Re: Warning for buildok/context_test.stp on utrace kernel

[Wenji Huang <wenji dot huang at oracle dot com>]

Mark Wielaard wrote:
> On Wed, 2010-02-10 at 17:42 +0800, Wenji Huang wrote:
> > Hi,
> >
> > There will be warning if we run the following test on
> > RHEL5, Fedora kernel or utrace kernel.
> >
> > $ stap -p4 testsuite/buildok/context_test.stp
> > WARNING: "uprobe_get_pc" 
> > [/tmp/staphrGioI/stap_0aef4ead3fec39f2dee30c60b1b684d2_8042.ko] undefined!
> > /home/wjhuang/.systemtap/cache/0a/stap_0aef4ead3fec39f2dee30c60b1b684d2_8042.ko
> >
> > There reason is from print_ubacktrace() in probe.
> >
> > In fact, if we run stap -ve 'probe begin(print_ubacktrace()){}' on
> > utrace kernel without loaded uprobes module.  We can see the error
> >
> > Error inserting module 
> > '/tmp/stap0t5pyH/stap_5bd1a1d6b83ff84cfa8e7c67a402fa40_1153.ko': Unknown 
> > symbol in module
> >
> > since no symbol uprobe_get_pc can be resolved.
>
> And that is because the uprobes module isn't pulled in?
> This seems to come from commit 4c5ce7 which introduced a call to
> uprobe_get_pc() in __stp_stack_print().
>
> How can we make sure that if that is used (through ubacktrace) the
> uprobes module gets automagically loaded?
Yes, the commit causes the possibility of error. IMHO, it's not good way
to mix generic stack code with uprobe although CONFIG_FOO can smooth it.

> > Yes, the test case is a little picky, but maybe we can
> > process it in a more graceful way.
> >
> > BTW: there is no manual for print_ubacktrace, ubacktrace(), etc.
>
> I got man pages for all of them. Since they are documented in
> tapset/ucontext_unwind.stp they should show up in the TapSet Reference
> Manual and as individual man pages.
Sorry, I can't find the stuff in stapfuncs.3stap, other place?

Regards,
Wenji