Re: irq lock inversion

[Jens Axboe <jens.axboe@oracle.com>]

On Mon, Nov 09 2009, Ingo Molnar wrote:
> 
> * Jens Axboe <jens.axboe@oracle.com> wrote:
> 
> > On Sun, Nov 08 2009, Ingo Molnar wrote:
> > > 
> > > * Jens Axboe <jens.axboe@oracle.com> wrote:
> > > 
> > > > On Fri, Nov 06 2009, Ingo Molnar wrote:
> > > > > Read the lockdep message please, consider all the backtraces it prints, 
> > > > > it says something different.
> > > > 
> > > > In all honesty, reading and parsing lockdep messages requires a 
> > > > special state of mind. IOW, readability is not its high point.
> > > 
> > > We frequently do patches to improve the messages but there's a hard 
> > > limit: generally the messages mirror the complexity of the underlying 
> > > locking scenario.
> > > 
> > > Unfortunately lockdep cannot pretend something is simple when it is not. 
> > > There are two ways out of that: either to simplify the underlying 
> > > locking rules, or to understand them.
> > 
> > I think the primary problem is that it tries to condense too much 
> > information, instead of just spelling it out. That may be obvious to a 
> > person intimately familiar with lockdep, but not to others. Things 
> > like the STATE line, for instance. It would read a lot easier if these 
> > things were just spelled out.
> > 
> > I know this message isn't really productive, just tossing it out 
> > there. I'll try to to back it up with a patch the next time it annoys 
> > me :-)
> 
> Well, previously lockdep spewed out a lot of info, which we condensed 
> down because people complained ;-)

Heh, can't win 'em all! Stack traces are so large anyway that I don't
think saving 1-2 lines per lock in the trace would make much of a
difference. It's debug output, after all.

-- 
Jens Axboe