[PATCH] ia64: including <asm/signal.h> alone causes compilation errors

[PATCH] ia64: including <asm/signal.h> alone causes compilation errors

[Jeff Mahoney]

 Including *just* <asm/signal.h> on ia64 will result in a compilation failure,
 where it will succeed on every other architecture.

 Every other arch includes <linux/compiler.h> either directly or via
 <linux/types.h> at the top of <asm-*/signal.h>. ia64 includes
 <linux/types.h> after including <asm-generic/signal.h>, which causes
 the __user in <asm-generic/signal.h> to get passed through untouched, causing
 compilation errors.

 This patch moves the #include <linux/types.h> up to the beginning of signal.h,
 as found on every other arch.

 A specific example of where this behavior is observed is the recent addition
 of OCFS2. fs/ocfs2/cluster/userdlm.c seems alone in only including
 <asm/signal.h>, which seems to be perfectly valid.

 While I could have just fixed this in the OCFS2 code, I thought that making
 ia64 more consistent with the rest of the kernel would be the better fix.

Signed-off-by: Jeff Mahoney <jeffm@suse.com>

diff -ruNpX dontdiff linux-2.6.15/include/asm-ia64/signal.h linux-2.6.15-ocfs2/include/asm-ia64/signal.h
--- linux-2.6.15/include/asm-ia64/signal.h	2006-01-02 22:21:10.000000000 -0500
+++ linux-2.6.15-ocfs2/include/asm-ia64/signal.h	2006-01-09 11:08:16.404700640 -0500
@@ -1,6 +1,8 @@
 #ifndef _ASM_IA64_SIGNAL_H
 #define _ASM_IA64_SIGNAL_H
 
+#include <linux/types.h>
+
 /*
  * Modified 1998-2001, 2003
  *	David Mosberger-Tang <davidm@hpl.hp.com>, Hewlett-Packard Co
@@ -122,8 +124,6 @@
 
 # ifndef __ASSEMBLY__
 
-#  include <linux/types.h>
-
 /* Avoid too many header ordering problems.  */
 struct siginfo;
 
-- 
Jeff Mahoney
SuSE Labs

Re: [PATCH] ia64: including <asm/signal.h> alone causes compilation errors

[Matthew Wilcox]

On Mon, Jan 09, 2006 at 12:15:14PM -0500, Jeff Mahoney wrote:
> +++ linux-2.6.15-ocfs2/include/asm-ia64/signal.h	2006-01-09 11:08:16.404700640 -0500
> @@ -1,6 +1,8 @@
>  #ifndef _ASM_IA64_SIGNAL_H
>  #define _ASM_IA64_SIGNAL_H
>  
> +#include <linux/types.h>
> +
>  /*
>   * Modified 1998-2001, 2003
>   *	David Mosberger-Tang <davidm@hpl.hp.com>, Hewlett-Packard Co
> @@ -122,8 +124,6 @@
>  
>  # ifndef __ASSEMBLY__
>  
> -#  include <linux/types.h>
> -
>  /* Avoid too many header ordering problems.  */
>  struct siginfo;

Is it still possible to include this file from assembly?  Do we still
need to do that?

Re: [PATCH] ia64: including <asm/signal.h> alone causes compilation errors

[Christoph Hellwig]

On Mon, Jan 09, 2006 at 12:15:14PM -0500, Jeff Mahoney wrote:
> 
>  Including *just* <asm/signal.h> on ia64 will result in a compilation failure,
>  where it will succeed on every other architecture.
> 
>  Every other arch includes <linux/compiler.h> either directly or via
>  <linux/types.h> at the top of <asm-*/signal.h>. ia64 includes
>  <linux/types.h> after including <asm-generic/signal.h>, which causes
>  the __user in <asm-generic/signal.h> to get passed through untouched, causing
>  compilation errors.
> 
>  This patch moves the #include <linux/types.h> up to the beginning of signal.h,
>  as found on every other arch.
> 
>  A specific example of where this behavior is observed is the recent addition
>  of OCFS2. fs/ocfs2/cluster/userdlm.c seems alone in only including
>  <asm/signal.h>, which seems to be perfectly valid.

