svn commit: r206878 - head/sys/kern

[Marius Strobl]

On Mon, May 10, 2010 at 12:27:40AM +0200, Attilio Rao wrote:
> 2010/5/9 Marius Strobl <marius at alchemy.franken.de>:
> > On Mon, Apr 19, 2010 at 11:27:54PM +0000, Attilio Rao wrote:
> >> Author: attilio
> >> Date: Mon Apr 19 23:27:54 2010
> >> New Revision: 206878
> >> URL: http://svn.freebsd.org/changeset/base/206878
> >>
> >> Log:
> >> Â  Fix a deadlock in the shutdown code:
> >> Â  When performing a smp_rendezvous() or more likely, on amd64 and i386,
> >> Â  a smp_tlb_shootdown() the caller will end up with the smp_ipi_mtx
> >> Â  spinlock held, busy-waiting for other CPUs to acknowledge the operation.
> >> Â  As long as CPUs are suspended (via cpu_reset()) between the active mask
> >> Â  read and IPI sending there can be a deadlock where the caller will wait
> >> Â  forever for a dead CPU to acknowledge the operation.
> >> Â  Please note that on CPU0 that is going to be someway heavier because of
> >> Â  the spinlocks being disabled earlier than quitting the machine.
> >>
> >> Â  Fix this bug by calling cpu_reset() with the smp_ipi_mtx held.
> >> Â  Note that it is very likely that a saner offline/online CPUs mechanism
> >> Â  will help heavilly in fixing similar cases as it is likely more bugs
> >> Â  of this type may arise in the future.
> >>
> >
> > This change causes a hang for me when running an SMP kernel on an
> > UP machine or an MP machine with SMP disabled as in these cases
> > smp_ipi_mtx isn't initialized (see mp_start()). The below patch
> > fixes this for me.
> 
> Marius,
> thanks a lot for reporting this.
> However, I think that a better fix is to always initialize the mutex.
> Do you think the following patch is fine for you?:
> http://www.freebsd.org/~attilio/smpipi_fixup.diff
> 

Yes, this fixes it as well.

Marius