[RFQ] make witness panic an option

[Attilio Rao]

On 11/15/12, Adrian Chadd <adrian at freebsd.org> wrote:
> On 15 November 2012 10:01, Attilio Rao <attilio at freebsd.org> wrote:
>> I think that your worries are focused more around the latter than the
>> former, which can be easilly shut down already today.
>>
>> And frankly I will never be in favor of a patch that automatically
>> shutdowns lock assertion. Please patch your local code to do so but
>> don't add any generic/upstream/all-around mechanism for that.
>
> Would a comprimise be ok? Ie, if I pushed everything but the sysctl
> upstream, and just defaulted it to always panic?
>
> That way my diff wouldn't have to be a big thing; I'd just add the sysctl.

I cannot forbid you from doing anything, I'm just giving you my
opinion as the person who co-authored current WITNESS and locking
primitives code.

I think what you want to do is dangerous and highly abusable, so I'm
not in favor of it at all, in whatever form it is. I understand you
want to minimize your development patchset with upstream, but I think
this is certainly not the way to go.
That's also why I never formalized the BLESSING mechanism in WITNESS,
for example. I already see WITNESS_KDB as an abuse, but at least until
we have a way to make specific LOR (based on file/line, unfortunately)
to be marked as "harmless" there is no way we can get rid of
WITNESS_KDB.

Said all that, you are free to do what you want, but if you commit
anything in this area make sure your patch is reviewed by someone else
and to state my firm disagreement with this approach in the commit
message.

Attilio


-- 
Peace can only be achieved by understanding - A. Einstein