RE: [PATCH] fixes for building kernel using Intel compiler

[Ulrich Weigand]

Jun Nakajima wrote:

>Having said that, one occasion where people might be surprised by gcc (this
>might be a known issue, though) is: typedef + __attribute__; it ignores
>__attribute__. For example,
>#include <stdio.h>
>
>struct foo_16 {
>        char xxx[3];
>        short yyy;
>} __attribute__ ((aligned (16)));
>
>typedef struct bar_16 {
>        char xxx[3];
>        short yyy;
>} bar_16_t __attribute__ ((aligned (16)));

This is a user error (sort of); you're supposed to write:

typedef struct bar_16 {
        char xxx[3];
        short yyy;
} __attribute__ ((aligned (16))) bar_16_t;

[The attribute modifies the original struct type, which gets then
assigned the typedef name.  This is similar to the case where a
variable definition follows:

struct { ... } __attribute__() var;
vs.
struct { ... ] var __attribute__();

In the first case, the attribute modifies the struct type itself,
while in the second case the attribute applies only to the one
instance var.]

A warning would still be nice; we got bitten by that one a couple
of times ...

>In the kernel, there are several device drivers (ftape-bsm.h, e100.h, for
>example) are doing this kind of thing (i.e. typedef + attribute).

Well, I guess in those files the attribute((packed)) is a no-op
anyway as the structs are already packed according to the default
rules, so it doesn't really matter.  It should probably still be
fixed ...

Mit freundlichen Gruessen / Best Regards

Ulrich Weigand

--
  Dr. Ulrich Weigand
  Linux for S/390 Design & Development
  IBM Deutschland Entwicklung GmbH, Schoenaicher Str. 220, 71032 Boeblingen
  Phone: [phone]   ---   Email: [e-mail]