Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
MIME-Version: 1.0
References: <20200208084022.193231-1-gch981213@gmail.com> <1581323455.2213.6.camel@mtksdaap41>
In-Reply-To: <1581323455.2213.6.camel@mtksdaap41>
From:   Chuanhong Guo <gch981213@gmail.com>
Date:   Tue, 11 Feb 2020 14:55:17 +0800
Message-ID: <CAJsYDVLzwWfT24NGDJMJShwoG0Qrq06mLqamHbH0xedmMLdrAQ@mail.gmail.com>
Subject: Re: [PATCH v2] mtd: mtk-quadspi: add support for DMA reading
To:     Yingjoe Chen <yingjoe.chen@mediatek.com>
Cc:     linux-mtd@lists.infradead.org,
        Vignesh Raghavendra <vigneshr@ti.com>,
        Tudor Ambarus <tudor.ambarus@microchip.com>,
        Richard Weinberger <richard@nod.at>,
        open list <linux-kernel@vger.kernel.org>,
        linux-mediatek@lists.infradead.org,
        Miquel Raynal <miquel.raynal@bootlin.com>,
        Matthias Brugger <matthias.bgg@gmail.com>,
        linux-arm-kernel@lists.infradead.org
Content-Type: text/plain; charset="UTF-8"
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

Hi!

On Mon, Feb 10, 2020 at 4:31 PM Yingjoe Chen <yingjoe.chen@mediatek.com> wrote:
> > DMA busy checking is implemented with readl_poll_timeout because
> > I don't have access to IRQ-related docs. The speed increment comes
> > from those saved cmd+addr+dummy clocks.
>
> Hi Chuanhong,
>
> Thanks for your patch, I'm checking with Guochun to see if we could
> release IRQ related information to you.

Thanks for the info.
I'd like to keep using polling mode in this patch for easier reviewing.
It's already a pretty lengthy patch now. I may implement IRQ support
in future patches.

>
> > This controller requires that DMA source/destination address and
> > reading length should be 16-byte aligned. We use a bounce buffer if
> > one of them is not aligned, read more than what we need, and copy
> > data from corresponding buffer offset.
>
> I've checked with our HW guys. The limitation is on DRAM only.
> So for read we should check buffer and length to make sure it is
> aligned, but don't need to check from.

My previous test on mt7629 shows that from address also needs to
be aligned. e.g. If I perform a DMA read from 0x2 I actually got data
starting from 0x0 instead.

Regards,
Chuanhong Guo