On Sat, Jun 19, 2010 at 04:23:12AM -0400, Jean Delvare wrote:
> On Fri, 18 Jun 2010 13:56:53 -0700, Guenter Roeck wrote:
> > [...]
> > > > + /*
> > > > + * Algorithm for reading ADC, per SMM665 datasheet
> > > > + *
> > > > + * {[S][addr][W][Ack]} {[offset][Ack]} {[S][addr][R][Nack]}
> > > > + * [wait 70 uS]
> > > > + * {[S][addr][R][Ack]} {[datahi][Ack]} {[datalo][Ack][P]}
> > > > + *
> > > > + * To implement the first part of this exchange,
> > > > + * do a full read transaction and expect a failure/Nack.
> > > > + * This sets up the address pointer on the SMM665
> > > > + * and starts the ADC conversion.
> > > > + * Then do a two-byte read transaction.
> > > > + */
> > > Is there no better way of handling this? There are protocol mangling hacks
> > > to tell the i2c core to ignore a NAKs under some circumstances.
>
> This is only available on raw I2C messages, not on the higher-level
> i2c_smbus_*() API. And not all bus drivers support it. And that's not
> what is needed here anyway: ignoring the nack means we would continue
> reading from the chip, while it really doesn't want to talk to us at
> that time.
>
> > > > + rv = i2c_smbus_read_byte_data(client, adc << 3);
> > > > + if (rv >= 0) {
>
> You should check for -ENXIO explicitly. According to
> Documentation/i2c/fault-codes, this is the value bus drivers should
> return on missing Ack.
>
Ok, I'll do that. I'll submit a separate patch for the sibyte
i2c bus driver since it returns -1 - which is wrong anyway.
> > > > + /* No error, something is wrong. Retry. */
> > > > + rv = -1;
> > > > + continue;
> > > > + }
> >
> > I looked through the core i2c code, but did not find anything I can
> > use.
> >
> > Problem is that per smm665 specification, the first NACK is expected. So
> > we do not just want to ignore this NACK, we want to actively check if
> > the command "failed" as expected, and report an error if it did _not_
> > fail.
>
> Do you really have to trigger the Nack for the ADC conversion to start?
> Can't you just use i2c_smbus_read_byte() (no _data) for the first part
> of the transaction?
>
Unfortunately, yes. It is the only command that works.
Guenter
_______________________________________________
lm-sensors mailing list
lm-sensors@xxxxxxxxxxxxxx
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors
