[REGRESSION] - Multiple userspace implementations of battery estimate broken after "ACPI: battery: negate current when discharging"

[REGRESSION] - Multiple userspace implementations of battery estimate broken after "ACPI: battery: negate current when discharging"

[Matthew Schwartz]

Hello,

I installed kernel 6.15.4 to find that my battery estimate on my handheld gaming device was completely inaccurate, instead giving negative values and an unknown estimated battery life in multiple places. 

After bisecting, I landed on "ACPI: battery: negate current when discharging” as the bad commit. This commit breaks not one but several userspace implementations of battery monitoring: Steam and MangoHud. Perhaps it breaks more, but those are the two I have noticed so far. 

Thanks,
Matthew Schwartz

Re: [REGRESSION] - Multiple userspace implementations of battery estimate broken after "ACPI: battery: negate current when discharging"

[Rafael J. Wysocki]

Hi,

On Thu, Jul 3, 2025 at 3:55 AM Matthew Schwartz
<matthew.schwartz@linux.dev> wrote:
>
> Hello,
>
> I installed kernel 6.15.4 to find that my battery estimate on my handheld gaming device was completely inaccurate, instead giving negative values and an unknown estimated battery life in multiple places.
>
> After bisecting, I landed on "ACPI: battery: negate current when discharging” as the bad commit. This commit breaks not one but several userspace implementations of battery monitoring: Steam and MangoHud. Perhaps it breaks more, but those are the two I have noticed so far.

Thanks for letting me know, I'm queuing up a revert of this one.

Re: [REGRESSION] - Multiple userspace implementations of battery estimate broken after "ACPI: battery: negate current when discharging"

[Hans de Goede]

Hi Matthew,

On 3-Jul-25 3:54 AM, Matthew Schwartz wrote:
> Hello,
> 
> I installed kernel 6.15.4 to find that my battery estimate on my handheld gaming device was completely inaccurate, instead giving negative values and an unknown estimated battery life in multiple places. 
> 
> After bisecting, I landed on "ACPI: battery: negate current when discharging” as the bad commit. This commit breaks not one but several userspace implementations of battery monitoring: Steam and MangoHud. Perhaps it breaks more, but those are the two I have noticed so far. 

Thank you for reporting this.

As Rafael already indicated this patch will need to be reverted to
unbreak userspace.

But, the patch was actually doing the right thing, according to:

Documentation/ABI/testing/sysfs-class-power

What:           /sys/class/power_supply/<supply_name>/current_avg
Date:           May 2007
Contact:        linux-pm@vger.kernel.org
Description:
                Battery:
...
                Access: Read

                Valid values: Represented in microamps. Negative values are
                used for discharging batteries, positive values for charging
                batteries and for USB IBUS current.

(and the same for current_now)

and there are many power_supply fuel-gauge drivers (1) under
drivers/power/supply/ which do adhere to this specification
and report a negative current for discharging.

So if any of the userspace consumers of this API you mention
were to run on hw with these drivers the same problem will
be hit. Can you please file bugs against these userspace
projects so that they can fix this?

Regards,

Hans



1) For directly accessing fuel-gauge chips on devices where these
are directly accessible instead of being exposed through ACPI

Re: [REGRESSION] - Multiple userspace implementations of battery estimate broken after "ACPI: battery: negate current when discharging"

[Matthew Schwartz]

> On Jul 3, 2025, at 7:47 AM, Hans de Goede <hansg@kernel.org> wrote:
> 
> Hi Matthew,
> 
>> On 3-Jul-25 3:54 AM, Matthew Schwartz wrote:
>> Hello,
>> 
>> I installed kernel 6.15.4 to find that my battery estimate on my handheld gaming device was completely inaccurate, instead giving negative values and an unknown estimated battery life in multiple places.
>> 
>> After bisecting, I landed on "ACPI: battery: negate current when discharging” as the bad commit. This commit breaks not one but several userspace implementations of battery monitoring: Steam and MangoHud. Perhaps it breaks more, but those are the two I have noticed so far.
> 
> Thank you for reporting this.
> 
> As Rafael already indicated this patch will need to be reverted to
> unbreak userspace.
> 
> But, the patch was actually doing the right thing, according to:
> 
> Documentation/ABI/testing/sysfs-class-power
> 
> What:           /sys/class/power_supply/<supply_name>/current_avg
> Date:           May 2007
> Contact:        linux-pm@vger.kernel.org
> Description:
>                Battery:
> ...
>                Access: Read
> 
>                Valid values: Represented in microamps. Negative values are
>                used for discharging batteries, positive values for charging
>                batteries and for USB IBUS current.
> 
> (and the same for current_now)
> 
> and there are many power_supply fuel-gauge drivers (1) under
> drivers/power/supply/ which do adhere to this specification
> and report a negative current for discharging.
> 
> So if any of the userspace consumers of this API you mention
> were to run on hw with these drivers the same problem will
> be hit. Can you please file bugs against these userspace
> projects so that they can fix this?

Sure, I filed a bug with MangoHud last night about the kernel change breaking the current battery logic so I will mention this concern when updating that issue, and I will also file a new bug with Steam.

