Re: possible deadlock in lru_add_drain_all

[Byungchul Park]

On Tue, Oct 31, 2017 at 04:10:24PM +0100, Peter Zijlstra wrote:
> On Tue, Oct 31, 2017 at 03:58:04PM +0100, Michal Hocko wrote:
> > On Tue 31-10-17 15:52:47, Peter Zijlstra wrote:
> > [...]
> > > If we want to save those stacks; we have to save a stacktrace on _every_
> > > lock acquire, simply because we never know ahead of time if there will
> > > be a new link. Doing this is _expensive_.
> > > 
> > > Furthermore, the space into which we store stacktraces is limited;
> > > since memory allocators use locks we can't very well use dynamic memory
> > > for lockdep -- that would give recursive and robustness issues.

I agree with all you said.

But, I have a better idea, that is, to save only the caller's ip of each
acquisition as an additional information? Of course, it's not enough in
some cases, but it's cheep and better than doing nothing.

For example, when building A->B, let's save not only full stack of B,
but also caller's ip of A together, then use them on warning like:

-> #3 aa_mutex:
   a()
   b()
   c()
   d()
   ---
   while holding bb_mutex at $IP <- additional information I said

-> #2 bb_mutex:
   e()
   f()
   g()
   h()
   ---
   while holding cc_mutex at $IP <- additional information I said

-> #1 cc_mutex:
   i()
   j()
   k()
   l()
   ---
   while holding xxx at $IP <- additional information I said

and so on.

Don't you think this is worth working it?