Skip to contentSkip to conversation

Allow percpu variables to be page-aligned

PATCH In review Jeremy Fitzhardinge 2 participants 3 messages Apr 10, 2007

Subsystems (9)
Arm (32-bit) PowerPC MIPS s390 +5 more subsystems
lore
Share & tools
lore

Jeremy Fitzhardinge

author
#0 lore raw

PATCH 2/2 Allow percpu variables to be page-aligned

Let's allow page-alignment in general for per-cpu data (wanted by Xen, and Ingo suggested KVM as well).

Because larger alignments can use more room, we increase the max per-cpu memory to 64k rather than 32k: it's getting a little tight.

Signed-off-by Rusty Russell <[email protected]>
Signed-off-by Jeremy Fitzhardinge <[email protected]>
Acked-by Ingo Molnar <[email protected]>
Cc Andi Kleen <[email protected]>
Signed-off-by Andrew Morton <[email protected]>
---
 arch/alpha/kernel/vmlinux.lds.S   |    2 +-
 arch/arm/kernel/vmlinux.lds.S     |    2 +-
 arch/cris/arch-v32/vmlinux.lds.S  |    1 +
 arch/frv/kernel/vmlinux.lds.S     |    1 +
 arch/i386/kernel/vmlinux.lds.S    |    2 +-
 arch/m32r/kernel/vmlinux.lds.S    |    2 +-
 arch/mips/kernel/vmlinux.lds.S    |    2 +-
 arch/parisc/kernel/vmlinux.lds.S  |    2 +-
 arch/powerpc/kernel/setup_64.c    |    4 ++--
 arch/powerpc/kernel/vmlinux.lds.S |    6 +-----
 arch/ppc/kernel/vmlinux.lds.S     |    2 +-
 arch/s390/kernel/vmlinux.lds.S    |    2 +-
 arch/sh/kernel/vmlinux.lds.S      |    2 +-
 arch/sh64/kernel/vmlinux.lds.S    |    2 +-
 arch/sparc/kernel/vmlinux.lds.S   |    2 +-
 arch/sparc64/kernel/smp.c         |    6 +++---
 arch/x86_64/kernel/setup64.c      |    4 ++--
 arch/x86_64/kernel/vmlinux.lds.S  |    2 +-
 arch/xtensa/kernel/vmlinux.lds.S  |    2 +-
 init/main.c                       |    4 ++--
 kernel/module.c                   |    8 ++++----
 21 files changed, 29 insertions(+), 31 deletions(-)

===================================================================

arch/alpha/kernel/vmlinux.lds.S +1 -1
@@ -69,7 +69,7 @@ SECTIONS
   . = ALIGN(8);
   SECURITY_INIT
 
-  . = ALIGN(64);
+  . = ALIGN(8192);
   __per_cpu_start = .;
   .data.percpu : { *(.data.percpu) }
   __per_cpu_end = .;

===================================================================

arch/arm/kernel/vmlinux.lds.S +1 -1
@@ -59,7 +59,7 @@ SECTIONS
 			usr/built-in.o(.init.ramfs)
 		__initramfs_end = .;
 #endif
-		. = ALIGN(64);
+		. = ALIGN(4096);
 		__per_cpu_start = .;
 			*(.data.percpu)
 		__per_cpu_end = .;

===================================================================

arch/cris/arch-v32/vmlinux.lds.S +1 -0
@@ -91,6 +91,7 @@ SECTIONS
 	}
 	SECURITY_INIT
 
+	. =  ALIGN (8192);
 	__per_cpu_start = .;
 	.data.percpu  : { *(.data.percpu) }
 	__per_cpu_end = .;

===================================================================

arch/frv/kernel/vmlinux.lds.S +1 -0
@@ -57,6 +57,7 @@ SECTIONS
   __alt_instructions_end = .;
  .altinstr_replacement : { *(.altinstr_replacement) }
 
+  . = ALIGN(4096);
   __per_cpu_start = .;
   .data.percpu  : { *(.data.percpu) }
   __per_cpu_end = .;

===================================================================

arch/i386/kernel/vmlinux.lds.S +1 -1
@@ -194,7 +194,7 @@ SECTIONS
 	__initramfs_end = .;
   }
 #endif
-  . = ALIGN(L1_CACHE_BYTES);
+  . = ALIGN(4096);
   .data.percpu  : AT(ADDR(.data.percpu) - LOAD_OFFSET) {
 	__per_cpu_start = .;
 	*(.data.percpu)

===================================================================

arch/m32r/kernel/vmlinux.lds.S +1 -1
@@ -110,7 +110,7 @@ SECTIONS
   __initramfs_end = .;
 #endif
 
-  . = ALIGN(32);
+  . = ALIGN(4096);
   __per_cpu_start = .;
   .data.percpu  : { *(.data.percpu) }
   __per_cpu_end = .;

===================================================================

arch/mips/kernel/vmlinux.lds.S +1 -1
@@ -119,7 +119,7 @@ SECTIONS
   .init.ramfs : { *(.init.ramfs) }
   __initramfs_end = .;
 #endif
-  . = ALIGN(32);
+  . = ALIGN(_PAGE_SIZE);
   __per_cpu_start = .;
   .data.percpu  : { *(.data.percpu) }
   __per_cpu_end = .;

