Re: 2.6.14-rc3-rt2

[Steven Rostedt <rostedt@goodmis.org>]

First, Please don't send to my kihontech.com account.  That's my email
that my customers use and I would really like to keep the spam noise down
an that account.  Thanks!

On Tue, 4 Oct 2005, Thomas Gleixner wrote:

> On Tue, 2005-10-04 at 11:58 -0700, Mark Knecht wrote:
> > > I guess its related to the priority leak I'm tracking down right now.
> > > Can you please set following config options and check if you get a bug
> > > similar to this ?
> > >
> > > BUG: init/1: leaked RT prio 98 (116)?
> > >
> > > Steven, it goes away when deadlock detection is enabled. Any pointers
>
> Thats actually a red hering caused by asymetric accounting which only
> happens when
>
> CONFIG_DEBUG_PREEMPT=y
> and
> # CONFIG_RT_DEADLOCK_DETECT is not set
>

Yep, I was going let you know but it seems you already figured it out :-)
The CONFIG_RT_DEADLOCK_DETECT turns on the trace_lock which makes all
locks run serially.  Without CONFIG_RT_DEADLOCK_DETECT, the importance of
the pi_lock of the task is greater.  So If something was changed that
didn't properly lock the pi_lock, then there could be problems with the
locking.

So looking at the patch you sent, are you saying that the leak was a false
positive?

I'm just starting to look at -rt7, and will be testing it today.

-- Steve