Re: [PATCH v3 3/4] leds: pca955x: Optimize probe led selection

[Eddie James <eajames@linux.ibm.com>]

On 5/4/22 12:24, Pavel Machek wrote:
> Hi!
>
>> Previously, the probe function might do up to 32 reads and writes
>> to the same 4 registers to program the led selection. Reduce this to
>> a maximum of 8 operations by accumulating the changes to the led
>> selection and comparing with the previous value to write the
>> selection if different.
> We have regmap APIs. You are free to use them if you really care about
> those few reads. Reimplementing them by hand is not acceptable. How big is
> the seedup here?


Hi,

To be honest this was not about speed, it's about a problem I've 
observed where many operations in quick succession caused the PCA chip 
to NAK the i2c transfer. I'm not sure of the root cause yet. So reducing 
the number of operations helped. We are also probably going to carry a 
throttling patch, unless I can prove the PCA chip itself is at fault and 
needs slower transfers.

I can try the regmap API and see if it helps though.

Thanks,

Eddie


>
> Best regards,
> 								Pavel
>
>> @@ -554,6 +556,15 @@ static int pca955x_probe(struct i2c_client *client)
>>   	init_data.devname_mandatory = false;
>>   	init_data.devicename = "pca955x";
>>   
>> +	nls = pca955x_num_led_regs(chip->bits);
>> +	for (i = 0; i < nls; ++i) {
>> +		err = pca955x_read_ls(pca955x, i, &ls1[i]);
>> +		if (err)
>> +			return err;
>> +
>> +		ls2[i] = ls1[i];
>> +	}
>> +
>>   	for (i = 0; i < chip->bits; i++) {
>>   		pca955x_led = &pca955x->leds[i];
>>   		pca955x_led->led_num = i;
>> @@ -624,6 +634,14 @@ static int pca955x_probe(struct i2c_client *client)
>>   		}
>>   	}
>>   
>> +	for (i = 0; i < nls; ++i) {
>> +		if (ls1[i] != ls2[i]) {
>> +			err = pca955x_write_ls(pca955x, i, ls2[i]);
>> +			if (err)
>> +				return err;
>> +		}
>> +	}
>> +
>>   	/* PWM0 is used for half brightness or 50% duty cycle */
>>   	err = pca955x_write_pwm(pca955x, 0, 255 - LED_HALF);
>>   	if (err)