Nicola Corna <nic...@corna.info> writes: > As reported here > http://www.advamation.com/knowhow/raspberrypi/rpi-i2c-bug.html > the BCM2835 has a bug in its i2c implementation which prevents a correct > clock stretching. > > Signed-off-by: Nicola Corna <nic...@corna.info> > --- > drivers/i2c/busses/i2c-bcm2835.c | 5 +++++ > 1 file changed, 5 insertions(+) > > diff --git a/drivers/i2c/busses/i2c-bcm2835.c > b/drivers/i2c/busses/i2c-bcm2835.c > index 3032b89..38d7102 100644 > --- a/drivers/i2c/busses/i2c-bcm2835.c > +++ b/drivers/i2c/busses/i2c-bcm2835.c > @@ -222,6 +222,10 @@ static const struct i2c_algorithm bcm2835_i2c_algo = { > .functionality = bcm2835_i2c_func, > }; > > +static const struct i2c_adapter_quirks bcm2835_i2c_quirks = { > + .flags = I2C_AQ_NO_CLK_STRETCH, > +}; > + > static int bcm2835_i2c_probe(struct platform_device *pdev) > { > struct bcm2835_i2c_dev *i2c_dev; > @@ -293,6 +297,7 @@ static int bcm2835_i2c_probe(struct platform_device *pdev) > adap->algo = &bcm2835_i2c_algo; > adap->dev.parent = &pdev->dev; > adap->dev.of_node = pdev->dev.of_node; > + adap->quirks = &bcm2835_i2c_quirks; > > bcm2835_i2c_writel(i2c_dev, BCM2835_I2C_C, 0);
Looking at the driver, we're not setting the REDL/FEDL fields, which are involved in the delays following clock stretching. Looks like I2C_DEL's low 16 bits (rising edge delay) should be I2C_DIV / 4 (wait half the period for the slave to update its data before we sample it), and the high 16 bits (falling edge delay) should be I2C_DIV / 16 (put our data out before we expet the slave to sample the value). Note: this is not a detailed analysis, nor am I asserting that this is the bug, just something that we seem to be missing in the Linux driver. (Added cc to linux-rpi-kernel, in case anyone on that list is interested in I2C issues on the platform)
signature.asc
Description: PGP signature
