[patch 04/14] ia64: Remove the __SMALL_ADDR_AREA attribute for per cpu access

[patch 04/14] ia64: Remove the __SMALL_ADDR_AREA attribute for per cpu access

[Christoph Lameter]

The model(small) attribute is not supported by gcc 4.X. The tests
will always be negative today.

Cc: linux-ia64@vger.kernel.org
Cc: tony.luck@intel.com
Signed-off-by: Christoph Lameter <clameter@sgi.com>

---
 arch/ia64/scripts/check-model.c   |    1 -
 arch/ia64/scripts/toolchain-flags |    6 ------
 include/asm-ia64/percpu.h         |   12 +++---------
 3 files changed, 3 insertions(+), 16 deletions(-)

Index: linux-2.6/include/asm-ia64/percpu.h
=================================--- linux-2.6.orig/include/asm-ia64/percpu.h	2007-11-22 15:55:47.634454755 -0800
+++ linux-2.6/include/asm-ia64/percpu.h	2007-11-22 15:56:15.974704716 -0800
@@ -15,24 +15,18 @@
 
 #include <linux/threads.h>
 
-#ifdef HAVE_MODEL_SMALL_ATTRIBUTE
-# define __SMALL_ADDR_AREA	__attribute__((__model__ (__small__)))
-#else
-# define __SMALL_ADDR_AREA
-#endif
-
 #define DECLARE_PER_CPU(type, name)				\
-	extern __SMALL_ADDR_AREA __typeof__(type) per_cpu__##name
+	extern __typeof__(type) per_cpu__##name
 
 /* Separate out the type, so (int[3], foo) works. */
 #define DEFINE_PER_CPU(type, name)				\
 	__attribute__((__section__(".data.percpu")))		\
-	__SMALL_ADDR_AREA __typeof__(type) per_cpu__##name
+	__typeof__(type) per_cpu__##name
 
 #ifdef CONFIG_SMP
 #define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)			\
 	__attribute__((__section__(".data.percpu.shared_aligned")))	\
-	__SMALL_ADDR_AREA __typeof__(type) per_cpu__##name		\
+	__typeof__(type) per_cpu__##name				\
 	____cacheline_aligned_in_smp
 #else
 #define DEFINE_PER_CPU_SHARED_ALIGNED(type, name)	\
Index: linux-2.6/arch/ia64/scripts/check-model.c
=================================--- linux-2.6.orig/arch/ia64/scripts/check-model.c	2007-11-22 15:56:40.890455063 -0800
+++ /dev/null	1970-01-01 00:00:00.000000000 +0000
@@ -1 +0,0 @@
-int __attribute__ ((__model__ (__small__))) x;
Index: linux-2.6/arch/ia64/scripts/toolchain-flags
=================================--- linux-2.6.orig/arch/ia64/scripts/toolchain-flags	2007-11-22 15:57:07.329204964 -0800
+++ linux-2.6/arch/ia64/scripts/toolchain-flags	2007-11-22 15:57:27.229018356 -0800
@@ -35,12 +35,6 @@ if [ $res -eq 0 ]; then
     CPPFLAGS="$CPPFLAGS -DHAVE_WORKING_TEXT_ALIGN"
 fi
 
-if ! $CC -c $dir/check-model.c -o $out 2>&1 | grep  __model__ | grep -q attrib
-then
-    CPPFLAGS="$CPPFLAGS -DHAVE_MODEL_SMALL_ATTRIBUTE"
-fi
-rm -f $out
-
 # Check whether assembler supports .serialize.{data,instruction} directive.
 
 $CC -c $dir/check-serialize.S -o $out 2>/dev/null

-- 

Re: [patch 04/14] ia64: Remove the __SMALL_ADDR_AREA attribute for per cpu access

[David Mosberger-Tang]

On 11/26/07, Christoph Lameter <clameter@sgi.com> wrote:
> The model(small) attribute is not supported by gcc 4.X. The tests
> will always be negative today.

What was the rationale for removing this attribute?

  --david

Re: [patch 04/14] ia64: Remove the __SMALL_ADDR_AREA attribute for

[Christoph Lameter]

On Mon, 26 Nov 2007, David Mosberger-Tang wrote:

> On 11/26/07, Christoph Lameter <clameter@sgi.com> wrote:
> > The model(small) attribute is not supported by gcc 4.X. The tests
> > will always be negative today.
> 
> What was the rationale for removing this attribute?

