[PATCH] ppc32: Fix sector_t definition with CONFIG_LBD

[PATCH] ppc32: Fix sector_t definition with CONFIG_LBD

[Benjamin Herrenschmidt]

Hi !

sector_t depends on CONFIG_LBD but include/config.h may not be there
thus causing interesting breakage in some places...

Here's the fix for ppc32 (problem found by Roman Zippel, other archs
need a similar fix).

diff -urN linux-2.5/include/asm-ppc/types.h linuxppc-2.5-benh/include/asm-ppc/types.h
--- linux-2.5/include/asm-ppc/types.h	2004-03-01 18:13:06.000000000 +1100
+++ linuxppc-2.5-benh/include/asm-ppc/types.h	2004-03-29 13:14:50.000000000 +1000
@@ -37,6 +37,8 @@
 
 #ifndef __ASSEMBLY__
 
+#include <linux/config.h>
+
 typedef signed char s8;
 typedef unsigned char u8;
 

Re: [PATCH] ppc32: Fix sector_t definition with CONFIG_LBD

[Andrew Morton]

Benjamin Herrenschmidt <benh@kernel.crashing.org> wrote:
>
>  sector_t depends on CONFIG_LBD but include/config.h may not be there
>  thus causing interesting breakage in some places...

Nasty.

>  Here's the fix for ppc32 (problem found by Roman Zippel, other archs
>  need a similar fix).

Three of them.

 25-akpm/include/asm-s390/types.h   |    2 ++
 25-akpm/include/asm-sh/types.h     |    2 ++
 25-akpm/include/asm-x86_64/types.h |    2 ++
 3 files changed, 6 insertions(+)

diff -puN include/asm-s390/types.h~types_h-needs-config_h include/asm-s390/types.h
--- 25/include/asm-s390/types.h~types_h-needs-config_h	2004-03-28 23:02:57.481365480 -0800
+++ 25-akpm/include/asm-s390/types.h	2004-03-28 23:02:57.486364720 -0800
@@ -50,6 +50,8 @@ typedef __signed__ long saddr_t;
  */
 #ifdef __KERNEL__
 
+#include <linux/config.h>
+
 #ifndef __s390x__
 #define BITS_PER_LONG 32
 #else
diff -puN include/asm-sh/types.h~types_h-needs-config_h include/asm-sh/types.h
--- 25/include/asm-sh/types.h~types_h-needs-config_h	2004-03-28 23:02:57.482365328 -0800
+++ 25-akpm/include/asm-sh/types.h	2004-03-28 23:02:57.486364720 -0800
@@ -31,6 +31,8 @@ typedef unsigned long long __u64;
  */
 #ifdef __KERNEL__
 
+#include <linux/config.h>
+
 #define BITS_PER_LONG 32
 
 #ifndef __ASSEMBLY__
diff -puN include/asm-x86_64/types.h~types_h-needs-config_h include/asm-x86_64/types.h
--- 25/include/asm-x86_64/types.h~types_h-needs-config_h	2004-03-28 23:02:57.484365024 -0800
+++ 25-akpm/include/asm-x86_64/types.h	2004-03-28 23:02:57.486364720 -0800
@@ -29,6 +29,8 @@ typedef unsigned long long  __u64;
  */
 #ifdef __KERNEL__
 
+#include <linux/config.h>
+
 #define BITS_PER_LONG 64
 
 #ifndef __ASSEMBLY__

_

Re: [PATCH] ppc32: Fix sector_t definition with CONFIG_LBD

[Andi Kleen]

Andrew Morton <akpm@osdl.org> writes:

> Benjamin Herrenschmidt <benh@kernel.crashing.org> wrote:
> >
> >  sector_t depends on CONFIG_LBD but include/config.h may not be there
> >  thus causing interesting breakage in some places...
> 
> Nasty.
> 
> >  Here's the fix for ppc32 (problem found by Roman Zippel, other archs
> >  need a similar fix).
> 
> Three of them.
> 
>  25-akpm/include/asm-s390/types.h   |    2 ++
>  25-akpm/include/asm-sh/types.h     |    2 ++
>  25-akpm/include/asm-x86_64/types.h |    2 ++

Please use this change for x86-64 instead.

-Andi

diff -u linux-2.6.5rc2-amd64/include/asm-x86_64/types.h-o linux-2.6.5rc2-amd64/include/asm-x86_64/types.h
--- linux-2.6.5rc2-amd64/include/asm-x86_64/types.h-o	2004-03-21 21:11:54.000000000 +0100
+++ linux-2.6.5rc2-amd64/include/asm-x86_64/types.h	2004-03-29 04:44:24.000000000 +0200
@@ -48,10 +48,8 @@
 typedef u64 dma64_addr_t;
 typedef u64 dma_addr_t;
 
-#ifdef CONFIG_LBD
 typedef u64 sector_t;
 #define HAVE_SECTOR_T
-#endif
 
 #endif /* __ASSEMBLY__ */
 

[PATCH] small change to fget()

[Eric Dumazet]

