| Message ID | 20200119201124.29620-1-tony@atomide.com (mailing list archive) |
|---|---|
| State | Not Applicable, archived |
| Headers | show |
| Series | [1/3] RFC: power: supply: cpcap-battery: Add helper for rough capacity | expand |
Hi! > Get a rough battery charge estimate until we've seen a high or low > battery level. After that we can use the coulomb counter to calculate > the battery capacity. > > Note that I should probably update this to support ocv-capacity-table > before this makes sense to apply. With ocv-capacity-table we should be > able to estimate battery state as described in the documentation for > Documentation/devicetree/bindings/power/supply/battery.txt. Maybe we should let userspace do that? https://github.com/pavelmachek/libbattery > +struct cpcap_battery_capacity { > + int capacity; > + int voltage; > + int percentage; > +}; > + > +#define CPCAP_CAP(l, v, p) \ > +{ \ > + .capacity = (l), \ > + .voltage = (v), \ > + .percentage = (p), \ > +}, > + > +/* Pessimistic battery capacity mapping before high or low value is seen */ > +static const struct cpcap_battery_capacity cpcap_battery_cap[] = { > + CPCAP_CAP(POWER_SUPPLY_CAPACITY_LEVEL_UNKNOWN, 0, 0) > + CPCAP_CAP(POWER_SUPPLY_CAPACITY_LEVEL_CRITICAL, 3100000, 0) > + CPCAP_CAP(POWER_SUPPLY_CAPACITY_LEVEL_LOW, 3300000, 2) > + CPCAP_CAP(POWER_SUPPLY_CAPACITY_LEVEL_NORMAL, 3700000, 50) > + CPCAP_CAP(POWER_SUPPLY_CAPACITY_LEVEL_HIGH, 4000000, 75) > + CPCAP_CAP(POWER_SUPPLY_CAPACITY_LEVEL_FULL, 4200000 - 18000, 100) > +}; No, not like this; it is simultaneously too complex and not complex enough. Too complex: we don't really want a table, when there's well know formula implementing it: http://git.goldelico.com/?p=gta04-kernel.git;a=patch;h=22ab047ae296e998379c1aa29fe1210043cfa040 Not complex enough: we know the current (and there are going to be big differences between CPU idle and CPU loaded with GSM active), and there's dependency on temperature, too... Best regards, Pavel
On 21/01/2020 10:57, Pavel Machek wrote: > Hi! > >> Get a rough battery charge estimate until we've seen a high or low >> battery level. After that we can use the coulomb counter to calculate >> the battery capacity. >> >> Note that I should probably update this to support ocv-capacity-table >> before this makes sense to apply. With ocv-capacity-table we should be >> able to estimate battery state as described in the documentation for >> Documentation/devicetree/bindings/power/supply/battery.txt. > > Maybe we should let userspace do that? > > https://github.com/pavelmachek/libbattery Unless this gets integrated in UPower, My point of view is that userspace probably wants some uniform approach. Kernel interface definitely is one uniform interface that would also get picked up by UPower. Cheers, Merlijn
Hi! > >> Get a rough battery charge estimate until we've seen a high or low > >> battery level. After that we can use the coulomb counter to calculate > >> the battery capacity. > >> > >> Note that I should probably update this to support ocv-capacity-table > >> before this makes sense to apply. With ocv-capacity-table we should be > >> able to estimate battery state as described in the documentation for > >> Documentation/devicetree/bindings/power/supply/battery.txt. > > > > Maybe we should let userspace do that? > > > > https://github.com/pavelmachek/libbattery > > Unless this gets integrated in UPower, My point of view is that > userspace probably wants some uniform approach. Kernel interface > definitely is one uniform interface that would also get picked up by UPower. I see that it is convenient to have it in kernel. I just assumed Sebastian would not take that, because it does not _need_ to be in kernel... And maybe it is easier in userspace, since userspace has access to persistent storage, so it can store detailed battery calibration parameters? (And Gnome/Mate even does something like that, see mate-power-statistics). So... not my decision. Don't care too much either way. I have been doing this in userspace, but... What I'd like to see is some kind of indication that "this attribute is not from hardware, kernel computed it for you" so clever userspace can ignore it/compute it in more advanced way. Best regards, Pavel
