might look there first.
>
> I believe there is an ongoing effort to clean up kernel/sysctl.c, but it wasn't
> clear to me if this entry (and perhaps the other panic related entries too)
> should be placed on kernel/panic.c. I wanted to verify first that this change
> would be welcomed before doing additional refactoring work.
>
> I'd appreciate any comments or suggestions.
>
> Thank you,
> Alejandro

Hi Alejandro, thanks for you patch. I have a "selfish" concern though,
I'll expose it here.

I'm working a panic refactor, in order to split the panic notifiers in
more lists - good summary of this discussion at [0].
I'm in the half of the patches, hopefully next 2 weeks I have something
ready to submit (I'll be out next week).

As part of this effort, I plan to have a more fine-grained control of
this parameter, and it's going to be a sysctl, but not
"crash_kexec_post_notifiers" - this one should be kept I guess due to
retro-compatibility, but it'd be a layer on top oh the new one.
With that said, unless you have urgent needs for this patch to be
reviewed/merged , I'd like to ask you to wait the series and I can loop
you there, so you may review/comment and see if it fits your use case.

Thanks,


Guilherme


[0] https://lore.kernel.org/lkml/YfPxvzSzDLjO5ldp@alley/


2022-04-05 03:19:40

by Alejandro Jimenez

[permalink] [raw]
Subject: Re: [RFC 0/1] Add sysctl entry for controlling crash_kexec_post_notifiers

Hi Guilherme,

On 4/2/2022 10:01 AM, Guilherme G. Piccoli wrote:
> On 01/04/2022 17:22, Alejandro Jimenez wrote:
>> I noticed that in contrast to other kernel core parameters (e.g. kernel.panic,
>> kernel.panic_on_warn, kernel.panic_print) crash_kexec_post_notifiers is not
>> available as a sysctl tunable. I am aware that because it is a kernel core
>> parameter, there is already an entry under:
>>
>> /sys/module/kernel/parameters/crash_kexec_post_notifiers
>>
>> and that allows us to read/modify it at runtime. However, I thought it should
>> also be available via sysctl, since users that want to read/set this value at
>> runtime might look there first.
>>
>> I believe there is an ongoing effort to clean up kernel/sysctl.c, but it wasn't
>> clear to me if this entry (and perhaps the other panic related entries too)
>> should be placed on kernel/panic.c. I wanted to verify first that this change
>> would be welcomed before doing additional refactoring work.
>>
>> I'd appreciate any comments or suggestions.
>>
>> Thank you,
>> Alejandro
> Hi Alejandro, thanks for you patch. I have a "selfish" concern though,
> I'll expose it here.
>
> I'm working a panic refactor, in order to split the panic notifiers in
> more lists - good summary of this discussion at [0].
> I'm in the half of the patches, hopefully next 2 weeks I have something
> ready to submit (I'll be out next week).
>
> As part of this effort, I plan to have a more fine-grained control of
> this parameter, and it's going to be a sysctl, but not
> "crash_kexec_post_notifiers" - this one should be kept I guess due to
> retro-compatibility, but it'd be a layer on top oh the new one.
It would be great to provide finer control and isolate the riskier
modifiers.

I am using crash_kexec_post_notifiers to control behavior of pvpanic and
pstore and there is not an urgent need for my change, so I don't mind
waiting for the new interface to evolve. Please copy me if possible on
future submissions.

Thank you,
Alejandro

> With that said, unless you have urgent needs for this patch to be
> reviewed/merged , I'd like to ask you to wait the series and I can loop
> you there, so you may review/comment and see if it fits your use case.
>
> Thanks,
>
>
> Guilherme
>
>
> [0] https://lore.kernel.org/lkml/YfPxvzSzDLjO5ldp@alley/