[PATCH 0/5] section name cleanup for parisc

[PATCH 0/5] section name cleanup for parisc

[Tim Abbott]

This patch series cleans up the section names on the parisc
architecture.  It requires the architecture-independent macro
definitions from this patch series:

<http://www.spinics.net/lists/mips/msg33499.html>

The long-term goal here is to add support for building the kernel with
-ffunction-sections -fdata-sections.  This requires renaming all the
magic section names in the kernel of the form .text.foo, .data.foo,
.bss.foo, and .rodata.foo to not have collisions with sections
generated for code like:

static int nosave = 0; /* -fdata-sections places in .data.nosave */
static void head(); /* -ffunction-sections places in .text.head */

Note that these patches have not been boot-tested (aside from testing
the analogous changes on x86), since I don't have access to the
appropriate hardware.

	-Tim Abbott


Tim Abbott (5):
  parisc: use NOSAVE_DATA macro for .data.nosave section.
  parisc: use new macro for .data.cacheline_aligned section.
  parisc: use new macros for .data.init_task.
  parisc: use new macro for .data.read_mostly section.
  parisc: convert to new generic read_mostly support.

 arch/parisc/Kconfig              |    3 +++
 arch/parisc/include/asm/cache.h  |    2 --
 arch/parisc/kernel/head.S        |    3 ++-
 arch/parisc/kernel/init_task.c   |    2 +-
 arch/parisc/kernel/vmlinux.lds.S |   36 +++++-------------------------------
 5 files changed, 11 insertions(+), 35 deletions(-)

[PATCH 1/5] parisc: use NOSAVE_DATA macro for .data.nosave section.

[Tim Abbott]

.data.nosave should not need a separate output section; this change
moves it into the .data section.

Signed-off-by: Tim Abbott <tabbott@mit.edu>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: Helge Deller <deller@gmx.de>
Cc: linux-parisc@vger.kernel.org
---
 arch/parisc/kernel/vmlinux.lds.S |   12 +-----------
 1 files changed, 1 insertions(+), 11 deletions(-)

diff --git a/arch/parisc/kernel/vmlinux.lds.S b/arch/parisc/kernel/vmlinux.lds.S
index fd2cc4f..0cc9658 100644
--- a/arch/parisc/kernel/vmlinux.lds.S
+++ b/arch/parisc/kernel/vmlinux.lds.S
@@ -103,6 +103,7 @@ SECTIONS
 	. = ALIGN(L1_CACHE_BYTES);
 	/* Data */
 	.data : {
+		NOSAVE_DATA
 		DATA_DATA
 		CONSTRUCTORS
 	}
@@ -118,17 +119,6 @@ SECTIONS
 		*(.data.lock_aligned)
 	}
 
-	/* nosave data is really only used for software suspend...it's here
-	 * just in case we ever implement it
-	 */
-	. = ALIGN(PAGE_SIZE);
-	__nosave_begin = .;
-	.data_nosave : {
-		*(.data.nosave)
-	}
-	. = ALIGN(PAGE_SIZE);
-	__nosave_end = .;

[PATCH 2/5] parisc: use new macro for .data.cacheline_aligned section.

[Tim Abbott]

.data.cacheline_aligned should not need a separate output section;
this change moves it into the .data section.

Signed-off-by: Tim Abbott <tabbott@mit.edu>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: Helge Deller <deller@gmx.de>
Cc: linux-parisc@vger.kernel.org
---
 arch/parisc/kernel/vmlinux.lds.S |    6 +-----
 1 files changed, 1 insertions(+), 5 deletions(-)

diff --git a/arch/parisc/kernel/vmlinux.lds.S b/arch/parisc/kernel/vmlinux.lds.S
index 0cc9658..b5936c9 100644
--- a/arch/parisc/kernel/vmlinux.lds.S
+++ b/arch/parisc/kernel/vmlinux.lds.S
@@ -104,15 +104,11 @@ SECTIONS
 	/* Data */
 	.data : {
 		NOSAVE_DATA
+		CACHELINE_ALIGNED_DATA(L1_CACHE_BYTES)
 		DATA_DATA
 		CONSTRUCTORS
 	}
 
-	. = ALIGN(L1_CACHE_BYTES);
-	.data.cacheline_aligned : {
-		*(.data.cacheline_aligned)
-	}

[PATCH 3/5] parisc: use new macros for .data.init_task.

[Tim Abbott]

.data.init_task should not need a separate output section; this change
moves it into the .data section.

