Re: [PATCH v2 5/7] mm: task_mmu: use a folio in smaps_account()

From: Matthew Wilcox <willy@infradead.org>
To: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: Andrew Morton <akpm@linux-foundation.org>,
	linux-kernel@vger.kernel.org, linux-mm@kvack.org,
	David Hildenbrand <david@redhat.com>,
	Gregory Price <gregory.price@memverge.com>
Subject: Re: [PATCH v2 5/7] mm: task_mmu: use a folio in smaps_account()
Date: Fri, 10 Nov 2023 18:29:29 +0000	[thread overview]
Message-ID: <ZU52ifwZcNhmz0rA@casper.infradead.org> (raw)
In-Reply-To: <20231110033324.2455523-6-wangkefeng.wang@huawei.com>

On Fri, Nov 10, 2023 at 11:33:22AM +0800, Kefeng Wang wrote:
> Replace seven implicit calls to compound_head() with one page_folio().

You're so focused on trying to accomplish your goal (killing off the
page_idle and page_young functions) that you're doing a poor job thinkig
about the conversions you're doing along the way.

> +++ b/fs/proc/task_mmu.c
> @@ -445,23 +445,25 @@ static void smaps_account(struct mem_size_stats *mss, struct page *page,
>  {
>  	int i, nr = compound ? compound_nr(page) : 1;
>  	unsigned long size = nr * PAGE_SIZE;
> +	struct folio *folio = page_folio(page);

Stop right here.  The use of compound_nr() should give you pause.

After looking at how smaps_account() is used, it seems to me that the
two callers should each pass in PAGE_SIZE or PMD_SIZE instead of doing
this calculation.

More generally, step back from this series.  Do a good job of
transforming fs/proc/task_mmu.c to use folios.  Once you've done
that, you can approach the young/idle work again.