Re: + mm-balance_dirty_pages-reduce-calls-to-global_page_state-to-reduce-c ache-references.patch added to -mm tree

[Wu Fengguang <fengguang.wu@intel.com>]

On Thu, Sep 03, 2009 at 10:22:23AM +0800, Wu Fengguang wrote:
> So I suggested just remove clip_bdi_dirty_limit(). To be sure, could
> run with the following patch and check if big numbers are showed.
> 

Simple tests show that clip_bdi_dirty_limit() is no-op with one single
disk and 1-2 dirtier(s):

[...]
[  720.630917] clip_bdi_dirty_limit 0
[  720.753865] clip_bdi_dirty_limit 0
[  720.915198] clip_bdi_dirty_limit 0
[  721.149511] clip_bdi_dirty_limit 0
[...]

Thanks,
Fengguang

> --- linux-mm.orig/mm/page-writeback.c	2009-09-02 17:16:51.000000000 +0800
> +++ linux-mm/mm/page-writeback.c	2009-09-03 10:19:00.000000000 +0800
> @@ -258,6 +258,7 @@ static void clip_bdi_dirty_limit(struct 
>  		unsigned long dirty, unsigned long *pbdi_dirty)
>  {
>  	unsigned long avail_dirty;
> +	unsigned long delta;
>  
>  	avail_dirty = global_page_state(NR_FILE_DIRTY) +
>  		 global_page_state(NR_WRITEBACK) +
> @@ -272,6 +273,11 @@ static void clip_bdi_dirty_limit(struct 
>  	avail_dirty += bdi_stat(bdi, BDI_RECLAIMABLE) +
>  		bdi_stat(bdi, BDI_WRITEBACK);
>  
> +	delta = *pbdi_dirty - min(*pbdi_dirty, avail_dirty);
> +	delta *= 1024;
> +	delta /= *pbdi_dirty + 1;
> +	printk("clip_bdi_dirty_limit %lu\n", delta);
> +
>  	*pbdi_dirty = min(*pbdi_dirty, avail_dirty);
>  }
>