Re: [PATCH] [RFC][Patch x86-tip] add notifier before kdump

[ebiederm@xmission.com (Eric W. Biederman)]

Jin Dongming <jin.dongming@np.css.fujitsu.com> writes:

> Hi Vivek
>
> Vivek Goyal wrote:
>> - Other thing, I think Eric Biederman suggested was that we need to see
>>   the code that will be executed after crash so that it can be audited and
>>   keep it in kernel instead of blindly exporting a hook to module and let
>>   module do whatever it want to do.
>> 
>
> I am sorry for that I could not catch the meaning of above words. In my
> understanding:
>     If the code is already in current kernel and no matter whether it is built
>     or modulized, the code could be added before crash_kexec. Is it right?

No.  crash_kexec skips the calling the notifiers intentionally to increase
it's robustness.

> If my understanding is not right, please let me ask a question before we
> provide the code.
>     What kind of code is considered as nice code and could be added into
>     current kernel? Could you give me some requirements or examples
>     about nice code?

In general code that is necessary.  If we can find a way to avoid running
code in a broken kernel we should.

The recent addition of code to disable processor vitalization features
on a crash is an example of code that is necessary.  A small function
that we can always call that is not provided by a module.

Eric