Signed-off-by: Tim Abbott <tabbott@mit.edu>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: Helge Deller <deller@gmx.de>
Cc: linux-parisc@vger.kernel.org
---
 arch/parisc/kernel/init_task.c   |    2 +-
 arch/parisc/kernel/vmlinux.lds.S |   10 ++--------
 2 files changed, 3 insertions(+), 9 deletions(-)

diff --git a/arch/parisc/kernel/init_task.c b/arch/parisc/kernel/init_task.c
index 1e25a45..8ee17ea 100644
--- a/arch/parisc/kernel/init_task.c
+++ b/arch/parisc/kernel/init_task.c
@@ -48,7 +48,7 @@ EXPORT_SYMBOL(init_mm);
  * "init_task" linker map entry..
  */
 union thread_union init_thread_union
-	__attribute__((aligned(128))) __attribute__((__section__(".data.init_task"))) =
+	__attribute__((aligned(128))) __init_task_data =
 		{ INIT_THREAD_INFO(init_task) };
 
 #if PT_NLEVELS == 3
diff --git a/arch/parisc/kernel/vmlinux.lds.S b/arch/parisc/kernel/vmlinux.lds.S
index b5936c9..c8a528d 100644
--- a/arch/parisc/kernel/vmlinux.lds.S
+++ b/arch/parisc/kernel/vmlinux.lds.S
@@ -103,6 +103,8 @@ SECTIONS
 	. = ALIGN(L1_CACHE_BYTES);
 	/* Data */
 	.data : {
+		/* assembler code expects init_task to be 16k aligned */
+		INIT_TASK_DATA(16384)
 		NOSAVE_DATA
 		CACHELINE_ALIGNED_DATA(L1_CACHE_BYTES)
 		DATA_DATA
@@ -133,14 +135,6 @@ SECTIONS
 	}
 	__bss_stop = .;
 
-
-	/* assembler code expects init_task to be 16k aligned */
-	. = ALIGN(16384);
-	/* init_task */
-	.data.init_task : {
-		*(.data.init_task)
-	}

[PATCH 4/5] parisc: use new macro for .data.read_mostly section.

[Tim Abbott]

.data.read_mostly should not need a separate output section; this
change moves it into the .data section.

I used alignment L1_CACHE_BYTES instead of 16 since the whole point is
for read-mostly data to have different cache lines than other stuff.

Signed-off-by: Tim Abbott <tabbott@mit.edu>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: Helge Deller <deller@gmx.de>
Cc: linux-parisc@vger.kernel.org
---
 arch/parisc/kernel/vmlinux.lds.S |    8 +-------
 1 files changed, 1 insertions(+), 7 deletions(-)

diff --git a/arch/parisc/kernel/vmlinux.lds.S b/arch/parisc/kernel/vmlinux.lds.S
index c8a528d..3176407 100644
--- a/arch/parisc/kernel/vmlinux.lds.S
+++ b/arch/parisc/kernel/vmlinux.lds.S
@@ -94,19 +94,13 @@ SECTIONS
 		__stop___unwind = .;
 	}
 
-	/* rarely changed data like cpu maps */
-	. = ALIGN(16);
-	.data.read_mostly : {
-		*(.data.read_mostly)
-	}

[PATCH 5/5] parisc: convert to new generic read_mostly support.

[Tim Abbott]

Signed-off-by: Tim Abbott <tabbott@mit.edu>
Cc: Kyle McMartin <kyle@mcmartin.ca>
Cc: Helge Deller <deller@gmx.de>
Cc: linux-parisc@vger.kernel.org
---
 arch/parisc/Kconfig             |    3 +++
 arch/parisc/include/asm/cache.h |    2 --
 arch/parisc/kernel/head.S       |    3 ++-
 3 files changed, 5 insertions(+), 3 deletions(-)

diff --git a/arch/parisc/Kconfig b/arch/parisc/Kconfig
index 9038f39..cc113d0 100644
--- a/arch/parisc/Kconfig
+++ b/arch/parisc/Kconfig
@@ -101,6 +101,9 @@ config ARCH_MAY_HAVE_PC_FDC
 	depends on BROKEN
 	default y
 
+config HAVE_READ_MOSTLY_DATA
+	def_bool y
+
 source "init/Kconfig"
 
 source "kernel/Kconfig.freezer"
diff --git a/arch/parisc/include/asm/cache.h b/arch/parisc/include/asm/cache.h
index 32c2cca..21e0865 100644
--- a/arch/parisc/include/asm/cache.h
+++ b/arch/parisc/include/asm/cache.h
@@ -28,8 +28,6 @@
 
