Re: [PATCH v5 20/37] mm: fix non-compound multi-order memory accounting in __free_pages

[Matthew Wilcox <willy@infradead.org>]

On Wed, Mar 06, 2024 at 10:24:18AM -0800, Suren Baghdasaryan wrote:
> When a non-compound multi-order page is freed, it is possible that a
> speculative reference keeps the page pinned. In this case we free all
> pages except for the first page, which will be freed later by the last
> put_page(). However put_page() ignores the order of the page being freed,
> treating it as a 0-order page. This creates a memory accounting imbalance
> because the pages freed in __free_pages() do not have their own alloc_tag
> and their memory was accounted to the first page. To fix this the first
> page should adjust its allocation size counter when "tail" pages are freed.

It's not "ignored".  It's not available!

Better wording:

However the page passed to put_page() is indisinguishable from an
order-0 page, so it cannot do the accounting, just as it cannot free
the subsequent pages.  Do the accounting here, where we free the pages.

(I'm sure further improvements are possible)

> +static inline void pgalloc_tag_sub_bytes(struct alloc_tag *tag, unsigned int order)
> +{
> +	if (mem_alloc_profiling_enabled() && tag)
> +		this_cpu_sub(tag->counters->bytes, PAGE_SIZE << order);
> +}

This is a terribly named function.  And it's not even good for what we
want to use it for.

static inline void pgalloc_tag_sub_pages(struct alloc_tag *tag, unsigned int nr)
{
	if (mem_alloc_profiling_enabled() && tag)
		this_cpu_sub(tag->counters->bytes, PAGE_SIZE * nr);
}

> +++ b/mm/page_alloc.c
> @@ -4697,12 +4697,21 @@ void __free_pages(struct page *page, unsigned int order)
>  {
>  	/* get PageHead before we drop reference */
>  	int head = PageHead(page);
> +	struct alloc_tag *tag = pgalloc_tag_get(page);
>  
>  	if (put_page_testzero(page))
>  		free_the_page(page, order);
>  	else if (!head)
> -		while (order-- > 0)
> +		while (order-- > 0) {
>  			free_the_page(page + (1 << order), order);
> +			/*
> +			 * non-compound multi-order page accounts all allocations
> +			 * to the first page (just like compound one), therefore
> +			 * we need to adjust the allocation size of the first
> +			 * page as its order is ignored when put_page() frees it.
> +			 */
> +			pgalloc_tag_sub_bytes(tag, order);

-	else if (!head
+	else if (!head) {
+		pgalloc_tag_sub_pages(1 << order - 1);
		while (order-- > 0)
			free_the_page(page + (1 << order), order);
+	}

It doesn't need a comment, it's obvious what you're doing.