[PATCH v8 00/43] arm64: Add support for LPA2 and WXN at stage 1

[Ard Biesheuvel]

On Fri, 16 Feb 2024 at 19:23, Ard Biesheuvel <ardb at kernel.org> wrote:
>
> On Fri, 16 Feb 2024 at 18:35, Catalin Marinas <catalin.marinas at arm.com> wrote:
> >
> > On Wed, 14 Feb 2024 13:28:46 +0100, Ard Biesheuvel wrote:
> > > This v8 covers the remaining changes that implement support for LPA2 and
> > > WXN at stage 1, now that some of the prerequisites are in place.
> > >
> > > v4: https://lore.kernel.org/r/20230912141549.278777-63-ardb@google.com/
> > > v5: https://lore.kernel.org/r/20231124101840.944737-41-ardb@google.com/
> > > v6: https://lore.kernel.org/r/20231129111555.3594833-43-ardb@google.com/
> > > v7: https://lore.kernel.org/r/20240123145258.1462979-52-ardb%2Bgit%40google.com/
> > >
> > > [...]
> >
> > I queued this series via the arm64 tree (for-next/stage1-lpa2). I tried
> > a couple of releases ago but for some reason my tests started failing at
> > it was very close to the merging window, so dropped. This time around,
> > if anything goes wrong, we have a bit of time to fix (it might as well
> > have been my test scripts and nothing to do with these patches).
> >
> > The last patch introducing WXN has ABI implications but it's default
> > off. I think we should keep the patch as certain markets will likely
> > turn it on.
> >
> > Surprisingly, there are no conflicts with Ryan's contpte series AFAICT
> > (I did a merge locally).
> >
>
> No *lexical* conflicts, right? :-)
>
> I built for-next/core with 16k pages/52-bits, and ended up with the
> splat below. Unfortunately, it is intermittent, and I haven't been
> able to reproduce it, so I have no idea whether it is my code, Ryan's
> code or an inadvertent interaction between the two. (Or perhaps some
> other code in the tree)
>

It's my code: on 16k without LPA2, the PUDs are folded at runtime, so
set_pud() is writing to swapper_pg_dir, which is mapped read-only.
I'll have a fix out shortly.