Re: [PATCH v2] gfp.h: GFP_THISNODE can go to other nodes if some are unpopulated

[Nishanth Aravamudan <nacc@us.ibm.com>]

On 11.06.2007 [11:29:14 -0700], Christoph Lameter wrote:
> On Mon, 11 Jun 2007, Nishanth Aravamudan wrote:
> 
> > These are the exact semantics, I expected. so I'll be happy to test/work
> > on these fixes.
> > 
> > This would also make it unnecessary to add the populated checks in
> > various places, I think, as THISNODE will mean ONLYTHISNODE (and perhaps
> > should be renamed in the series).
> 
> Here is a draft on how this could work:

<snip>

> Index: linux-2.6/mm/mempolicy.c
> ===================================================================
> --- linux-2.6.orig/mm/mempolicy.c	2007-06-11 11:13:09.000000000 -0700
> +++ linux-2.6/mm/mempolicy.c	2007-06-11 11:19:03.000000000 -0700
> @@ -1125,9 +1125,11 @@ static unsigned interleave_nodes(struct 
>  	struct task_struct *me = current;
> 
>  	nid = me->il_next;
> -	next = next_node(nid, policy->v.nodes);
> -	if (next >= MAX_NUMNODES)
> -		next = first_node(policy->v.nodes);
> +	do {
> +		next = next_node(nid, policy->v.nodes);
> +		if (next >= MAX_NUMNODES)
> +			next = first_node(policy->v.nodes);
> +	} while (!NODE_DATA(node)->present_pages);
>  	me->il_next = next;
>  	return nid;
>  }

So, I'm splitting up the populated_map patch in two, so that these bits
or the hugetlbfs bits could be put on top of having that nodemask.

*but*, if this change occurs in mempolicy.c, I think we still have a
problem, where me->il_next could be initialized in do_set_mempolicy() to
a memoryless node:

	if (new && new->policy == MPOL_INTERLEAVE)
		current->il_next = first_node(new->v.nodes);

Since we return nid in mm/mempolicy.c, we've fix the problem for
subsequent intereaves, but not the first one. So should it be:

	unsigned nid;

	if (new && new->policy == MPOL_INTERLEAVE) {
		nid = first_node(new->v.nodes);
		while (!node_populated(nid)) {
			nid = next_node(nid, new->v.nodes);
			if (nid >= MAX_NUMNODES) {
				mpol_free(current->mempolicy);
				current->mempolicy = NULL;
				mpol_set_task_struct_flag();
				return -EINVAL;
			}
		}
	}

??

Sorry if I'm way off here, just trying to get it right.

Thanks,
Nish

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>