Re: [PATCH 1/2] lockdep: Introduce CROSSRELEASE_STACK_TRACE and make it not unwind as default

[Ingo Molnar]

* ïÚïÃ/ïïïÓïïïïï/SW Platform(ïï)AOTïï(byungchul.park@xxxxxxx) <byungchul.park@xxxxxxx> wrote:

> I don't want to pretend I'm perfect. Of course, I can make mistakes.
> I'm just saying that *I have not seen* any crash by cross-release.
> 
> In that case you pointed out, likewise, the crash was caused by ae813308f:
> lockdep: Avoid creating redundant links, which is not related to the feature
> actually. It was also falsely accused at the time again...
> 
> Of course, it's my fault not to have made the design more robust so that
> others can modify lockdep code caring less after cross-release commit.
> That's what I'm sorry for.
> 
> I already mentioned the above in the thread talking about the issue you
> are pointing now. Of course, I basically appreciate all comments and
> suggestions you have given, but you seem to have mis-understood some
> issues wrt cross-release feature.

Two different cross-release commits got bisected to with kernel crashes:

  Sep 30 kernel test rob    | ce07a9415f ("locking/lockdep: Make check_prev_add() able to .."):  BUG: unable to handle kernel NULL pointer dereference at 00000020
  Oct 03 Fengguang Wu       | [lockdep] b09be676e0 BUG: unable to handle kernel NULL pointer dereference at 000001f2

The first crash was bisected to:

  ce07a9415f26: locking/lockdep: Make check_prev_add() able to handle external stack_trace

The second crash was bisected to:

  b09be676e0ff: locking/lockdep: Implement the 'crossrelease' feature

... and unless your argument that both bisections were bad, it doesn't matter 
where the root cause ended up being, fact is that it was not a problem free series 
and let's not pretend it was.

Note that to me it *really* does not matter that a commit causes a crash: bugs 
happen, they are part of software development done by humans - so as long as it's 
not a pattern of underlying carelessness or some development process error it's 
not something to get emotional about.

Ok?

Thanks,

	Ingo