* Pavel Machek <pavel@ucw.cz> [200121 22:26]: > Hi! > > > >> Get a rough battery charge estimate until we've seen a high or low > > >> battery level. After that we can use the coulomb counter to calculate > > >> the battery capacity. > > >> > > >> Note that I should probably update this to support ocv-capacity-table > > >> before this makes sense to apply. With ocv-capacity-table we should be > > >> able to estimate battery state as described in the documentation for > > >> Documentation/devicetree/bindings/power/supply/battery.txt. > > > > > > Maybe we should let userspace do that? > > > > > > https://github.com/pavelmachek/libbattery > > > > Unless this gets integrated in UPower, My point of view is that > > userspace probably wants some uniform approach. Kernel interface > > definitely is one uniform interface that would also get picked up by UPower. > > I see that it is convenient to have it in kernel. I just assumed > Sebastian would not take that, because it does not _need_ to be in > kernel... I think we should find a way where we can keep the existing kernel interface as all the apps rely on that. If we allow userspace to optionally set the estimated battery full and battery empty coulomb counter values we should be all set. For temperature compensation it seems that the ocv-capacity-table is the way to go to describe the battery if the battery data is unsuable or does not exist. > And maybe it is easier in userspace, since userspace has access to > persistent storage, so it can store detailed battery calibration > parameters? (And Gnome/Mate even does something like that, see > mate-power-statistics). Agreed, at minimum the userspace should save estimated coulomb counter full and empty values. After that calculations are trivial as seen with this series. > So... not my decision. Don't care too much either way. I have been > doing this in userspace, but... > > What I'd like to see is some kind of indication that "this attribute > is not from hardware, kernel computed it for you" so clever userspace > can ignore it/compute it in more advanced way. Well we do have that via IIO already. Not sure if that is needed for power supply. Regards, Tony
Hi, Tony. I used the kernel with the patch serieas applied for a while. What I expected from the userspace perspective was having /sys/class/power_supply/battery/capacity undefined until kernel calculates more or less accurate values for it. Until then, the userspace should estimate percentage on its own using voltage and current values provided by the kernel. Like it's already done with bq27200 on Nokia N900. Right now the values which the kernel provides with /sys/class/power_supply/battery/capacity after a system boot are confusing. The user can see the battery plugin doesn't change its gauge gradually like it was done with previous kernel versions. Sometimes it gets suddenly empty, sometimes it changes from empty to half full. And it always reports "battery full" in advance - when the battery is being charged with relatively high current. The following part of mail dedicated to what I think should be fixed in the commits. 