Skip to contentSkip to conversation

align per-cpu section to configured cache bytes

PATCH In review Zach Brown 3 participants 6 messages Jan 27, 2006 – Jan 28, 2006

lore
Share & tools
lore

Zach Brown

author
#0 lore raw

PATCH 1/2 align per-cpu section to configured cache bytes

[x86-64] align per-cpu section to configured cache bytes

Align the start of the per-cpu section to the configured number of bytes in a cache line. This stops a BUG_ON() from triggering in load_module() when DEFINE_PER_CPU() is used in a module and the section isn't cacheline-aligned. Rusty also found this and sent a patch in a while ago
(http://lkml.org/lkml/2004/10/19/17), I don't know what came of that.

---

 arch/x86_64/kernel/vmlinux.lds.S |    2 +-
 1 files changed, 1 insertion(+), 1 deletion(-)
arch/x86_64/kernel/vmlinux.lds.S +1 -1
Index: 2.6.16-rc1-mm3-per-cpu-alignment/arch/x86_64/kernel/vmlinux.lds.S
===================================================================
@@ -173,7 +173,7 @@
   __initramfs_start = .;
   .init.ramfs : AT(ADDR(.init.ramfs) - LOAD_OFFSET) { *(.init.ramfs) }
   __initramfs_end = .;	
-  . = ALIGN(32);
+  . = ALIGN(CONFIG_X86_L1_CACHE_BYTES);
   __per_cpu_start = .;
   .data.percpu  : AT(ADDR(.data.percpu) - LOAD_OFFSET) { *(.data.percpu) }
   __per_cpu_end = .;
author
#1 lore raw

PATCH 2/2 align per-cpu section to configured cache bytes

[x86] align per-cpu section to configured cache bytes

This matches the fix for a bug seen on x86-64. Test booted on old hardware that had 32 byte cachelines to begin with.

---

 arch/i386/kernel/vmlinux.lds.S |    3 ++-
 1 files changed, 2 insertions(+), 1 deletion(-)
arch/i386/kernel/vmlinux.lds.S +2 -1
Index: 2.6.16-rc1-per-cpu-align/arch/i386/kernel/vmlinux.lds.S
===================================================================
@@ -7,6 +7,7 @@
 #include <asm-generic/vmlinux.lds.h>
 #include <asm/thread_info.h>
 #include <asm/page.h>
+#include <asm/cache.h>
 
 OUTPUT_FORMAT("elf32-i386", "elf32-i386", "elf32-i386")
 OUTPUT_ARCH(i386)
@@ -115,7 +116,7 @@
   __initramfs_start = .;
   .init.ramfs : AT(ADDR(.init.ramfs) - LOAD_OFFSET) { *(.init.ramfs) }
   __initramfs_end = .;
-  . = ALIGN(32);
+  . = ALIGN(L1_CACHE_BYTES);
   __per_cpu_start = .;
   .data.percpu  : AT(ADDR(.data.percpu) - LOAD_OFFSET) { *(.data.percpu) }
   __per_cpu_end = .;
#2 lore raw
Quoted 6 lines · Zach Brown

[x86] align per-cpu section to configured cache bytes

This matches the fix for a bug seen on x86-64. Test booted on old hardware that had 32 byte cachelines to begin with.

I've applied this verbatim.
The asm-generic part suggested in other mail was not straightforward due to subtle differences and I have not yet my cross compile environment running on this box.

Sam

#3 lore raw
In reply to Zach Brown

arch/x86_64/kernel/vmlinux.lds.S line 176

Quoted 2 lines · Zach Brown
arch/x86_64/kernel/vmlinux.lds.S +1 -1
-  . = ALIGN(32);
+  . = ALIGN(CONFIG_X86_L1_CACHE_BYTES);

Grepping other arch's than just x86 and x86_64 it looks like a common thing.
Is this fix really only relevant for x86 + x86_64 or should it be done for all arch's?

If we do it for all archs we may as well create:
#define PERCPU(aling) ...
macro in asm-generic/vmlinux.lds.h

Sam

author
#4 lore raw
Quoted 10 lines · Sam Ravnborg

arch/x86_64/kernel/vmlinux.lds.S line 176

Quoted 2 lines · Zach Brown
arch/x86_64/kernel/vmlinux.lds.S +1 -1
-  . = ALIGN(32);
+  . = ALIGN(CONFIG_X86_L1_CACHE_BYTES);

Grepping other arch's than just x86 and x86_64 it looks like a common thing.
Is this fix really only relevant for x86 + x86_64 or should it be done for all arch's?

I think it'd be needed if other archs had situations where C's (load_module()'s, in particular) notion of the cacheline size differed from vmlinux.lds.S's. I didn't want to go screwing around with archs that I couldn't immediately test :)

Quoted 3 lines

If we do it for all archs we may as well create:
#define PERCPU(aling) ...
macro in asm-generic/vmlinux.lds.h

Sounds reasonable to me, should I leave that in your capable hands?

- z

#5 lore raw
In reply to Zach Brown
Quoted 7 lines · Zach Brown

[x86-64] align per-cpu section to configured cache bytes

Align the start of the per-cpu section to the configured number of bytes in a cache line. This stops a BUG_ON() from triggering in load_module() when DEFINE_PER_CPU() is used in a module and the section isn't cacheline-aligned. Rusty also found this and sent a patch in a while ago
(http://lkml.org/lkml/2004/10/19/17), I don't know what came of that.

Added.

-Andi