[GSD-12710] Missing zes_pwr_handle_t for iGPUs on Linux

[ProjectPhysX]

Pre-submission Checklist

• I am using the latest GPU driver version (releases)
• I have searched for similar issues and found none

GPU Hardware

Intel iGPUs (UHD 770, Alder Lake, Meteor Lake, Lunar Lake)

DRI Devices Information

GPU Detailed Information (lspci output)

Driver Version

26.14.37833.4

Installed GPU Driver Packages

No response

Driver Installation Details

Followed installation instructions from here: https://github.com/intel/compute-runtime/releases/tag/26.14.37833.4

Linux Distribution

Ubuntu 24.04 LTS

Other Linux Distribution

No response

Kernel Version & Boot Parameters

kernel 6.17.0-22-generic

Actual Behavior

Issue moved from oneapi-src/level-zero#441

Sysman power reporting on iGPUs (Alder Lake, Meteor Lake, Lunar Lake) works on Windows but is broken on Linux:

counter on Intel iGPUs	Windows	Linux
zesDeviceGetCardPowerDomain()	❌ returns ZE_RESULT_ERROR_UNSUPPORTED_FEATURE	❌ returns ZE_RESULT_ERROR_UNSUPPORTED_FEATURE
zesDeviceEnumPowerDomains()	✅ returns ZE_RESULT_SUCCESS and returns one valid zes_pwr_handle_t	❌ returns ZE_RESULT_SUCCESS but no zes_pwr_handle_t

Expected Behavior

iGPU power counters should also be available on Linux.

Reproduction Rate

Always reproduces - 100%

Steps to Reproduce

For debugging you may use https://github.com/ProjectPhysX/hw-smi

Is this a regression?

• Yes, this is a regression - functionality that previously worked is now broken

Last Known Working Driver Version

No response

First Known Failing Driver Version

No response

API Call Logs

No response

strace Logs

No response

System Logs / dmesg Output

No response

Backtrace (if crash or hang occurred)

No response

Source Code / Reproducer

No response

Command Line / Application Details

No response

oneAPI Version (if applicable)

No response

Screenshots / Video

No response

Additional Notes

No response

[SilentSobs]

Hey @ProjectPhysX,

I can confirm the Windows behavior - just tested hw-smi on my Windows laptop with Intel UHD Graphics (13th gen) and power monitoring works fine (showing 6/45W).

Quick diagnostic question - could you check if the hwmon directory exists on your iGPU systems?

# For each affected system (UHD 770, Meteor Lake, Lunar Lake):
ls -la /sys/class/drm/card*/device/hwmon/
cat /sys/class/drm/card*/device/hwmon/hwmon*/name 2>/dev/null

I'm curious if the Xe driver creates hwmon entries for integrated GPUs, since that's where the Linux power code looks for energy counters.

Thanks!

[ProjectPhysX]

Hi @SilentSobs,

on the UHD 770 system (with also A750 and B580 in the same system):

moritz@opencl-pc:~$ ls -la /sys/class/drm/card*/device/hwmon/
/sys/class/drm/card2/device/hwmon/:
total 0
drwxr-xr-x  3 root root 0 Mai 15 12:06 .
drwxr-xr-x 15 root root 0 Mai 15 12:01 ..
drwxr-xr-x  3 root root 0 Mai 15 12:01 hwmon4

/sys/class/drm/card3/device/hwmon/:
total 0
drwxr-xr-x  3 root root 0 Mai 15 12:06 .
drwxr-xr-x 22 root root 0 Mai 15 12:01 ..
drwxr-xr-x  3 root root 0 Mai 15 12:01 hwmon6
moritz@opencl-pc:~$ cat /sys/class/drm/card*/device/hwmon/hwmon*/name 2>/dev/null
i915
xe

[SilentSobs]

@ProjectPhysX

Thanks for checking! So the hwmon directory exists with i915, but power still fails. I suspect the hwmon might only have temperature files but not the actual power/energy monitoring files.

Could you run these commands to see what's actually in there?