 #define SMP_CACHE_BYTES L1_CACHE_BYTES
 
-#define __read_mostly __attribute__((__section__(".data.read_mostly")))
-
 void parisc_cache_init(void);	/* initializes cache-flushing */
 void disable_sr_hashing_asm(int); /* low level support for above */
 void disable_sr_hashing(void);   /* turns off space register hashing */
diff --git a/arch/parisc/kernel/head.S b/arch/parisc/kernel/head.S
index 0e3d9f9..712d29d 100644
--- a/arch/parisc/kernel/head.S
+++ b/arch/parisc/kernel/head.S
@@ -19,6 +19,7 @@
 #include <asm/assembly.h>
 #include <asm/pgtable.h>
 
+#include <linux/cache.h>
 #include <linux/linkage.h>
 #include <linux/init.h>
 
@@ -345,7 +346,7 @@ smp_slave_stext:
 ENDPROC(stext)
 
 #ifndef CONFIG_64BIT
-	.section .data.read_mostly
+	__READ_MOSTLY
 
 	.align	4
 	.export	$global$,data
-- 
1.6.2.1

Re: [PATCH 3/5] parisc: use new macros for .data.init_task.

[Helge Deller]

Tim Abbott wrote:
> .data.init_task should not need a separate output section; this change
> moves it into the .data section.
> 
> Signed-off-by: Tim Abbott <tabbott@mit.edu>
> Cc: Kyle McMartin <kyle@mcmartin.ca>
> Cc: Helge Deller <deller@gmx.de>
> Cc: linux-parisc@vger.kernel.org


I think this patch is wrong, although it is theoretically correct.

IIRC, gcc on hppa is not able to provide an alignment >= 8k, which is
why we have done the 16k alignment inside the linker script.
So, I think this change will prevent the parisc kernel to boot up.
Needs testing.

Helge

> ---
>  arch/parisc/kernel/init_task.c   |    2 +-
>  arch/parisc/kernel/vmlinux.lds.S |   10 ++--------
>  2 files changed, 3 insertions(+), 9 deletions(-)
> 
> diff --git a/arch/parisc/kernel/init_task.c b/arch/parisc/kernel/init_task.c
> index 1e25a45..8ee17ea 100644
> --- a/arch/parisc/kernel/init_task.c
> +++ b/arch/parisc/kernel/init_task.c
> @@ -48,7 +48,7 @@ EXPORT_SYMBOL(init_mm);
>   * "init_task" linker map entry..
>   */
>  union thread_union init_thread_union
> -	__attribute__((aligned(128))) __attribute__((__section__(".data.init_task"))) =
> +	__attribute__((aligned(128))) __init_task_data =
>  		{ INIT_THREAD_INFO(init_task) };
>  
>  #if PT_NLEVELS == 3
> diff --git a/arch/parisc/kernel/vmlinux.lds.S b/arch/parisc/kernel/vmlinux.lds.S
> index b5936c9..c8a528d 100644
> --- a/arch/parisc/kernel/vmlinux.lds.S
> +++ b/arch/parisc/kernel/vmlinux.lds.S
> @@ -103,6 +103,8 @@ SECTIONS
>  	. = ALIGN(L1_CACHE_BYTES);
>  	/* Data */
>  	.data : {
> +		/* assembler code expects init_task to be 16k aligned */
> +		INIT_TASK_DATA(16384)
>  		NOSAVE_DATA
>  		CACHELINE_ALIGNED_DATA(L1_CACHE_BYTES)
>  		DATA_DATA
> @@ -133,14 +135,6 @@ SECTIONS
>  	}
>  	__bss_stop = .;
>  
> -
> -	/* assembler code expects init_task to be 16k aligned */
> -	. = ALIGN(16384);
> -	/* init_task */
> -	.data.init_task : {
> -		*(.data.init_task)
> -	}
> -
>  #ifdef CONFIG_64BIT
>  	. = ALIGN(16);
>  	/* Linkage tables */

Re: [PATCH 3/5] parisc: use new macros for .data.init_task.

[John David Anglin]

> IIRC, gcc on hppa is not able to provide an alignment >= 8k, which is
> why we have done the 16k alignment inside the linker script.

