Re: [PATCH v3 04/11] i2c: riic: Enable runtime PM autosuspend support

[claudiu beznea <claudiu.beznea@tuxon.dev>]

Hi, Biju,

On 12.07.2024 10:15, Biju Das wrote:
> Hi Claudiu,
> 
>> -----Original Message-----
>> From: Claudiu <claudiu.beznea@tuxon.dev>
>> Sent: Thursday, July 11, 2024 12:52 PM
>> Subject: [PATCH v3 04/11] i2c: riic: Enable runtime PM autosuspend support
>>
>> From: Claudiu Beznea <claudiu.beznea.uj@bp.renesas.com>
>>
>> Enable runtime PM autosuspend support for the RIIC driver. With this, in case there are consecutive
>> xfer requests the device wouldn't be runtime enabled/disabled after each consecutive xfer but after
>> the the delay configured by user. With this, we can avoid touching hardware registers involved in
>> runtime PM suspend/resume saving in this way some cycles. The default chosen autosuspend delay is
>> zero to keep the previous driver behavior.
> 
> On the other hand, you are saving power. Currently the driver is highly optimized for 
> Power usage.
> 
> Before transfer turn on the clock
> After transfer turn off the clock, this is the optimal power usage correspond to suspend delay.
> 
> By adding suspend delay, you are consuming power corresponding to
> that delay.

The default delay is zero, see the following diff in this patch:

@@ -479,6 +481,8 @@ static int riic_i2c_probe(struct platform_device *pdev)

 	i2c_parse_fw_timings(dev, &i2c_t, true);

+	pm_runtime_set_autosuspend_delay(dev, 0);

With this, the previous behavior of the driver is preserved, nothing is
changed.

Having it this way preserves the previous behavior and give the the user
the possibility to adjust the autosuspend delay (e.g., though sysfs) to
avoid enable/disable the power to this IP in scenarios where this might be
a bottleneck.

Thank you,
Claudiu Beznea

> 
> Cheers,
> Biju