Re: amd64 -current crashs at boot

[Sverre Froyen <sverre%viewmark.com@localhost>]

On Tuesday 23 December 2008, Christoph Egger wrote:
> Sverre Froyen wrote:
> > On Tuesday 23 December 2008, Patrick Welche wrote:
> >> On Mon, Dec 22, 2008 at 12:10:59PM +0000, Andrew Doran wrote:
> >>> It should be fixed now, I'm about to check.
> >>
> >> Just quickly grabbed ffs_vnops.c v 1.107 - the panic I had has gone, and
> >> now I hit the same on as Christoph:
> >>
> >> 0xc01df968 in ffs_update (vp=0xcb420000, acc=0x0, mod=0x0, updflags=0)
> >>     at ../../../../ufs/ffs/ffs_inode.c:121
> >> 121             if (vp->v_mount->mnt_flag & MNT_RDONLY)
> >> (gdb) bt
> >> #0  0xc01df968 in ffs_update (vp=0xcb420000, acc=0x0, mod=0x0,
> >> updflags=0) at ../../../../ufs/ffs/ffs_inode.c:121
> >> #1  0xc01ec57d in ffs_full_fsync (vp=0xcb420000, flags=520)
> >>     at ../../../../ufs/ffs/ffs_vnops.c:580
> >> #2  0xc03b843b in spec_fsync (v=0xcb445cd8)
> >>     at ../../../../miscfs/specfs/spec_vnops.c:781
> >
> > spec_vnops.c:781 says:
> >
> > error = VFS_FSYNC(mp, vp, ap->a_flags | FSYNC_VFS);
> >
> > Does this, i.e., "| FSYNC_VFS" imply more crack smoking?
>
> I tried this w/o FSYNC_VFS flag. I trigger the KASSERT in
> sys/ufs/ffs/ffs_vnops.c, line 421 then.
>
> backtrace:
> breakpoint+0x5
> panic+0x24d
> __kernassert+0x2d
> ffs_full_fsync+0x1d7
> spec_fsync+0x53
> VOP_FSYNC+0x71
> sched_sync+0x14f

And "ffsino" is set too -- resulting in a call to ffs_update without the 
assert.  Sounds like we have to wait for Andrew to clarify.

-- Sverre

PS.  The assert checks "vp->v_tag".  There is a comment in sys/sys/vnode.h 
that vnode tags "are for the benefit of external programs only (e.g., pstat) 
and should NEVER be inspected by the kernel".  Perhaps that comment should be 
modified?