Re: [PATCH v11 3/4] x86/cpu: Do a sanity check on required feature bits

[H. Peter Anvin]

On March 23, 2026 2:50:49 PM PDT, Borislav Petkov <bp@xxxxxxxxx> wrote:
>On Mon, Mar 23, 2026 at 02:40:47PM -0700, H. Peter Anvin wrote:
>> I did: the PAE case, which was not hypothetical. I'm sure I can dig up more.
>
>So let's start our justification from this POV: we have use cases X and Y and they
>would profit from getting these warning checks about required features not
>being set.
>
>I still am not persuaded that we want Linux to boot on those. Because those
>CPUs are clearly "out of whack" and they would work by pure luck.
>
>But I don't know anything about the particular snafus that happened there.
>
>To take your example, the BIOS vendor disabled PAE. Do you really wanna deal
>with fixing that? I mean, what else is b0rked there...?
>
>So frankly, I'd turn that taint into a panic.
>
>Because this CPU sounds to me like an all-bets-are-off thing.
>

Yes, hence the tainting.

The PAE case is historic, but it was a concrete example I had in my head where this issue You also of course mention BIOS settings and there are VMs ...