Re: [PATCH V3 2/2] x86/tdx: Emit warning if IRQs are enabled during HLT #VE handling
From: Vishal Annapurve
Date: Mon Feb 10 2025 - 13:08:26 EST
On Fri, Feb 7, 2025 at 7:22 AM Sean Christopherson <seanjc@xxxxxxxxxx> wrote:
>
> On Thu, Feb 06, 2025, Vishal Annapurve wrote:
> ...
> > ---
> > arch/x86/coco/tdx/tdx.c | 5 +++++
> > 1 file changed, 5 insertions(+)
> >
> > diff --git a/arch/x86/coco/tdx/tdx.c b/arch/x86/coco/tdx/tdx.c
> > index 5e68758666a4..ed6738ea225c 100644
> > --- a/arch/x86/coco/tdx/tdx.c
> > +++ b/arch/x86/coco/tdx/tdx.c
> > @@ -393,6 +393,11 @@ static int handle_halt(struct ve_info *ve)
> > {
> > const bool irq_disabled = irqs_disabled();
> >
> > + if (!irq_disabled) {
>
> Wrap the check with WARN_ONCE(), doing so adds an unlikely to the <drum roll>
> unlikely scenario.
>
> > + WARN_ONCE(1, "HLT instruction emulation unsafe with irqs enabled\n");
>
> Newline is redundant, the WARN does that for you. IMO, it's also worth adding
> a comment, because this is like the fifth time "safe halt" has come up in the
> context of TDX.
>
> E.g.
>
> /*
> * HLT with IRQs enabled is unsafe, as an IRQ that is intended to be a
> * wake event may be consumed before requesting HLT emulation, leaving
> * the vCPU blocking indefinitely.
> */
> if (WARN_ONCE(!irq_disabled, "HLT emulation with IRQs enabled"))
> return -EIO;
Ack, will integrate this feedback in v4.
Thanks,
Vishal