testing on devel kernel

[Roland McGrath <roland at redhat dot com>]

Frank said he'd hit some crashes using recent utrace branch kernels
and asked about folks like me doing systemtap test suite runs on the
development kernels.

I ran tests on a Fedora 10 x86-64 machine.  For baseline I did:

# rpm -q systemtap-testsuite
systemtap-testsuite-0.9.5-1.fc10.x86_64
# make -C /usr/share/systemtap/testsuite installcheck
...

using the kernel-2.6.27.21-170.2.56.fc10.x86_64 kernel.

I didn't actually save the results, but the "unexpected failures"
number was 27, I think.

Then I booted my usual utrace development hacking kernel and did:

# ln -s /net/ma/home/roland/build/linux-utrace /lib/modules/`uname -r`
	^^^^ my build directory, has the vmlinux with -g
# stap -e 'probe kernel.function("sys_exit_group") { printf("%s %d exits\n", execname(), pid()) }' -c /bin/true
true 3205 exits
	^^^^ smoke test worked
# rpm -q systemtap-testsuite
systemtap-testsuite-0.9.5-1.fc10.x86_64
# make -C /usr/share/systemtap/testsuite installcheck
...

>From that:

# of unexpected failures	34

So not far off.  There was nothing like a kernel crash at all.
However, there was a lockdep complaint from inside uprobes.ko
code.  It looks probably harmless, I presume that is a known thing.

Is that the sort of testing you had in mind?

BTW, it would be nice if there were a handy way to run the whole suite
giving -r /something (or -p5 -r /something or whatever it has to be).
i.e., "make installcheck RUNNING_KERNEL=/my/build" instead of the
/lib/modules symlink hack.

I haven't tried it with git systemtap.  I assume just "make install"
and "make installcheck" on the test machine would be enough for that.
(The systemtap-testsuite rpm was just so temptingly easy to use!)

Since I see no indication of any crash like Frank talked about,
I don't plan to put more time now into stap testing.  Holler if
you actually hit something I should look at.


Thanks,
Roland