Re: [PATCH 1/2] hwmon: (pmbus/isl68137) add debugfs config and black box endpoints

[Guenter Roeck]

On Tue, Apr 21, 2020 at 04:49:17PM -0500, Grant Peltier wrote:
> On Tue, Apr 21, 2020 at 11:58:51AM -0700, Guenter Roeck wrote:
> > Normally this is emulated for such controllers. I don't recall seeing
> > such a need before. The code below duplicates similar code in
> > i2c_smbus_xfer_emulated(), which is much more sophisticated.
> > Are you sure this is needed ? Can you point me to an affected
> > controller ?
> > 
> > > +static s32 raa_smbus_read40(const struct i2c_client *client, u8 command,
> > > +			    unsigned char *data)
> > > +{
> > > +	int status;
> > > +	unsigned char msgbuf[1];
> > > +	struct i2c_msg msg[2] = {
> > > +		{
> > > +			.addr = client->addr,
> > > +			.flags = client->flags,
> > > +			.len = 1,
> > > +			.buf = msgbuf,
> > > +		},
> > > +		{
> > > +			.addr = client->addr,
> > > +			.flags = client->flags | I2C_M_RD,
> > > +			.len = 5,
> > > +			.buf = data,
> > > +		},
> > > +	};
> > > +
> > > +	msgbuf[0] = command;
> > > +	status = i2c_transfer(client->adapter, msg, 2);
> > > +	if (status != 2)
> > > +		return status;
> > 
> > i2c_transfer() can return 1 if only one of the two messages was sent.
> > 
> > > +	return 0;
> > > +}
> I have been using BCM2835 for most of my testing. I originally tried using
> i2c_smbus_read_block_data() but that was returning errors. However, from your
> email, I went back and tried i2c_smbus_read_i2c_block_data() and that appears
> to be working so I will switch to that instead.
> 
This is odd, since the function should do a SMBus block read. Did you pass a
buffer that was too small, by any chance ?

> > > +
> > > +/**
> > > + * Helper function required since linux SMBus implementation does not currently
> > > + * (v5.6) support the SMBus 3.0 "Read 32" protocol
> > > + */
> > > +static s32 raa_dmpvr2_smbus_read32(const struct i2c_client *client, u8 command,
> > > +				   unsigned char *data)
> > > +{
> > > +	int status;
> > > +	unsigned char msgbuf[1];
> > > +	struct i2c_msg msg[2] = {
> > > +		{
> > > +			.addr = client->addr,
> > > +			.flags = client->flags,
> > > +			.len = 1,
> > > +			.buf = msgbuf,
> > > +		},
> > > +		{
> > > +			.addr = client->addr,
> > > +			.flags = client->flags | I2C_M_RD,
> > > +			.len = 4,
> > > +			.buf = data,
> > > +		},
> > > +	};
> > > +
> > > +	msgbuf[0] = command;
> > > +	status = i2c_transfer(client->adapter, msg, 2);
> > > +	if (status != 2)
> > > +		return status;
> > > +	return 0;
> > > +}
> > 
> > Maybe it would be worthwhile to consider implementing it ?
> > 
> I could add these functions to i2c-core-smbus instead if that is desired.
> However, I am unsure if it would be proper to only partially update the SMBus
> implementation to match the 3.0 spec. Is there perhaps a better forum to
> discuss adding all of the 3.0 changes?
> 
Never mind, just keep it local. I don't like it too much, but getting patches
accepted into the i2c core might take several releases, and it would be unfair
to impose that on you.

Guenter