Re: [GIT PULL] UBI changes for 2.6.40

[Artem Bityutskiy <dedekind1@gmail.com>]

Hi Markus,

yes, sorry, this patch was probably a mistake - I just stupidly followed
the new trend.

On Sun, 2011-06-12 at 08:40 +0200, Markus Trippelsdorf wrote:
> Either reverting or using something like this fixes the problem:

However, I am a bit reluctant to revert it now because this would only
hide the issue. And some googling shows that networking people had a
similar error in the past and solved it by reverting before the change
went upstream.

> diff --git a/include/mtd/ubi-user.h b/include/mtd/ubi-user.h
> index 3c41097..47a6c21 100644
> --- a/include/mtd/ubi-user.h
> +++ b/include/mtd/ubi-user.h
> @@ -23,6 +23,10 @@
>  
>  #include <linux/types.h>
>  
> +#ifndef __packed
> +#define __packed __attribute__((packed))
> +#endif
> +
>  /*
>   * UBI device creation (the same as MTD device attachment)
>   * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

This would be a dirty hack. I think a better solution is to enhance the
"scripts/headers_install.pl" script and teach it to substitute __packed
with __attribute__((packed)). Then at "make headers_install" time this
issue would be fixed up.

Or may be the definition of __packed has to be somehow included in
ubi-user.h ...

Would you please try to make a patch and send it to lkml to hopefully
generate some discussion and the right way to go would probably be
found?

-- 
Best Regards,
Artem Bityutskiy