1) RFC: power: supply: cpcap-battery: Add helper for rough capacity CPCAP_CAP(POWER_SUPPLY_CAPACITY_LEVEL_FULL, 4200000 - 18000, 100) This line seems wrong, because Droid4 full battery is 4.35V, not 4.2V. 2.1) RFC: power: supply: cpcap-battery: Implement capacity percentage val->intval = (ddata->config.info.charge_full_design - delta) * 100; val->intval /= ddata->config.info.charge_full_design; IMHO, charge_full_design should never be used in battery charge percentage calculations. Li-ion batteries loose their capacity with age. Therefore such calculations are likely to always be wrong. I'd prefer to have the full charge value to be dynamically calculated by the kernel with a formula like this: charge_full = high->counter_uah - low->counter_uah; This which will give us accurate estimation for the battery charge, esti- mated time left on the battery and so on. It would be good, if we allow the userspace to store the full charge value between reboots and to feed it to the kernel after, so it could start providing accurated data faster: the battery should only go full or empty once and we are ready. I'd like to mention explicitly: I think the kernel should return -ENODATA for capacity values until the battery is "calibrated". By "calibration" I mean having high and low counter_uah values initialized with the data collected when the battery went a full charge or discharge cycle (from empty to full or vice versa). -- Best regards, Spinal
* Arthur D. <spinal.by@gmail.com> [200310 04:55]: > Hi, Tony. > > I used the kernel with the patch serieas applied for a while. > > What I expected from the userspace perspective was having > /sys/class/power_supply/battery/capacity undefined until > kernel calculates more or less accurate values for it. > > Until then, the userspace should estimate percentage on its own > using voltage and current values provided by the kernel. > Like it's already done with bq27200 on Nokia N900. > > Right now the values which the kernel provides with > /sys/class/power_supply/battery/capacity after a system boot > are confusing. OK yeah I like the idea of not showing anything until userspace as configured estimated high and low values. > The user can see the battery plugin doesn't change its gauge > gradually like it was done with previous kernel versions. > Sometimes it gets suddenly empty, sometimes it changes from empty > to half full. And it always reports "battery full" in advance - > when the battery is being charged with relatively high current. > > The following part of mail dedicated to what I think should be fixed > in the commits. > > 1) RFC: power: supply: cpcap-battery: Add helper for rough capacity > > CPCAP_CAP(POWER_SUPPLY_CAPACITY_LEVEL_FULL, 4200000 - 18000, 100) > > This line seems wrong, because Droid4 full battery is 4.35V, not 4.2V. Hmm yeah this is now configurable and defaults to 4.2V so we should use the confgured value. The reason we default to 4.2V now is because we suspect that if left connected to the charger at higher than 4.2V the battery ages faster. > 2.1) RFC: power: supply: cpcap-battery: Implement capacity percentage > > val->intval = (ddata->config.info.charge_full_design - delta) * 100; > val->intval /= ddata->config.info.charge_full_design; > > IMHO, charge_full_design should never be used in battery charge > percentage calculations. Li-ion batteries loose their capacity > with age. Therefore such calculations are likely to always be wrong. > > I'd prefer to have the full charge value to be dynamically calculated by > the kernel with a formula like this: > charge_full = high->counter_uah - low->counter_uah; OK makes sense. I guess we can assuem if both are set to the the same number like 0 on boot, there's no estimate available. > This which will give us accurate estimation for the battery charge, esti- > mated time left on the battery and so on. It would be good, if we allow > the userspace to store the full charge value between reboots and to feed > it to the kernel after, so it could start providing accurated data faster: > the battery should only go full or empty once and we are ready. Yes userspace is the only place that can eventually generate accurate estimate based on battery wear, temperature etc. > I'd like to mention explicitly: I think the kernel should return -ENODATA > for capacity values until the battery is "calibrated". By "calibration" > I mean having high and low counter_uah values initialized with the data > collected when the battery went a full charge or discharge cycle (from > empty to full or vice versa). Yes I like this idea. That leaves out the need for poor estimates in the kernel. I guess we should have the following new properties: POWER_SUPPLY_PROP_CHARGE_COUNTER_FULL POWER_SUPPLY_PROP_CHARGE_COUNTER_EMPTY And these would be 0 on boot and then userspace can update these based on battery data. Regards, Tony
