[PATCH] ARM: supplementing IO accessors with 64 bit capability

[Arnd Bergmann]

On Thursday 23 October 2014 16:15:19 Nicolas Pitre wrote:
> On Thu, 23 Oct 2014, Russell King - ARM Linux wrote:
> 
> > On Thu, Oct 23, 2014 at 03:47:32PM -0400, Nicolas Pitre wrote:
> > > On Wed, 22 Oct 2014, Catalin Marinas wrote:
> > > 
> > > > On Wed, Oct 22, 2014 at 05:06:23PM +0100, mathieu.poirier at linaro.org wrote:
> > > > > +#if __LINUX_ARM_ARCH__ >= 5
> > > > 
> > > > My old ARMv5 book does not list LDRD/STRD. It looks like they only come
> > > > with ARMv5TE. Are there any processors prior to this supported by the
> > > > kernel?
> > > 
> > > We still supports ARMv4 targets.
> > > 
> > > As far as I know, all the ARMv5 targets we support are also TE capable.
> > 
> > Not quite.  We have ARM1020, which according to our proc-*.S files is
> > only ARMv5T, not ARMv5TE.

Does this actually work when we are building with -march=armv5te?

The Makefile contains this line:

arch-$(CONFIG_CPU_32v5) =-D__LINUX_ARM_ARCH__=5 $(call cc-option,-march=armv5te,-march=armv4t)

which looks like it would break for ARM1020.


On a related note, I also wonder about this part:

tune-$(CONFIG_CPU_ARM946E)      =$(call cc-option,-mtune=arm9e,-mtune=arm9tdmi)
tune-$(CONFIG_CPU_ARM920T)      =-mtune=arm9tdmi
tune-$(CONFIG_CPU_ARM922T)      =-mtune=arm9tdmi
tune-$(CONFIG_CPU_ARM925T)      =-mtune=arm9tdmi
tune-$(CONFIG_CPU_ARM926T)      =-mtune=arm9tdmi

I stumbled over this a while ago and couldn't figure it out. Does ARM926T
actually exist, or is that a mistake that should actually be ARM926E?

If this is always ARM926E, shouldn't we build with -mtune=arm9e as we do
for ARM946E?

> Oh well.  Never saw such a beast in the field though.

The only ARM10 implementation aside from integrator/realview that I'm aware
of is an ARM1026E based Conexant/Ikanos DSL modem SoC (CX94xxx), and that
is of course ARMv5TE.

> Maybe to be on the very safe side, given that no ARMV5TE is likely to 
> need 64-bit IO accessors at this point, this could simply be 
> __LINUX_ARM_ARCH__ >= 6 instead.

Which drivers need that support anyway? We definitely need to ensure
that we don't try to build them on architectures without this support
when CONFIG_COMPILE_TEST is set.

	Arnd