# Show all files in UHD 770 hwmon
ls -la /sys/class/drm/card2/device/hwmon/hwmon4/

# Check specifically for power/energy files
ls -la /sys/class/drm/card2/device/hwmon/hwmon4/energy* 2>/dev/null || echo "NO energy files"
ls -la /sys/class/drm/card2/device/hwmon/hwmon4/power* 2>/dev/null || echo "NO power files"

# Compare with B580 (for reference)
ls -la /sys/class/drm/card3/device/hwmon/hwmon6/energy* 2>/dev/null
ls -la /sys/class/drm/card3/device/hwmon/hwmon6/power* 2>/dev/null

My guess is the i915 hwmon only exposes temperature monitoring for iGPUs, not power. If that's the case, the code would need a fallback to RAPL for package-level energy:

# Check if RAPL is available as alternative
ls -la /sys/class/powercap/intel-rapl/intel-rapl:0/energy_uj

This would explain why Windows works (different power query mechanism) while Linux fails.

[ProjectPhysX]

@SilentSobs somehow, after a reboot, /sys/class/drm/card2/device/hwmon/hwmon4 is gone. Or it moved to /sys/class/drm/card3/device/hwmon/hwmon5?
The two hwmon's are for A750 and B580.

moritz@opencl-pc:~$ lspci | grep "VGA"
00:02.0 VGA compatible controller: Intel Corporation Raptor Lake-S GT1 [UHD Graphics 770] (rev 04)
03:00.0 VGA compatible controller: Intel Corporation DG2 [Arc A750] (rev 08)
05:00.0 VGA compatible controller: NVIDIA Corporation GP102 [TITAN Xp] (rev a1)
76:00.0 VGA compatible controller: Intel Corporation Battlemage G21 [Arc B580]

moritz@opencl-pc:~$ ls -la /sys/class/drm/card*/device/hwmon/
/sys/class/drm/card3/device/hwmon/:
total 0
drwxr-xr-x  3 root root 0 Mai 15 13:45 .
drwxr-xr-x 15 root root 0 Mai 15 13:45 ..
drwxr-xr-x  3 root root 0 Mai 15 13:45 hwmon5

/sys/class/drm/card4/device/hwmon/:
total 0
drwxr-xr-x  3 root root 0 Mai 15 13:45 .
drwxr-xr-x 22 root root 0 Mai 15 13:45 ..
drwxr-xr-x  3 root root 0 Mai 15 13:45 hwmon6
moritz@opencl-pc:~$ ls -la /sys/class/drm/card3/device/hwmon/hwmon5/
total 0
drwxr-xr-x 3 root root    0 Mai 15 13:45 .
drwxr-xr-x 3 root root    0 Mai 15 13:45 ..
lrwxrwxrwx 1 root root    0 Mai 15 13:40 device -> ../../../0000:03:00.0
-r--r--r-- 1 root root 4096 Mai 15 13:40 energy1_input
-r--r--r-- 1 root root 4096 Mai 15 13:40 fan1_input
-r--r--r-- 1 root root 4096 Mai 15 13:40 in0_input
-r--r--r-- 1 root root 4096 Mai 15 13:40 name
drwxr-xr-x 2 root root    0 Mai 15 13:45 power
-rw-rw-r-- 1 root root 4096 Mai 15 13:40 power1_max
-rw-rw-r-- 1 root root 4096 Mai 15 13:45 power1_max_interval
-r--r--r-- 1 root root 4096 Mai 15 13:45 power1_rated_max
lrwxrwxrwx 1 root root    0 Mai 15 13:45 subsystem -> ../../../../../../../../class/hwmon
-r--r--r-- 1 root root 4096 Mai 15 13:40 temp1_input
-rw-r--r-- 1 root root 4096 Mai 15 13:45 uevent
moritz@opencl-pc:~$ ls -la /sys/class/drm/card4/device/hwmon/hwmon6/
total 0
drwxr-xr-x 3 root root    0 Mai 15 13:45 .
drwxr-xr-x 3 root root    0 Mai 15 13:45 ..
lrwxrwxrwx 1 root root    0 Mai 15 13:40 device -> ../../../0000:76:00.0
-r--r--r-- 1 root root 4096 Mai 15 13:40 energy1_input
-r--r--r-- 1 root root 4096 Mai 15 13:46 energy1_label
-r--r--r-- 1 root root 4096 Mai 15 13:40 energy2_input
-r--r--r-- 1 root root 4096 Mai 15 13:46 energy2_label
-r--r--r-- 1 root root 4096 Mai 15 13:40 fan1_input
-r--r--r-- 1 root root 4096 Mai 15 13:40 fan2_input
-r--r--r-- 1 root root 4096 Mai 15 13:40 fan3_input
-r--r--r-- 1 root root 4096 Mai 15 13:40 name
drwxr-xr-x 2 root root    0 Mai 15 13:46 power
-rw-rw-r-- 1 root root 4096 Mai 15 13:40 power1_cap
-rw-rw-r-- 1 root root 4096 Mai 15 13:45 power1_cap_interval
-rw-r--r-- 1 root root 4096 Mai 15 13:40 power1_crit
-r--r--r-- 1 root root 4096 Mai 15 13:46 power1_label
lrwxrwxrwx 1 root root    0 Mai 15 13:46 subsystem -> ../../../../../../../../class/hwmon
-r--r--r-- 1 root root 4096 Mai 15 13:40 temp2_input
-r--r--r-- 1 root root 4096 Mai 15 13:46 temp2_label
-r--r--r-- 1 root root 4096 Mai 15 13:40 temp3_input
-r--r--r-- 1 root root 4096 Mai 15 13:46 temp3_label
-rw-r--r-- 1 root root 4096 Mai 15 13:46 uevent

