On Mon, 08 Feb 2021, Mark Jonas wrote: > From: Hubert Streidl <hubert.streidl@xxxxxxxxxxxx> > > By default the PMIC DA9063 2-wire interface is SMBus compliant. This > means the PMIC will automatically reset the interface when the clock > signal ceases for more than the SMBus timeout of 35 ms. > > If the I2C driver / device is not capable of creating atomic I2C > transactions, a context change can cause a ceasing of the clock signal. > This can happen if for example a real-time thread is scheduled. Then > the DA9063 in SMBus mode will reset the 2-wire interface. Subsequently > a write message could end up in the wrong register. This could cause > unpredictable system behavior. > > The DA9063 PMIC also supports an I2C compliant mode for the 2-wire > interface. This mode does not reset the interface when the clock > signal ceases. Thus the problem depicted above does not occur. > > This patch tests for the bus functionality "I2C_FUNC_I2C". It can > reasonably be assumed that the bus cannot obey SMBus timings if > this functionality is set. SMBus commands most probably are emulated > in this case which is prone to the latency issue described above. > > This patch enables the I2C bus mode if I2C_FUNC_I2C is set or > otherwise enables the SMBus mode for a native SMBus controller > which doesn't have I2C_FUNC_I2C set. > > Signed-off-by: Hubert Streidl <hubert.streidl@xxxxxxxxxxxx> > Signed-off-by: Mark Jonas <mark.jonas@xxxxxxxxxxxx> > --- > Changes in v4: > - Remove logging of selected 2-wire bus mode. > > Changes in v3: > - busmode now contains the correct bit DA9063_TWOWIRE_TO > > Changes in v2: > - Implement proposal by Adam Thomson and Wolfram Sang to check for > functionality I2C_FUNC_I2C instead of introducing a new DT property. > > drivers/mfd/da9063-i2c.c | 11 +++++++++++ > include/linux/mfd/da9063/registers.h | 3 +++ > 2 files changed, 14 insertions(+) > > diff --git a/drivers/mfd/da9063-i2c.c b/drivers/mfd/da9063-i2c.c > index 3781d0bb7786..9450c95a3741 100644 > --- a/drivers/mfd/da9063-i2c.c > +++ b/drivers/mfd/da9063-i2c.c > @@ -355,6 +355,7 @@ static int da9063_i2c_probe(struct i2c_client *i2c, > const struct i2c_device_id *id) > { > struct da9063 *da9063; > + unsigned int busmode; > int ret; > > da9063 = devm_kzalloc(&i2c->dev, sizeof(struct da9063), GFP_KERNEL); > @@ -442,6 +443,16 @@ static int da9063_i2c_probe(struct i2c_client *i2c, > return ret; > } > > + busmode = i2c_check_functionality(i2c->adapter, I2C_FUNC_I2C) ? > + 0 : DA9063_TWOWIRE_TO; Nit: I find ternaries like this tend to complicate matters and harm readability rather than the converse. > + ret = regmap_update_bits(da9063->regmap, DA9063_REG_CONFIG_J, > + DA9063_TWOWIRE_TO, busmode); > + if (ret < 0) { > + dev_err(da9063->dev, "Failed to set 2-wire bus mode.\n"); > + return -EIO; > + } I'm a little confused by this. It's likely just me, but I would still like some clarification. So you write to the TWOWIRE register despite whether I2C is operable not, which I guess is fine. But what if I2C is disabled and the update fails. You seem to complain to the user that a failure occurred and return an error even if the configuration is invalid in the first place. Would it not be better to encapsulate the update inside the functionality check? > return da9063_device_init(da9063, i2c->irq); > } > > diff --git a/include/linux/mfd/da9063/registers.h b/include/linux/mfd/da9063/registers.h > index 1dbabf1b3cb8..6e0f66a2e727 100644 > --- a/include/linux/mfd/da9063/registers.h > +++ b/include/linux/mfd/da9063/registers.h > @@ -1037,6 +1037,9 @@ > #define DA9063_NONKEY_PIN_AUTODOWN 0x02 > #define DA9063_NONKEY_PIN_AUTOFLPRT 0x03 > > +/* DA9063_REG_CONFIG_J (addr=0x10F) */ > +#define DA9063_TWOWIRE_TO 0x40 > + > /* DA9063_REG_MON_REG_5 (addr=0x116) */ > #define DA9063_MON_A8_IDX_MASK 0x07 > #define DA9063_MON_A8_IDX_NONE 0x00 -- Lee Jones [李琼斯] Senior Technical Lead - Developer Services Linaro.org │ Open source software for Arm SoCs Follow Linaro: Facebook | Twitter | Blog