Received-SPF: pass (google.com: domain of linux-kernel+bounces-183909-linux.lists.archive=gmail.com@vger.kernel.org designates 139.178.88.99 as permitted sender) client-ip=139.178.88.99;
Date: Mon, 20 May 2024 17:33:25 +0200
From: Miquel Raynal <miquel.raynal@bootlin.com>
To: Keguang Zhang <keguang.zhang@gmail.com>
Cc: Keguang Zhang via B4 Relay <devnull+keguang.zhang.gmail.com@kernel.org>,
 Richard Weinberger <richard@nod.at>, Vignesh Raghavendra <vigneshr@ti.com>,
 Rob Herring <robh@kernel.org>, Krzysztof Kozlowski
 <krzysztof.kozlowski+dt@linaro.org>, Conor Dooley <conor+dt@kernel.org>,
 linux-mtd@lists.infradead.org, linux-kernel@vger.kernel.org,
 linux-mips@vger.kernel.org, devicetree@vger.kernel.org
Subject: Re: [PATCH v7 2/3] mtd: rawnand: Enable monolithic read when
 reading subpages
Message-ID: <20240520173325.79fee6a5@xps-13>
In-Reply-To: <CAJhJPsVOx_AZmsRuZ5jy2-wJ+7-Wm+8RQAJ_LhKGLU3aFzrR7g@mail.gmail.com>
References: <20240430-loongson1-nand-v7-0-60787c314fa4@gmail.com>
	<20240430-loongson1-nand-v7-2-60787c314fa4@gmail.com>
	<20240506091748.18c120d5@xps-13>
	<CAJhJPsVOx_AZmsRuZ5jy2-wJ+7-Wm+8RQAJ_LhKGLU3aFzrR7g@mail.gmail.com>
Organization: Bootlin
Precedence: bulk
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: quoted-printable

Hi Keguang,

keguang.zhang@gmail.com wrote on Mon, 20 May 2024 18:42:30 +0800:

> On Mon, May 6, 2024 at 3:17=E2=80=AFPM Miquel Raynal <miquel.raynal@bootl=
in.com> wrote:
> >
> > Hi,
> >
> > devnull+keguang.zhang.gmail.com@kernel.org wrote on Tue, 30 Apr 2024
> > 19:11:11 +0800:
> > =20
> > > From: Keguang Zhang <keguang.zhang@gmail.com>
> > >
> > > nand_read_subpage() reads data and ECC data by two separate
> > > operations.
> > > This patch allows the NAND controllers who support
> > > monolithic page read to do subpage read by a single operation,
> > > which is more effective than nand_read_subpage(). =20
> >
> > I am a bit puzzled by this change. Usually nand_read_subpage is used
> > for optimizations (when less data than a full page must be retrieved).
> > I know it may be used in other cases (because it's easier for the core
> > in order to support a wide range of controllers). Can you please show a
> > speed test showing the results before I consider merging this patch?
> > =20
> With this patch:
> # flash_speed -c 128 -d /dev/mtd1
> scanning for bad eraseblocks
> scanned 128 eraseblocks, 0 are bad
> testing eraseblock write speed
> eraseblock write speed is 2112 KiB/s
> testing eraseblock read speed
> eraseblock read speed is 3454 KiB/s
> testing page write speed
> page write speed is 1915 KiB/s
> testing page read speed
> page read speed is 2999 KiB/s
> testing 2 page write speed
> 2 page write speed is 2000 KiB/s
> testing 2 page read speed
> 2 page read speed is 3207 KiB/s
> Testing erase speed
> erase speed is 72495 KiB/s
> Testing 2x multi-block erase speed
> 2x multi-block erase speed is 74135 KiB/s
> Testing 4x multi-block erase speed
> 4x multi-block erase speed is 74812 KiB/s
> Testing 8x multi-block erase speed
> 8x multi-block erase speed is 75502 KiB/s
> Testing 16x multi-block erase speed
> 16x multi-block erase speed is 75851 KiB/s
> Testing 32x multi-block erase speed
> 32x multi-block erase speed is 75851 KiB/s
> Testing 64x multi-block erase speed
> 64x multi-block erase speed is 76204 KiB/s
> finished
>=20
> Without this patch:
> # flash_speed -c 128 -d /dev/mtd1
> scanning for bad eraseblocks
> scanned 128 eraseblocks, 0 are bad
> testing eraseblock write speed
> eraseblock write speed is 2074 KiB/s
> testing eraseblock read speed
> eraseblock read speed is 2895 KiB/s
> testing page write speed
> page write speed is 998 KiB/s
> testing page read speed
> page read speed is 1499 KiB/s
> testing 2 page write speed
> 2 page write speed is 1002 KiB/s
> testing 2 page read speed
> 2 page read speed is 1554 KiB/s
> Testing erase speed
> erase speed is 76560 KiB/s
> Testing 2x multi-block erase speed
> 2x multi-block erase speed is 74019 KiB/s
> Testing 4x multi-block erase speed
> 4x multi-block erase speed is 74769 KiB/s
> Testing 8x multi-block erase speed
> 8x multi-block erase speed is 75149 KiB/s
> Testing 16x multi-block erase speed
> 16x multi-block erase speed is 75921 KiB/s
> Testing 32x multi-block erase speed
> 32x multi-block erase speed is 75921 KiB/s
> Testing 64x multi-block erase speed
> 64x multi-block erase speed is 75921 KiB/s
> finished
>=20
> The throughput of the former is twice that of the latter.

And what is your NAND controller driver?

subpage reads are used when you only want to read a subset of a NAND
page.

Otherwise the core may use the RNDOUT command to change the pointer in
the chip's SRAM to read from a different location, but I don't see what
is impacting so much, unless if the driver implementation is really
sub-optimized.

Thanks,
Miqu=C3=A8l