Hi Thomas, Thanks for your contribution :) On Tue, 3 Dec 2024 at 10:12, Thomas Antoine <t.antoine@xxxxxxxxxxxx> wrote: > > On 12/3/24 10:31, André Draszik wrote: > > On Tue, 2024-12-03 at 10:08 +0100, Thomas Antoine wrote: > >> On 12/3/24 07:47, André Draszik wrote: > >>>> From: Thomas Antoine <t.antoine@xxxxxxxxxxxx> > >>>> > >>>> The Maxim max77759 fuel gauge has the same interface as the Maxim max1720x > >>>> except for the non-volatile memory slave address which is not available. > >>> > >>> It is not fully compatible, and it also has a lot more registers. > >>> > >>> For example, the voltage now is not in register 0xda as this driver assumes. > >>> With these changes, POWER_SUPPLY_PROP_VOLTAGE_NOW just reads as 0. 0xda > >>> doesn't exist in max77759 > >>> > >>> I haven't compared in depth yet, though. > >> > >> Is the voltage necessary for the driver? If so, we could just not > >> read the voltage. If it is necessary, I can try to kook into it and try > >> to find in which register it is located (if there is one). > > > > Downstream reports it in > > https://android.googlesource.com/kernel/google-modules/bms/+/refs/heads/android-gs-raviole-mainline/max1720x_battery.c#2400 > > > > so upstream should do, too. > > I should have checked before answering. Indeed, I will try to see the > best way to choose the register based on the chip. From what I see, it > will also be necessary to change the translation of the reg value to microvolt > as downstream does *78125/1000 when it is currently *1250 in mainline: > https://android.googlesource.com/kernel/google-modules/bms/+/refs/heads/android-gs-raviole-mainline/max1720x_battery.h#49 > > >>>> static const char *const max17205_model = "MAX17205"; > >>>> +static const char *const max77759_model = "MAX77759"; > >>>> > >>>> struct max1720x_device_info { > >>>> struct regmap *regmap; > >>>> @@ -54,6 +57,21 @@ struct max1720x_device_info { > >>>> int rsense; > >>>> }; > >>>> > >>>> +struct chip_data { > >>>> + u16 default_nrsense; /* in regs in 10^-5 */ > >>>> + u8 has_nvmem; > >>>> +}; > >>>> + > >>>> +static const struct chip_data max1720x_data = { > >>>> + .default_nrsense = 1000, > >>>> + .has_nvmem = 1, > >>>> +}; > >>>> + > >>>> +static const struct chip_data max77759_data = { > >>>> + .default_nrsense = 500, > >>>> + .has_nvmem = 0, > >>>> +}; > >>> > >>> This should be made a required devicetree property instead, at least for > >>> max77759, as it's completely board dependent, 'shunt-resistor-micro-ohms' > >>> is widely used. > >>> > >>> I also don't think there should be a default. The driver should just fail > >>> to probe if not specified in DT (for max77759). > >> > >> I hesitated to do this but I didn't know what would be better. Will change > >> for v2. > > > > Just to clarify, has_nvmem can stay here in the driver, just rsense should > > go into DT is what I mean. > > It was clear don't worry. This answer is related to the same subject: > https://lore.kernel.org/linux-devicetree/20241202-b4-gs101_max77759_fg-v1-0-98d2fa7bfe30@xxxxxxxxxxxx/T/#ma55f41d42bf39be826d6cbf8987138bcc4eb52e3 > > >>>> + > >>>> /* > >>>> * Model Gauge M5 Algorithm output register > >>>> * Volatile data (must not be cached) > >>>> @@ -369,6 +387,8 @@ static int max1720x_battery_get_property(struct > >>>> power_supply *psy, > >>>> val->strval = max17201_model; > >>>> else if (reg_val == MAX172XX_DEV_NAME_TYPE_MAX17205) > >>>> val->strval = max17205_model; > >>>> + else if (reg_val == MAX172XX_DEV_NAME_TYPE_MAX77759) > >>>> + val->strval = max77759_model; > >>>> else > >>> > >>> This is a 16 bit register, and while yes, MAX172XX_DEV_NAME_TYPE_MASK only > >>> cares about the bottom 4 bits, the register is described as 'Firmware > >>> Version Information'. > >>> > >>> But maybe it's ok to do it like that, at least for now. > >> > >> I thought this method would be ok as long as there is no collision on > >> values. I hesitated to change the model evaluation method based on chip > >> model, where the max77759 would thus have an hard-coded value and the > >> max1720x would still evaluate the register value. I did not do it because > >> it led to a lot more changes for no difference. > > > > Downstream uses the upper bits for max77759: > > https://android.googlesource.com/kernel/google-modules/bms/+/refs/heads/android-gs-raviole-mainline/max_m5.h#135 > > > > I don't know what the original max17201/5 report in this register > > for those bits, though. Given for max77759 this register returns > > the firmware version, I assume the lower bits can change. > > Based on this datasheet of the max1720x, the upper bits are the revision > and the four lower bits are device. So it could change. > https://www.analog.com/media/en/technical-documentation/data-sheets/MAX17201-MAX17215.pdf#MAX17201%20DS.indd%3A.213504%3A15892 > > If the four lower bits are not always 0 for the max77759 then I guess it > is necessary to change this as it wouldn't work with all max77759. The definition of this register for max77759 is Register name(addr): DevName (0x21) Rest value: 0x6200 Bitfield: DevName Bits: 15:0 Description: Firmware Version Information So I don't think you can rely on the bottom bits always being zero regards, Peter