Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Date:   Fri, 13 Jan 2023 14:25:37 +0000
From:   Matthew Wilcox <willy@infradead.org>
To:     Kefeng Wang <wangkefeng.wang@huawei.com>
Cc:     akpm@linux-foundation.org, linux-mm@kvack.org,
        linux-kernel@vger.kernel.org
Subject: Re: [PATCH -next 1/7] mm: huge_memory: make
 __do_huge_pmd_anonymous_page() to take a folio
Message-ID: <Y8Fp4UBvO/i8/PYf@casper.infradead.org>
References: <20230112083006.163393-1-wangkefeng.wang@huawei.com>
 <20230112083006.163393-2-wangkefeng.wang@huawei.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20230112083006.163393-2-wangkefeng.wang@huawei.com>
Precedence: bulk

On Thu, Jan 12, 2023 at 04:30:00PM +0800, Kefeng Wang wrote:
> Let's __do_huge_pmd_anonymous_page() take a folio and convert related
> functions to use folios.

No, this is actively wrong!  Andrew, please drop this patch.

If we want to support folio sizes larger than PMD size (and I think we
do), we need to be able to specify precisely which page in the folio
is to be stored at this PTE.  The *interface* must remain struct page.
We can convert from page to folio within the function, but we *MUST NOT*
go the other way.

>  static vm_fault_t __do_huge_pmd_anonymous_page(struct vm_fault *vmf,
> -			struct page *page, gfp_t gfp)
> +			struct folio *folio, gfp_t gfp)
>  {
>  	struct vm_area_struct *vma = vmf->vma;
> +	struct page *page = &folio->page;

... ie this is bad and wrong.

> @@ -834,7 +835,7 @@ vm_fault_t do_huge_pmd_anonymous_page(struct vm_fault *vmf)
>  		count_vm_event(THP_FAULT_FALLBACK);
>  		return VM_FAULT_FALLBACK;
>  	}
> -	return __do_huge_pmd_anonymous_page(vmf, &folio->page, gfp);
> +	return __do_huge_pmd_anonymous_page(vmf, folio, gfp);
>  }
>  
>  static void insert_pfn_pmd(struct vm_area_struct *vma, unsigned long addr,

A reasonable person might ask "But Matthew, you allocated a folio here,
then you're converting back to a struct page to call
__do_huge_pmd_anonymous_page() so isn't this a sensible patch?"

And I would say "still no".  This is a question of interfaces, and
even though __do_huge_pmd_anonymous_page() is static and has precisely
one caller today that always allocates a folio of precisely PMD size,
I suspect it will need to be more visible in the future and the
conversion of the interface from page to folio misleads people.