Re: Linux 6.11-rc1

[Linus Torvalds]

On Wed, 31 Jul 2024 at 08:55, Peter Zijlstra <peterz@xxxxxxxxxxxxx> wrote:
>
> Right, so Thomas found that i386-pti fails to map the entire entry text.
> Specifically pti_clone_pgtable() hard relies -- and does not verify --
> that the start address is aligned to the given granularity.
>
> Now, i386 does not align __entry_text_start, and so the termination
> condition goes sideways and pte_clone_entry() does not always work right
> and it becomes a games of code layout roulette.

Lovely.

> Also, should we just kill PTI on 32bit perhaps?

I don't think there's much technical reason to keep it - I can't
imagine any security-conscious people actually use 32-bit x86 any more
- but apart from fixing this bug I wonder how much of a maintenance
burden it is? I think most of the code is shared with 64-bit, isn't
it? The 32-bit case in many ways is simpler, even if it happened to
hit this odd alignment issue because it's obviously also a lot less
tested.

I'd rather kill highmem and X86_PAE, but I also suspect that horror
has a much larger chance of still being used.

The day we finally get rid of HIGHMEM I will dance on its grave. I
have hated that thing for a long long time.

              Linus