Re: [PATCH v2 1/1] SecurityPkg/RngDxe: Fix Rng algo selection for Arm

Ard Biesheuvel

On Fri, 18 Nov 2022 at 11:32, PierreGondois <pierre.gondois@...> wrote:

On 11/18/22 11:14, Ard Biesheuvel wrote:
On Fri, 18 Nov 2022 at 11:10, Sami Mujawar <Sami.Mujawar@...> wrote:

Hi Ard,

Please find my response inline marked [SAMI].


Sami Mujawar

´╗┐On 18/11/2022, 09:56, "Ard Biesheuvel" <ardb@...> wrote:

On Wed, 16 Nov 2022 at 16:02, PierreGondois <pierre.gondois@...> wrote:
> From: Pierre Gondois <pierre.gondois@...>
> BZ:
> The EFI_RNG_PROTOCOL can advertise multiple algorithms through
> Guids. The PcdCpuRngSupportedAlgorithm contains a Guid that
> can be configured. It represents the algorithm used in RngLib.
> PcdCpuRngSupportedAlgorithm is set to the Zero Guid for KvmTool.
> When running KvmTool on a platform platform only having the RngLib,
> the only Guid available for EFI_RNG_PROTOCOL will be the zero Guid.
> To select the default algorithm in EFI_RNG_PROTOCOL.GetRng():
> a. Zero Guids are skipped
> b. If no algorithm is found, an ASSERT is triggered
> To allow using the RngLib to be used for the case above, Zero Guids
> should not be skipped (a.).
> If no algorithm is found, don't prevent from booting on DEBUG builds
> (b.).
> Allow Zero Guids to be selected and don't ASSERT if no algorithm is
> found. Also simplify the selection of the Rng algorithm when the
> default one is selected by just picking up the first element of
> mAvailableAlgoArray.
> Reported-by: Sami Mujawar <sami.mujawar@...>
> Signed-off-by: Pierre Gondois <Pierre.Gondois@...>

I am still confused by this.

Does this mean we might register the RNG protocol if we don't have
anything to back it up?
[SAMI] From a Guest firmware implementation perspective, we do not know the available RNG source.
I would assume either one of CPU RNG or Arm FW TRNG would be implemented on the host platform. If none of these are present, we would want to fall back to VIRTIO RNG.

Considering this, I think we should not register the EFI_RNG_PRTOCOL if no supported algorithms are present.

The other argument would be that the protocol allows discovery of supported RNG source. But looking how this is consumed in Linux, I think it is better to not register EFI_RNG_PRTOCOL if no supported algorithms are present.

Please do let me know your thoughts.
Agreed. I am adding support for the TRNG to the QEMU firmware, and if
this is supported, there is really no point in attaching a driver to
virtio-rng as well.

This means that checking for the existence of the EFI_RNG_PROTOCOL
should be sufficient to decide whether or not install another one.
There would maybe be a case where the consumer explicitly requests the
gEfiRngAlgorithmRaw GUID and the RngDxe doesn't detect it. So there would be
two EFI_RNG_PRTOCOL registered on different handles with different
algorithms available. I am not sure of what should happen then.
Fair point. So in the QEMU case, I should test
- whether EFI_RNG_PROTOCOL exists
- whether it implements the raw flavor

and only in that case, avoid virtio-rng (which only supports raw as well)

But about the inital point, EFI_RNG_PRTOCOL should indeed not be registered
if no algorithm is available.

And btw, I noticed that the TrngLib has a whole bunch of ASSERT()s
that trigger when trying to use it on QEMU - can we rip those out as
well please?

Join to automatically receive all group messages.