Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Thu, 8 Jun 2023 22:55:26 +0100
From:   Will Deacon <will@kernel.org>
To:     Mostafa Saleh <smostafa@google.com>
Cc:     Oliver Upton <oliver.upton@linux.dev>, maz@kernel.org,
        linux-arm-kernel@lists.infradead.org, kvmarm@lists.linux.dev,
        linux-kernel@vger.kernel.org, tabba@google.com,
        kaleshsingh@google.com, catalin.marinas@arm.com,
        yuzenghui@huawei.com, suzuki.poulose@arm.com, james.morse@arm.com
Subject: Re: [PATCH] KVM: arm64: Use different pointer authentication keys
 for pKVM
Message-ID: <20230608215525.GA2742@willie-the-truck>
References: <20230516141531.791492-1-smostafa@google.com>
 <ZHEa+HAixbYijQTA@linux.dev>
 <ZHSJ38WATzgJF7SR@google.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <ZHSJ38WATzgJF7SR@google.com>
User-Agent: Mutt/1.10.1 (2018-07-13)
Precedence: bulk

On Mon, May 29, 2023 at 11:17:51AM +0000, Mostafa Saleh wrote:
> On Fri, May 26, 2023 at 08:47:52PM +0000, Oliver Upton wrote:
> > On Tue, May 16, 2023 at 02:15:31PM +0000, Mostafa Saleh wrote:
> > > When the kernel is compiled with CONFIG_ARM64_PTR_AUTH_KERNEL, it
> > > uses Armv8.3-Pauth for return address protection for the kernel code
> > > including nvhe code in EL2.
> > > 
> > > Same keys are used in both kernel(EL1) and nvhe code(EL2), this is
> > > fine for nvhe but not when running in protected mode(pKVM) as the host
> > > can't be trusted.
> > 
> > But we trust it enough to hand pKVM a fresh set of keys before firing
> > off? I understand there is some degree of initialization required to get
> > pKVM off the ground, but I question in this case if key handoff is
> > strictly necessary.
> >
> > There are potentially other sources of random directly available at EL2,
> > such as the SMCCC TRNG ABI or FEAT_RNG. Should pKVM prefer one of these
> > random implementations and only fall back to host-provided keys if
> > absolutely necessary?
> > 
> According to my understanding, the kernel is still completely trusted at
> this point (it sets the initial page table for the hypervisor), so I
> believe it should be fine to trust it for ptrauth keys. However, I agree,
> it would be better if the hypervisor can get its own keys through
> firmware/hardware if supported. I will add this in V2.

I appreciate the sentiment, but I think we should avoid adding additional
complexity here if it adds no security benefit. If nothing else, it adds
pointless overhead, but beyond that it gives the false illusion of a
security boundary.

Prior to deprivilege, the kernel can write to the hypervisor text, modify
its stage-1 page-table and change its data values. I think the pointer
auth keys are the least of our worries if it's compromised...

Will