Re: [RFC][PATCH 3/3] x86: Add workaround to NMI iret woes

From: Steven Rostedt
Date: Fri Dec 09 2011 - 10:25:45 EST


On Fri, 2011-12-09 at 16:10 +0100, Peter Zijlstra wrote:

> KVM does NMIs, the write_idt thing is a paravirt call. Then again, a
> vcpu is a single thread of execution, there can only ever be 1 hypercall
> at the same time.
>
> The only thing that remains is if the hypercall interface is NMI-safe,
> what if the NMI comes in just as another context is also starting a
> hypercall. I really don't know enough about all that crap to know.

Hmm, well we can detect that we are a paravirt guest or not (that's the
nature of paravirt). We can keep the current methods around, an for
paravirt guests, it will fall back to the old stop_machine() method.
Heck guests have issues with latencies anyway.

For running on bare-metal, we can switch to this method.

-- Steve



--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
Please read the FAQ at http://www.tux.org/lkml/