23.01.2019 2:26, Sowjanya Komatineni пишет:
>
>
>>>>> + if (i2c_dev->msg_err == I2C_ERR_ARBITRATION_LOST) {
>>>>> + if (!i2c_dev->is_multimaster_mode)
>>>>> + return tegra_i2c_issue_bus_clear(i2c_dev);
>>>>> + return -EAGAIN;
>>>>
>>>> This changes the returned errno from -EIO to -EAGAIN for the
>>>> supports_bus_clear=false case, is it okay and intentional?
>>>>
>>>
>>> Yes EAGAIN is intentional to allow for transfer retry.
>>> During single master mode, ARBITRATION LOST notification happens when
>>> 1. I2C Master sees the bus is occupied by some other device when a
>>> transfer is initiated 2. I2C Master lost the bus during arbitration
>>> incase if slave device pulls SDA line low continuously for some
>>> unknown reason If arbitration lost is due to cause 1, retry helps to
>>> continue with transfer once bus is released by the slave and it just
>>> added delay in communication due to bus release delay by slave. In
>>> case of 2nd cause, retry never succeeds in cases where bus clear is
>>> not supported.
>>
>> It's unclear whether the "never succeeds retry" may fail with the EAGAIN, causing an endless retry-loop. Could you please clarify this moment?
>
> during master transmit mode, on arbitration lost and if master doesn’t support bus clear to recover then transfer will return EAGAIN.
> I2c core base driver performs retries if return code from i2c_transfer is EAGAIN up to specified retries in i2c adapter and returns the ret code from the last retry.
> Retry is not endless as i2c core base performs retry only up to specified adapter retries.
> Following return code from documentation
> https://www.kernel.org/doc/Documentation/i2c/fault-codes
>
>
Could you please point at the code that sets the number of retries for the Tegra's I2C? Looks it is always 0 and hence EAGAIN won't do anything useful.
