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;
Message-ID: <7458df77-2721-71a5-893f-8247f05be765@baylibre.com>
Date:   Mon, 11 Apr 2022 09:30:39 +0200
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101
 Thunderbird/91.5.0
Subject: Re: [PATCH v2 0/2] Ensure High and Low periods of SCL are correct
Content-Language: en-US
To:     Lucas Tanure <tanure@linux.com>,
        Kevin Hilman <khilman@baylibre.com>,
        Jerome Brunet <jbrunet@baylibre.com>,
        Martin Blumenstingl <martin.blumenstingl@googlemail.com>
Cc:     linux-i2c@vger.kernel.org, linux-arm-kernel@lists.infradead.org,
        linux-amlogic@lists.infradead.org, linux-kernel@vger.kernel.org
References: <20220409164334.157664-1-tanure@linux.com>
From:   Neil Armstrong <narmstrong@baylibre.com>
Organization: Baylibre
In-Reply-To: <20220409164334.157664-1-tanure@linux.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk

Hi,

On 09/04/2022 18:43, Lucas Tanure wrote:
> The default duty cycle of 33% is less than the required
> by the I2C specs for the LOW period of the SCL clock at
> 100KHz bus speed.
> 
> So, for 100Khz or less, use 50%H/50%L duty cycle, and
> for the clock above 100Khz, use 40%H/60%L duty cycle.
> That ensures the low period of SCL is always more than
> the minimum required by the specs at any given frequency.
> 
> I did a few measures on the Khadas Vim3 board:
> 
> i2c_AO (i2c@5000):
> 
> Before the patchset, I got:
> - 100KHz: 3.338us HIGH, 6.746us LOW, 33%/67%, Freq 99KHz (Not Valid tHIGH < 4.0us)
> - 400KHz: 860ns HIGH, 1.734us LOW, 33.15%/62.85%, Freq 385.505KHz (Valid)
> - 1000KHz: 362ns HIGH, 732ns LOW, 33.09%/66.91%, Freq 914.077KHz (Valid)
> 
> With the patchset
> - 100KHz: 4.952us HIGH, 5.134us LOW, 49%/51%, Freq 99KHz (Valid)
> - 400KHz: 966ns HIGH, 1.628us LOW, 37.24%/62.76%, Freq 385.505KHz (Valid)
> - 1000KHz: 372ns HIGH, 720ns LOW, 34.07%/65.93%, Freq 915.741KHz (Valid)
> 
> i2c3 (i2c@1c000):
> 
> Before the patchset, I got:
> - 100KHz: 3.348us HIGH, 6.704us LOW, 33%/67%, Freq 99.5KHz (Not Valid tHIGH < 4.0us)
> - 400KHz: 864ns HIGH, 1.69us LOW, 33.83%/62.17%, Freq 391.543KHz (Valid)
> - 1000KHz: 360ns HIGH, 690ns LOW, 34.29%/65.71%, Freq 952.381KHz (Valid)
> 
> With the patchset
> - 100KHz: 4.958us HIGH, 5.092us LOW, 49%/51%, Freq 99KHz (Valid)
> - 400KHz: 970ns HIGH, 1.582us LOW, 38%/62%, Freq 391.85KHz (Valid)
> - 1000KHz: 370ns HIGH, 680ns LOW, 35.24%/64.76%, Freq 952.57KHz (Valid)
> 
> v2 changelog:
>   - Keep the previous calculation for Meson6
>   - Use I2C_MAX_STANDARD_MODE_FREQ
>   - move the comment before the if()
>   - use FIELD_PREP for setting div_l
>   - Drop removal of meson_i2c_data
> 
> Previous versions:
>   V1: https://lkml.org/lkml/2022/3/26/109
> 
> Lucas Tanure (2):
>    i2c: meson: Use _SHIFT and _MASK for register definitions
>    i2c: meson: Use 50% duty cycle for I2C clock
> 
>   drivers/i2c/busses/i2c-meson.c | 111 ++++++++++++++++++++++++---------
>   1 file changed, 82 insertions(+), 29 deletions(-)
> 

Thanks a lot for the timings !

I think it's ok to be applied,

Neil