Hi. >> 1) RFC: power: supply: cpcap-battery: Add helper for rough capacity >> >> CPCAP_CAP(POWER_SUPPLY_CAPACITY_LEVEL_FULL, 4200000 - 18000, 100) >> >> This line seems wrong, because Droid4 full battery is 4.35V, not 4.2V. > > Hmm yeah this is now configurable and defaults to 4.2V so we > should use the confgured value. The reason we default to 4.2V > now is because we suspect that if left connected to the charger > at higher than 4.2V the battery ages faster. 1. How to configure this value? I tried: echo 4350000 > /sys/class/power_supply/battery/constant_charge_voltage ...but it didn't change the fully charged battery voltage. 2. If the value is configurable, it probably shouldn't be hardcoded like this in the code: PCAP_CAP(POWER_SUPPLY_CAPACITY_LEVEL_FULL, 4200000 - 18000, 100) >> I think the kernel should return -ENODATA >> for capacity values until the battery is "calibrated". By "calibration" >> I mean having high and low counter_uah values initialized with the data >> collected when the battery went a full charge or discharge cycle (from >> empty to full or vice versa). > > Yes I like this idea. That leaves out the need for poor estimates > in the kernel. > > I guess we should have the following new properties: > > POWER_SUPPLY_PROP_CHARGE_COUNTER_FULL > POWER_SUPPLY_PROP_CHARGE_COUNTER_EMPTY > > And these would be 0 on boot and then userspace can > update these based on battery data. My idea is to only have one extra property which will store the delta of counter values for full and empty battery states. In this case, it's going to be more accurate than allowing the userspace to store/restore both full and empty counters. Some users may have a dual boot setup. In this case restoring full & empty counters after reboot can make the kernel to report wrong battery percentage values. Let's imagine the following situation: 1. The user charges the battery to the full state and discharges it to be almost depleted, so the kernel has "calibrated" values for counter_full and counter_empty. 2. The user decides to reboot to another OS. Let's assume he has a fully charged battery before doing the reboot. For example the kernel provides the following values: POWER_SUPPLY_PROP_CHARGE_COUNTER_FULL = -250000 POWER_SUPPLY_PROP_CHARGE_COUNTER_EMPTY = 950000 /sys/class/power_supply/battery/charge_counter = -250000 A daemon saves these values to a permanent storage. 3. The user boots another OS, works there for a while, the battery charge becomes 50%. 4. The user reboots back to our system, we restore the values: POWER_SUPPLY_PROP_CHARGE_COUNTER_FULL <- -250000 POWER_SUPPLY_PROP_CHARGE_COUNTER_EMPTY <- 950000 5. /sys/class/power_supply/battery/charge_counter is initialized to 0 or -250000, it doesn't matter, because in both cases we will get the wrong battery percentage. We should report 50%, remember? In any case it can't be calculated to this value. And here's how I suggest it to be implemented: 1. Same as above 2. Same as above, except the kernel provides a single value that the user should store between reboots (it's the delta I mentioned above): POWER_SUPPLY_PROP_CHARGE_FULL = 1200000 3. Same as above 4. The user reboots back to our system, we restore the value: POWER_SUPPLY_PROP_CHARGE_FULL <- 1200000 5. The kernel reports -ENODATA as battery capacity value until the battery goes to a reference point (fully charged or almost discharged state). When the battery is fully charged or discharged, the kernel can immediately calculate the opposite counter value. And no matter what was the initial value of /sys/class/power_supply/battery/charge_counter when the system booted. Of course, the value of the delta will not be constant, it should be recalculated over time. The good news is that this value will be changed slowly over time. That is, it can vary from 2% to 10% per year. -- Best regards, Spinal