The code is then similar across all architectures and can be moved into 
generic code.

Re: [patch 04/14] ia64: Remove the __SMALL_ADDR_AREA attribute for per cpu access

[David Mosberger-Tang]

On 11/27/07, Christoph Lameter <clameter@sgi.com> wrote:
> On Mon, 26 Nov 2007, David Mosberger-Tang wrote:
>
> > On 11/26/07, Christoph Lameter <clameter@sgi.com> wrote:
> > > The model(small) attribute is not supported by gcc 4.X. The tests
> > > will always be negative today.
> >
> > What was the rationale for removing this attribute?
>
> The code is then similar across all architectures and can be moved into
> generic code.

Uniformity for the sake of uniformity?  The small data addressing is
really elegant and I don't think it should be dropped just for the
sake of uniformity.

  --david

Re: [patch 04/14] ia64: Remove the __SMALL_ADDR_AREA attribute for

[Christoph Lameter]

On Tue, 27 Nov 2007, David Mosberger-Tang wrote:

> Uniformity for the sake of uniformity?  The small data addressing is
> really elegant and I don't think it should be dropped just for the
> sake of uniformity.

Uniformity for the sake of code size reduction and easier maintenance. 

Yes I think it would be great to have this feature on all arches if 
possible. If someone could work with the gcc /linker folks to get this 
done that would be great.

But the feature has been removed from gcc and so its not usable for IA64 
with a current compiler anymore. This is basically removing useless code.

Re: [patch 04/14] ia64: Remove the __SMALL_ADDR_AREA attribute for per cpu access

[David Mosberger-Tang]

On 11/27/07, Christoph Lameter <clameter@sgi.com> wrote:
> On Tue, 27 Nov 2007, David Mosberger-Tang wrote:
>
> > Uniformity for the sake of uniformity?  The small data addressing is
> > really elegant and I don't think it should be dropped just for the
> > sake of uniformity.
>
> Uniformity for the sake of code size reduction and easier maintenance.

Code-size reduction?  You must be talking *source* code size
reduction.  Surely the small-data access-method decreases object code
size.

  --david

Re: [patch 04/14] ia64: Remove the __SMALL_ADDR_AREA attribute for

[Christoph Lameter]

On Tue, 27 Nov 2007, David Mosberger-Tang wrote:

> Code-size reduction?  You must be talking *source* code size
> reduction.  Surely the small-data access-method decreases object code
> size.

Yes source code reduction. I just added the attribute back but in such a 
way that any arch can add attributes to per cpu definitions.

Re: [patch 04/14] ia64: Remove the __SMALL_ADDR_AREA attribute for per cpu access

[Andreas Schwab]

Christoph Lameter <clameter@sgi.com> writes:

> The model(small) attribute is not supported by gcc 4.X.

Which gcc 4.X are you talking about?

Andreas.

-- 
Andreas Schwab, SuSE Labs, schwab@suse.de
SuSE Linux Products GmbH, Maxfeldstraße 5, 90409 Nürnberg, Germany
PGP key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: [patch 04/14] ia64: Remove the __SMALL_ADDR_AREA attribute for

[Christoph Lameter]

On Tue, 27 Nov 2007, Andreas Schwab wrote:

> Christoph Lameter <clameter@sgi.com> writes:
> 
> > The model(small) attribute is not supported by gcc 4.X.
> 
> Which gcc 4.X are you talking about?

All. Last gcc that supported this was 3.4.

Re: [patch 04/14] ia64: Remove the __SMALL_ADDR_AREA attribute for per cpu access

[Andreas Schwab]

Christoph Lameter <clameter@sgi.com> writes:

> On Tue, 27 Nov 2007, Andreas Schwab wrote:
>
>> Christoph Lameter <clameter@sgi.com> writes:
>> 
>> > The model(small) attribute is not supported by gcc 4.X.
>> 
>> Which gcc 4.X are you talking about?
>
> All. Last gcc that supported this was 3.4.

Strange.  Works fine here.

