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:   Wed, 9 Nov 2022 17:21:01 +0000
From:   Matthew Wilcox <willy@infradead.org>
To:     Daniel Golle <daniel@makrotopia.org>
Cc:     Jens Axboe <axboe@kernel.dk>,
        Miquel Raynal <miquel.raynal@bootlin.com>,
        Richard Weinberger <richard@nod.at>,
        Vignesh Raghavendra <vigneshr@ti.com>,
        Davidlohr Bueso <dave@stgolabs.net>,
        "Martin K. Petersen" <martin.petersen@oracle.com>,
        Chaitanya Kulkarni <kch@nvidia.com>,
        Ming Lei <ming.lei@redhat.com>, linux-block@vger.kernel.org,
        linux-kernel@vger.kernel.org, linux-mtd@lists.infradead.org,
        linux-efi@vger.kernel.org
Subject: Re: [PATCH v4 2/5] block: add partition parser for U-Boot uImage.FIT
Message-ID: <Y2vhfdhvw0PADeIF@casper.infradead.org>
References: <Y2rgNEQlxKgiIDdk@makrotopia.org>
 <Y2uyBYyQOC4O5iep@casper.infradead.org>
 <Y2u62z4DQVENz69L@makrotopia.org>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <Y2u62z4DQVENz69L@makrotopia.org>
Precedence: bulk

On Wed, Nov 09, 2022 at 02:36:11PM +0000, Daniel Golle wrote:
> On Wed, Nov 09, 2022 at 01:58:29PM +0000, Matthew Wilcox wrote:
> > ... actually, why can't you call read_part_sector() and avoid all of
> > this?
> 
> I've tried that before and the problem is that read_part_sector()
> returns a pointer to one sector (typically 512 bytes) of data.
> And this pointer should not be accesses beyond sector boundaries,
> right? You'd have to call read_part_sector() again for the next
> sector.
> 
> The FIT structure, however, usually exceeds the size of one sector,
> and having a continous memory area covering the structure as a whole
> is crucial for libfdt to do its job.
> 
> I could, of course, use read_part_sector() to copy all sectors
> covering the FIT structure into a buffer, but that seemed strange
> given that read_part_sector() actually used read_mapping_page()
> (and now uses read_mapping_folio()) internally and then returns a
> pointer to the offset within the page/folio. So why not read it in one
> piece in first place instead of having it first split up to sectors
> by read_part_sector() just to then having to reassemble it into a
> continous buffer again.

Are you guaranteed that it's "sufficiently" aligned on storage so
that it fits entirely within a single page?  If not, you'll have
to copy it, vmap it, or fix libfdt to handle a segmented buffer.