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;
Subject: Re: [PATCH 3/3] mmc: tegra: prevent ACMD23 on Tegra 3
To:     Stefan Agner <stefan@agner.ch>
Cc:     adrian.hunter@intel.com, ulf.hansson@linaro.org,
        thierry.reding@gmail.com, jonathanh@nvidia.com,
        marcel.ziswiler@toradex.com, linux-mmc@vger.kernel.org,
        linux-tegra@vger.kernel.org, linux-kernel@vger.kernel.org
References: <20180712073904.4705-1-stefan@agner.ch>
 <20180712073904.4705-3-stefan@agner.ch>
 <430c718d-9ede-d47a-44c0-b8e47e297720@gmail.com>
 <1553c44e99b96a813af129d1c4169222@agner.ch>
From:   Peter Geis <pgwipeout@gmail.com>
Message-ID: <553f9ed6-8f7c-775b-d731-fd9732f74b07@gmail.com>
Date:   Thu, 26 Jul 2018 11:12:23 -0400
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:52.0) Gecko/20100101
 Thunderbird/52.9.1
MIME-Version: 1.0
In-Reply-To: <1553c44e99b96a813af129d1c4169222@agner.ch>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 07/26/2018 10:47 AM, Stefan Agner wrote:
> On 26.07.2018 15:56, Peter Geis wrote:
>> On 07/12/2018 03:39 AM, Stefan Agner wrote:
>>> It seems that SD3.0 advertisement needs to be set for higher eMMC
>>> speed modes (namely DDR52) as well. The TRM states that the SD3.0
>>> advertisement bit should be set for all controller instances, even
>>> for those not supporting UHS-I mode...
>>>
>>> When specifying vqmmc-supply as a fixed 1.8V regulator on a Tegra
>>> SD/MMC instance which is connected to a eMMC device, the stack
>>> enables SD3.0. However, enabling it has consequences: If SDHCI 3.0
>>> support is advertised the stack enables Auto-CMD23. Unfortunately
>>> Auto-CMD23 seems not to work well with Tegra 3 currently. It leads
>>> to regular warnings:
>>>     mmc2: Got command interrupt 0x00010000 even though no command operation was in progress.
>>>
>>> It is not entirely clear why those errors happens. It seems that
>>> a Linux 3.1 based downstream kernel which has Auto-CMD23 support
>>> does not show those warnings.
>>>
>>> Use quirk SDHCI_QUIRK2_ACMD23_BROKEN to prevent Auto-CMD23 being
>>> used for now. With this the eMMC works stable on high-speed mode
>>> while still announcing SD3.0.
>>>
>>> This allows to use mmc-ddr-1_8v to enables DDR52 mode. In DDR52
>>> mode read speed improves from about 42MiB/s to 72MiB/s on an
>>> Apalis T30.
>>>
>>> Signed-off-by: Stefan Agner <stefan@agner.ch>
>>> ---
>>>    drivers/mmc/host/sdhci-tegra.c | 10 +++++++++-
>>>    1 file changed, 9 insertions(+), 1 deletion(-)
>>>
>>> diff --git a/drivers/mmc/host/sdhci-tegra.c b/drivers/mmc/host/sdhci-tegra.c
>>> index 888a1ad511db..11c0b2069c7c 100644
>>> --- a/drivers/mmc/host/sdhci-tegra.c
>>> +++ b/drivers/mmc/host/sdhci-tegra.c
>>> @@ -336,7 +336,15 @@ static const struct sdhci_pltfm_data sdhci_tegra30_pdata = {
>>>    		  SDHCI_QUIRK_BROKEN_ADMA_ZEROLEN_DESC |
>>>    		  SDHCI_QUIRK_CAP_CLOCK_BASE_BROKEN,
>>>    	.quirks2 = SDHCI_QUIRK2_PRESET_VALUE_BROKEN |
>>> -		   SDHCI_QUIRK2_BROKEN_HS200,
>>> +		   SDHCI_QUIRK2_BROKEN_HS200 |
>>> +		   /*
>>> +		    * Auto-CMD23 leads to "Got command interrupt 0x00010000 even
>>> +		    * though no command operation was in progress."
>>> +		    *
>>> +		    * The exact reason is unknown, as the same hardware seems
>>> +		    * to support Auto CMD23 on a downstream 3.1 kernel.
>>> +		    */
>>> +		   SDHCI_QUIRK2_ACMD23_BROKEN,
>>>    	.ops  = &tegra_sdhci_ops,
>>>    };
>>>
>>
>> I finally got around to testing this on the Ouya (Tegra 3).
> 
> Thanks for testing!
> 
>>
>> I found that the "Got command interrupt 0x00010000 even though no
>> command operation was in progress." error occurred when the interface
>> is unstable.
>> I've had a lot of problems with sdmmc4 stability on the Ouya above 34
>> Mhz, probably due to the fact that they are using the internal cmd and
>> clock pull-up resistors, against the TRM's instruction.
>> At 39Mhz, I saw the error this patch corrects.
>> With the patch, the error went away, but the interface is still
>> unstable under load.
> 
> How does this instability manifest exactly?
> 

At the very edge of stability, you see write errors under heavy load.
As clock rate increases, the write errors occur more frequently.
At a certain point, you start getting read errors.
Following that you get constant io errors during card probing.
Eventually the emmc will fail to initialize, with errors 87 or 110.

I've been tweaking the pull up/down values to try and improve the 
stability, but without access to anything but the TRM it's a lot of 
trial and error.

>>
>> Lowering down to 32Mhz, without the patch there are no errors.
> 
> So the patch does not make it less stable right?
> 

No, it did not affect stability.
Although I'd conduct some performance testing to check for degradation.
Of course I'm nowhere near the limits of the controller, so it is 
doubtful I'd see a hit.

> --
> Stefan
>