moritz@opencl-pc:~$ ls -la /sys/class/powercap/intel-rapl/intel-rapl:0/energy_uj
-r-------- 1 root root 4096 Mai 15 13:52 /sys/class/powercap/intel-rapl/intel-rapl:0/energy_uj

[SilentSobs]

Okay, so that explains it! After the reboot, the UHD 770 (card2/00:02.0) doesn't have a hwmon directory at all. So the i915 driver just isn't creating hwmon for the integrated GPU on your system.

Here's what we're seeing:

• A750 (i915 discrete): Has hwmon with energy1_input, power1_max - works fine
• B580 (xe discrete): Has hwmon with energy1_input, energy2_input, full power1_* files - also works
• UHD 770 (i915 iGPU): Nothing. No hwmon at all.

But there's a potential workaround - RAPL is there (/sys/class/powercap/intel-rapl/intel-rapl:0/energy_uj). That file gives package-level energy which includes both CPU and iGPU power.

One quick check - can you read it without sudo?

cat /sys/class/powercap/intel-rapl/intel-rapl:0/energy_uj

If it's accessible, we could add a fallback in LinuxPowerImp::init() - when the hwmon scan comes back empty for ZES_POWER_DOMAIN_PACKAGE on the main device, check if RAPL is available and read from there instead. That would basically replicate what Windows does (using alternate interfaces when the primary one isn't there).

Since RAPL reports package-level power anyway, it maps perfectly to what ZES_POWER_DOMAIN_PACKAGE is meant to represent - the whole package including the iGPU.

Does that sound like a reasonable direction? It should fix the UHD 770 and probably help with Meteor Lake and Lunar Lake too if they have the same issue.

[ProjectPhysX]

@SilentSobs sounds great, thank you a lot!
sudo is required:

moritz@opencl-pc:~$ cat /sys/class/powercap/intel-rapl/intel-rapl:0/energy_uj
cat: '/sys/class/powercap/intel-rapl/intel-rapl:0/energy_uj': Permission denied
moritz@opencl-pc:~$ sudo cat /sys/class/powercap/intel-rapl/intel-rapl:0/energy_uj
43260544220

[SilentSobs]

Thanks for testing! So RAPL does need root by default, which is a problem.

I looked into the i915 driver code and found why - the kernel explicitly disables hwmon power monitoring for integrated GPUs. Only discrete GPUs (DG1/DG2) get the power registers enabled. For iGPUs, all power/energy registers are set to INVALID.

Before proposing a workaround, could you check what RAPL domains are available on your system?