$ arch/ia64/scripts/toolchain-flags gcc objdump readelf
-DHAVE_WORKING_TEXT_ALIGN -DHAVE_MODEL_SMALL_ATTRIBUTE -DHAVE_SERIALIZE_DIRECTIVE
$ gcc --version | head -n 1
gcc (GCC) 4.2.1 (SUSE Linux)
$ grep ia64_handle_model_attribute config/ia64/*.c
config/ia64/ia64.c:static tree ia64_handle_model_attribute (tree *, tree, tree, int, bool *);
config/ia64/ia64.c:  { "model",	       1, 1, true, false, false, ia64_handle_model_attribute },
config/ia64/ia64.c:ia64_handle_model_attribute (tree *node, tree name, tree args,
$ grep small_addr_symbolic_operand config/ia64/*.md
config/ia64/constraints.md:  (match_operand 0 "small_addr_symbolic_operand"))
config/ia64/predicates.md:(define_predicate "small_addr_symbolic_operand" 

Andreas.

-- 
Andreas Schwab, SuSE Labs, schwab@suse.de
SuSE Linux Products GmbH, Maxfeldstraße 5, 90409 Nürnberg, Germany
PGP key fingerprint = 58CA 54C7 6D53 942B 1756  01D3 44D5 214B 8276 4ED5
"And now for something completely different."

Re: [patch 04/14] ia64: Remove the __SMALL_ADDR_AREA attribute for

[Christoph Lameter]

On Tue, 27 Nov 2007, Andreas Schwab wrote:

> Strange.  Works fine here.
> 
> $ arch/ia64/scripts/toolchain-flags gcc objdump readelf
> -DHAVE_WORKING_TEXT_ALIGN -DHAVE_MODEL_SMALL_ATTRIBUTE -DHAVE_SERIALIZE_DIRECTIVE
> $ gcc --version | head -n 1
> gcc (GCC) 4.2.1 (SUSE Linux)
> $ grep ia64_handle_model_attribute config/ia64/*.c
> config/ia64/ia64.c:static tree ia64_handle_model_attribute (tree *, tree, tree, int, bool *);
> config/ia64/ia64.c:  { "model",	       1, 1, true, false, false, ia64_handle_model_attribute },
> config/ia64/ia64.c:ia64_handle_model_attribute (tree *node, tree name, tree args,
> $ grep small_addr_symbolic_operand config/ia64/*.md
> config/ia64/constraints.md:  (match_operand 0 "small_addr_symbolic_operand"))
> config/ia64/predicates.md:(define_predicate "small_addr_symbolic_operand" 

Hmmm...

http://www.ohse.de/uwe/articles/gcc-attributes.html

says:

model (MODEL-NAME)
    Found in versions: 2.8-3.4

But true my compiler still takes it. Ok, I am going to add an option to 
add attributes to percpu definitions.

Re: [patch 04/14] ia64: Remove the __SMALL_ADDR_AREA attribute for

[Adrian Bunk]

On Tue, Nov 27, 2007 at 01:38:02PM -0800, Christoph Lameter wrote:
>...
> Hmmm...
> 
> http://www.ohse.de/uwe/articles/gcc-attributes.html
> 
> says:
> 
> model (MODEL-NAME)
>     Found in versions: 2.8-3.4
>...

This site says at the top it used gcc versions up to 3.4, so it 
obviously can't find anything in gcc >= 4.0 ...

cu
Adrian

-- 

       "Is there not promise of rain?" Ling Tan asked suddenly out
        of the darkness. There had been need of rain for many days.
       "Only a promise," Lao Er said.
                                       Pearl S. Buck - Dragon Seed

[patch 13/14] ia64: Use generic percpu

[Christoph Lameter]

ia64 has a special processor specific mapping that can be used to locate the
offset for the current per cpu area.

