Re: [PATCH v1] KVM: arm64: selftests: Test pointer authentication support in KVM guest

From: Oliver Upton <oliver.upton@linux.dev>
To: Jing Zhang <jingzhangos@google.com>
Cc: Marc Zyngier <maz@kernel.org>, KVM <kvm@vger.kernel.org>,
	KVMARM <kvmarm@lists.linux.dev>,
	ARMLinux <linux-arm-kernel@lists.infradead.org>,
	Paolo Bonzini <pbonzini@redhat.com>,
	James Morse <james.morse@arm.com>,
	Suzuki K Poulose <suzuki.poulose@arm.com>,
	Zenghui Yu <yuzenghui@huawei.com>
Subject: Re: [PATCH v1] KVM: arm64: selftests: Test pointer authentication support in KVM guest
Date: Wed, 2 Aug 2023 20:17:16 +0000	[thread overview]
Message-ID: <ZMq5zDJ16xav7NPa@linux.dev> (raw)
In-Reply-To: <CAAdAUthi6oZ6oRQDFLeOFVwm2dtcTK2ERJm316x0bdn5TQObYw@mail.gmail.com>

Hi Jing,

Nothing serious, but when you're replying on a thread can you add a
leading and trailing line of whitespace between the quotation and your
reply? Otherwise threads get really dense and hard to read.

On Wed, Aug 02, 2023 at 10:19:30AM -0700, Jing Zhang wrote:
> > > +                     case FAIL_KVM:
> > > +                             TEST_FAIL("KVM doesn't support guest PAuth!\n");
> >
> > Why is that a hard failure? The vast majority of the HW out there
> > doesn't support PAuth...
> Since previous TEST_REQUIRES have passed, KVM should be able to
> support guest PAuth. The test will be skipped on those HW without
> PAuth.

So then what is the purpose of this failure mode? The only case where
this would happen is if KVM is if KVM screwed up the emulation somehow,
took a trap on a PAC instruction or register and reflected that back
into the guest as an UNDEF.

That's a perfectly valid thing to test for, but the naming and failure
messages should indicate what actually happened.

> > As I mentioned above, another thing I'd like to see is a set of
> > reference results for a given set of keys and architected algorithm
> > (QARMA3, QARMA5) so that we can compare between implementations
> > (excluding the IMPDEF implementations, of course).
> Sure. Will do.

I was initially hesitant towards testing PAC like this since it is
entirely a hardware issue besides KVM context switching, but you could
spin this off as a way to test if vCPU save/restore works correctly by
priming the vCPU from userspace.

Marc, is there something else here you're interested in exercising I
may've missed?

-- 
Thanks,
Oliver