On Sun, Jul 15, 2007 at 06:54:07AM +0200, Erik Bertelsen wrote:
> >
> >Erik, can you "ls -lisa" the kernels, in addition to the info David asked
> >for? I wonder if this is an OFW bug and some kernels live beyound some
> >magic sector number.
> >
> >Martin
> >
> 
> I think that we're closing in now:
> 
> ls -lisa netbsd.*
> 353468 2544 -rwxr-xr-x  1 root  wheel  2584633 Mar  6 00:44 netbsd.030
> 353013 2560 -rwxr-xr-x  1 root  wheel  2589578 Mar 21 21:47 netbsd.031
> 352959 2560 -rwxr-xr-x  1 root  wheel  2590610 Mar 24 12:13 netbsd.032
> 352900 2560 -rwxr-xr-x  1 root  wheel  2601703 May 21 18:23 netbsd.033
> 352879 2576 -rwxr-xr-x  1 root  wheel  2617243 May 22 22:25 netbsd.034
> 352827 2576 -rwxr-xr-x  1 root  wheel  2618896 May 30 19:16 netbsd.035
> 352807 2576 -rwxr-xr-x  1 root  wheel  2618734 May 31 21:20 netbsd.036
> 352803 2592 -rwxr-xr-x  1 root  wheel  2625001 Jul  7 15:05 netbsd.037
> 499550 7648 -rwxr-xr-x  1 root  wheel  7810373 Jul 14 17:29 netbsd.generic
> 406929 6992 -rwxr-xr-x  1 root  wheel  7137645 Jul 13 11:27 netbsd.u2
> 406937 6992 -rwxr-xr-x  1 root  wheel  7137044 Jul 12 03:47 netbsd.u2.001
> 636152 6992 -rwxr-xr-x  1 root  wheel  7134381 Jul 13 13:07 netbsd.u3
> 636188 6992 -rwxr-xr-x  1 root  wheel  7133780 Jul 12 15:11 netbsd.u3.001
> 545785 6768 -rwxr-xr-x  2 root  wheel  6909426 Jul 12 13:18 netbsd.u4
> 656247 6752 -rwxr-xr-x  1 root  wheel  6882774 Jul 13 14:55 netbsd.u6
> 523027 7472 -rwxr-xr-x  1 root  wheel  7619973 Jul 13 22:38 netbsd.u7
> 675829 7632 -rwxr-xr-x  1 root  wheel  7791191 Jul 14 15:38 netbsd.u8
> 679577 7520 -rwxr-xr-x  1 root  wheel  7679388 Jul 14 22:00 netbsd.u9
>     4 7520 -rwxr-xr-x  1 root  wheel  7679388 Jul 15 05:58 netbsd.xx
> 
> 
> netbsd.generic and netbsd.u2{.001} do boot, the remainder of the
> netbsd.u? kernels don't.
> netbsd.xx was made by cp /netbsd.u9 /netbsd.xx. netbsd.u9 does not
> boot, but netbsd.xx does.
> The netbsd.03? kernels are pre-4.99.23 and did all boot.
> 
> Is there something magical about loading kernels with inodes being
> before or after 512K ?

Doesn't seem likely, the 'obvious' error would be failing to locate the
2nd cylinder group - but that would happen much earlier.
More likely is that the copy is in low-numbered sectors - possibly because
it got a low inode number.

If you generate additional hard links to netbsd.xx, you should (eventually)
get one with an inode number in the 5xxxxx range.  If that boots it isn't
an inode number problem.

However, since the fs code tries to allocate blocks in the same CG
as the inode, if there is a problem with big sector numbers it would tend
to affect big inode numbers.
The output of 'dumpfs -s /' might be informative (if terse).

	David

-- 
David Laight: david@l8s.co.uk