ARM big-endian on current kernels for linux-3.8

[catalin.marinas@arm.com (Catalin Marinas)]

On Wed, Feb 13, 2013 at 11:18:35AM +0000, Ben Dooks wrote:
> On 13/02/13 10:24, Matthieu CASTET wrote:
> > Ben Dooks a ?crit :
> >> On 12/02/13 17:13, Russell King - ARM Linux wrote:
> >>> On Fri, Feb 08, 2013 at 11:17:30PM +0000, Ben Dooks wrote:
> >>>> I have been working on getting big-endian kernels working, mainly from
> >>>> little-endian boot envrionments. The following patch series is what I
> >>>> have been working on, mainly on the highbank and axp systems.
> >>> What is missing from this is the justification about why we need this
> >>> additional pain, given that all the supporting userspaces today are all
> >>> LE based.
> >>>
> >>> Sure, I know that telcos have an endless love of big endian and don't
> >>> understand anything else, but getting this working on non-telco socs
> >>> seems to be a little odd.
> >>
> >> Our problem is we have some code that works on a lot of big endian data
> >> but is not easy to re-build to work on ARM little endian. The current
> >> solution is to change to running the system big endian.
> >>
> >> Unfortunately we cannot just run user-space big endian as the MMU is
> >> fetched in the same endian mode as the processor's data.
> >>
> > Which architecture are you using ?
> >
> > On armv7 the mmu endian mode is selected by SCTLR.EE [1], but userspace can
> > change it's endian with setend instruction [2].
> >
> > As the endian state is stored in cpsr, it should be preserved across context switch.
> 
> Ah yes, I got confused as there is also an endian mode for the page
> tables that needs to match the endian mode that the kernel is doing
> loads/stores in.
> 
> However if we had BE userland with LE kernel then things are going to
> get difficult with a BE<>LE syscall issue.

I had an (insane) idea once but no time to pursue. You can enable the
'compat' layer for a 32-bit ARM kernel and define all the compat_* types
to be the same as the native ones. The compat layer has several handlers
for syscalls which pretty much do the conversion between compat and
native structures. The compat structures are read/written from/to user
using get_user/put_user on each member. You then need change the ARM
get_user/put_user code to test a new TIF_BE flag and do a 'rev' on the
data. At this point the native kernel structures would have the correct
little endianness.

I reckon the above would cover 70-80% of the syscalls. You need to chase
other syscalls and update the binfmt_elf.c to cope with BE ELF files.
Probably there are other issues as well.

-- 
Catalin