svn commit: r358363 - head/sys/vm

[Mark Johnston]

On Sun, Mar 01, 2020 at 10:58:17PM +0100, Jan Beich wrote:
> Mark Johnston <markj at freebsd.org> writes:
> 
> > On Sat, Feb 29, 2020 at 09:24:11PM +0100, Emmanuel Vadot wrote:
> >
> >> 
> >>  Hi Jeff,
> >> 
> >> On Thu, 27 Feb 2020 02:37:27 +0000 (UTC)
> >> Jeff Roberson <jeff at FreeBSD.org> wrote:
> >> 
> >> > Author: jeff
> >> > Date: Thu Feb 27 02:37:27 2020
> >> > New Revision: 358363
> >> > URL: https://svnweb.freebsd.org/changeset/base/358363
> >> > 
> >> > Log:
> >> >   Add unlocked grab* function variants that use lockless radix code to
> >> >   lookup pages.  These variants will fall back to their locked counterparts
> >> >   if the page is not present.
> >> >   
> >> >   Discussed with:	kib, markj
> >> >   Differential Revision:	https://reviews.freebsd.org/D23449
> >> > 
> >> > Modified:
> >> >   head/sys/vm/vm_page.c
> >> >   head/sys/vm/vm_page.h
> >> > 
> >> 
> >>  We're (jbeich@ and me) seeing wired memory leak since this commit.
> >>  A simple way to reproduce is to have drm-kmod (either 4.16 or 5.0) and
> >> liba-intel-driver/libva-intel-media-driver and start mpv in a loop :
> >> 
> >>  while mpv --hwdec=vaapi --start=10.0 --end=10.1 /path/to/file ; do
> >>  done
> >> 
> >>  Reverting this commit on one of my test machine shows that this is the
> >> culprit.
> >>  If you need anymore info/debug let me know.
> >> 
> >>  Thanks.
> >
> > Could you give this patch a try?
> >
> > diff --git a/sys/vm/vm_page.c b/sys/vm/vm_page.c
> > index 9d50881ed27e..41e7a5bb3099 100644
> > --- a/sys/vm/vm_page.c
> > +++ b/sys/vm/vm_page.c
> > @@ -4535,7 +4535,8 @@ vm_page_grab_valid(vm_page_t *mp, vm_object_t object, vm_pindex_t pindex, int al
> >  	    (VM_ALLOC_NOWAIT | VM_ALLOC_WAITFAIL | VM_ALLOC_ZERO)) == 0,
> >  	    ("vm_page_grab_valid: Invalid flags 0x%X", allocflags));
> >  	VM_OBJECT_ASSERT_WLOCKED(object);
> > -	pflags = allocflags & ~(VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY);
> > +	pflags = allocflags & ~(VM_ALLOC_NOBUSY | VM_ALLOC_SBUSY |
> > +	    VM_ALLOC_WIRED);
> >  	pflags |= VM_ALLOC_WAITFAIL;
> >  
> >  retrylookup:
> 
> I confirm, the patch fixes the issue. After trying to reproduce for a
> few minutes Wired doesn't grow several times larger than ARC Total.

Thanks, committed in r358505.