Re: [PATCH 2/2] kernel: add support for live patching

[Josh Poimboeuf <jpoimboe@redhat.com>]

On Fri, Nov 07, 2014 at 02:13:37PM +0100, Jiri Kosina wrote:
> On Fri, 7 Nov 2014, Josh Poimboeuf wrote:
> 
> > > Also, lpc_create_object(), lpc_create_func(), lpc_create_patch(), 
> > > lpc_create_objects(), lpc_create_funcs(), ... they all are pretty much 
> > > alike, and are asking for some kind of unification ... perhaps iterator 
> > > for generic structure initialization?
> > 
> > The allocation and initialization code is very simple and
> > straightforward.  I really don't see a problem there.
> 
> This really boils down to the question I had in previous mail, whether 
> three-level hierarchy (patch->object->funcs), which is why there is a lot 
> of very alike initialization code, is not a bit over-designed.

Oh sorry, I missed that point :-)  See below.
> 
> > > I am not also really fully convinced that we need the 
> > > patch->object->funcs abstraction hierarchy (which also contributes to 
> > > the structure allocation being rather a spaghetti copy/paste code) ... 
> > > wouldn't patch->funcs be suffcient, with the "object" being made just 
> > > a property of the function, for example?

The patched object represents the module being patched (or "vmlinux").
It is much more than a property of the function.  Multiple functions can
be patched in the same object.  There are several things we do on a
per-object basis, including try_module_get(), deferred module patching
(patching from the module notifier), and dynamic relocations.

> > > 
> > > > Plus, I show that kernel/kgraft.c + kernel/kgraft_files.c is
> > > > 906+193=1099.  I'd say they are about the same size :)
> > > 
> > > Which is still seem to me to be a ratio worth thinking about improving 
> > > :)
> > 
> > Yes, this code doesn't have a consistency model, but it does have some
> > other non-kGraft things like dynamic relocations, 
> 
> BTW we need to put those into arch/x86/ as they are unfortunately not 
> generic. But more on this later independently.
> 
> > deferred module patching,
> 
> FWIW kgraft supports that as well.
> 
> > and a unified API.  There's really no point in comparing lines of code.
> 
> Oh, sure, I didn't mean that this is any kind of metrics that should be 
> taken too seriously at all. I was just expressing my surprise that 
> unification of the API would bring so much code that it makes the result 
> comparably sized to "the whole thing" :)
> 
> Thanks,
> 
> -- 
> Jiri Kosina
> SUSE Labs

-- 
Josh