Re: lock order reversal

On Wednesday 30 April 2003 18:28, Georg Funk wrote:
> hm, the patch doesn't seem to work:
>
> --------------------------------------------------
> #patch < pcm.diff
> Hmm...  Looks like a normal diff to me...
> The text leading up to this was:
> --------------------------
>
> |cvs diff: Diffing sys/dev/sound/pcm
> |Index: sys/dev/sound/pcm/dsp.c
> |===================================================================
> |RCS file:
>
> /old/pub/FreeBSD/development/FreeBSD-CVS/src/sys/dev/sound/pcm/dsp.c,v
>
> |retrieving revision 1.62
> |diff -r1.62 dsp.c
>
> --------------------------
> Patching file dsp.c using Plan A...
> Hunk #1 failed at 265.
> Hunk #2 failed at 270.
> Hunk #3 failed at 287.
> Hunk #4 succeeded at 304.
> 3 out of 4 hunks failed--saving rejects to dsp.c.rej
> done
>
> On Wednesday 30 April 2003 16:33, you wrote:
> > On Wednesday 30 April 2003 14:55, Georg Funk wrote:
> > > Hi!
> > > I'm running the FreeBSD 5.0-RELEASE. When I boot up, this appears:
> > >
> > > 1st 0xc12a7230 process lock (process lock) _at_
> > > ../../../kern/kern_descrip.c:2112 lock order reversal
> > > I haven't noticed any disfunctions, but what does this mean?
> > >
> > > I get a similar error when I use my soundcard:
> > >
> > > lock order reversal
> > >  1st 0xc120a680 pcm0 (sound softc) _at_ ../../../dev/sound/isa/mss.c:177
> > >  2nd 0xc120a740 pcm0:play:0 (pcm channel) _at_
> > > ../../../dev/sound/pcm/channel.c:441
> > >
> > > I have to restart the KDE artsd before the soundcard works correct.
> > > Please give me advice how to solve this.
> > >
> > > Thanks in advance, Georg Funk
> >
> > I have also seen that problem, and this patch seems to work here..
> > It is probably not very optimized, but the lock order reversal didn't
> > show up again...
> >
> > cvs diff: Diffing sys/dev/sound/pcm
> > Index: sys/dev/sound/pcm/dsp.c
> > ===================================================================
> > RCS file:
> > /old/pub/FreeBSD/development/FreeBSD-CVS/src/sys/dev/sound/pcm/dsp.c,v
> > retrieving revision 1.62
> > diff -r1.62 dsp.c
> > 265,266c265
> > <       pcm_unlock(d);
> > <       /* finished with snddev, new channels still locked */
> > ---
> >
> > >       /* keep the snddev lock to avoid a lock order reversal, new
> > > channels still locked */
> >
> > 271d269
> > <                       pcm_lock(d);
> > 289d286
> > <                       pcm_lock(d);
> > 306a304
> >
> > >       pcm_unlock(d);
> >

Sorry, I was in a hurry. I was supposed to be at work 30 minutes later...

This one should work:

--- dsp.c       Wed Apr 30 22:15:50 2003
+++ /home/sigsegv/pcm/dsp.c     Wed Apr 30 21:55:38 2003
_at__at_ -262,13 +262,11 _at__at_

        i_dev->si_drv1 = rdch;
        i_dev->si_drv2 = wrch;
-       pcm_unlock(d);
-       /* finished with snddev, new channels still locked */
+       /* keep the snddev lock to avoid a lock order reversal, new channels 
still locked */

        /* bump refcounts, reset and unlock any channels that we just opened 
*/
        if (flags & FREAD) {
                if (chn_reset(rdch, fmt)) {
-                       pcm_lock(d);
                        pcm_chnrelease(rdch);
                        i_dev->si_drv1 = NULL;
                        if (wrch && (flags & FWRITE)) {
_at__at_ -286,7 +284,6 _at__at_
        }
        if (flags & FWRITE) {
                if (chn_reset(wrch, fmt)) {
-                       pcm_lock(d);
                        pcm_chnrelease(wrch);
                        i_dev->si_drv2 = NULL;
                        if (flags & FREAD) {
_at__at_ -304,6 +301,7 _at__at_
                pcm_chnref(wrch, 1);
                CHN_UNLOCK(wrch);
        }
+       pcm_unlock(d);
        splx(s);
        return 0;
 }