# Show all RAPL domains and their names
sudo ls -la /sys/class/powercap/intel-rapl*/
for d in /sys/class/powercap/intel-rapl*/*/name; do
    echo "$d: $(sudo cat $d 2>/dev/null)"
done

I'm checking if your UHD 770 exposes a separate GPU energy domain (sometimes called PP1 or "uncore"). If it does, we could use that for GPU-only power monitoring. If not, we're limited to package-level (CPU+GPU combined).

The permissions issue would require a one-time udev rule setup, which isn't ideal but is what tools like turbostat use. Want to see what domains exist first before discussing workarounds.

[ProjectPhysX]

@SilentSobs

moritz@opencl-pc:~$ sudo ls -la /sys/class/powercap/intel-rapl*/
/sys/class/powercap/intel-rapl/:
total 0
drwxr-xr-x 4 root root    0 Mai 15 21:22 .
drwxr-xr-x 3 root root    0 Mai 15 21:22 ..
-rw-r--r-- 1 root root 4096 Mai 15 21:22 enabled
drwxr-xr-x 5 root root    0 Mai 15 21:22 intel-rapl:0
drwxr-xr-x 2 root root    0 Mai 15 21:22 power
lrwxrwxrwx 1 root root    0 Mai 15 21:21 subsystem -> ../../../../class/powercap
-rw-r--r-- 1 root root 4096 Mai 15 21:22 uevent

'/sys/class/powercap/intel-rapl:0/':
total 0
drwxr-xr-x 5 root root    0 Mai 15 21:22 .
drwxr-xr-x 4 root root    0 Mai 15 21:22 ..
-r--r--r-- 1 root root 4096 Mai 15 21:22 constraint_0_max_power_uw
-r--r--r-- 1 root root 4096 Mai 15 21:22 constraint_0_name
-rw-r--r-- 1 root root 4096 Mai 15 21:22 constraint_0_power_limit_uw
-rw-r--r-- 1 root root 4096 Mai 15 21:22 constraint_0_time_window_us
-r--r--r-- 1 root root 4096 Mai 15 21:22 constraint_1_max_power_uw
-r--r--r-- 1 root root 4096 Mai 15 21:22 constraint_1_name
-rw-r--r-- 1 root root 4096 Mai 15 21:22 constraint_1_power_limit_uw
-rw-r--r-- 1 root root 4096 Mai 15 21:22 constraint_1_time_window_us
-r--r--r-- 1 root root 4096 Mai 15 21:22 constraint_2_max_power_uw
-r--r--r-- 1 root root 4096 Mai 15 21:22 constraint_2_name
-rw-r--r-- 1 root root 4096 Mai 15 21:22 constraint_2_power_limit_uw
-rw-r--r-- 1 root root 4096 Mai 15 21:22 constraint_2_time_window_us
lrwxrwxrwx 1 root root    0 Mai 15 21:22 device -> ../../intel-rapl
-rw-r--r-- 1 root root 4096 Mai 15 21:22 enabled
-r-------- 1 root root 4096 Mai 15 21:22 energy_uj
drwxr-xr-x 3 root root    0 Mai 15 21:22 intel-rapl:0:0
drwxr-xr-x 3 root root    0 Mai 15 21:22 intel-rapl:0:1
-r--r--r-- 1 root root 4096 Mai 15 21:22 max_energy_range_uj
-r--r--r-- 1 root root 4096 Mai 15 21:22 name
drwxr-xr-x 2 root root    0 Mai 15 21:22 power
lrwxrwxrwx 1 root root    0 Mai 15 21:21 subsystem -> ../../../../../class/powercap
-rw-r--r-- 1 root root 4096 Mai 15 21:22 uevent

