Re: [PATCH v2] x86/cpu: Fix migration safety with X86_BUG_NULL_SEL

From: Borislav Petkov <bp@alien8.de>
To: "H. Peter Anvin" <hpa@zytor.com>
Cc: Jane Malalane <jane.malalane@citrix.com>,
	LKML <linux-kernel@vger.kernel.org>,
	x86@kernel.org, Thomas Gleixner <tglx@linutronix.de>,
	Ingo Molnar <mingo@redhat.com>, Pu Wen <puwen@hygon.cn>,
	Paolo Bonzini <pbonzini@redhat.com>,
	Sean Christopherson <seanjc@google.com>,
	Peter Zijlstra <peterz@infradead.org>,
	Andrew Cooper <andrew.cooper3@citrix.com>,
	Yazen Ghannam <Yazen.Ghannam@amd.com>,
	Brijesh Singh <brijesh.singh@amd.com>,
	Huang Rui <ray.huang@amd.com>, Andy Lutomirski <luto@kernel.org>,
	Kim Phillips <kim.phillips@amd.com>,
	stable@vger.kernel.org
Subject: Re: [PATCH v2] x86/cpu: Fix migration safety with X86_BUG_NULL_SEL
Date: Mon, 18 Oct 2021 21:31:19 +0200	[thread overview]
Message-ID: <YW3Lhwzux0K1P72e@zn.tnic> (raw)
In-Reply-To: <35CA4584-4CFD-4E47-9825-6438D9ED4ECC@zytor.com>

On Mon, Oct 18, 2021 at 12:10:20PM -0700, H. Peter Anvin wrote:
> AFAIK no Intel CPU has ever had that behavior, and always cleared the
> segments; I don't Intel has any plans of supporting such a CPUID bit
> (although I'd certainly be willing to take such a request back to the
> CPU teams on request.)

No need - we can always set or clear a flag on Intel, depending on what
we do.

> That being said, this sounds like an ideal use for the hypervisor CPU
> feature flag.

Yap, it uses it.

> Maybe we should consider a migration hypervisor flag too to explicitly
> tell the kernel not to rely on hardware probing that breaks migration
> in general.

Meh, migration-specific flag calls for all kinds of nasty when each
HV would want different things to happen in the guest, for migration.
And then the patch flood will come. I mean, we already do a bunch of
X86_FEATURE_HYPERVISOR all over the place and apparently it is enough
here too...

> Now, with a CPUID but being introduced, the right thing would be to
> use the CPUID bit as a feature instead of using a bug flag, and add
> whitelisting in the vendor-specific code as applicable.

I guess we can flip all that logic checking X86_BUG_NULL_SEG... it
sounds like a lot of churn to me, though and I don't see a pressing need
for it unless someone is bored and wants to do some kernel patching
exercises but whatever...

-- 
Regards/Gruss,
    Boris.

https://people.kernel.org/tglx/notes-about-netiquette