RFC: Aligning data in buffers with a macro?

["Carlos O'Donell" <carlos at systemhalted dot org>]

We don't have any reasonable macros for aligning data inside a buffer
to a given minimum alignment.

What we have is a collection of manual C trickery to achieve either
continually growing up alignment or continually growing down
alignment.

Take for example the following snippets:

resolv/nss_dns/dns-host.c
 601   uintptr_t pad = -(uintptr_t) buffer % __alignof__ (struct host_data);
 602   buffer += pad;

(I was shocked this one worked since I'd never seen a construct like this)

nptl/allocatestack.c
 470       /* Make sure the size of the stack is enough for the guard and
 471          eventually the thread descriptor.  */
 472       guardsize = (attr->guardsize + pagesize_m1) & ~pagesize_m1;
 473       if (__builtin_expect (size < ((guardsize + __static_tls_size
 474                                      + MINIMAL_REST_STACK + pagesize_m1)
 475                                     & ~pagesize_m1),
 476                             0))
 477         /* The stack is too small (or the guard too large).  */
 478         return EINVAL;
...
 628 # elif _STACK_GROWS_UP
 629           char *guard = (char *) (((uintptr_t) pd - guardsize) &
~pagesize_m1);
 630 #endif

So thus far we've got:

(1) Negative modulo
* Align with round-up semantics.
pad = -(uintptr_t) address % alignment;
address += pad;

(2) Bitwise AND of the negation
* Align with round-down semantics.
address = address & ~alignment;

(3) Add and bitwise AND of the negation
* Align with round-up semantics.
address = (address + alignment) & ~alignment;

What do people think about creating some generic macros we can use for
alignment?

That way it's crystal clear what you intend with the code.

For exmaple this...

 628 # elif _STACK_GROWS_UP
 629           char *guard = (char *) (((uintptr_t) pd - guardsize) &
~pagesize_m1);
 630 #endif

... would become:

 628 # elif _STACK_GROWS_UP
 629           char *guard = (char *) ALIGN_DOWN((uintptr_t) pd -
guardsize, pagesize_m1);
 630 #endif

Do the kernel headers have something we can crib or use here?

Cheers,
Carlos.