New subject: [Bug 58397] Missing power_supply class attributes for NtPowerInformation blocks threads and causes fps stutter

20 Jun 2025

      http://bugs.winehq.org/show_bug.cgi?id=58397

            Bug ID: 58397
           Summary: Missing power_supply class attributes for
                    NtPowerInformation blocks threads and causes fps
                    stutter
           Product: Wine
           Version: unspecified
          Hardware: x86-64
                OS: Linux
            Status: UNCONFIRMED
          Severity: normal
          Priority: P2
         Component: ntdll
          Assignee: wine-bugs(a)winehq.org
          Reporter: rtdiasmendes(a)gmail.com
      Distribution: Mint

Created attachment 78805
  --> http://bugs.winehq.org/attachment.cgi?id=78805
logs mentioned in the bug Description

Affected Application:

 - Tainted Grail: Fall of Avalon

Observed Symptoms:

 - Frametime spikes from ~5ms to ~100ms occurring at ~1Hz intervals.

 - GPU frame dispatch (via DXVK) is delayed due to thread blocking in Wine’s
battery status polling logic.

Performance Analysis (see attached logs):

 - strace shows constant, repeated attempts to read nonexistent files in
/sys/class/power_supply/BAT1/

 - perf stat reveals ~8,270 openat() syscalls in 10 second window (on game
title screen).

 - perf record/report shows significant thread time consumed inside
NtPowerInformation-related logic in ntdll.

Root Cause:

 - In dlls/ntdll/unix/system.c, the function fill_battery_state() (Linux
version) hardcodes the use of attributes: "charge_now" "charge_full" and
"current_now".

 - These attributes are not universally present in /sys/class/power_supply/. On
some systems, particularly those using energy-based reporting (µWh), the
ABI-compliant alternatives are: "energy_now", "energy_full" and "power_now".

 - Since Wine looks only for the "charge_*" files and fails to return expected
information, the game loops the query constantly which results in openat()
being used in a tight polling loop - blocking threads and causing intermittent
CPU stalls that propagate to the GPU rendering pipeline.

Solution 1:
 - Verify what attributes are present in the power_supply class.
 - Implement logic with supplied attributes.

Solution 2:
 - Attempt to read the already-implemented files.
 - Fallback to "energy_*" if those are not present.

I gather solution 1 has a more consistent impact across different systems,
while solution 2 is more lightweight on systems with "charge_*" and heavier on
others.

Further reading:
 - https://docs.kernel.org/power/power_supply_class.html
 - https://www.kernel.org/doc/Documentation/ABI/testing/sysfs-devices-power
 - https://github.com/torvalds/linux/blob/master/include/linux/power_supply.h
 -
https://github.com/torvalds/linux/blob/master/drivers/power/supply/power_sup...

Note:

Attribute "scope", while implemented in the kernel, is undocumented in the
power_supply_class docs. Will later check if the rest of the documentation is
up to date and raise the appropriate issue upstream. Until then, do not assume
it correctly reflects the kernel implementation.

Disclaimer:

This issue was identified and reproduced using Proton-GE, Proton (Steam
Runtime), Soda and Flatpak's build. I am reporting it here for visibility and
discussion since it appears to originate from WineHQ’s upstream implementation
of NtPowerInformation.

Out-of-scope commentary:

While Wine's behavior of querying /sys/class/power_supply/ on every
NtPowerInformation() call correctly emulates the Windows API semantics, it has
unintended performance implications on Linux. This is due to the fact that
retrieving battery information from sysfs involves real filesystem I/O,
including frequent failed lookups on systems that expose energy_* instead of
charge_* attributes.

Therefore, applications may call NtPowerInformation() without prioritizing
caching or proper fallback logic as they expect the operation to have a
lower-cost then it does on Linux - causing significant and unnecessary
performance degradation.

Will raise this for an enchancement in the proper channels.

-- 
Do not reply to this email, post in Bugzilla using the
above URL to reply.
You are receiving this mail because:
You are watching all bug changes.

[Bug 58397] New: Missing power_supply class attributes for NtPowerInformation blocks threads and causes fps stutter

WineHQ Bugzilla

WineHQ Bugzilla

WineHQ Bugzilla

WineHQ Bugzilla

WineHQ Bugzilla

WineHQ Bugzilla

tags

participants (1)