On Wed, Jun 15, 2022 at 08:37:07AM +0200, hch@lst.de wrote:

On Tue, Jun 14, 2022 at 03:32:38PM +0300, jarkko@kernel.org wrote:

quoted

quoted

Like say for a next step we moved prog pack out of bpf into core code,
gave it it's own copy of module_alloc(), and then made kprobes use it.
Then we would have something with improved W^X guard rails, and kprobes
would not depend on modules anymore. I think maybe it's a step in the
right direction, even if it's not perfect.

So you're saying that I should (as a first step) basically clone
module_alloc() implementation for kprobes, and future for BPF 
use, in order to get a clean starting point?

I don't think cloning the code helps anyone.  The fact that except
for the eBPF mess everyone uses module_alloc and the related
infrastructure is a feature and not a bug.  The interface should
become better than what we have right now, but there is few enough
users that this can be done in one go.

So assuming we really care deeply enough about fancy tracing without
modules (and I'm not sure we do, even if you don't use modules it
doesn't hurt to just build the modules code, I do that all the time
for my test machines), the general approach in your series is the
right one.

OK, thanks for the elaboration!

However I bake it, I doubt that next version is going to be the final
version, given all the angles. Therefore, I mostly Christophe's
suggestions on compilation flags, and also split this into per-arch
patches.

That should be at least to the right direction.

BR, Jarkko