Re: [PATCH bpf-next v2 0/5] execmem_alloc for BPF programs

[Mike Rapoport <rppt@kernel.org>]

On Thu, Nov 17, 2022 at 10:36:43AM -0800, Song Liu wrote:
> On Thu, Nov 17, 2022 at 12:50 AM Mike Rapoport <rppt@kernel.org> wrote:
> >
> > On Mon, Nov 14, 2022 at 12:30:49PM -0800, Song Liu wrote:
> > > On Sun, Nov 13, 2022 at 2:35 AM Mike Rapoport <rppt@kernel.org> wrote:
> > > >
> > > > My concern is that the proposed execmem_alloc() cannot be used for
> > > > centralized handling of loading text. I'm not familiar enough with
> > > > modules/ftrace/kprobes/BPF to clearly identify the potential caveats, but
> > > > my gut feeling is that the proposed execmem_alloc() won't be an improvement
> > > > but rather a hindrance for moving to centralized handling of loading text.
> > >
> > > I don't follow why this could ever be a hindrance. Luis is very excited about
> > > this, and I am very sure it works for ftrace, kprobe, and BPF.
> >
> > Again, it's a gut feeling. But for execmem_alloc() to be a unified place of
> > code allocation, it has to work for all architectures. If architectures
> > have to override it, then where is the unification?
> >
> > The implementation you propose if great for x86, but to see it as unified
> > solution it should be good at least for the major architectures.
> 
> As I mentioned earlier, folks are working on using bpf_prog_pack for BPF
> JIT on powerpc. We will also work on something similar for ARM.

Does "something similar" mean that it won't use execmem_alloc() as is?

> I guess these are good enough for major architectures?

Sorry if I wasn't clear, I referred for unified solution for all code
allocations, not only BPF, so that execmem_alloc() will eventually replace
module_alloc(). And that means it has to be able to deal with with
architecture specific requirements at least on ARM and powerpc, probably
others as well.

> > > > It feels to me that a lot of ground work is needed to get to the point
> > > > where we can use centralized handling of loading text.
> > >
> > > Could you please be more specific on what is needed?
> >
> > The most obvious one to implement Peter's suggestion with VM_TOPDOWN_VMAP
> > so that execmem_alloc() can be actually used by modules.
> 
> Current implementation is an alternative to VM_TOPDOWN_VMAP. I am
> very sure it works for modules just like VM_TOPDOWN_VMAP solution.

It might, but it still does not. And until they do I consider these
patches as an optimization for BFP rather than unification of code
allocations.
 
> Thanks,
> Song

-- 
Sincerely yours,
Mike.