On Fri, Dec 07, 2012 at 12:28:09PM +0000, Russell King - ARM Linux wrote:

[...]

Having thought about this for quite a while, I think there's three
options:

1. The Sharp Zaurus code is unmaintained.  Declare it to be broken, and
   remove all the Sharp code from the kernel so that people aren't tempted
   to think it's something that the kernel supports.  (In light of the
   lack of maintainer for these platforms which almost no one knows much
   about, that will lower the maintanence burden on everyone else.)

2. Revert the hyp-mode support and say that we can't accept _any_ change
   whatsoever to the decompressor head.S through fear that it will cause
   a regression for Sharp Zaurus platforms.  (This effectively means we
   can't support virtualization on ARM - which is not acceptable.)

3. Revert the hyp-mode support and make it conditional on CPUs that have
   hyp-mode support.

Out of those three, (3) is the best way forward.  So, unless I hear any
objections, I'm going to revert 424e5994e63 in mainline, and wait for a
patch to safe_svcmode_maskall so that it can be switched between the
hyp-mode version and the non-hyp-mode.

This shouldn't be a problem for the single-zImage people; the break-point
is the ARMv5/ARMv6 boundary which we're already keeping as separate kernels.

I've posted a separate patch which ought to accomplish the appropriate
change to safe_svc_maskall
([PATCH] ARM: head: Remove boot-time HYP mode check for v5 and below)

I've done a compile-disassemble sanity-check for v5 and v7 to make sure
that the results are sane, but I can't check whether it fixed the
problem for the affected hardware.

Marko, can you give it a try?

Cheers
---Dave