Re: [PATCH] i2c/busses: (mpc) Add support for SMBUS_READ_BLOCK_DATA

[Guenter Roeck <guenter.roeck-IzeFyvvaP7pWk0Htik3J/w@public.gmane.org>]

Hi Jean,

On Tue, 2011-11-15 at 03:54 -0500, Jean Delvare wrote:
> Hi Guenter,
> 
> On Mon, 14 Nov 2011 22:27:42 -0800, Guenter Roeck wrote:
> > Add support for SMBUS_READ_BLOCK_DATA to the i2c-mpc bus driver.
> > Required to support the PMBus zl6100 driver with i2c-mpc.
> > 
> > Reported-by: Tang Yuantian <B29983-KZfg59tc24xl57MIdRCFDg@public.gmane.org>
> > Cc: Tang Yuantian <B29983-KZfg59tc24xl57MIdRCFDg@public.gmane.org>
> > Signed-off-by: Guenter Roeck <guenter.roeck-IzeFyvvaP7pWk0Htik3J/w@public.gmane.org>
> > ---
> >  drivers/i2c/busses/i2c-mpc.c |   33 +++++++++++++++++++++++++--------
> >  1 files changed, 25 insertions(+), 8 deletions(-)
> > 
> > diff --git a/drivers/i2c/busses/i2c-mpc.c b/drivers/i2c/busses/i2c-mpc.c
> > index 107397a..77aade7 100644
> > --- a/drivers/i2c/busses/i2c-mpc.c
> > +++ b/drivers/i2c/busses/i2c-mpc.c
> > @@ -454,7 +454,7 @@ static int mpc_write(struct mpc_i2c *i2c, int target,
> >  }
> >  
> >  static int mpc_read(struct mpc_i2c *i2c, int target,
> > -		    u8 *data, int length, int restart)
> > +		    u8 *data, int length, int restart, bool block)
> 
> bool block would be better named bool recv_len IMHO. It will be set to
> 0 for I2C block reads, which is confusing.
> 
Ok.

> >  {
> >  	unsigned timeout = i2c->adap.timeout;
> >  	int i, result;
> > @@ -470,7 +470,7 @@ static int mpc_read(struct mpc_i2c *i2c, int target,
> >  		return result;
> >  
> >  	if (length) {
> > -		if (length == 1)
> > +		if (length == 1 && !block)
> >  			writeccr(i2c, CCR_MIEN | CCR_MEN | CCR_MSTA | CCR_TXAK);
> >  		else
> >  			writeccr(i2c, CCR_MIEN | CCR_MEN | CCR_MSTA);
> > @@ -479,17 +479,28 @@ static int mpc_read(struct mpc_i2c *i2c, int target,
> >  	}
> >  
> >  	for (i = 0; i < length; i++) {
> > +		u8 byte;
> > +
> >  		result = i2c_wait(i2c, timeout, 0);
> >  		if (result < 0)
> >  			return result;
> >  
> > +		byte = readb(i2c->base + MPC_I2C_DR);
> > +		/*
> > +		 * Adjust length if first received byte is length
> > +		 */
> > +		if (i == 0 && block) {
> > +			if (byte == 0 || byte > I2C_SMBUS_BLOCK_MAX)
> > +				return -EPROTO;
> > +			length += byte;
> > +		}
> > +		data[i] = byte;
> >  		/* Generate txack on next to last byte */
> >  		if (i == length - 2)
> >  			writeccr(i2c, CCR_MIEN | CCR_MEN | CCR_MSTA | CCR_TXAK);
> >  		/* Do not generate stop on last byte */
> >  		if (i == length - 1)
> >  			writeccr(i2c, CCR_MIEN | CCR_MEN | CCR_MSTA | CCR_MTX);
> > -		data[i] = readb(i2c->base + MPC_I2C_DR);
> >  	}
> 
> This needs careful testing (which I can't do.) There may have been a
> reason why the read was done after the writes. Swapping the commands
> may be the wrong thing to do. The dummy read earlier in this function
> suggests that maybe changes to CCR do not take effect until you read
> from (or write to) the DR register.
> 
Interestingly there is not always a read after a write to ccr. If a read
is necessary, I'd rather add a dummy read after writeccr().

> Can't the above be rewritten to keep the order of the commands as it
> was before? AFAICS it would only take one or two extra tests.
> 
The resulting code would not support 1-byte block reads. That seems to
be unnecessary and undesirable.

> Note that the hardware implementation may make it difficult or even
> impossible to properly support SMBus block reads of 1 byte. Not sure
> what should be done when this can be supported and still happens...
> Returning -EOPNOTSUPP I guess, and then probably the I2C engine needs
> some form of reset.
> 
The code generates stop after returning from mpc_read(). That should
hopefully take care of error conditions. I could add 
	writeccr(i2c, 0);
into the error path, as is done for timeouts, but I am not sure if that
would be helpful or not.

Either case, I'd like to avoid that case. I think it would be better to
get some test coverage from someone who has access to a board, or even
better feedback from someone who knows the chip.

Yuantian indicated that my raw patch worked with the zl6100 driver, and
the PMBus driver does a lot of accesses, so we do have some test
coverage already.

> >  
> >  	return length;
> > @@ -532,12 +543,17 @@ static int mpc_xfer(struct i2c_adapter *adap, struct i2c_msg *msgs, int num)
> >  			"Doing %s %d bytes to 0x%02x - %d of %d messages\n",
> >  			pmsg->flags & I2C_M_RD ? "read" : "write",
> >  			pmsg->len, pmsg->addr, i + 1, num);
> > -		if (pmsg->flags & I2C_M_RD)
> > -			ret =
> > -			    mpc_read(i2c, pmsg->addr, pmsg->buf, pmsg->len, i);
> > -		else
> > +		if (pmsg->flags & I2C_M_RD) {
> > +			bool block = pmsg->flags & I2C_M_RECV_LEN;
> 
> Here again I'd rather name it bool recv_len for clarity.
> 
Ok.

> > +
> > +			ret = mpc_read(i2c, pmsg->addr, pmsg->buf, pmsg->len, i,
> > +				       block);
> 
> That's a lot of parameters, most coming from pmsg. It would be more
> efficient to pass pmsg itself. Not directly related to your patch,
> admittedly, but it makes the problem more obvious. Maybe a cleanup for
> later.
> 
later ...

> > +			if (block && ret > 0)
> > +				pmsg->len = ret;
> > +		} else {
> >  			ret =
> >  			    mpc_write(i2c, pmsg->addr, pmsg->buf, pmsg->len, i);
> > +		}
> >  	}
> >  	mpc_i2c_stop(i2c);
> >  	return (ret < 0) ? ret : num;
> > @@ -545,7 +561,8 @@ static int mpc_xfer(struct i2c_adapter *adap, struct i2c_msg *msgs, int num)
> >  
> >  static u32 mpc_functionality(struct i2c_adapter *adap)
> >  {
> > -	return I2C_FUNC_I2C | I2C_FUNC_SMBUS_EMUL;
> > +	return I2C_FUNC_I2C | I2C_FUNC_SMBUS_EMUL
> > +	  | I2C_FUNC_SMBUS_READ_BLOCK_DATA;
> 
> You could add I2C_FUNC_SMBUS_BLOCK_PROC_CALL too, even though I
> don't know of any slave driver using it.
> 
Sure.

> >  }
> >  
> >  static const struct i2c_algorithm mpc_algo = {
> 
> 
Thanks,
Guenter