question about ARM11MP cpu architecture identification

[Russell King - ARM Linux]

On Tue, Jan 12, 2010 at 07:03:35PM +0200, saeed bishara wrote:
> On Tue, Jan 12, 2010 at 6:55 PM, Catalin Marinas
> <catalin.marinas at arm.com> wrote:
> > On Tue, 2010-01-12 at 16:42 +0000, Russell King - ARM Linux wrote:
> >> On Tue, Jan 12, 2010 at 04:26:12PM +0000, Catalin Marinas wrote:
> >> > On Tue, 2010-01-12 at 16:21 +0000, saeed bishara wrote:
> >> > >     The ARM11MP (rev r2p0) cpu architecture is ARMv6, and according to
> >> > > the http://infocenter.arm.com, its Main ID register is 0x410FB024, and
> >> > > the MMFR[3:0] is 3. but it looks to me that the cpu_architecture()
> >> > > will identify it as ARMv7, and if the kernel was configured with
> >> > > CPU_V7 only, then the lookup_processor_type() function also will
> >> > > identify it as ARMv7. Am I missing something here?
> >> >
> >> > cpu_architecture() returns the architecture version of the supported MMU
> >> > which on ARM11MPCore is the same as on ARMv7.
> >> >
> >> > Even if it is possible to configure both v6 and v7 in the same kernel, I
> >> > don't think it is currently possible to successfully run such kernel.
> >>
> >> What makes you think that?  I've been building v6+v7 kernels for Realview
> >> (with a patch to provide the right build options for the ARMv7 arch/arm/mm
> >> files) and ISTR it works fine on my Realview board.
> >
> > I don't remember exactly but it didn't work for me some time ago. I
> > think it failed to build (though I could have worked around this with
> > something like -march=all).
> >
> I think you need the attached patch from Lennert (it was planned to be
> sent later by me)

Using -Wa,-march= for assembler options is actually a better idea.
I'll update my patch with that (which actually reflects what I did
for Integrator/CP with ARM926 + v6).