Re: [mdadm PATCH 1/3] fix gcc warnings about strict-aliasing rules

[Neil Brown <neilb@suse.de>]

On Tue, 2 Mar 2010 08:48:11 +0100
Luca Berra <bluca@comedia.it> wrote:

> On Tue, Mar 02, 2010 at 10:40:21AM +0300, Michael Tokarev wrote:
> >Luca Berra wrong:
> >> Signed-off-by: Luca Berra <bluca@comedia.it>
> >> ---
> >>   util.c |    4 ++--
> >>   1 files changed, 2 insertions(+), 2 deletions(-)
> >> 
> >> diff --git a/util.c b/util.c
> >> index 68f048d..1def2a0 100644
> >> --- a/util.c
> >> +++ b/util.c
> >> @@ -1160,7 +1160,7 @@ static int get_gpt_last_partition_end(int fd,
> >> unsigned long long *endofpart)
> >>       entry_size = __le32_to_cpu(buf[GPT_ENTRY_SIZE_OFFSET]);
> >>         /* Check GPT signature*/
> >> -    if (*((__u64*)buf) != GPT_SIGNATURE_MAGIC)
> >> +    if ((__u64)buf[0] != GPT_SIGNATURE_MAGIC)
> >
> >This looks wrong.
> >
> >buf is an array of unsigned char.  Before, we converted the
> >whole thing to a pointer to u64 and took the first element
> >at that address, u64 size.  Now after the change, we take
> >first _byte_ of the array, convert it to u64 (adding leading
> >zeros) and compare with a large number.
> 
> you are correct, this is obviously wrong, i wonder what i was thinking.
> 
> any idea on how to get this right?
> sscanf? union?

I fixed it with:

 __u64* buf64;

 ....
 buf64 = (__u64*)buf;
 if (buf64[0] != ....).....

NeilBrown