On 05/26/2013 01:28 PM, Simon Glass wrote:
On Wed, May 22, 2013 at 11:48 AM, Stephen Warren <swarren@wwwdotorg.org mailto:swarren@wwwdotorg.org> wrote:
From: Stephen Warren <swarren@nvidia.com <mailto:swarren@nvidia.com>> Initialized character arrays on the stack can cause gcc to emit code that performs unaligned accessess. Make the data static to avoid this. Note that the unaligned accesses are made when copying data to prefix[] on the stack from .rodata. By making the data static, the copy is completely avoided. All explicitly written code treats the data as u8[], so will never cause any unaligned accesses. Signed-off-by: Stephen Warren <swarren@nvidia.com <mailto:swarren@nvidia.com>>Acked-by: Simon Glass <sjg@chromium.org mailto:sjg@chromium.org>
Thanks for fixing.
I hit this with gcc 4.7. I wonder if previous revisions would not make this assumption?
IIRC, gcc-4.7 introduces the emission of native unaligned accesses, and it's been back-ported to Linaro gcc-4.6.
Another problem I have is that the 'linux' in 'linux,keymap' in the device compile turns into '1' since gcc predefines 'linux' to 1:
I think I'm going to add a -Ulinux to dts/Makefile.
I forget the exact details, but if you check the Linux makefiles for dtc+cpp, they don't suffer from this issue any more; it may have been due to use of -x assembler-with-cpp. I do also have a bug filed internally to NVIDIA to fix that, which is assigned to Tom. But, I'm sure he'd be glad if you fixed it:-)