Re: [PATCH v5 13/38] bpf: arena: mark arena_map_mmap() mappings VM_MIXEDMAP

["Lorenzo Stoakes (ARM)" <ljs@kernel.org>]

On Mon, Oct 12, 2026 at 08:56:23PM +0200, David Hildenbrand (Arm) wrote:
> On 10/12/26 17:55, Lorenzo Stoakes (ARM) wrote:
> > On Mon, Oct 12, 2026 at 05:08:09PM +0200, David Hildenbrand (Arm) wrote:
> >> On 10/9/26 20:04, Lorenzo Stoakes (ARM) wrote:
> >
> >>> diff --git a/kernel/bpf/arena.c b/kernel/bpf/arena.c
> >>> index 7b6847200b43..b69fe5e34339 100644
> >>> --- a/kernel/bpf/arena.c
> >>> +++ b/kernel/bpf/arena.c
> >>> @@ -620,8 +620,9 @@ static int arena_map_mmap(struct bpf_map *map, struct vm_area_struct *vma)
> >>>  	 * clears VM_MAYEXEC. Set VM_DONTEXPAND to avoid potential change
> >>>  	 * of user_vm_start. Set VM_DONTCOPY to prevent arena VMA from
> >>>  	 * being copied into the child process on fork.
> >>> +	 * This is a kernel page so set VM_MIXEDMAP.
> >>>  	 */
> >>> -	vm_flags_set(vma, VM_DONTEXPAND | VM_DONTCOPY);
> >>> +	vm_flags_set(vma, VM_MIXEDMAP | VM_DONTEXPAND | VM_DONTCOPY);
> >>>  	vma->vm_ops = &arena_vm_ops;
> >>>  	return 0;
> >>>  }
> >>>
> >>
> >> Historically, VM_MIXEDMAP really just means: can contain refcounted and
> >> non-refcounted things. Everything with a struct page is refcounted.
> >>
> >> I think we are now changing the semantics to mean "maps kernel pages". I think
> >
> > Yes.
> >
> > It was sort of implicit before given insert_page() etc. but still really meaning
> > that it _might_ be both.
>
> Yeah, it's in the name. Mixed: pfn and/or ordinary.

Well the name is confusing as a great many users never mix anything, and it's
far from obvious you need to set one.

Anyway, one for later. Maybe a mask like VMA_NON_MM_MANAGED_MASK or something
better than that...

>
> >
> >> this is fine, but we really have to update the documentation (and possibly
> >> rename this thing to something more expressive?)
> >
> > Yeah I completely agree.
> >
> > I will rename it, but I think that's best for 7.5 obviously! :)
>
> Yes.
>
> VM_PFNMAP / VM_MIXEDMAP - These contain kernel-mapped data which cannot be
> 			  reasonably reconstructed on page fault.
>
> the description is already pretty clear. just not that kernel-mapped pages
> *must* use one of these.

I think mixed map could do with a rename because it implies it _is_ mixed rather
than could be.

But again one for the future!

>
> (again, the ->fault thing returning allcoated kernel memory is nasty)

Yeah added to my todo to fix that. Hateful.

--
Cheers, Lorenzo