Re: _cpu_down deadlock [was Re: 2.6.19-rc1-mm1]

[Neil Brown]

On Wednesday October 11, akpm@xxxxxxxx wrote:
> > 
> > So A waits on B and C, C waits on B, B waits on A.
> > Deadlock.
> 
> Except the entire operation is serialised by the the two top-level callers
> (cpu_up() and cpu_down()) taking mutex_lock(&cpu_add_remove_lock).  Can
> lockdep be taught about that?

So you are saying that even though we have locking sequences
  A -> B  and B -> A,
that cannot - in this case - cause a deadlock as both sequences only
ever happen under a third exclusive lock C,
So when lockdep records a lock-dependency A -> B, it should also
record a list of locks that are *always* held when that dependency
occurs.
Then, when it finds a new dependency and does loop detection, it
should exclude from the path any dependency which is always under a
lock that some other dependency in the path is always under.
Also, loop checking as to happen both when a new dependency is found,
and when a lock is removed from the set of locks that protect the
dependency.

Recording stack traces might be interesting as you potentially need to
record a trace for ever minimal set of locks that the dependency is
created under.

So the ball is back in Ingo's court ?

Though it is odd that the warning doesn't trigger every time....


> 
> > Who do we blame this on?  Are you still the cpu-hot-plug guy Rusty?
> 
> It's fun blaming Rusty for stuff, but he can dodge this one with
> more-than-usual ease, I'm afraid.

In that case,  I was never dreaming of blaming him, only letting him
know that there is a lock-dep warning in code that he might be seen as
responsible for - just in case anyone does blame him.
Yes.  That's what I was doing.  Definitely.

NeilBrown
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/