===================================================================

arch/parisc/kernel/vmlinux.lds.S +1 -1
@@ -181,7 +181,7 @@ SECTIONS
   .init.ramfs : { *(.init.ramfs) }
   __initramfs_end = .;
 #endif
-  . = ALIGN(32);
+  . = ALIGN(ASM_PAGE_SIZE);
   __per_cpu_start = .;
   .data.percpu  : { *(.data.percpu) }
   __per_cpu_end = .;

===================================================================

arch/powerpc/kernel/setup_64.c +2 -2
@@ -583,14 +583,14 @@ void __init setup_per_cpu_areas(void)
 	char *ptr;
 
 	/* Copy section for each CPU (we discard the original) */
-	size = ALIGN(__per_cpu_end - __per_cpu_start, SMP_CACHE_BYTES);
+	size = ALIGN(__per_cpu_end - __per_cpu_start, PAGE_SIZE);
 #ifdef CONFIG_MODULES
 	if (size < PERCPU_ENOUGH_ROOM)
 		size = PERCPU_ENOUGH_ROOM;
 #endif
 
 	for_each_possible_cpu(i) {
-		ptr = alloc_bootmem_node(NODE_DATA(cpu_to_node(i)), size);
+		ptr = alloc_bootmem_pages_node(NODE_DATA(cpu_to_node(i)), size);
 		if (!ptr)
 			panic("Cannot allocate cpu data for CPU %d\n", i);

===================================================================

arch/powerpc/kernel/vmlinux.lds.S +1 -5
@@ -139,11 +139,7 @@ SECTIONS
 		__initramfs_end = .;
 	}
 #endif
-#ifdef CONFIG_PPC32
-	. = ALIGN(32);
-#else
-	. = ALIGN(128);
-#endif
+	. = ALIGN(PAGE_SIZE);
 	.data.percpu : {
 		__per_cpu_start = .;
 		*(.data.percpu)

===================================================================

arch/ppc/kernel/vmlinux.lds.S +1 -1
@@ -130,7 +130,7 @@ SECTIONS
   __ftr_fixup : { *(__ftr_fixup) }
   __stop___ftr_fixup = .;
 
-  . = ALIGN(32);
+  . = ALIGN(4096);
   __per_cpu_start = .;
   .data.percpu  : { *(.data.percpu) }
   __per_cpu_end = .;

===================================================================

arch/s390/kernel/vmlinux.lds.S +1 -1
@@ -99,7 +99,7 @@ SECTIONS
   . = ALIGN(2);
   __initramfs_end = .;
 #endif
-  . = ALIGN(256);
+  . = ALIGN(4096);
   __per_cpu_start = .;
   .data.percpu  : { *(.data.percpu) }
   __per_cpu_end = .;

===================================================================

arch/sh/kernel/vmlinux.lds.S +1 -1
@@ -54,7 +54,7 @@ SECTIONS
   . = ALIGN(PAGE_SIZE);
   .data.page_aligned : { *(.data.page_aligned) }
 
-  . = ALIGN(L1_CACHE_BYTES);
+  . = ALIGN(PAGE_SIZE);
   __per_cpu_start = .;
   .data.percpu : { *(.data.percpu) }
   __per_cpu_end = .;

===================================================================

arch/sh64/kernel/vmlinux.lds.S +1 -1
@@ -85,7 +85,7 @@ SECTIONS
   . = ALIGN(PAGE_SIZE);
   .data.page_aligned : C_PHYS(.data.page_aligned) { *(.data.page_aligned) }
 
-  . = ALIGN(L1_CACHE_BYTES);
+  . = ALIGN(PAGE_SIZE);
   __per_cpu_start = .;
   .data.percpu : C_PHYS(.data.percpu) { *(.data.percpu) }
   __per_cpu_end = . ;

===================================================================

arch/sparc/kernel/vmlinux.lds.S +1 -1
@@ -65,7 +65,7 @@ SECTIONS
   __initramfs_end = .;
 #endif
 
-  . = ALIGN(32);
+  . = ALIGN(4096);
   __per_cpu_start = .;
   .data.percpu  : { *(.data.percpu) }
   __per_cpu_end = .;

===================================================================

arch/sparc64/kernel/smp.c +3 -3
@@ -1449,11 +1449,11 @@ void __init setup_per_cpu_areas(void)
 	/* Copy section for each CPU (we discard the original) */
 	goal = PERCPU_ENOUGH_ROOM;
 
-	__per_cpu_shift = 0;
-	for (size = 1UL; size < goal; size <<= 1UL)
+	__per_cpu_shift = PAGE_SHIFT;
+	for (size = PAGE_SIZE; size < goal; size <<= 1UL)
 		__per_cpu_shift++;
 
-	ptr = alloc_bootmem(size * NR_CPUS);
+	ptr = alloc_bootmem_pages(size * NR_CPUS);
 