Thanks,
Matt

> 
> Regards,
> 
> Hans
> 
> 
> 
> 1) For directly accessing fuel-gauge chips on devices where these
> are directly accessible instead of being exposed through ACPI

Re: [REGRESSION] - Multiple userspace implementations of battery estimate broken after "ACPI: battery: negate current when discharging"

[Sebastian Reichel]

[-- Attachment #1: Type: text/plain, Size: 2490 bytes --]

Hi,

On Thu, Jul 03, 2025 at 08:51:10AM -0700, Matthew Schwartz wrote:
> > On Jul 3, 2025, at 7:47 AM, Hans de Goede <hansg@kernel.org> wrote:
> > Hi Matthew,
> > 
> >> On 3-Jul-25 3:54 AM, Matthew Schwartz wrote:
> >> Hello,
> >> 
> >> I installed kernel 6.15.4 to find that my battery estimate on
> >> my handheld gaming device was completely inaccurate, instead
> >> giving negative values and an unknown estimated battery life in
> >> multiple places.
> >> 
> >> After bisecting, I landed on "ACPI: battery: negate current
> >> when discharging” as the bad commit. This commit breaks not one
> >> but several userspace implementations of battery monitoring:
> >> Steam and MangoHud. Perhaps it breaks more, but those are the
> >> two I have noticed so far.
> > 
> > Thank you for reporting this.
> > 
> > As Rafael already indicated this patch will need to be reverted to
> > unbreak userspace.
> > 
> > But, the patch was actually doing the right thing, according to:
> > 
> > Documentation/ABI/testing/sysfs-class-power
> > 
> > What:           /sys/class/power_supply/<supply_name>/current_avg
> > Date:           May 2007
> > Contact:        linux-pm@vger.kernel.org
> > Description:
> >                Battery:
> > ...
> >                Access: Read
> > 
> >                Valid values: Represented in microamps. Negative values are
> >                used for discharging batteries, positive values for charging
> >                batteries and for USB IBUS current.
> > 
> > (and the same for current_now)
> > 
> > and there are many power_supply fuel-gauge drivers (1) under
> > drivers/power/supply/ which do adhere to this specification
> > and report a negative current for discharging.
> > 
> > So if any of the userspace consumers of this API you mention
> > were to run on hw with these drivers the same problem will
> > be hit. Can you please file bugs against these userspace
> > projects so that they can fix this?
> 
> Sure, I filed a bug with MangoHud last night about the kernel
> change breaking the current battery logic so I will mention this
> concern when updating that issue, and I will also file a new bug
> with Steam.

I also asked the Collabora team supporting Valve with the Steamdeck
to look into it. So I hope that it will be fixed soon :)

Thanks for coordinating this Hans. I fully agree with your analysis
and middle term plans to revert the revert :)

Greetings,

-- Sebastian

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 833 bytes --]

Re: [REGRESSION] - Multiple userspace implementations of battery estimate broken after "ACPI: battery: negate current when discharging"

[Peter Marheine]

I'm not surprised that there exist a number of userspace programs that
assume the buggy ACPI battery behavior is the only one, but this does
leave us in the previous situation where there's a clear bug in the
ACPI driver.

> But, the patch was actually doing the right thing, according to:
>
> Documentation/ABI/testing/sysfs-class-power

This is the key issue, since it's entirely plausible for a program
assuming non-negative battery current to run on a non-ACPI platform
and misbehave in the same way. If we're not going to fix the ACPI
driver to behave as specified for the kernel ABI, then the ABI needs
to be redefined to reflect the actual behavior. It's either that or we
give userspace an opportunity to fix itself (and I'm not sure exactly
how that would be done such that the clients which need to be fixed
discover that they need to be) and correct the driver's behavior
later.

Re: [REGRESSION] - Multiple userspace implementations of battery estimate broken after "ACPI: battery: negate current when discharging"

[Hans de Goede]

Hi Peter,

On 4-Jul-25 6:56 AM, Peter Marheine wrote:
> I'm not surprised that there exist a number of userspace programs that
> assume the buggy ACPI battery behavior is the only one, but this does
> leave us in the previous situation where there's a clear bug in the
> ACPI driver.
> 
>> But, the patch was actually doing the right thing, according to:
>>
>> Documentation/ABI/testing/sysfs-class-power
> 
> This is the key issue, since it's entirely plausible for a program
> assuming non-negative battery current to run on a non-ACPI platform
> and misbehave in the same way. If we're not going to fix the ACPI
> driver to behave as specified for the kernel ABI, then the ABI needs
> to be redefined to reflect the actual behavior. It's either that or we
> give userspace an opportunity to fix itself (and I'm not sure exactly
> how that would be done such that the clients which need to be fixed
> discover that they need to be) and correct the driver's behavior
> later.

Right, this is why I asked for bugs to be filed against the problematic
userspace programs, so that we can try to fix the ACPI driver again in
say 1 or 2 years from now.

In the mean time you could submit a patch to document the known broken
behavior of the ACPI battery driver in:

Documentation/ABI/testing/sysfs-class-power

with a big warning that userspace should not rely on this behavior.

You could even document how to work around this, e.g.:

"Negative currents are always discharging. Because of the broken ACPI
battery driver behaviour positive currents should be seen as discharging
rather then charging when the "status" sysfs attribute reports discharging."

Regards,

Hans