On Fri, Aug 08, 2014 at 10:20:58AM -0700, Josh Stone wrote:
> On 08/07/2014 11:33 PM, Stefan Hajnoczi wrote:
> > The initscript currently fails for user-space probing scripts on systems where
> > uprobes.ko is built from source by stap(1).  This is because the initscript
> > uses a two-phase "compile and then run" approach:
> > 
> > The uprobes.ko module is generated during the compile phase but not copied into
> > the cache directory where modules are placed for the run phase.  The staprun(8)
> > command fails because the script module cannot be loaded without uprobes.ko.
> 
> This confused me at first, because uprobes.ko *is* cached with a
> kernel-hashed name -- see uprobes_pass() in buildrun.cxx.  But that's in
> SYSTEMTAP_DIR (default ~/.systemtap), and the initscript is talking
> about its own CACHE_PATH in /var/cache/systemtap.  So, ok.
> 
> I'm not terribly keen on using -k to find uprobes.  But at a minimum, if
> we do this, that "Keeping temp..." string is translatable, so you need
> to ensure stap runs in English.
> 
> It's already using -m, which triggers systemtap_session::save_module to
> copy the script module to $PWD at the end of passes_0_4() -- maybe this
> should also save uprobes.ko if needed?  Or we could add an explicit
> option to request this behavior.  Then the initscript can simply look in
> its own tmpdir to see if uprobes.ko was created.

An explicit stap option makes this cleaner since we don't have to manage
the temporary directory.

Will fix in v2.

Stefan