[OpenAFS] Re: kernel panics with 1.6.0 and 1.6.1pre2 on openindiana

Andrew Deason adeason@sinenomine.net
Sun, 29 Jan 2012 16:52:29 -0600


On Fri, 27 Jan 2012 17:25:52 -0800
"Logan O'Sullivan Bruns" <logan@gedanken.org> wrote:

> I can't think of anything weird about it. At this point other then
> OpenAFS it is still pretty close the basic 151a server install. I did
> install Solaris Studio 12.3 to build OpenAFS.
> 
> $ /opt/SUNWspro/bin/cc -V
> cc: Sun C 5.12 SunOS_i386 2011/11/16

I think this is the important difference.

$ /opt/SUNWspro/bin/cc -V
cc: Sun C 5.11 SunOS_i386 2010/08/13

I also forgot to ask which kernel module you were using. I assume you
used libafs64.nonfs.o.

> What did you use for your test system? If you used a VM did you
> configure it as 64 bit with multiple processors and greater then a
> gigabyte of memory (to trigger the 64 bit kernel install.)

It's single-proc, but if that were the problem I'd expect this to be
less consistent. It doesn't have that much memory, but it's easy enough
to force it to boot 64-bit anyway.

> Maybe you could email me the kernel module and afsd you used?

This is what I was going to suggest depending on the details you
provided. However, though I thought the machine was fine through a few
reboots on Friday, when I went to get these files, it started panic'ing
on boot every time. The panic location was the same, and in a very
trivial section of code (copying some constants around to populate an
error table). However, I noticed that there were some SSE-like
instructions around that code segment that the compiler put in, so I was
guessing that this may be due to compiler optimizations (I'm guessing
maybe we can't use SSE stuff this earlier in the boot process or
something? I didn't look into it, since we don't need them.)

We've had trouble in the past with newer Studio performing optimizations
that are not valid for kernel code (though this is the first I'm aware
of such a thing causing a _panic_), and I think as new releases come
out, more and more optimizations are done by default. I'll look at
changing the flags we give for libafs compiles, but in the meantime...

If I configure with --disable-optimize-kernel, the issue for me goes
away. Can you try that? (please save the build commands/output) If that
does not help, try the binaries I just sent you.

-- 
Andrew Deason
adeason@sinenomine.net