Skip to contentSkip to conversation

section name cleanup for mn10300

PATCH In review Tim Abbott 3 participants 13 messages Apr 30, 2009 – May 1, 2009

lore
Share & tools
lore

Tim Abbott

author
#0 lore raw

cover section name cleanup for mn10300

This patch series cleans up the section names on the mn10300 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 (6):
mn10300: Use macros for .bss.page_aligned section.
mn10300: Drop unused .data.idt section.
mn10300: use NOSAVE_DATA macro for .data.nosave section.
mn10300: use new macro for .data.cacheline_aligned section.
mn10300: use new macros for .data.init_task.
mn10300: use new macro for .data.read_mostly section.

 arch/mn10300/kernel/init_task.c   |    3 +--
 arch/mn10300/kernel/vmlinux.lds.S |   29 ++++++-----------------------
 2 files changed, 7 insertions(+), 25 deletions(-)
author
#1 lore raw

PATCH 1/6 mn10300: Use macros for .bss.page_aligned section.

---
 arch/mn10300/kernel/vmlinux.lds.S |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)
arch/mn10300/kernel/vmlinux.lds.S +1 -1
index 24de6b9..6ad0fa8 100644
@@ -131,7 +131,7 @@ SECTIONS
 
   __bss_start = .;		/* BSS */
   .bss : {
-	*(.bss.page_aligned)
+	PAGE_ALIGNED_BSS
 	*(.bss)
   }
   . = ALIGN(4);
Signature · 1 line
1.6.2.1
author
#2 lore raw

PATCH 2/6 mn10300: Drop unused .data.idt section.

Since nothing gets put into .data.idt on this architecture, eliminate it from the linker script.

This change was apparently proposed previously:
<http://lkml.indiana.edu/hypermail/linux/kernel/0802.2/2538.html>

CCing the author and people who acked that patch in case there was a reason it wasn't applied.

---
 arch/mn10300/kernel/vmlinux.lds.S |    3 ---
 1 files changed, 0 insertions(+), 3 deletions(-)
arch/mn10300/kernel/vmlinux.lds.S +0 -3
index 6ad0fa8..364250f 100644
@@ -59,9 +59,6 @@ SECTIONS
   . = ALIGN(PAGE_SIZE);
   __nosave_end = .;
 
-  . = ALIGN(PAGE_SIZE);
-  .data.page_aligned : { *(.data.idt) }
-
   . = ALIGN(32);
   .data.cacheline_aligned : { *(.data.cacheline_aligned) }
 
Signature · 1 line
1.6.2.1
author
#3 lore raw

PATCH 3/6 mn10300: use NOSAVE_DATA macro for .data.nosave section.

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

Cc David Howells <[email protected]>
---
 arch/mn10300/kernel/vmlinux.lds.S |    7 +------
 1 files changed, 1 insertions(+), 6 deletions(-)
arch/mn10300/kernel/vmlinux.lds.S +1 -6
index 364250f..5e825e8 100644
@@ -49,16 +49,11 @@ SECTIONS
 
   /* writeable */
   .data : {			/* Data */
+	NOSAVE_DATA
 	DATA_DATA
 	CONSTRUCTORS
 	}
 
-  . = ALIGN(PAGE_SIZE);
-  __nosave_begin = .;
-  .data_nosave : { *(.data.nosave) }
-  . = ALIGN(PAGE_SIZE);
-  __nosave_end = .;
-
   . = ALIGN(32);
   .data.cacheline_aligned : { *(.data.cacheline_aligned) }
 
Signature · 1 line
1.6.2.1
author
#4 lore raw

PATCH 4/6 mn10300: use new macro for .data.cacheline_aligned section.

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

Cc David Howells <[email protected]>
---
 arch/mn10300/kernel/vmlinux.lds.S |    4 +---
 1 files changed, 1 insertions(+), 3 deletions(-)
arch/mn10300/kernel/vmlinux.lds.S +1 -3
index 5e825e8..cb43406 100644
@@ -50,13 +50,11 @@ SECTIONS
   /* writeable */
   .data : {			/* Data */
 	NOSAVE_DATA
+	CACHELINE_ALIGNED_DATA(32)
 	DATA_DATA
 	CONSTRUCTORS
 	}
 
-  . = ALIGN(32);
-  .data.cacheline_aligned : { *(.data.cacheline_aligned) }
-
   /* rarely changed data like cpu maps */
   . = ALIGN(32);
   .data.read_mostly : AT(ADDR(.data.read_mostly)) {
Signature · 1 line
1.6.2.1
author
#5 lore raw

PATCH 5/6 mn10300: use new macros for .data.init_task.

