On 02/05/2018 09:23 AM, John Paul Adrian Glaubitz wrote:
On 02/04/2018 09:58 PM, Frank Scheiner wrote:
* libcc1-0 (8-20180130-1 => 7.3.0-1+b1)
* libgcc1 (1:8-20180130-1 => 1:7.3.0-1+b1)
* libstdc++6 (8-20180130-1 => 7.3.0-1+b1)
...to see if this makes a difference and after recreating the initramfs and
rebooting, I again got segfaults when trying to login.
So better ignore the *7.3.0-1+b1 versions of these packages and use the ones from the
"experimental" suite or wait for a fix, if you depend on serial logins.
The 7.3.0-1+b1 versions are actually supposed to work because those have
been rebuilt with the fixed version of binutils. 7.3.0-1 is expected to
crash, 7.3.0-1+b1 is *supposed* to work. I am really surprised that you
are having issues with 7.3.0-1+b1.
But I will just trigger a rebuild of gcc-7 once more to make sure there
were no artifacts left when building with the previously broken compiler.
Please wait for 7.3.0-1+b2 to show up in the archives and then try again
to see if that resolves the issue.
Sorry, but my tests this morning after I again upgraded to the *8-*
versions of the above libs plus rebuilding initramfs again and reboot
showed segfaults when trying to login via serial console.
So I think I have to revert my conclusion from yesterday, it now more
looks like a "sporadic" issue or a local problem. For a sproadic issues
it'd be strange that this hit both T1000 and T5220. And as I used the
same disk on both machines, maybe it's the disk that creates the issues,
e.g. bad blocks. This could make sense, because netbooting the exact
same kernel and initramfs (w/on-disk root FS) that allowed successful
serial logins yesterday, today also has issues.
I'll have to check with a new disk to be sure. Well, gives me the
opportunity to test another machine with the SPARC64 installer ISO.
Sorry for the noise. :-/
Frank