Re: [PATCH 2/2] mm/mempolicy: Skip extra call to __alloc_pages_bulk in weighted interleave

[Joshua Hahn <joshua.hahnjy@gmail.com>]

On Mon, 30 Jun 2025 22:05:48 +0200 Kees Bakker <kees@ijzerbout.nl> wrote:

> >   mm/mempolicy.c | 39 ++++++++++++++++++++-------------------
> >   1 file changed, 20 insertions(+), 19 deletions(-)
> >
> > diff --git a/mm/mempolicy.c b/mm/mempolicy.c
> > index 78ad74a0e249..0d693f96cf66 100644
> > --- a/mm/mempolicy.c
> > +++ b/mm/mempolicy.c
> > @@ -2569,7 +2569,7 @@ static unsigned long alloc_pages_bulk_weighted_interleave(gfp_t gfp,
> >   	unsigned long node_pages, delta;
> >   	u8 *weights, weight;
> >   	unsigned int weight_total = 0;
> > -	unsigned long rem_pages = nr_pages;
> > +	unsigned long rem_pages = nr_pages, carryover = 0;
> >   	nodemask_t nodes;
> >   	int nnodes, node;
> >   	int resume_node = MAX_NUMNODES - 1;
> > @@ -2594,18 +2594,12 @@ static unsigned long alloc_pages_bulk_weighted_interleave(gfp_t gfp,
> >   	node = me->il_prev;
> >   	weight = me->il_weight;
> >   	if (weight && node_isset(node, nodes)) {
> > -		node_pages = min(rem_pages, weight);
> > -		nr_allocated = __alloc_pages_bulk(gfp, node, NULL, node_pages,
> > -						  page_array);
> > -		page_array += nr_allocated;
> > -		total_allocated += nr_allocated;
> > -		/* if that's all the pages, no need to interleave */
> >   		if (rem_pages <= weight) {
> > -			me->il_weight -= rem_pages;
> > -			return total_allocated;
> > +			node_pages = rem_pages;
> > +			me->il_weight -= node_pages;
> > +			goto allocate;

Hello Kees,

Thank you for reviewing my code!

> This is a goto into the middle of a for-loop.
> What do you think is going to happen at the end of that loop?
> 
> I think (only tested with a small C program) it will go to the start of
> the loop, do the i++, check i<nnodes, and possibly do the loop again.
> Variable i is uninitialized at that point. In the loop it hits several
> uninitialized variables.

From what I can see from my code, I think the only the goto statement leads
to a second iteration of the for loop is if allocation fails.
But otherwise, it should be ok since we always hit

if (total_allocated == nr_pages)
	break;

within the loop. For the branch that takes the goto, we set
node_pages = rem_pages, then jump to the label and allocate.
So nr_allocated = node_pages, and total_allocated = 0 + nr_allocated
so total_allocated = node_pages

total_allocated == node_pages == rem_pages == nr_pages, so we will break. Phew!

To cover the case where allocation fails, I think we should be breaking
anyways, so I can definitely add a new check for this.

> Even if this is legal C code, it is pretty obscure.

I agree that it not very clean. I did this to reduce the amount of repeated
code there is. Even if this code works, it could definitely be written
better to make it more readable and maintainable. As I noted in my second
response to Gregory, I'm not planning on pursuing this version anymore,
so if I decide to send a second version, I'll keep this in mind.

Thank you again for taking the time to review this, and also testing it on
your end! I hope you have a great day : -)
Joshua

Sent using hkml (https://github.com/sjp38/hackermail)