Re: [PATCH v8 0/6] Introduce CET supervisor state support

[Sean Christopherson <seanjc@google.com>]

On Fri, May 23, 2025, Dave Hansen wrote:
> On 5/23/25 09:57, Sean Christopherson wrote:
> > Side topic, and *probably* unrelated to this series, I tripped the following
> > WARN when running it through the KVM tests (though I don't think it has anything
> > to do with KVM?).  The WARN is the version of xfd_validate_state() that's guarded
> > by CONFIG_X86_DEBUG_FPU=y.
> > 
> >    WARNING: CPU: 232 PID: 15391 at arch/x86/kernel/fpu/xstate.c:1543 xfd_validate_state+0x65/0x70
> 
> Huh, and the two processes getting hit by it:
> 
>    CPU: 232 UID: 0 PID: 15391 Comm: DefaultEventMan ...
>    CPU: 77  UID: 0 PID: 14821 Comm: futex-default-S ...
> 
> don't _look_ like KVM test processes.

Yeah, that's why I haven't dug into it, I don't really know where to start, and
I don't even really know what triggered it.

> My guess would be it's some mixture of KVM and a signal handler fighting with
> XFD state.
> 
> I take it this is a Sapphire Rapids system?

Emerald Rapids

> Is there anything interesting about the config other than CONFIG_X86_DEBUG_FPU?

The only thing I can think of that's remotely interesting is CONFIG_PROVE_LOCKING=y.
Other than that, it's a pretty vanilla config.