Re: [PATCH v4 5/8] hugetlb: call update_and_free_page without hugetlb_lock

[Oscar Salvador <osalvador@suse.de>]

On Mon, Apr 05, 2021 at 04:00:40PM -0700, Mike Kravetz wrote:
> With the introduction of remove_hugetlb_page(), there is no need for
> update_and_free_page to hold the hugetlb lock.  Change all callers to
> drop the lock before calling.
> 
> With additional code modifications, this will allow loops which decrease
> the huge page pool to drop the hugetlb_lock with each page to reduce
> long hold times.
> 
> The ugly unlock/lock cycle in free_pool_huge_page will be removed in
> a subsequent patch which restructures free_pool_huge_page.
> 
> Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>

Without looking too close at the changes made to alloc_and_dissolve_huge_page():

Reviewed-by: Oscar Salvador <osalvador@suse.de>

One question below:

> @@ -2671,22 +2682,34 @@ static void try_to_free_low(struct hstate *h, unsigned long count,
>  						nodemask_t *nodes_allowed)
>  {
>  	int i;
> +	struct page *page, *next;
> +	LIST_HEAD(page_list);
>  
>  	if (hstate_is_gigantic(h))
>  		return;
>  
> +	/*
> +	 * Collect pages to be freed on a list, and free after dropping lock
> +	 */
>  	for_each_node_mask(i, *nodes_allowed) {
> -		struct page *page, *next;
>  		struct list_head *freel = &h->hugepage_freelists[i];
>  		list_for_each_entry_safe(page, next, freel, lru) {
>  			if (count >= h->nr_huge_pages)
> -				return;
> +				goto out;
>  			if (PageHighMem(page))
>  				continue;
>  			remove_hugetlb_page(h, page, false);
> -			update_and_free_page(h, page);
> +			list_add(&page->lru, &page_list);
>  		}
>  	}
> +
> +out:
> +	spin_unlock(&hugetlb_lock);
> +	list_for_each_entry_safe(page, next, &page_list, lru) {
> +		update_and_free_page(h, page);
> +		cond_resched();
> +	}
> +	spin_lock(&hugetlb_lock);

Can we get here with an empty list? Maybe if someone raced with us manipulating
nr_huge_pages? AFAICS, this gets called under the lock, and the adjusting in
remove_hugetlb_page() gets also done under the lock, so I guess this is not
possible to happen.
The reason I am asking is whether we want to check for the list to be empty before
we do the unacquire/acquire lock dancing.


-- 
Oscar Salvador
SUSE L3