Hello all

This one liner patch permits to save 2 instructions in the fget() 
function, at least on x86_64 SMP (only one ref to current->files)

diff -Nru linux-2.6.2/fs/file_table.c linux-2.6.2-custom/fs/file_table.c
--- linux-2.6.2/fs/file_table.c     2004-02-04 17:59:43.000000000 +0100
+++ linux-2.6.2-custom/fs/file_table.c        2004-03-29 
11:26:42.170555000 +0200
@@ -198,7 +198,7 @@
        struct files_struct *files = current->files;

        spin_lock(&files->file_lock);
-       file = fcheck(fd);
+       file = fcheck_files(files,fd) ;
        if (file)
                get_file(file);
        spin_unlock(&files->file_lock);


The assembly code was (oprofile data included) :

ffffffff8016c3d0 <fget>: /* fget total: 921168  0.1624 */
 51191  0.0090 :ffffffff8016c3d0:       mov    %gs:0x0,%rax
 76621  0.0135 :ffffffff8016c3d9:       mov    0x7e8(%rax),%rsi
121592  0.0214 :ffffffff8016c3e0:       lock decb 0x4(%rsi)
 11278  0.0020 :ffffffff8016c3e4:       js     ffffffff8016c5e8 
<.text.lock.file_table+0x20>
122737  0.0216 :ffffffff8016c3ea:       mov    %gs:0x0,%rax
174870  0.0308 :ffffffff8016c3f3:       mov    0x7e8(%rax),%rcx
  2274 4.0e-04 :ffffffff8016c3fa:       xor    %eax,%eax
 77418  0.0137 :ffffffff8016c3fc:       cmp    0x8(%rcx),%edi
  3417 6.0e-04 :ffffffff8016c3ff:       jae    ffffffff8016c40b <fget+0x3b>
  8184  0.0014 :ffffffff8016c401:       mov    0x18(%rcx),%rax
 37585  0.0066 :ffffffff8016c405:       mov    %edi,%edx
  7357  0.0013 :ffffffff8016c407:       mov    (%rax,%rdx,8),%rax
  5817  0.0010 :ffffffff8016c40b:       test   %rax,%rax
 57910  0.0102 :ffffffff8016c40e:       je     ffffffff8016c414 <fget+0x44>
  3401 6.0e-04 :ffffffff8016c410:       lock incl 0x28(%rax)
  7729  0.0014 :ffffffff8016c414:       movb   $0x1,0x4(%rsi)
151787  0.0268 :ffffffff8016c418:       retq



And it is now :

 380:   65 48 8b 04 25 00 00    mov    %gs:0x0,%rax
 387:   00 00
 389:   48 8b 88 e8 07 00 00    mov    0x7e8(%rax),%rcx
 390:   f0 fe 49 04             lock decb 0x4(%rcx)
 394:   0f 88 ee 01 00 00       js     588 <.text.lock.file_table+0x20>
 39a:   31 c0                   xor    %eax,%eax
 39c:   3b 79 08                cmp    0x8(%rcx),%edi
 39f:   73 0a                   jae    3ab <fget+0x2b>
 3a1:   48 8b 41 18             mov    0x18(%rcx),%rax
 3a5:   89 fa                   mov    %edi,%edx
 3a7:   48 8b 04 d0             mov    (%rax,%rdx,8),%rax
 3ab:   48 85 c0                test   %rax,%rax
 3ae:   74 04                   je     3b4 <fget+0x34>
 3b0:   f0 ff 40 28             lock incl 0x28(%rax)
 3b4:   c6 41 04 01             movb   $0x1,0x4(%rcx)
 3b8:   c3                      retq


Thanks

Re: [PATCH] ppc32: Fix sector_t definition with CONFIG_LBD

[Benjamin Herrenschmidt]

>  
> -#ifdef CONFIG_LBD
>  typedef u64 sector_t;
>  #define HAVE_SECTOR_T
> -#endif

Do you need that at all then ? The default unsigned long should
be just fine...

Ben.

Re: [PATCH] ppc32: Fix sector_t definition with CONFIG_LBD

[Andi Kleen]

On Mon, 29 Mar 2004 23:18:23 +1000
Benjamin Herrenschmidt <benh@kernel.crashing.org> wrote:

> 
> >  
> > -#ifdef CONFIG_LBD
> >  typedef u64 sector_t;
> >  #define HAVE_SECTOR_T
> > -#endif
> 
> Do you need that at all then ? The default unsigned long should
> be just fine...

True, it could be completely dropped too.

-Andi

Re: [PATCH] ppc32: Fix sector_t definition with CONFIG_LBD

[Anton Blanchard]

On Sun, Mar 28, 2004 at 11:03:51PM -0800, Andrew Morton wrote:
> Benjamin Herrenschmidt <benh@kernel.crashing.org> wrote:
> >
> >  sector_t depends on CONFIG_LBD but include/config.h may not be there
> >  thus causing interesting breakage in some places...
> 
> Nasty.

Dont forget trusty old linux/scripts/checkconfig.pl.

Anton

Re: [PATCH] ppc32: Fix sector_t definition with CONFIG_LBD

[Christoph Hellwig]

On Sun, Mar 28, 2004 at 11:03:51PM -0800, Andrew Morton wrote:
> >  Here's the fix for ppc32 (problem found by Roman Zippel, other archs
> >  need a similar fix).
> 
> Three of them.

Why do we keep thios in asm-*/types.h instead of linux/types.h, btw?