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

Cc David Howells <[email protected]>
---
 arch/mn10300/kernel/init_task.c   |    3 +--
 arch/mn10300/kernel/vmlinux.lds.S |    4 +---
 2 files changed, 2 insertions(+), 5 deletions(-)
arch/mn10300/kernel/init_task.c +1 -2
index 5ac3566..7a89fc3 100644
@@ -30,8 +30,7 @@ EXPORT_SYMBOL(init_mm);
  * way process stacks are handled. This is done by having a special
  * "init_task" linker map entry..
  */
-union thread_union init_thread_union
-	__attribute__((__section__(".data.init_task"))) =
+union thread_union init_thread_union __init_task_data =
 		{ INIT_THREAD_INFO(init_task) };
 
 /*
arch/mn10300/kernel/vmlinux.lds.S +1 -3
index cb43406..4dc63e6 100644
@@ -49,6 +49,7 @@ SECTIONS
 
   /* writeable */
   .data : {			/* Data */
+	INIT_TASK_DATA(THREAD_SIZE)
 	NOSAVE_DATA
 	CACHELINE_ALIGNED_DATA(32)
 	DATA_DATA
@@ -62,9 +63,6 @@ SECTIONS
 	_edata = .;		/* End of data section */
   }
 
-  . = ALIGN(THREAD_SIZE);	/* init_task */
-  .data.init_task : { *(.data.init_task) }
-
   /* might get freed after init */
   . = ALIGN(PAGE_SIZE);
   .smp_locks : AT(ADDR(.smp_locks) - LOAD_OFFSET) {
Signature · 1 line
1.6.2.1
author
#6 lore raw

PATCH 6/6 mn10300: use new macro for .data.read_mostly section.

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

Cc David Howells <[email protected]>
---
 arch/mn10300/kernel/vmlinux.lds.S |    9 ++-------
 1 files changed, 2 insertions(+), 7 deletions(-)
arch/mn10300/kernel/vmlinux.lds.S +2 -7
index 4dc63e6..3da48d5 100644
@@ -52,16 +52,11 @@ SECTIONS
 	INIT_TASK_DATA(THREAD_SIZE)
 	NOSAVE_DATA
 	CACHELINE_ALIGNED_DATA(32)
+	READ_MOSTLY_DATA(32)
 	DATA_DATA
 	CONSTRUCTORS
-	}
-
-  /* rarely changed data like cpu maps */
-  . = ALIGN(32);
-  .data.read_mostly : AT(ADDR(.data.read_mostly)) {
-	*(.data.read_mostly)
 	_edata = .;		/* End of data section */
-  }
+	}
 
   /* might get freed after init */
   . = ALIGN(PAGE_SIZE);
Signature · 1 line
1.6.2.1
#7 lore raw

Re: [PATCH 5/6] mn10300: use new macros for .data.init_task.

In reply to Tim Abbott
Quoted 2 lines · Tim Abbott

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

Hmmm... This causes the kernel I'm building to make an extra page of RAM into dead space.

David

author
#8 lore raw

Re: [PATCH 5/6] mn10300: use new macros for .data.init_task.

Quoted 7 lines · David Howells
Quoted 2 lines · Tim Abbott

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

Hmmm... This causes the kernel I'm building to make an extra page of RAM into dead space.

I would try reversing the order of the things inside the .data output section, since the stuff after .data at the end of this patch series is page-aligned, while the stuff before .data isn't.

-Tim Abbott

#9 lore raw

Re: [PATCH 5/6] mn10300: use new macros for .data.init_task.

Quoted 7 lines · Tim Abbott
Quoted 5 lines · David Howells
Quoted 2 lines · Tim Abbott

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

Quoted 6 lines
Quoted 2 lines

Hmmm... This causes the kernel I'm building to make an extra page of RAM into dead space.

I would try reversing the order of the things inside the .data output section, since the stuff after .data at the end of this patch series is page-aligned, while the stuff before .data isn't.

I'm looking into creating a WRDATA macro like we have RODATA today. If that flys then we will hide a lot in that section, including the correct order.

But it will take a few hours before I'm ready.

Sam

#10 lore raw
In reply to Tim Abbott
Quoted 3 lines · Tim Abbott

This patch series cleans up the section names on the mn10300 architecture. It requires the architecture-independent macro definitions from this patch series:

Interesting... Whilst it compiles and links with no problems, this cleanup has the following detrimental effects:

None of these patches applied

	   text    data     bss     dec     hex filename
	1879329   82980  321772 2284081  22da31 vmlinux

All of these patches applied:

	   text    data     bss     dec     hex filename
	1879329   87076  321772 2288177  22ea31 vmlinux

Somewhere it eats an extra page.

David

#11 lore raw
Quoted 14 lines · David Howells
Quoted 3 lines · Tim Abbott

This patch series cleans up the section names on the mn10300 architecture. It requires the architecture-independent macro definitions from this patch series:

Interesting... Whilst it compiles and links with no problems, this cleanup has the following detrimental effects:

None of these patches applied

	   text    data     bss     dec     hex filename
	1879329   82980  321772 2284081  22da31 vmlinux
Quoted 6 lines

All of these patches applied:

	   text    data     bss     dec     hex filename
	1879329   87076  321772 2288177  22ea31 vmlinux

Somewhere it eats an extra page.

This is likely due to moving around the sections and we have got them in less perfect order.

I also thing 5/6 is wrong as it moves
init_task befoe _edata.

Could you try to back out them one by one
to see which one causes the extra memory waste.

Sam

#12 lore raw
Quoted 5 lines · Sam Ravnborg

I also thing 5/6 is wrong as it moves
init_task befoe _edata.

Could you try to back out them one by one
to see which one causes the extra memory waste.

Moving init_task around is the culprit.

I guess the extra page has a 50% chance of being leaked anyway, as it depends on the alignment of the previous section.

I can probably make the kernel reclaim it by sticking a label either side of the alignment and if they're more than a page apart, reclaim the page that ends at the second label.

David