Jeremy Huntwork wrote:
Why are we doing this? I would think there are other things (and easier)
to test first before jumping straight away to the conclusion that it is
unionfs and building an entire CD with another poorly tested concept.
This concept won't lose its poorly-tested status until one tests it.
And, it is not entirely different. The init code differs only in the
following places:
1) loop setup code factored out to a function
2) added some calls to the dm library functions
3) different order of loop setup and fs mounting
The CD detection and init argument passing are exactly the same in trunk
and dm. The kernel only differs by including dm, zisofs and ext2 as
non-modues and not having unionfs at all.
OTOH, this "probably unionfs bug" conclusion seems logical enough to me.
1) Certain older unionfs version exhibited crashes/oopsed when editing
xorg.conf. I noticed that by sedding out /sbin/hotplug reference from
the kernel, I was able to hide that crash. That's why I say that early
/sbin/hotplug processes are a good stress-test.
2) At the 1.0.14 time, some unionfs bug reports on the list indicated
lockups without anything printed, as opposed to oopses.
What I think was done suboptimally here is the order in which Justin
built the CDs. The "just trunk with /sbin/hotplug sedded out of the
kernel" is certainly faster to build and would hide (but not really fix)
the unionfs stress issue, if any. My fault is that I have not correctly
indicated priorities to Justin.
In fact, we shouldn't be building an entire CD just to test the init.
Are you 100% sure that it is init? The "reproduced something similar on
Gentoo" comment makes me think it is not init.
What we should be doing is working to isolate the problem and trying to
reproduce it ourselves, using a systematic approach.
I have already reproduced a similar (but not identical) problem. See my
comment about the LiteOn CD-ROM. Unfortunately, I can't touch that box
now, it is a production server.
Just to save another mail: your proposal to build minimal isos with the
init that prints "hello world" is 99% good. I say 99% because it cannot
properly incorporate the "early /sbin/hotplug" stress-test. That doesn't
make it useless.
--
Alexander E. Patrakov
--
http://linuxfromscratch.org/mailman/listinfo/livecd
FAQ: http://www.linuxfromscratch.org/faq/
Unsubscribe: See the above information page
