Re: [PATCH 1/3] regulator: twl: Make sure we have access to powerbus before trying to write to it

[Ivaylo Dimitrov]

Hi,

On 24.04.2016 19:10, Pavel Machek wrote:
> Hi!
>
> > According to the TRM, we need to enable i2c access to powerbus before
> > writing to it. Also, a new write to powerbus should not be attempted if
> > there is a pending transfer. The current code does not implement that
> > functionality and while there are no known problems caused by that, it is
> > better to follow what TRM says.
> >
> > Signed-off-by: Ivaylo Dimitrov <ivo.g.dimitrov.75@xxxxxxxxx>
> > ---
> >   drivers/regulator/twl-regulator.c | 78 +++++++++++++++++++++++++++++++++++----
> >   1 file changed, 70 insertions(+), 8 deletions(-)
> >
> > diff --git a/drivers/regulator/twl-regulator.c b/drivers/regulator/twl-regulator.c
> > index 955a6fb..aad748b0 100644
> > --- a/drivers/regulator/twl-regulator.c
> > +++ b/drivers/regulator/twl-regulator.c
> > @@ -21,7 +21,7 @@
> >   #include <linux/regulator/machine.h>
> >   #include <linux/regulator/of_regulator.h>
> >   #include <linux/i2c/twl.h>
> > -
> > +#include <linux/delay.h>
> >
> >   /*
> >    * The TWL4030/TW5030/TPS659x0/TWL6030 family chips include power management, a
> > @@ -188,6 +188,74 @@ static int twl6030reg_is_enabled(struct regulator_dev *rdev)
> >   	return grp && (val == TWL6030_CFG_STATE_ON);
> >   }
> >
> > +#define PB_I2C_BUSY	BIT(0)
> > +#define PB_I2C_BWEN	BIT(1)
> > +
> > +/* Wait until buffer empty/ready to send a word on power bus. */
> > +static int twl4030_wait_pb_ready(void)
> > +{
> > +
> > +	int	ret;
> > +	int	timeout = 10;
> > +	u8	val;
> > +
>
> Can we do this plain
>
>      while (timeout--) {
>      }...

Now looking at the code, yes, while(timeout--) looks prettier, but as 
the $subject patch is in the linux-next for a couple of weeks already, 
that change will need another patch. I'll put that in my TODO, right 
after the RFC for the N900 cameras :) .

> ? Also... if the bit is not immediately available, it will wait for
> 1msec. Would it make sense to have timeout = 1000 but wait only 10usec
> each time or something?

Well, anyway these look like a kind of arbitrary values, I guess it will 
work both ways. But, I borrowed the code in the patch from stock Nokia 
kernel, it has been field tested on tens of thousands of devices, so, 
unless you have hard values giving a reason to do it the other way, I 
prefer to keep it as it is.

Regards,
Ivo

> > +	do {
> > +		ret = twl_i2c_read_u8(TWL_MODULE_PM_MASTER, &val,
> > +				      TWL4030_PM_MASTER_PB_CFG);
> > +		if (ret < 0)
> > +			return ret;
> > +
> > +		if (!(val & PB_I2C_BUSY))
> > +			return 0;
> > +
> > +		mdelay(1);
> > +		timeout--;
> > +	} while (timeout);
> > +
> > +	return -ETIMEDOUT;
> > +}