Re: [patch] sched: fix scheduling latencies for !PREEMPT kernels

[Nick Piggin <nickpiggin@yahoo.com.au>]

Ingo Molnar wrote:

> --- linux/mm/vmscan.c.orig	
> +++ linux/mm/vmscan.c	
> @@ -361,6 +361,8 @@ static int shrink_list(struct list_head 
>  		int may_enter_fs;
>  		int referenced;
>  
> +		cond_resched();
> +
>  		page = lru_to_page(page_list);
>  		list_del(&page->lru);
>  
> @@ -710,6 +712,7 @@ refill_inactive_zone(struct zone *zone, 
>  		reclaim_mapped = 1;
>  
>  	while (!list_empty(&l_hold)) {
> +		cond_resched();
>  		page = lru_to_page(&l_hold);
>  		list_del(&page->lru);
>  		if (page_mapped(page)) {


Could these ones go up a level? We break down scanning into 32 page
chunks, so I don't think it needs to be checked every page.

in shrink_zone(), maybe stick them:

         while (nr_active || nr_inactive) {
                 if (nr_active) {
---> here
                         sc->nr_to_scan = min(nr_active,
                                         (unsigned long)SWAP_CLUSTER_MAX);
                         nr_active -= sc->nr_to_scan;
                         refill_inactive_zone(zone, sc);
                 }

                 if (nr_inactive) {
---> and here
                         sc->nr_to_scan = min(nr_inactive,
                                         (unsigned long)SWAP_CLUSTER_MAX);
                         nr_inactive -= sc->nr_to_scan;
                         shrink_cache(zone, sc);
                         if (sc->nr_to_reclaim <= 0)
                                 break;
                 }
         }

Does that work for you?