> ok - with the last fix from steve xfs on ppc starts to be useable
> (but it definitely has bugs) ... for instance the following is
> possible:
>
> [root@aqua /root]# uname -a
> Linux aqua 2.4.0-test5 #5 Wed Aug 30 04:08:43 CEST 2000 ppc unknown
> [root@aqua /root]# mount ^^^
> /dev/hda8 on / type xfs (rw) <---
> none on /proc type proc (rw)
> /dev/hda5 on /boot type hfs (ro)
> /dev/hda9 on /home type ext2 (rw)
> none on /dev/pts type devpts (rw,gid=5,mode=620)
> [root@aqua /root]#
>
> so it works quite well (i at least got a 2gb filesystem copied over
> to xfs and was able to boot from it :-) ... also the journal replay
> code seems to work somehow (got it mounted clean after an unclean
> umount) ... but the system did not survive a reboot after the
> xfs root experiment ... also i still get invalid inode numbers
> (it started at the end of copying the disk over which is close
> to completely full) at
>
> aqua kernel: XFS assertion failed: xfs_dir_ino_validate(mp,
> INT_GET(dep->inumber, ARCH_CONVERT)) == 0, file: xfs_dir2_data.c,
> line: 175
>
> which is:
>
> ASSERT(xfs_dir_ino_validate(mp, INT_GET(dep->inumber, ARCH_CONVERT)) == 0);
>
> in xfs_dir2_data.c (btw. i have changed asserts to not result in
> a BUG() )
All I can say here is that with debug turned on this function is called all
over the place to scan all the entries in a directory and validate their
consistency. The xfs_dir_ino_validate function will print out the
offending inode number (as a 64 bit hex number), could you possibly
modify this to be just a printk and capture the offending number.
>
> one thing related to this seems to be that the root-dir of the
> xfs filesystem does not have a ".." entry visible (but like with
> the test entry in the last weeks it's there)
Are you saying not visible from user space, or not visible from xfs_db?
I would not be surprised if there is not another endian conversion macro
where the wrong half of a 64 bit value is being copied around in the big
endian case.
>
> also now it would be really good to have a working non-sim
> xfs_repair ...
Does the non-sim version build on the ppc? If it does then xfs_repair -n
output would be a good thing to see.
Steve
|