On Mon, 2014-04-07 at 18:22 +0200, Albert ARIBAUD wrote:

Hi Charles,

On Wed, 5 Mar 2014 11:30:26 +1300, Charles Manning cdhmanning@gmail.com wrote:

...

Hello All

I am currently reworking a socfpga signer patch which is part of mkimage.

One thing I need to do is to stuff some bytes into a header ensuring the endianism is correct.

This is a trivial thing to do in C, but I have been instructed to use the existing functions put_unaligned_le32() etc.

These are part of Linux in linux/unaligned/access_ok.h. But this includes <asm/*>

It seems to me that this mixing of host space and target space is potentially problematic.

Hmm... Where exactly is there a mixing of host and target?

arch/<whatever>/include/asm are target-specific headers. I suppose the headers in arch/sandbox would work, but is that currently what will be used if a host tool references an asm/ header? Plus, it's Linux-specific.

Or if you mean to include the host's asm/ headers, that's also Linux-specific.

...

Is this a safe thing to do?

In tools/relocate-rela.c I open-coded endian swapping because I didn't see a better way to do so portably (outside of the networking functions which didn't do what I needed). Perhaps that could be factored out (and extended to other sizes) to somewhere common.

-Scott