Re: ffs_alloccg: map corrupted

[Izumi Tsutsui <tsutsui%ceres.dti.ne.jp@localhost>]

> I've attempted a fresh releng HEAD install on an Amiga 1200,
> equipped with 68030+68882. The kernel panicked during extraction
> of the sets:
> 
> http://c0ff33.net/drop/ffs_alloccg_panic.jpg

Even after clean newfs?

I saw similar ffs panic on sh3 ports (due to bus_dma cache bugs)
and I had to newfs(8) again to recover it even after the bug was fixed
in kernel.

> Also, currently it's not possible to install the system with 68060 CPU.
> There seems to be a separate problem with 68060 FP emulation, which
> results in segfaults in pax, but I'm still analyzing this issue.
> Was the recent change to 68k FP emulation code tested on 68060?

Not yet confirmed very well, but I saw the following problems on
misc m68k ports recently:

 - at least 20110823 kernels on atari TT030 seems stable enough
   (can build pkgsrc ruby18 (requires ~300MB swap) and misc other packages)

 - with recent (around 20111013) userland ruby18 *sometimes* gets
   illegal instruction errors both on luna68k and x68k (both have 68030),
   but the same ruby18 binary works fine on userland around 20110823
   even with recent kernel

 - as mentioned in PR/45519 log, x68k kernel on 030 gets weird panic
   (no stack trace except cpu_Debugger)

So it might be worth to build release sets from sources around 20110823.

I'm afraid MI UVM might also have some bugs after COLORMATCH was
implemented around 20110901, and I also wonder if newer binutils
imported around 20110924 could cause other errors.
(too many MI changes to confirm on slower m68k machines...)

---
Izumi Tsutsui