[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <ZgKKcurtndJaIwAV@linux.dev>
Date: Tue, 26 Mar 2024 01:42:26 -0700
From: Oliver Upton <oliver.upton@...ux.dev>
To: Sebastian Ene <sebastianene@...gle.com>
Cc: catalin.marinas@....com, james.morse@....com, jean-philippe@...aro.org,
maz@...nel.org, qperret@...gle.com, qwandor@...gle.com,
suzuki.poulose@....com, tabba@...gle.com, will@...nel.org,
yuzenghui@...wei.com, kvmarm@...ts.linux.dev,
linux-arm-kernel@...ts.infradead.org, linux-kernel@...r.kernel.org,
kernel-team@...roid.com
Subject: Re: [PATCH 2/2] KVM: arm64: Allow only the specified FF-A calls to
be forwarded to TZ
On Mon, Mar 25, 2024 at 11:29:39AM +0000, Sebastian Ene wrote:
> On Fri, Mar 22, 2024 at 07:07:52PM -0700, Oliver Upton wrote:
> > On Fri, Mar 22, 2024 at 12:43:03PM +0000, Sebastian Ene wrote:
> > > The previous logic used a deny list to filter the FF-A calls. Because of
> > > this, some of the calls escaped the check and they were forwarded by
> > > default to Trustzone. (eg. FFA_MSG_SEND_DIRECT_REQ was denied but the 64
> > > bit version of the call was not).
> > > Modify the logic to use an allowlist and allow only the calls specified in
> > > the filter function to be proxied to TZ from the hypervisor.
>
> Hi Oliver,
>
> >
> > I had discussed this with Will back when the feature was upstreamed and
> > he said there's a lot of off-label calls that necessitate a denylist
> > implementation. Has anything changed to give us confidence that we can
> > be restrictive, at least on the FF-A range?
> >
>
> I remember your proposal for having an allowlist instead. The current change makes
> sense if we have https://lore.kernel.org/kvmarm/20240322124303.309423-1-sebastianene@google.com/
> which opens the window for more FF-A calls to be forwarded to TZ.
Got it. Last time I didn't catch the level of abuse we expect to endure
from vendors, but now it seems we will not support non-conforming calls
that appear in standardized SMC ranges.
Adding mention of this to the changelog might be a good idea then.
--
Thanks,
Oliver
Powered by blists - more mailing lists