 	__per_cpu_base = ptr - __per_cpu_start;

===================================================================

arch/x86_64/kernel/setup64.c +2 -2
@@ -103,9 +103,9 @@ void __init setup_per_cpu_areas(void)
 		if (!NODE_DATA(cpu_to_node(i))) {
 			printk("cpu with no node %d, num_online_nodes %d\n",
 			       i, num_online_nodes());
-			ptr = alloc_bootmem(size);
+			ptr = alloc_bootmem_pages(size);
 		} else { 
-			ptr = alloc_bootmem_node(NODE_DATA(cpu_to_node(i)), size);
+			ptr = alloc_bootmem_pages_node(NODE_DATA(cpu_to_node(i)), size);
 		}
 		if (!ptr)
 			panic("Cannot allocate cpu data for CPU %d\n", i);

===================================================================

arch/x86_64/kernel/vmlinux.lds.S +1 -1
@@ -194,7 +194,7 @@ SECTIONS
   __initramfs_end = .;
 #endif
 
-    . = ALIGN(CONFIG_X86_L1_CACHE_BYTES);
+  . = ALIGN(4096);
   __per_cpu_start = .;
   .data.percpu  : AT(ADDR(.data.percpu) - LOAD_OFFSET) { *(.data.percpu) }
   __per_cpu_end = .;

===================================================================

arch/xtensa/kernel/vmlinux.lds.S +1 -1
@@ -198,7 +198,7 @@ SECTIONS
   __ftr_fixup : { *(__ftr_fixup) }
   __stop___ftr_fixup = .;
 
-  . = ALIGN(32);
+  . = ALIGN(4096);
   __per_cpu_start = .;
   .data.percpu  : { *(.data.percpu) }
   __per_cpu_end = .;

===================================================================

init/main.c +2 -2
@@ -370,8 +370,8 @@ static void __init setup_per_cpu_areas(v
 
 	/* Copy section for each CPU (we discard the original) */
 
-	size = ALIGN(PERCPU_ENOUGH_ROOM, SMP_CACHE_BYTES);
-	ptr = alloc_bootmem(size * nr_possible_cpus);
+	size = ALIGN(PERCPU_ENOUGH_ROOM, PAGE_SIZE);
+	ptr = alloc_bootmem_pages(size * nr_possible_cpus);
 
 	for_each_possible_cpu(i) {
 		__per_cpu_offset[i] = ptr - __per_cpu_start;

===================================================================

kernel/module.c +4 -4
@@ -346,10 +346,10 @@ static void *percpu_modalloc(unsigned lo
 	unsigned int i;
 	void *ptr;
 
-	if (align > SMP_CACHE_BYTES) {
-		printk(KERN_WARNING "%s: per-cpu alignment %li > %i\n",
-		       name, align, SMP_CACHE_BYTES);
-		align = SMP_CACHE_BYTES;
+	if (align > PAGE_SIZE) {
+		printk(KERN_WARNING "%s: per-cpu alignment %li > %li\n",
+		       name, align, PAGE_SIZE);
+		align = PAGE_SIZE;
 	}
 
 	ptr = __per_cpu_start;
#1 lore raw
Quoted 5 lines · Jeremy Fitzhardinge

Let's allow page-alignment in general for per-cpu data (wanted by Xen, and Ingo suggested KVM as well).

Because larger alignments can use more room, we increase the max per-cpu memory to 64k rather than 32k: it's getting a little tight.

The second paragraph of the comment is dated. You are not changing the per-cpu memory size at all now.

The code itself looks sane.

Although we should probably fail the module load if the requested alignment is to great, but that is just picking nits, and is no worse than what we do today.

Eric

Quoted 18 lines

===================================================================

kernel/module.c +4 -4
@@ -346,10 +346,10 @@ static void *percpu_modalloc(unsigned lo
 	unsigned int i;
 	void *ptr;
 
-	if (align > SMP_CACHE_BYTES) {
-		printk(KERN_WARNING "%s: per-cpu alignment %li > %i\n",
-		       name, align, SMP_CACHE_BYTES);
-		align = SMP_CACHE_BYTES;
+	if (align > PAGE_SIZE) {
+		printk(KERN_WARNING "%s: per-cpu alignment %li > %li\n",
+		       name, align, PAGE_SIZE);
+		align = PAGE_SIZE;
 	}
 
 	ptr = __per_cpu_start;
author
#2 lore raw
Quoted 6 lines · Eric W. Biederman
Quoted 3 lines · Jeremy Fitzhardinge

Let's allow page-alignment in general for per-cpu data (wanted by Xen, and Ingo suggested KVM as well).

Quoted 7 lines
Quoted 3 lines

Because larger alignments can use more room, we increase the max per-cpu memory to 64k rather than 32k: it's getting a little tight.

The second paragraph of the comment is dated. You are not changing the per-cpu memory size at all now.

Yup.

Quoted 4 lines

Although we should probably fail the module load if the requested alignment is to great, but that is just picking nits, and is no worse than what we do today.

Yes. And we barely support a module's use of page-aligned data, since it would potentially chew through the percpu space very quickly.

J