Re: instrumenting vs. module loading

[Ananth N Mavinakayanahalli <amavin at redhat dot com>]

Frank Ch. Eigler wrote:
> Hi -
>
> On Fri, May 13, 2005 at 10:29:09AM -0400, William Cohen wrote:
>
>
> > [...]  Hmmm, instrumenting each inline function site? For the kernel
> > that is known.  What happens when a module is loaded that uses an
> > instrumented inline function?
>
>
> The way the translator probe points are going to be specified, the
> target software is identified clearly (a particular module, or the
> whole kernel).  If a probe point is for a module, and it's not loaded
> at probe startup time, the initialization should abort.  Let's not
> support a wildcard syntax for kernel-module selection until the
> implications are better analyzed.
>
>
> > Or unload for that matter?
>
>
> I imagine the translator arranging to increment the use-count of
> modules that it places instrumentation into, to prevent their
> premature removal.  (How does raw kprobes deal with this case?)

Please correct me if my understanding is wrong: Are you speaking of a
case where a kprobe is placed in a module text and than the module is 
unloaded?

We will anyway not have any subsequent hits of such a kprobe. There is
currently no "scavenging" done for such kprobes and in any case, kprobes
core doesn't have the knowledge if the kp.addr is a module text address
or not.

I don't have much idea about in kernel details of module loading and
unloading, but I'd imagine we'll encounter "interesting" issues if a
different module is loaded in the same text range with such a stray
kprobe.

Ananth