* Arthur D. <spinal.by@gmail.com> [200310 18:36]: > Hi. > > > > 1) RFC: power: supply: cpcap-battery: Add helper for rough capacity > > > > > > CPCAP_CAP(POWER_SUPPLY_CAPACITY_LEVEL_FULL, 4200000 - 18000, 100) > > > > > > This line seems wrong, because Droid4 full battery is 4.35V, not 4.2V. > > > > Hmm yeah this is now configurable and defaults to 4.2V so we > > should use the confgured value. The reason we default to 4.2V > > now is because we suspect that if left connected to the charger > > at higher than 4.2V the battery ages faster. > > 1. How to configure this value? I tried: > echo 4350000 > /sys/class/power_supply/battery/constant_charge_voltage > > ...but it didn't change the fully charged battery voltage. If you really want to use it, first enable it for the battery in /sys/class/power/supply/usb to allow it, then also enable it for the charger in /sys/class/power/supply/usb to actually use it. But as discussed earlier, it really seems the battery ages faster at the higher charge rates. Especially if left connected to the charger for longer times as also seen with Android earlier. > 2. If the value is configurable, it probably shouldn't be hardcoded like > this in the code: > PCAP_CAP(POWER_SUPPLY_CAPACITY_LEVEL_FULL, 4200000 - 18000, 100) Right, we should now use the configured voltage :) > > > I think the kernel should return -ENODATA > > > for capacity values until the battery is "calibrated". By "calibration" > > > I mean having high and low counter_uah values initialized with the data > > > collected when the battery went a full charge or discharge cycle (from > > > empty to full or vice versa). > > > > Yes I like this idea. That leaves out the need for poor estimates > > in the kernel. > > > > I guess we should have the following new properties: > > > > POWER_SUPPLY_PROP_CHARGE_COUNTER_FULL > > POWER_SUPPLY_PROP_CHARGE_COUNTER_EMPTY > > > > And these would be 0 on boot and then userspace can > > update these based on battery data. > > My idea is to only have one extra property which will store the delta of > counter values for full and empty battery states. In this case, it's going > to be more accurate than allowing the userspace to store/restore both full > and empty counters. Hmm OK. Sounds like you might have some patches coming then? > Some users may have a dual boot setup. In this case restoring full & empty > counters after reboot can make the kernel to report wrong battery > percentage > values. Yeah no idea how that is handled. Also changing the battery can cause issues with this. I was expecting the 1-wire EEPROM to contain some battery specific data like a serial number, but it seems this data is the same for all the same type batteries. > Let's imagine the following situation: > > 1. The user charges the battery to the full state and discharges it to be > almost depleted, so the kernel has "calibrated" values for counter_full and > counter_empty. > 2. The user decides to reboot to another OS. Let's assume he has a fully > charged battery before doing the reboot. > For example the kernel provides the following values: > POWER_SUPPLY_PROP_CHARGE_COUNTER_FULL = -250000 > POWER_SUPPLY_PROP_CHARGE_COUNTER_EMPTY = 950000 > /sys/class/power_supply/battery/charge_counter = -250000 > A daemon saves these values to a permanent storage. > 3. The user boots another OS, works there for a while, the battery charge > becomes 50%. > 4. The user reboots back to our system, we restore the values: > POWER_SUPPLY_PROP_CHARGE_COUNTER_FULL <- -250000 > POWER_SUPPLY_PROP_CHARGE_COUNTER_EMPTY <- 950000 > 5. /sys/class/power_supply/battery/charge_counter is initialized to 0 or > -250000, it doesn't matter, because in both cases we will get the wrong > battery percentage. We should report 50%, remember? In