How does this arise?  We have the following define for MAX_OFILE_ALIGNMENT
in elfos.h:

#define MAX_OFILE_ALIGNMENT (((unsigned int) 1 << 28) * 8)

If you are correct, MAX_OFILE_ALIGNMENT should be changed.

Dave
-- 
J. David Anglin                                  dave.anglin@nrc-cnrc.gc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6602)

Re: [PATCH 3/5] parisc: use new macros for .data.init_task.

[Kyle McMartin]

On Sat, May 02, 2009 at 07:13:37AM +0200, Helge Deller wrote:
> I think this patch is wrong, although it is theoretically correct.
> 
> IIRC, gcc on hppa is not able to provide an alignment >= 8k, which is
> why we have done the 16k alignment inside the linker script.
> So, I think this change will prevent the parisc kernel to boot up.
> Needs testing.
> 

I think you're confusing this with the 8-byte maximum alignment from
kmalloc and on-stack that prevents us from just using a 16-byte aligned
word as a lock on pa1.1?

The patch I trimmed from this mail looks correct to me.

regards, Kyle

Re: [PATCH 0/5] section name cleanup for parisc

[Kyle McMartin]

On Thu, Apr 30, 2009 at 07:55:14PM -0400, Tim Abbott wrote:
> This patch series cleans up the section names on the parisc
> architecture.  It requires the architecture-independent macro
> definitions from this patch series:
> 
> <http://www.spinics.net/lists/mips/msg33499.html>
> 
> The long-term goal here is to add support for building the kernel with
> -ffunction-sections -fdata-sections.  This requires renaming all the
> magic section names in the kernel of the form .text.foo, .data.foo,
> .bss.foo, and .rodata.foo to not have collisions with sections
> generated for code like:
> 
> static int nosave = 0; /* -fdata-sections places in .data.nosave */
> static void head(); /* -ffunction-sections places in .text.head */
> 
> Note that these patches have not been boot-tested (aside from testing
> the analogous changes on x86), since I don't have access to the
> appropriate hardware.
> 
> 	-Tim Abbott
> 
> 
> Tim Abbott (5):
>   parisc: use NOSAVE_DATA macro for .data.nosave section.
>   parisc: use new macro for .data.cacheline_aligned section.
>   parisc: use new macros for .data.init_task.
>   parisc: use new macro for .data.read_mostly section.
>   parisc: convert to new generic read_mostly support.
> 

Not sure what the easiest way to merge these will be... I'll apply them
to a section-cleanup branch in the parisc git tree until the
architecture independent portions are upstream.

regards, Kyle

Re: [PATCH 3/5] parisc: use new macros for .data.init_task.

[Helge Deller]

Kyle McMartin wrote:
> On Sat, May 02, 2009 at 07:13:37AM +0200, Helge Deller wrote:
>> I think this patch is wrong, although it is theoretically correct.
>>
>> IIRC, gcc on hppa is not able to provide an alignment >= 8k, which is
>> why we have done the 16k alignment inside the linker script.
>> So, I think this change will prevent the parisc kernel to boot up.
>> Needs testing.
>>
> 
> I think you're confusing this with the 8-byte maximum alignment from
> kmalloc and on-stack that prevents us from just using a 16-byte aligned
> word as a lock on pa1.1?

No, I was not confusing it with the 8byte-alignment.

I really meant that a > 8k alignment was not possible.
I tried exactly the same stuff once and failed.
I think the restriction came from hpux compatibility or some old gas...

Anyway, I just tried some assembly and it seems to work.

> The patch I trimmed from this mail looks correct to me.

If you apply it and it boots OK for you, I'm fine.

Helge

Re: [PATCH 0/5] section name cleanup for parisc

[Sam Ravnborg]