Cc: linux-ia64@vger.kernel.org
Cc: tony.luck@intel.com
Signed-off-by: Christoph Lameter <clameter@sgi.com>

---
 include/asm-ia64/percpu.h |   24 ++++++++----------------
 1 file changed, 8 insertions(+), 16 deletions(-)

Index: linux-2.6/include/asm-ia64/percpu.h
=================================--- linux-2.6.orig/include/asm-ia64/percpu.h	2007-11-24 10:27:31.088350556 -0800
+++ linux-2.6/include/asm-ia64/percpu.h	2007-11-24 10:31:54.053187674 -0800
@@ -9,10 +9,9 @@
 #define PERCPU_ENOUGH_ROOM PERCPU_PAGE_SIZE
 
 #ifdef __ASSEMBLY__
-# define THIS_CPU(var)	(per_cpu__##var)  /* use this to mark accesses to per-CPU variables... */
+# define THIS_CPU(var)	(PERCPU_ADDR + per_cpu__##var)  /* use this to mark accesses to per-CPU variables... */
 #else /* !__ASSEMBLY__ */
 
-
 #include <linux/threads.h>
 
 #define DECLARE_PER_CPU(type, name)				\
@@ -20,24 +19,12 @@
 
 #ifdef CONFIG_SMP
 
-extern unsigned long __per_cpu_offset[NR_CPUS];
-#define per_cpu_offset(x) (__per_cpu_offset[x])
-
-/* Equal to __per_cpu_offset[smp_processor_id()], but faster to access: */
-DECLARE_PER_CPU(unsigned long, local_per_cpu_offset);
-
-#define per_cpu(var, cpu)  (*RELOC_HIDE(&per_cpu__##var, __per_cpu_offset[cpu]))
-#define __get_cpu_var(var) (*RELOC_HIDE(&per_cpu__##var, __ia64_per_cpu_var(local_per_cpu_offset)))
-#define __raw_get_cpu_var(var) (*RELOC_HIDE(&per_cpu__##var, __ia64_per_cpu_var(local_per_cpu_offset)))
+#define __my_cpu_offset	__ia64_per_cpu_var(local_per_cpu_offset)
 
-extern void setup_per_cpu_areas (void);
 extern void *per_cpu_init(void);
 
 #else /* ! SMP */
 
-#define per_cpu(var, cpu)			(*((void)(cpu), &per_cpu__##var))
-#define __get_cpu_var(var)			per_cpu__##var
-#define __raw_get_cpu_var(var)			per_cpu__##var
 #define per_cpu_init()				(__phys_per_cpu_start)
 
 #endif	/* SMP */
@@ -48,7 +35,12 @@ extern void *per_cpu_init(void);
  * On the positive side, using __ia64_per_cpu_var() instead of __get_cpu_var() is slightly
  * more efficient.
  */
-#define __ia64_per_cpu_var(var)	(per_cpu__##var)
+#define __ia64_per_cpu_var(var)	(*SHIFT_PTR(&per_cpu__##var, PERCPU_ADDR))
+
+#include <asm-generic/percpu.h>
+
+/* Equal to __per_cpu_offset[smp_processor_id()], but faster to access: */
+DECLARE_PER_CPU(unsigned long, local_per_cpu_offset);
 
 #endif /* !__ASSEMBLY__ */
 

-- 

Re: [patch 13/14] ia64: Use generic percpu

[Christoph Lameter]

Duh. This particular patch assumes already relocated per cpu areas which 
does not work with ia64's per cpu area mapings. This fix is needed:

---
 include/asm-ia64/percpu.h |    4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

Index: linux-2.6/include/asm-ia64/percpu.h
=================================--- linux-2.6.orig/include/asm-ia64/percpu.h	2007-11-26 17:14:22.823022434 -0800
+++ linux-2.6/include/asm-ia64/percpu.h	2007-11-26 17:18:34.063021793 -0800
@@ -9,7 +9,7 @@
 #define PERCPU_ENOUGH_ROOM PERCPU_PAGE_SIZE
 
 #ifdef __ASSEMBLY__
-# define THIS_CPU(var)	(PERCPU_ADDR + per_cpu__##var)  /* use this to mark accesses to per-CPU variables... */
+# define THIS_CPU(var)	per_cpu__##var  /* use this to mark accesses to per-CPU variables... */
 #else /* !__ASSEMBLY__ */
 
 #include <linux/threads.h>
@@ -35,7 +35,7 @@ extern void *per_cpu_init(void);
  * On the positive side, using __ia64_per_cpu_var() instead of __get_cpu_var() is slightly
  * more efficient.
  */
-#define __ia64_per_cpu_var(var)	(*SHIFT_PTR(&per_cpu__##var, PERCPU_ADDR))
+#define __ia64_per_cpu_var(var)	per_cpu__##var
 
 #include <asm-generic/percpu.h>