Re: MPC5200 does not boot

[Nicholas Piggin <npiggin@gmail.com>]

On Tue, 02 Aug 2016 12:44:33 +1000
Benjamin Herrenschmidt <benh@kernel.crashing.org> wrote:

> On Tue, 2016-08-02 at 00:44 +0200, Michal Sojka wrote:
> > Hi Benjamin,
> >=20
> > the following commit causes my MPC5200 not to boot.
> >=20
> > =C2=A0=C2=A0=C2=A0=C2=A0commit 9402c684613163888714df0955fa1f17142b08bf=
 =20
> > > =C2=A0=C2=A0=C2=A0=C2=A0Author: Benjamin Herrenschmidt <benh@kernel.c=
rashing.org> =20
> > =C2=A0=C2=A0=C2=A0=C2=A0Date:=C2=A0=C2=A0=C2=A0Tue Jul 5 15:03:41 2016 =
+1000
> > =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0
> > =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0powerpc: Factor do_feat=
ure_fixup calls
> > =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0
> > =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A032 and 64-bit do a simi=
lar set of calls early on, we move it all to
> > =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0a single common functio=
n to make the boot code more readable.
> > =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0
> > =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0Signed-off-by: Benjamin=
 Herrenschmidt <benh@kernel.crashing.org> =20
> > > =C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0Signed-off-by: Michae=
l Ellerman <mpe@ellerman.id.au> =20
> >=20
> > I suspect that the cause will be similar as described in commit
> > 1cd03890ea64795e53f17a94928cca22495acb2a. Unfortunately, I don't have
> > much time to debug this, but I can easily test patches. =20
>=20
> Are you sure of your bisection ? Did you verify that reverting that one
> patch fixes it ? Because all this does is move code to a function,
> the code is functionally the same and called in the same place...

+	struct cpu_spec *spec =3D *PTRRELOC(&cur_cpu_spec);
+
+	/*
+	 * Apply the CPU-specific and firmware specific fixups to kernel text
+	 * (nop out sections not relevant to this CPU or this firmware).
+	 */
+	do_feature_fixups(spec->cpu_features,
+			  PTRRELOC(&__start___ftr_fixup),
+			  PTRRELOC(&__stop___ftr_fixup));

Shouldn't these be PTRRELOC(spec)->cpu_features ? You are relocating
access to the pointer word, but not the address it contains.
identify_cpu() returns the relocated pointer which is what 32-bit used
to use.

Thanks,
Nick