On Sat, May 02, 2009 at 10:19:53AM -0400, Kyle McMartin wrote:
> On Thu, Apr 30, 2009 at 07:55:14PM -0400, Tim Abbott wrote:
> > This patch series cleans up the section names on the parisc
> > architecture.  It requires the architecture-independent macro
> > definitions from this patch series:
> > 
> > <http://www.spinics.net/lists/mips/msg33499.html>
> > 
> > The long-term goal here is to add support for building the kernel with
> > -ffunction-sections -fdata-sections.  This requires renaming all the
> > magic section names in the kernel of the form .text.foo, .data.foo,
> > .bss.foo, and .rodata.foo to not have collisions with sections
> > generated for code like:
> > 
> > static int nosave = 0; /* -fdata-sections places in .data.nosave */
> > static void head(); /* -ffunction-sections places in .text.head */
> > 
> > Note that these patches have not been boot-tested (aside from testing
> > the analogous changes on x86), since I don't have access to the
> > appropriate hardware.
> > 
> > 	-Tim Abbott
> > 
> > 
> > Tim Abbott (5):
> >   parisc: use NOSAVE_DATA macro for .data.nosave section.
> >   parisc: use new macro for .data.cacheline_aligned section.
> >   parisc: use new macros for .data.init_task.
> >   parisc: use new macro for .data.read_mostly section.
> >   parisc: convert to new generic read_mostly support.
> > 
> 
> Not sure what the easiest way to merge these will be... I'll apply them
> to a section-cleanup branch in the parisc git tree until the
> architecture independent portions are upstream.

I expect the parisc stuff to be redone. I'm working at the arch
independent stuff atm and will post a new patch later this weekend.
That will make the current patch obsolete.

	Sam

Re: [PATCH 3/5] parisc: use new macros for .data.init_task.

[John David Anglin]

> I really meant that a > 8k alignment was not possible.
> I tried exactly the same stuff once and failed.
> I think the restriction came from hpux compatibility or some old gas...

The maximum alignment on hpux is 4k bytes.  This was implemented in
2004 in the PA backend for hpux.  Prior to that, the default for
MAX_OFILE_ALIGNMENT (BIGGEST_ALIGNMENT) was in effect for all PA
targets.  It looks like Carlos added the define to elfos.h in November
2006.  So unless this was back ported, 4.2.0 would be the first gcc
release to support large alignments on linux.

Dave
-- 
J. David Anglin                                  dave.anglin@nrc-cnrc.gc.ca
National Research Council of Canada              (613) 990-0752 (FAX: 952-6602)

Re: [PATCH 3/5] parisc: use new macros for .data.init_task.

[Sam Ravnborg]

On Sat, May 02, 2009 at 07:13:37AM +0200, Helge Deller wrote:
> Tim Abbott wrote:
> > .data.init_task should not need a separate output section; this change
> > moves it into the .data section.
> > 
> > Signed-off-by: Tim Abbott <tabbott@mit.edu>
> > Cc: Kyle McMartin <kyle@mcmartin.ca>
> > Cc: Helge Deller <deller@gmx.de>
> > Cc: linux-parisc@vger.kernel.org
> 
> 
> I think this patch is wrong, although it is theoretically correct.
> 
> IIRC, gcc on hppa is not able to provide an alignment >= 8k, which is
> why we have done the 16k alignment inside the linker script.
> So, I think this change will prevent the parisc kernel to boot up.
> Needs testing.

The patch does not do much...

> 
> Helge
> 
> > ---
> >  arch/parisc/kernel/init_task.c   |    2 +-
> >  arch/parisc/kernel/vmlinux.lds.S |   10 ++--------
> >  2 files changed, 3 insertions(+), 9 deletions(-)
> > 
> > diff --git a/arch/parisc/kernel/init_task.c b/arch/parisc/kernel/init_task.c
> > index 1e25a45..8ee17ea 100644
> > --- a/arch/parisc/kernel/init_task.c
> > +++ b/arch/parisc/kernel/init_task.c
> > @@ -48,7 +48,7 @@ EXPORT_SYMBOL(init_mm);
> >   * "init_task" linker map entry..
> >   */
> >  union thread_union init_thread_union
> > -	__attribute__((aligned(128))) __attribute__((__section__(".data.init_task"))) =
> > +	__attribute__((aligned(128))) __init_task_data =
> >  		{ INIT_THREAD_INFO(init_task) };
This is a simple replacement with a nicer way to say "this belongs to
the .data.init_task section - no functional difference.


> >  
> >  #if PT_NLEVELS == 3
> > diff --git a/arch/parisc/kernel/vmlinux.lds.S b/arch/parisc/kernel/vmlinux.lds.S
> > index b5936c9..c8a528d 100644
> > --- a/arch/parisc/kernel/vmlinux.lds.S
> > +++ b/arch/parisc/kernel/vmlinux.lds.S
> > @@ -103,6 +103,8 @@ SECTIONS
> >  	. = ALIGN(L1_CACHE_BYTES);
> >  	/* Data */
> >  	.data : {
> > +		/* assembler code expects init_task to be 16k aligned */
> > +		INIT_TASK_DATA(16384)
> >  		NOSAVE_DATA
> >  		CACHELINE_ALIGNED_DATA(L1_CACHE_BYTES)
> >  		DATA_DATA
> > @@ -133,14 +135,6 @@ SECTIONS
> >  	}
> >  	__bss_stop = .;
> >  
> > -
> > -	/* assembler code expects init_task to be 16k aligned */
> > -	. = ALIGN(16384);
> > -	/* init_task */
> > -	.data.init_task : {
> > -		*(.data.init_task)
> > -	}
> > -
This part moves away from a specific output section to including this in the
.data output section - with the _same_ alignmnet.
So the only issue here is that we move init_task before NOSAVE_DATA etc.

