Re: [PATCH 10/11] tracing/perf: Fix lock events recursions in the fast path

[Frederic Weisbecker <fweisbec@gmail.com>]

On Sat, Feb 06, 2010 at 03:17:47PM +0100, Peter Zijlstra wrote:
> On Sat, 2010-02-06 at 12:40 +0100, Frederic Weisbecker wrote:
> > On Sat, Feb 06, 2010 at 12:24:02PM +0100, Peter Zijlstra wrote:
> > > On Sat, 2010-02-06 at 12:12 +0100, Frederic Weisbecker wrote:
> > > 
> > > > That said, I think this is good for a first step, but we can't continue
> > > > to force the lock events -> lockdep dependency in the long term. We
> > > > can't have a serious lock profiling if we are doomed to suffer the
> > > > slowness due to lockdep checks at the same time.
> > > > 
> > > > Sure we can continue to support having both, but I think we should also
> > > > think about a solution to handle lock events without it in the future.
> > > > That will require some minimal lockdep functionalities (keeping the
> > > > lockdep map, and class hashes).
> > > 
> > > You mean like building without CONFIG_PROVE_LOCKING, or boot with
> > > lockdep.prove_locking=0, or use echo 0
> > > > /sys/modules/lockdep/prove_locking ?
> > > 
> > > That keeps the lock tracking but does away with all the dependency
> > > analysis and was created for just such an use case as you are looking
> > > at, namely lockstat.
> > 
> > 
> > Looks pretty what I'm looking for. Except that it still continues
> > to fill and keep track of the locks held by the current thread,
> > namely the copies in curr->held_locks.
> 
> Which is exactly what you need for that lock hierarchy recording you
> wanted :-)


Well, the lock hierarchy should probably be retrieved from the traces,
using state machines.
Otherwise we would need yet other lock events for that, which is going
to add even more overhead.