Re: [PATCH RFC] kprobes: Remove MODULES dependency

[Jarkko Sakkinen]

On Fri, Jul 10, 2020 at 11:03:44AM +0200, Peter Zijlstra wrote:
> On Fri, Jul 10, 2020 at 02:45:19AM +0300, Jarkko Sakkinen wrote:
> > Remove MODULES dependency and migrate from module_alloc to vmalloc().
> > According to Andi, the history with this dependency is that kprobes
> > originally required custom LKM's, which does not hold today anymore.
> > 
> > Right now one has to compile LKM support only to enable kprobes.  With
> > this change applied, it is somewhat easier to create custom test
> > kernel's with a proper debugging capabilities, thus making Linux more
> > developer friendly.
> > 
> > Cc: Andi Kleen <ak@xxxxxxxxxxxxxxx>
> > Signed-off-by: Jarkko Sakkinen <jarkko.sakkinen@xxxxxxxxxxxxxxx>
> 
> NAK
> 
> this patch is horrific, it sprinkles a metric ton of #ifdef and silently
> disables a lot of kprobe features (like all the opt stuff).

Perfectly nderstandable. I just drafted something quick andy dirty
together for idea's sake (and put RFC tag to state that).

The application where I use this chhange, is when I refactor large patch
set that I'm working on (namely SGX patch set in my case). I just want
squeece all the extra out from the kernel build and still have means for
instrumentation. A static kernel is very convenient for this kind of
purpose, as with EFI stub and statically linked user space you can have
a single test binary.

> How about unconditionally providing module_alloc() instead?

I believe so, yes.

Just so that I know (and learn), what did exactly disable optprobes?
Not too familiar with this part of the kernel - that's why I'm asking.
Does the module_alloc to vmalloc change disable it?

/Jarkko