I do not see why you think this changes alignmnet?

	Sam

Re: [PATCH 3/5] parisc: use new macros for .data.init_task.

[Helge Deller]

Sam Ravnborg wrote:
> On Sat, May 02, 2009 at 07:13:37AM +0200, Helge Deller wrote:
>> Tim Abbott wrote:
>>> .data.init_task should not need a separate output section; this change
>>> moves it into the .data section.
>>>
>>> Signed-off-by: Tim Abbott <tabbott@mit.edu>
>>> Cc: Kyle McMartin <kyle@mcmartin.ca>
>>> Cc: Helge Deller <deller@gmx.de>
>>> Cc: linux-parisc@vger.kernel.org
>>
>> I think this patch is wrong, although it is theoretically correct.
>>
>> IIRC, gcc on hppa is not able to provide an alignment >= 8k, which is
>> why we have done the 16k alignment inside the linker script.
>> So, I think this change will prevent the parisc kernel to boot up.
>> Needs testing.
> 
> The patch does not do much...
> 
>> Helge
>>
>>> ---
>>>  arch/parisc/kernel/init_task.c   |    2 +-
>>>  arch/parisc/kernel/vmlinux.lds.S |   10 ++--------
>>>  2 files changed, 3 insertions(+), 9 deletions(-)
>>>
>>> diff --git a/arch/parisc/kernel/init_task.c b/arch/parisc/kernel/init_task.c
>>> index 1e25a45..8ee17ea 100644
>>> --- a/arch/parisc/kernel/init_task.c
>>> +++ b/arch/parisc/kernel/init_task.c
>>> @@ -48,7 +48,7 @@ EXPORT_SYMBOL(init_mm);
>>>   * "init_task" linker map entry..
>>>   */
>>>  union thread_union init_thread_union
>>> -	__attribute__((aligned(128))) __attribute__((__section__(".data.init_task"))) =
>>> +	__attribute__((aligned(128))) __init_task_data =
>>>  		{ INIT_THREAD_INFO(init_task) };
> This is a simple replacement with a nicer way to say "this belongs to
> the .data.init_task section - no functional difference.
> 
> 
>>>  
>>>  #if PT_NLEVELS == 3
>>> diff --git a/arch/parisc/kernel/vmlinux.lds.S b/arch/parisc/kernel/vmlinux.lds.S
>>> index b5936c9..c8a528d 100644
>>> --- a/arch/parisc/kernel/vmlinux.lds.S
>>> +++ b/arch/parisc/kernel/vmlinux.lds.S
>>> @@ -103,6 +103,8 @@ SECTIONS
>>>  	. = ALIGN(L1_CACHE_BYTES);
>>>  	/* Data */
>>>  	.data : {
>>> +		/* assembler code expects init_task to be 16k aligned */
>>> +		INIT_TASK_DATA(16384)
>>>  		NOSAVE_DATA
>>>  		CACHELINE_ALIGNED_DATA(L1_CACHE_BYTES)
>>>  		DATA_DATA
>>> @@ -133,14 +135,6 @@ SECTIONS
>>>  	}
>>>  	__bss_stop = .;
>>>  
>>> -
>>> -	/* assembler code expects init_task to be 16k aligned */
>>> -	. = ALIGN(16384);
>>> -	/* init_task */
>>> -	.data.init_task : {
>>> -		*(.data.init_task)
>>> -	}
>>> -
> This part moves away from a specific output section to including this in the
> .data output section - with the _same_ alignmnet.
> So the only issue here is that we move init_task before NOSAVE_DATA etc.
> 
> I do not see why you think this changes alignmnet?

Ugh. Of course you are correct. It doesn't change anything.
Patch is OK for me.
I missed the 16384 in INIT_TASK_DATA(16384).

Thanks,
Helge