On Wed, Apr 17, 2019 at 12:48:18PM -0700, Nicolin Chen wrote:
> On Wed, Apr 17, 2019 at 11:39:49AM -0700, Nicolin Chen wrote:
>
> > > Thinking about it ... does it even make sense to cache reg_config twice,
> > > or would it be better to just update the local copy and use regmap_write()
> > > to send it to the chip ?
> >
> > I remember the reason of adding the read-back was to prevent race
> > condition. But now we have mutex protections for all sysfs nodes,
> > maybe it's not necessary anymore. I will read the code carefully
> > and see if it's safe to remove it -- will do in a separate patch.
>
> I just recalled a second thought for the reason why I left them
> there as it'd logically require a copy to restore upon failure
> of regmap_write, that might not look so neat as the read-back:
>
> old_config = reg_config;
> reg_config &= mask;
> reg_config |= val;
> ret = regmap_write(reg_config);
> if (ret) {
> reg_config = old_config;
> return ret;
> }
reg = (reg_config & mask) | val;
ret = regmap_write(reg);
if (ret)
return ret;
reg_config = reg;
doesn't look that bad to me, and is much less costly.
Guenter
