Question about unaligned pointer

Question about unaligned pointer

[Bin.Cheng]

Hi,
With -Waddress-of-packed-member option, GCC gives warning message if
address of packed structure member is taken and assigned to pointer,
like:
struct foo {
    char a;
    int b;
} __attribute__((packed));

int main()
{
    struct foo foo;
    int *p;
    p = &foo.b;  // !!
    *p = 1234;
    return 0;
}

This is expected, however, I wonder if there is any way to let gcc
know that `p` is a pointer that might be unaligned, so that an
unaligned access instruction is generated if `p` is dereferenced?
IIRC, microsoft compiler has __unaligned keyword extended for pointer
declaration.

Thanks,
bin

Re: Question about unaligned pointer

[Alexander Monakov]

On Tue, 18 Jan 2022, Bin.Cheng via Gcc-help wrote:

> Hi,
> With -Waddress-of-packed-member option, GCC gives warning message if
> address of packed structure member is taken and assigned to pointer,
> like:
> struct foo {
>     char a;
>     int b;
> } __attribute__((packed));
> 
> int main()
> {
>     struct foo foo;
>     int *p;
>     p = &foo.b;  // !!
>     *p = 1234;
>     return 0;
> }
> 
> This is expected, however, I wonder if there is any way to let gcc
> know that `p` is a pointer that might be unaligned, so that an
> unaligned access instruction is generated if `p` is dereferenced?
> IIRC, microsoft compiler has __unaligned keyword extended for pointer
> declaration.

Sure, just add a new type with non-standard alignment. Instead of
'int *p' you can write e.g.

    typedef int i32u __attribute__((aligned(1)));
    i32u *p;

Alexander

Re: Question about unaligned pointer

[Bin.Cheng]

On Tue, Jan 18, 2022 at 3:20 PM Alexander Monakov <amonakov@ispras.ru> wrote:
>
> On Tue, 18 Jan 2022, Bin.Cheng via Gcc-help wrote:
>
> > Hi,
> > With -Waddress-of-packed-member option, GCC gives warning message if
> > address of packed structure member is taken and assigned to pointer,
> > like:
> > struct foo {
> >     char a;
> >     int b;
> > } __attribute__((packed));
> >
> > int main()
> > {
> >     struct foo foo;
> >     int *p;
> >     p = &foo.b;  // !!
> >     *p = 1234;
> >     return 0;
> > }
> >
> > This is expected, however, I wonder if there is any way to let gcc
> > know that `p` is a pointer that might be unaligned, so that an
> > unaligned access instruction is generated if `p` is dereferenced?
> > IIRC, microsoft compiler has __unaligned keyword extended for pointer
> > declaration.
>
> Sure, just add a new type with non-standard alignment. Instead of
> 'int *p' you can write e.g.
>
>     typedef int i32u __attribute__((aligned(1)));
>     i32u *p;
>
Hi Alexander,
Thanks very much for helping.  This works.
However, I wonder what the difference is between typedef and the following code?

int __attribute__((aligned(1))) *p;
// ...

Thanks,
bin

Re: Question about unaligned pointer

[Alexander Monakov]

On Wed, 19 Jan 2022, Bin.Cheng wrote:

> > Sure, just add a new type with non-standard alignment. Instead of
> > 'int *p' you can write e.g.
> >
> >     typedef int i32u __attribute__((aligned(1)));
> >     i32u *p;
> >
> Hi Alexander,
> Thanks very much for helping.  This works.
> However, I wonder what the difference is between typedef and the following code?
> 
> int __attribute__((aligned(1))) *p;

Here the attribute applies to the thing being declared, i.e. the variable p, not
the type it points to. You can verify this by e.g. printing __alignof(p) and
__alignof(*p).

Alexander