On 23:38-20130619, Mark Brown wrote: > On Wed, Jun 19, 2013 at 02:17:54PM -0500, Nishanth Menon wrote: > > > Account for step size accuracy when exact voltage requests are send for > > step based regulators. > > If the consumer can tolerate a different voltage why not just request > the range that can be tolerated? Your problem here is specifying an > exact voltage. I think you mean using regulator_get_linear_step > > > The specific example I faced was using cpufreq-cpu0 driver with voltages > > for OPPs for MPU rail and attempting the common definitions against voltages > > that are non-exact multiples of stepsize of PMIC. > > > The alternative would be implement custom set_voltage (as againsta simpler > > set_voltage_sel and using linear map/list functions) for the regulator which > > will account for the same. > > > Yet another alternative might be to introduce yet another custom function similar > > to regulator_set_voltage_tol which accounts for this. something like: > > regulator_set_voltage_floor(regulator, voltage, tol) or something to that effect. > > Or as I keep telling you guys the consumer can just do that directly > using the existing API; the whole point in specifying the voltage as a > range is to allow the consumer to cope with arbatrary regulators by > giving a range of voltages that it can accept. > > The API is deliberately very conservative in these matters since there > is a danger of physical damage if things really go wrong in some > applications, it makes sure that both the drivers and the system > integration are involved. I agree. The intent of this series was to start a conversation to see if we can make it simpler. Searching for the users of regulator_get_linear_step in 3.10-rc6 shows none. For a generic driver which needs to handle platforms which have tolerance, they'd use regulator_set_voltage_tol. But the implementation would allow for uV - tol to uV + tol as range, which in the case I mentioned(min voltage =uV) wont work. If the consumer wants to be aware of linear step regulator, they'd have to do: step_uV = regulator_get_linear_step(...); regulator_set_voltage(uV, uV + step_uV); Then this wont handle tolerance. So the solution seems to be (for the consumer): step_uV = regulator_get_linear_step(...); .. if (tol) regulator_set_voltage_tol(uV, tol); else regulator_set_voltage(uV, uV + step_uV); (with the required error checks for regulator being a linear regulator etc..). At least to me, there is no sane manner to handle "tolerance" and linear step accuracy for a defined voltage (Should tolerance be rounded off to step_uV? what about the border cases etc.) Would you agree? -- Regards, Nishanth Menon PS: Since I just looped in cpufreq list, discussion thread: http://marc.info/?t=137166954900005&r=1&w=2 -- To unsubscribe from this list: send the line "unsubscribe cpufreq" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
