Re: [PATCH 2/4] x86/entry: Use asm_noist_exc_nmi() for NMI in early booting stage

[Thomas Gleixner]

On Mon, May 03 2021 at 23:45, Thomas Gleixner wrote:
> The problem was introduced in 4.14 with b70543a0b2b6 ("x86/idt: Move
> regular trap init to tables").
>
> Before that trap_init() installed an IST gate right away, but looking
> deeper this was broken forever because there is a hen and egg problem.
>
> ISTs only work after TSS is initialized and the ordering here is:
>
> trap_init()
>   init_idt()
>   cpu_init()
>     init_tss()
>
> So the original code had a race window between init_idt() and
> init_tss(). Any IST using exception in that window goes south because
> TSS is not initialized.
>
> b70543a0b2b6 traded the above with that NMI issue. All other
> exceptions are fine...
>
> I'll think about it tomorrow some more...

It does not really matter which way around we do it. Even if we do that
noist dance then still any NMI hitting _before_ init_idt() is going to
lala land. So having this tiny step in between is more or less cosmetic.

And just for completness sake, I don't see a reason why we have to set
up the idt gates _before_ the TSS muck, i.e. before cpu_init().

The only thing cpu_init() needs working which is not installed in the
early_idt is #GP because some cpu init code uses rd/wrmsrl_safe(). But
that's pretty much all of it.

So this wants a proper cleanup and not some paper over it with an extra
step and I don't see a reason why any of this should be backported
simply because it does not matter at all whether the early idt which
only populates a few essential gates is active for a bit longer.

So what we need is a solution for that KVM wreckage but that can be
stand alone.

Thanks,

        tglx