lg_prealloc_list seems to cry out for a per-cpu data structure; on a large
smp system I think this should be better. I've lightly tested this change
on a 4-cpu system. Comments welcome...
Signed-off-by: Eric Sandeen <[email protected]>
---
ndex: linux-2.6/fs/ext4/mballoc.c
===================================================================
--- linux-2.6.orig/fs/ext4/mballoc.c 2008-08-04 15:30:30.000000000 -0500
+++ linux-2.6/fs/ext4/mballoc.c 2008-08-13 13:48:33.224165751 -0500
@@ -2540,17 +2540,16 @@ int ext4_mb_init(struct super_block *sb,
sbi->s_mb_history_filter = EXT4_MB_HISTORY_DEFAULT;
sbi->s_mb_group_prealloc = MB_DEFAULT_GROUP_PREALLOC;
- i = sizeof(struct ext4_locality_group) * nr_cpu_ids;
- sbi->s_locality_groups = kmalloc(i, GFP_KERNEL);
+ sbi->s_locality_groups = alloc_percpu(struct ext4_locality_group);
if (sbi->s_locality_groups == NULL) {
clear_opt(sbi->s_mount_opt, MBALLOC);
kfree(sbi->s_mb_offsets);
kfree(sbi->s_mb_maxs);
return -ENOMEM;
}
- for (i = 0; i < nr_cpu_ids; i++) {
+ for_each_possible_cpu(i) {
struct ext4_locality_group *lg;
- lg = &sbi->s_locality_groups[i];
+ lg = per_cpu_ptr(sbi->s_locality_groups, i);
mutex_init(&lg->lg_mutex);
for (j = 0; j < PREALLOC_TB_SIZE; j++)
INIT_LIST_HEAD(&lg->lg_prealloc_list[j]);
@@ -2647,8 +2646,7 @@ int ext4_mb_release(struct super_block *
atomic_read(&sbi->s_mb_discarded));
}
- kfree(sbi->s_locality_groups);
Eric Sandeen wrote:
> lg_prealloc_list seems to cry out for a per-cpu data structure; on a large
> smp system I think this should be better. I've lightly tested this change
> on a 4-cpu system. Comments welcome...
Any thoughts on this? Does this seem to be the right way to go?
Thanks,
-Eric
> Signed-off-by: Eric Sandeen <[email protected]>
> ---
>
> ndex: linux-2.6/fs/ext4/mballoc.c
> ===================================================================
> --- linux-2.6.orig/fs/ext4/mballoc.c 2008-08-04 15:30:30.000000000 -0500
> +++ linux-2.6/fs/ext4/mballoc.c 2008-08-13 13:48:33.224165751 -0500
> @@ -2540,17 +2540,16 @@ int ext4_mb_init(struct super_block *sb,
> sbi->s_mb_history_filter = EXT4_MB_HISTORY_DEFAULT;
> sbi->s_mb_group_prealloc = MB_DEFAULT_GROUP_PREALLOC;
>
> - i = sizeof(struct ext4_locality_group) * nr_cpu_ids;
> - sbi->s_locality_groups = kmalloc(i, GFP_KERNEL);
> + sbi->s_locality_groups = alloc_percpu(struct ext4_locality_group);
> if (sbi->s_locality_groups == NULL) {
> clear_opt(sbi->s_mount_opt, MBALLOC);
> kfree(sbi->s_mb_offsets);
> kfree(sbi->s_mb_maxs);
> return -ENOMEM;
> }
> - for (i = 0; i < nr_cpu_ids; i++) {
> + for_each_possible_cpu(i) {
> struct ext4_locality_group *lg;
> - lg = &sbi->s_locality_groups[i];
> + lg = per_cpu_ptr(sbi->s_locality_groups, i);
> mutex_init(&lg->lg_mutex);
> for (j = 0; j < PREALLOC_TB_SIZE; j++)
> INIT_LIST_HEAD(&lg->lg_prealloc_list[j]);
> @@ -2647,8 +2646,7 @@ int ext4_mb_release(struct super_block *
> atomic_read(&sbi->s_mb_discarded));
> }
>
> - kfree(sbi->s_locality_groups);
> -
> + free_percpu(sbi->s_locality_groups);
> ext4_mb_history_release(sb);
> ext4_mb_destroy_per_dev_proc(sb);
>
> @@ -4055,8 +4053,7 @@ static void ext4_mb_group_or_file(struct
> * per cpu locality group is to reduce the contention between block
> * request from multiple CPUs.
> */
> - ac->ac_lg = &sbi->s_locality_groups[get_cpu()];
> - put_cpu();
> + ac->ac_lg = per_cpu_ptr(sbi->s_locality_groups, smp_processor_id());
>
> /* we're going to use group allocation */
> ac->ac_flags |= EXT4_MB_HINT_GROUP_ALLOC;
>
> --
> To unsubscribe from this list: send the line "unsubscribe linux-ext4" in
> the body of a message to [email protected]
> More majordomo info at http://vger.kernel.org/majordomo-info.html
On Mon, Aug 18, 2008 at 10:47:40AM -0500, Eric Sandeen wrote:
> Eric Sandeen wrote:
> > lg_prealloc_list seems to cry out for a per-cpu data structure; on a large
> > smp system I think this should be better. I've lightly tested this change
> > on a 4-cpu system. Comments welcome...
>
> Any thoughts on this? Does this seem to be the right way to go?
>
Acked-by: Aneesh Kumar K.V <[email protected]>
Aneesh Kumar K.V wrote:
> On Mon, Aug 18, 2008 at 10:47:40AM -0500, Eric Sandeen wrote:
>> Eric Sandeen wrote:
>>> lg_prealloc_list seems to cry out for a per-cpu data structure; on a large
>>> smp system I think this should be better. I've lightly tested this change
>>> on a 4-cpu system. Comments welcome...
>> Any thoughts on this? Does this seem to be the right way to go?
>>
>
> Acked-by: Aneesh Kumar K.V <[email protected]>
Thanks :) I'll toss it in the patch queue.
-Eric