On Wed, Jul 09, 2025 at 11:31:07AM -0300, K R wrote:
> On Wed, Jul 9, 2025 at 10:34 AM Claudio Jeker <[email protected]> 
> wrote:
> >
> > On Wed, Jul 09, 2025 at 10:06:59AM -0300, K R wrote:
> > > Please find below another debug session.
> > >
> > > The machine is amd64, 32 cpus, 128GB RAM. Can reproduce the hang every
> > > time now by doing:
> > >
> > > cd /usr/ports/lang/rust
> > > while :
> > > do
> > >   make clean=build
> > >   make
> > > done
> > >
> > > Tested with a -current kernel, compiled with WITNESS and MP_LOCKDEBUG
> > > and __mp_lock_spinout = 50L * INT_MAX.
> > >
> > > I've included traces on all the cpus with state "stopped".  Couldn't
> > > get the "stopping" cpus to stop with mach stopcpu N.  Trying to mach
> > > ddbcpu N on a "stopping" cpu will freeze ddb.
> >
> > You are not running the latest version of -current. This issue has been
> > fixed a few days ago.
> 
> You are right, I've just got the latest version, recompiled and
> rebooted.  Thanks for the warning.
> 
> >
> > Could you also include a dmesg of that box.
> 
> Attached below.
 
So you have a dual socket Intel box. So this may be indeed a similar issue
where the current mutex implementation is to unfair and starves cores
until they hit the MP_LOCKDEBUG trigger.

-- 
:wq Claudio

Reply via email to