Message-ID: <22172b58cd5353324d14b733aeac51e365597070.camel@linux.ibm.com>
Subject: Re: [PATCH v6 24/26] powerpc/pseries: Implement secvars for dynamic
 secure boot
From:   Andrew Donnellan <ajd@linux.ibm.com>
To:     Michael Ellerman <michaele@au1.ibm.com>,
        Stefan Berger <stefanb@linux.ibm.com>,
        linuxppc-dev@lists.ozlabs.org, linux-integrity@vger.kernel.org
Cc:     ruscur@russell.cc, bgray@linux.ibm.com, nayna@linux.ibm.com,
        gcwilson@linux.ibm.com, gjoyce@linux.ibm.com, brking@linux.ibm.com,
        sudhakar@linux.ibm.com, erichte@linux.ibm.com,
        gregkh@linuxfoundation.org, linux-kernel@vger.kernel.org,
        zohar@linux.ibm.com, joel@jms.id.au, npiggin@gmail.com
Date:   Tue, 14 Feb 2023 17:07:55 +1100
In-Reply-To: <87pmadvm0n.fsf@mpe.ellerman.id.au>
References: <20230210080401.345462-1-ajd@linux.ibm.com>
         <20230210080401.345462-25-ajd@linux.ibm.com>
         <f35e9ba1-5fdb-4cfa-5b41-cc55307dcd45@linux.ibm.com>
         <87pmadvm0n.fsf@mpe.ellerman.id.au>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: quoted-printable
User-Agent: Evolution 3.46.3 (3.46.3-1.fc37) 
MIME-Version: 1.0
Precedence: bulk

On Mon, 2023-02-13 at 22:32 +1100, Michael Ellerman wrote:
> > > +=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0memcpy(&flags, data, sizeo=
f(flags));
> >=20
> > conversion from bytestream to integer: I think in this case it
> > would be better to use
> >=20
> > flags =3D cpu_to_be64p((__u64*)data);
> >=20
> > so that the flags always in hypervisor/big endian format
>=20
> I don't think it's correct to byte swap the flags here. They must be
> in
> big endian format, but that's up to the caller.

That was what I initially thought, until I went and tested it properly
and found it was indeed broken (at least in our qemu environment, this
is slightly tricky for me to test right now on real hardware with real
PowerVM) depending on kernel endianness.

- Userspace writes the flags into the buffer in BE order

- The first 8 bytes of the buffer are memcpy()ed, in BE order, into
flags (a u64)

- plpar_hcall9() is called with flags as an argument, loaded into r9

- r9 is moved to r8 before jumping into the hypervisor

On a BE system, this works fine. On an LE system, this results in the
bytes in the flags variable being loaded into the register in LE order,
so the conversion is necessary.

> The powernv secvar backend doesn't byte swap the flags, if the
> pseries
> one did then the final content of the variable, written either by
> phyp
> or OPAL, would differ depending on which backend is active.
>=20
> Or am I missing something?

The powernv secvar backend doesn't have a notion of flags at all. (The
flags interface for pseries is a little ugly, but it gets the job done
- userspace can read the format attribute to discover what it needs to
do.)

--=20
Andrew Donnellan    OzLabs, ADL Canberra
ajd@linux.ibm.com   IBM Australia Limited