Re: Debian SID kernel results in "Fast Data Access MMU Miss" on Sun Ultra 30

[Stan Johnson <userm57@yahoo.com>]

Hi Adrian,

On 7/31/23 9:15 AM, John Paul Adrian Glaubitz wrote:
> 
> 
>> On Jul 31, 2023, at 4:33 PM, Stan Johnson <userm57@yahoo.com> wrote:
>>
>> I don't think a kernel bisect will identify the issue, since the kernel
>> boots ok. This has something to do with the initrd, and I don't know how
>> to troubleshoot that. Any recommendations?
> 
> It can still be a bug in the kernel, especially since we’re not seeing this issue on the T5220 and the SPARC-T4.
> 
> I bet if you rebuild the initrd for the older kernels, these will still boot fine which would prove my theory.
> ...

Booting an older working kernel (vmlinux-5.16.0-6-sparc64), I rebuilt
these initrd files:

# mkinitramfs -o initrd.img-5.16.0-6-sparc64 5.16.0-6-sparc64
# mkinitramfs -o initrd.img-6.1.0-7-sparc64 6.1.0-7-sparc64
# mkinitramfs -o initrd.img-6.4.0-1-sparc64 6.4.0-1-sparc64

Then I tried booting the following combinations of kernels and initrd files,

1) vmlinux-5.16.0-6-sparc64, initrd.img-5.16.0-6-sparc64 -- pass
2) vmlinux-6.1.0-7-sparc64, initrd.img-6.1.0-7-sparc64   -- fail
3) vmlinux-6.4.0-1-sparc64, initrd.img-6.4.0-1-sparc64   -- fail

Note:
 For these tests:
   fail -> "Fast Data Access MMU Miss" while loading the initrd
   pass -> normal boot to multiuser

I rebuilt the above initrd files again while booted into my custom 6.2
kernel and tried booting the following combinations:

4) vmlinux-5.16.0-6-sparc64, initrd.img-5.16.0-6-sparc64 -- pass
5) vmlinux-6.1.0-7-sparc64, initrd.img-6.1.0-7-sparc64   -- fail
6) vmlinux-6.4.0-1-sparc64, initrd.img-6.4.0-1-sparc64   -- fail

So that suggests the problem is not with mkinitramfs, since each initrd
behaves the same regardless of the kernel that was booted while each
initrd was built.

Next, I tried this:

7) vmlinux-6.4.0-1-sparc64, initrd.img-5.16.0-6-sparc64  -- fail

So I could try a kernel bisect next, but it would be a long process:
a) Cross-compile stock kernels and modules using the latest Debian SID
config file.
b) Copy new kernels and modules to the Ultra 30; boot a known working
kernel to build each corresponding initrd file.
c) Try booting the new kernel/initrd; a "Fast Data Access MMU Miss"
while loading the initrd is a failure; otherwise it passes.

Alternately, I could use a spare disk and install a default Debian SID
installation that includes GRUB. Perhaps there's a memory or some other
issue with SILO (I'm using SILO 1.4.14). I won't be using GRUB on my
Ultra-30 as long as the GRUB partition must remain mounted, since
there's a risk of data corruption if the system crashes. SILO works more
like yaboot in that it doesn't require its filesystem to remain mounted,
so it will not be affected by a crash, and once configured, there's no
need to update silo.conf, just the symbolic links to the kernel and
initrd. But I'll be happy to do a clean-install test if anyone thinks
that the "Fast Data Access MMU Miss" error might be related to SILO;
otherwise, I'll try a kernel bisect. Let me know if there's a preference.

-Stan