any case it can't > be calculated to this value. Maybe some charge counter offset limit can be used to validate the saved data in the userspace before updating values for kernel? > And here's how I suggest it to be implemented: > 1. Same as above > 2. Same as above, except the kernel provides a single value that the user > should store between reboots (it's the delta I mentioned above): > POWER_SUPPLY_PROP_CHARGE_FULL = 1200000 > 3. Same as above > 4. The user reboots back to our system, we restore the value: > POWER_SUPPLY_PROP_CHARGE_FULL <- 1200000 > 5. The kernel reports -ENODATA as battery capacity value until the battery > goes to a reference point (fully charged or almost discharged state). > When the battery is fully charged or discharged, the kernel can immediately > calculate the opposite counter value. And no matter what was the initial > value of /sys/class/power_supply/battery/charge_counter when the system > booted. > > Of course, the value of the delta will not be constant, it should be > recalculated over time. The good news is that this value will be changed > slowly over time. That is, it can vary from 2% to 10% per year. Sounds good to me :) Tony
diff --git a/drivers/power/supply/cpcap-battery.c b/drivers/power/supply/cpcap-battery.c --- a/drivers/power/supply/cpcap-battery.c +++ b/drivers/power/supply/cpcap-battery.c @@ -136,6 +136,29 @@ struct cpcap_battery_ddata { u16 vendor; }; +struct cpcap_battery_capacity { + int capacity; + int voltage; + int percentage; +}; + +#define CPCAP_CAP(l, v, p) \ +{ \ + .capacity = (l), \ + .voltage = (v), \ + .percentage = (p), \ +}, + +/* Pessimistic battery capacity mapping before high or low value is seen */ +static const struct cpcap_battery_capacity cpcap_battery_cap[] = { + CPCAP_CAP(POWER_SUPPLY_CAPACITY_LEVEL_UNKNOWN, 0, 0) + CPCAP_CAP(POWER_SUPPLY_CAPACITY_LEVEL_CRITICAL, 3100000, 0) + CPCAP_CAP(POWER_SUPPLY_CAPACITY_LEVEL_LOW, 3300000, 2) + CPCAP_CAP(POWER_SUPPLY_CAPACITY_LEVEL_NORMAL, 3700000, 50) + CPCAP_CAP(POWER_SUPPLY_CAPACITY_LEVEL_HIGH, 4000000, 75) + CPCAP_CAP(POWER_SUPPLY_CAPACITY_LEVEL_FULL, 4200000 - 18000, 100) +}; + #define CPCAP_NO_BATTERY -400 static struct cpcap_battery_state_data * @@ -411,6 +434,40 @@ static int cpcap_battery_update_status(struct cpcap_battery_ddata *ddata) return 0; } +static void cpcap_battery_get_rough(struct cpcap_battery_ddata *ddata, + int *level, int *percentage) +{ + struct cpcap_battery_state_data *latest; + const struct cpcap_battery_capacity *cap = NULL; + int voltage, i; + + latest = cpcap_battery_latest(ddata); + voltage = latest->voltage; + + for (i = ARRAY_SIZE(cpcap_battery_cap) - 1; i >=0; i--) { + cap = &cpcap_battery_cap[i]; + if (voltage >= cap->voltage) + break; + } + + if (!cap) + return; + + if (level) + *level = cap->capacity; + if (percentage) + *percentage = cap->percentage; +} + +static int cpcap_battery_get_rough_capacity(struct cpcap_battery_ddata *ddata) +{ + int capacity = 0; + + cpcap_battery_get_rough(ddata, &capacity, NULL); + + return capacity; +} + static enum power_supply_property cpcap_battery_props[] = { POWER_SUPPLY_PROP_STATUS, POWER_SUPPLY_PROP_PRESENT, @@ -516,18 +573,7 @@ static int cpcap_battery_get_property(struct power_supply *psy, val->intval = div64_s64(tmp, 100); break; case POWER_SUPPLY_PROP_CAPACITY_LEVEL: - if (cpcap_battery_full(ddata)) - val->intval = POWER_SUPPLY_CAPACITY_LEVEL_FULL; - else if (latest->voltage >= 3750000) - val->intval = POWER_SUPPLY_CAPACITY_LEVEL_HIGH; - else if (latest->voltage >= 3300000) - val->intval = POWER_SUPPLY_CAPACITY_LEVEL_NORMAL; - else if (latest->voltage > 3100000) - val->intval = POWER_SUPPLY_CAPACITY_LEVEL_LOW; - else if (latest->voltage <= 3100000) - val->intval = POWER_SUPPLY_CAPACITY_LEVEL_CRITICAL; - else - val->intval = POWER_SUPPLY_CAPACITY_LEVEL_UNKNOWN; + val->intval = cpcap_battery_get_rough_capacity(ddata); break; case POWER_SUPPLY_PROP_CHARGE_FULL_DESIGN: val->intval = ddata->config.info.charge_full_design;