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;
MIME-Version: 1.0
References: <20221205105931.410686-1-vadym.kochan@plvision.eu>
 <20221205105931.410686-4-vadym.kochan@plvision.eu> <CACRpkdaXQqrCEqu9HTMZMMWbnkQxXEmoJNtiH-HENZWeDqjqTw@mail.gmail.com>
 <BN9PR18MB4251300124E30A16B3F5C521DB1D9@BN9PR18MB4251.namprd18.prod.outlook.com>
In-Reply-To: <BN9PR18MB4251300124E30A16B3F5C521DB1D9@BN9PR18MB4251.namprd18.prod.outlook.com>
From:   Linus Walleij <linus.walleij@linaro.org>
Date:   Thu, 8 Dec 2022 22:35:16 +0100
Message-ID: <CACRpkdayw4JPY_3HFvgDpfPaDwOZHKKmyaahpEbKR6DaO_VJuA@mail.gmail.com>
Subject: Re: [EXT] Re: [PATCH v3 3/3] mmc: xenon: Fix 2G limitation on AC5 SoC
To:     Elad Nachman <enachman@marvell.com>
Cc:     Vadym Kochan <vadym.kochan@plvision.eu>,
        Hu Ziji <huziji@marvell.com>,
        Ulf Hansson <ulf.hansson@linaro.org>,
        Rob Herring <robh+dt@kernel.org>,
        Krzysztof Kozlowski <krzysztof.kozlowski+dt@linaro.org>,
        Adrian Hunter <adrian.hunter@intel.com>,
        "linux-mmc@vger.kernel.org" <linux-mmc@vger.kernel.org>,
        "devicetree@vger.kernel.org" <devicetree@vger.kernel.org>,
        "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
        Chris Packham <chris.packham@alliedtelesis.co.nz>
Content-Type: text/plain; charset="UTF-8"
Precedence: bulk

Hi Elad,

I get it, I think. I was a bit confused by the 3G/4G terminology.

On Thu, Dec 8, 2022 at 11:20 AM Elad Nachman <enachman@marvell.com> wrote:

> The lower 31-bits of the address placed in the ADMA is passed through the interconnect, and remapped to the base of the DDR.
>
> Hence only addressing of the lower 2GB of the DDR memory is supported for eMMC in this device family (AC5/X).
>
> So the quirk needs to kick in above 2GB of physical memory accessed from the base of the DDR.

How "clever" to skip bit 32. This should be in the patch description.

> This is why a quirk which only kicks in above 4GB is not sufficient.

So the author of the patch should create a new quirk that kicks in above 2GB,
devised to be similar in style of the 4GB quirk we already have.

> Furthermore, SDHCI_QUIRK_32BIT_DMA_ADDR is checked in sdhci_prepare_data() as a way to
> disable DMA when the offset of the scatter-list DMA address is not 32-bit aligned. If the address is
> aligned, this quirk does not disable the DMA, and will not solve our problem.

That's right.

Let's just create a new quirk:

SDHCI_QUIRK_31BIT_DMA_ROOF

Define the semantics such that this will allow DMA for buffers that are below
the 31st bit, but does not have the semantics to limit scatter-gather buffers to
be 32-bit aligned.

Yours,
Linus Walleij