On Dienstag, 30. November 2021 19:33:02 CET you wrote:

On Tue, Nov 30, 2021 at 07:12:11PM +0100, Nicolas Frattaroli wrote:

quoted

On Dienstag, 30. November 2021 17:10:21 CET Chris Morgan wrote:

quoted

On Tue, Nov 30, 2021 at 03:03:03AM +0100, Nicolas Frattaroli wrote:

quoted

On Donnerstag, 16. September 2021 21:42:04 CET Chris Morgan wrote:

quoted

From: Chris Morgan <redacted>

This series is to add support for the Rockchip rk817 battery charger
which is present in all Rockchip RK817 PMICs. The driver was written
as a joint effort by Maya Matuszczyk [off-list ref] and
myself Chris Morgan [off-list ref].

Hi Chris and Maya,

Gave this a whirl on my Quartz64 Model A. I noticed that this will
happily let me discharge past voltage_min_design:

 $ cat /sys/class/power_supply/rk817-battery/voltage_min_design 
 3625000
 $ cat /sys/class/power_supply/rk817-battery/voltage_avg 
 3381360

Is this normal? It went all the way to under 3V before the
board finally locked up.

Does the minimum voltage not affect some sort of cutout on
the RK817? Does it even have one? Is it the driver's job to
do something here or not?

Hi Chris,

quoted

It does not look like I coded that, but I can. The PMIC has a
selectable register (RK817_PMIC_SYS_CFG0 = 0xf1) to automatically shut
down the system at a certain voltage (between 2.7v and 3.4v in
increments of 100mv; bits 6-4), a register to set a low voltage value
(between 2.8v and 3.5v in increments of 100mv; bits 2-0), and a
register to set a low voltage action (either shut down the machine or
trigger an interrupt; bit 3 and then I believe the interrupt is read at
bit 7 of RK817_PMIC_INT_STS0 - 0xf8).

Excellent, I think shutting down the system is definitely the way to
go here, as by reaching this voltage we can assume userspace has gone
missing and we don't know if the kernel is still alive.

quoted

I guess I just assumed userspace would handle it, but that's probably
a bad assumption.

Yes, I didn't have upower installed, but over discharge protection
shouldn't be handled by userspace at all. Had I connected an unprotected
cell that didn't cut out at below 3V, it would've been irreversibly
damaged due to a simple forgotten package install. Even if unprotected
batteries are a bad idea, I think we best assume the worst case
situation here.

quoted

What if I set the low voltage value to either 3.5v
or the min design voltage, whichever is less; and then set the low
voltage trigger to shut down the system? This would prevent your
battery from dropping below 3.5v or the min design voltage (whichever
is less, sadly 3.5v is as high as I can go for the minimum) in the
event userspace doesn't take action first? We could also set the
shutdown voltage to be 100mv less than the min design voltage and use
the low voltage interrupt to trigger an action, but I'm not sure what
action would be appropriate (and if userspace isn't listening it would
be moot anyway).

A very simple solution would be to simply set the shutoff point to the
voltage reaching less than 3.1V. The RK817 only charges Lithium/Li-Po
batteries, and they should generally not be discharged below 3.0V. Of
the protected batteries I own, over-discharge protection generally
kicks in at the 3V threshold.

This gets rid of a lot of logic that could lead to bugs by simply setting
a known good value to begin with.

I think I can summarise how I understand this is supposed to work as
follows:

 - if battery reaches 0% and the battery alarm is rang, userspace should
   power down the system
     - this prevents unclean shutdowns
 - if the battery reaches a dangerously low voltage for its chemistry,
   the PMIC should cut power to the system as instructed by the kernel
     - last resort measure in case userspace is absent and the battery
       happens to be unprotected
 - there may or may not be battery protection which kicks in at 3V, but
   we should not rely on this working
      - some 18650 cases infamously aren't sized to fit cells with a
        protection circuit connected to it. I shan't name names.

In a freak accident where the kernel is completely locked up and the
battery is unprotected and being drained, I presume that having set
this PMIC register will still make it kick in, which lets me sleep
a little sounder at night.

Okay. The default value for automatic shutdown is 2.7v, but that's
probably too low. I'm wondering though if this should go in the main
MFD driver, rather than the battery driver? The reason why I ask is
that without a battery driver to raise the shutdown voltage to 3.1v,
the shutdown will not happen until 2.7v which risks damage to the
battery. This means devices in production today (such as anyone running
the Odroid Go Advance on a mainline kernel) on a battery will
experience this behavior. Setting that register value shouldn't do
anything harmful if you don't have a battery, so I assume if it's safe
to set it 100% of the time for battery users it's safe to set 100% of
the time in total. Which also begs the question, if this is such a
danger to the hardware, should we even be doing it in Linux? We could
be setting it in U-Boot, because if the MFD driver fails to load or the
kernel freezes before the MFD driver loads there is STILL the risk you
could undervolt the battery, regardless of fixing this in the MFD driver
or the battery driver.

I believe 2.7V is still fine. The protection circuits may cut off before
that because they want to keep the longevity of the battery and there
isn't much juice left between 2.7V and 3.0V anyway. If the PMIC always
shuts down at 2.7V then there isn't anything to fear really, that's
still above the 2.5V that Wikipedia states as being the lower end of
the safe voltage range. While that's not the source to end all sources,
the PMIC having a lower cutoff probably means that some engineer did
think of this and pulled up the correct numbers.

Now that I've learned of this, I suspect there isn't really an issue
at all here. If we shut off at 2.7V we're good. Userspace should shut
down before then, and if there's battery protection that kicks in
before that, it makes really no difference to the end result, as
long as there is some cutoff somewhere

Thanks for your quick responses, my concerns regarding this are
alleviated. Devices that are already out there should be fine
in this regard.

Thank you.

quoted

Regards,
Nicolas Frattaroli

quoted

Thank you.

quoted

Regards,
Nicolas Frattaroli