Generally you should only include <asm/signal.h> via <linux/signal.h>, which gets
<linux/compiler.h> via <linux/spinlock.h>

Re: [PATCH] ia64: including <asm/signal.h> alone causes compilation errors

[Jeff Mahoney]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Christoph Hellwig wrote:
> On Mon, Jan 09, 2006 at 12:15:14PM -0500, Jeff Mahoney wrote:
>>  Including *just* <asm/signal.h> on ia64 will result in a compilation failure,
>>  where it will succeed on every other architecture.
>>
>>  Every other arch includes <linux/compiler.h> either directly or via
>>  <linux/types.h> at the top of <asm-*/signal.h>. ia64 includes
>>  <linux/types.h> after including <asm-generic/signal.h>, which causes
>>  the __user in <asm-generic/signal.h> to get passed through untouched, causing
>>  compilation errors.
>>
>>  This patch moves the #include <linux/types.h> up to the beginning of signal.h,
>>  as found on every other arch.
>>
>>  A specific example of where this behavior is observed is the recent addition
>>  of OCFS2. fs/ocfs2/cluster/userdlm.c seems alone in only including
>>  <asm/signal.h>, which seems to be perfectly valid.
> 
> Generally you should only include <asm/signal.h> via <linux/signal.h>, which gets
> <linux/compiler.h> via <linux/spinlock.h>

Ok. I just did a quick grep and saw a number of places where
<asm/signal.h> was included. A closer look reveals most of those are in
arch/.

Ignore the noise.

- -Jeff

- --
Jeff Mahoney
SUSE Labs
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFDwp5DLPWxlyuTD7IRAoZBAKCf4MMQNC/rfcdXOH0Dto3RTa32qgCfSeFf
pX68kyrbj2ZdH23Mo+EmUZ4=
=y7xp
-----END PGP SIGNATURE-----

Re: [PATCH] ia64: including <asm/signal.h> alone causes compilation errors

[Jeff Mahoney]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Matthew Wilcox wrote:
> On Mon, Jan 09, 2006 at 12:15:14PM -0500, Jeff Mahoney wrote:
>> +++ linux-2.6.15-ocfs2/include/asm-ia64/signal.h	2006-01-09 11:08:16.404700640 -0500
>> @@ -1,6 +1,8 @@
>>  #ifndef _ASM_IA64_SIGNAL_H
>>  #define _ASM_IA64_SIGNAL_H
>>  
>> +#include <linux/types.h>
>> +
>>  /*
>>   * Modified 1998-2001, 2003
>>   *	David Mosberger-Tang <davidm@hpl.hp.com>, Hewlett-Packard Co
>> @@ -122,8 +124,6 @@
>>  
>>  # ifndef __ASSEMBLY__
>>  
>> -#  include <linux/types.h>
>> -
>>  /* Avoid too many header ordering problems.  */
>>  struct siginfo;
> 
> Is it still possible to include this file from assembly?  Do we still
> need to do that?
> 
> 

Yes, actually, it is. :(

Christoph also pointed out that including <linux/signal.h> is the better
solution, so I'm submitting that patch to the OCFS2 folks instead.

Sorry for the noise.

- -Jeff

- --
Jeff Mahoney
SUSE Labs
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.2 (GNU/Linux)
Comment: Using GnuPG with Thunderbird - http://enigmail.mozdev.org

iD8DBQFDwqCjLPWxlyuTD7IRAj8PAJwLv2EghfMR9yPuaVhpTLQMGysKpwCfcVBq
7ZT0LwuYEjKYVnJT/QeXmCw=
=Yt8o
-----END PGP SIGNATURE-----