'/sys/class/powercap/intel-rapl:0:0/':
total 0
drwxr-xr-x 3 root root    0 Mai 15 21:22 .
drwxr-xr-x 5 root root    0 Mai 15 21:22 ..
-r--r--r-- 1 root root 4096 Mai 15 21:22 constraint_0_max_power_uw
-r--r--r-- 1 root root 4096 Mai 15 21:22 constraint_0_name
-rw-r--r-- 1 root root 4096 Mai 15 21:22 constraint_0_power_limit_uw
-rw-r--r-- 1 root root 4096 Mai 15 21:22 constraint_0_time_window_us
lrwxrwxrwx 1 root root    0 Mai 15 21:22 device -> ../../intel-rapl:0
-rw-r--r-- 1 root root 4096 Mai 15 21:22 enabled
-r-------- 1 root root 4096 Mai 15 21:22 energy_uj
-r--r--r-- 1 root root 4096 Mai 15 21:22 max_energy_range_uj
-r--r--r-- 1 root root 4096 Mai 15 21:22 name
drwxr-xr-x 2 root root    0 Mai 15 21:22 power
lrwxrwxrwx 1 root root    0 Mai 15 21:22 subsystem -> ../../../../../../class/powercap
-rw-r--r-- 1 root root 4096 Mai 15 21:22 uevent

'/sys/class/powercap/intel-rapl:0:1/':
total 0
drwxr-xr-x 3 root root    0 Mai 15 21:22 .
drwxr-xr-x 5 root root    0 Mai 15 21:22 ..
-r--r--r-- 1 root root 4096 Mai 15 21:22 constraint_0_max_power_uw
-r--r--r-- 1 root root 4096 Mai 15 21:22 constraint_0_name
-rw-r--r-- 1 root root 4096 Mai 15 21:22 constraint_0_power_limit_uw
-rw-r--r-- 1 root root 4096 Mai 15 21:22 constraint_0_time_window_us
lrwxrwxrwx 1 root root    0 Mai 15 21:22 device -> ../../intel-rapl:0
-rw-r--r-- 1 root root 4096 Mai 15 21:22 enabled
-r-------- 1 root root 4096 Mai 15 21:22 energy_uj
-r--r--r-- 1 root root 4096 Mai 15 21:22 max_energy_range_uj
-r--r--r-- 1 root root 4096 Mai 15 21:22 name
drwxr-xr-x 2 root root    0 Mai 15 21:22 power
lrwxrwxrwx 1 root root    0 Mai 15 21:22 subsystem -> ../../../../../../class/powercap
-rw-r--r-- 1 root root 4096 Mai 15 21:22 uevent
moritz@opencl-pc:~$ for d in /sys/class/powercap/intel-rapl*/*/name; do
>     echo "$d: $(sudo cat $d 2>/dev/null)"
> done
/sys/class/powercap/intel-rapl:0:0/device/name: package-0
/sys/class/powercap/intel-rapl:0:1/device/name: package-0
/sys/class/powercap/intel-rapl:0/intel-rapl:0:0/name: core
/sys/class/powercap/intel-rapl:0/intel-rapl:0:1/name: uncore
/sys/class/powercap/intel-rapl/intel-rapl:0/name: package-0

[SilentSobs]

Perfect! The "uncore" domain exists - that's RAPL PP1 which should track the iGPU power specifically.

Can you measure it to confirm it's working?

# Measure uncore energy over 10 seconds (expecting ~60,000µJ increase for 6W idle)
echo "Reading uncore energy..."
sudo cat /sys/class/powercap/intel-rapl/intel-rapl:0/intel-rapl:0:1/energy_uj
sleep 10  
echo "After 10 seconds:"
sudo cat /sys/class/powercap/intel-rapl/intel-rapl:0/intel-rapl:0:1/energy_uj

If that shows reasonable values (~6W idle iGPU)

[ProjectPhysX]

@SilentSobs the energy counter increases every couple seconds

moritz@opencl-pc:~$ sudo cat /sys/class/powercap/intel-rapl/intel-rapl:0/intel-rapl:0:1/energy_uj
52673
moritz@opencl-pc:~$ sudo cat /sys/class/powercap/intel-rapl/intel-rapl:0/intel-rapl:0:1/energy_uj
55297
moritz@opencl-pc:~$ sudo cat /sys/class/powercap/intel-rapl/intel-rapl:0/intel-rapl:0:1/energy_uj
64697

[SilentSobs]

Perfect, the uncore counter is working and shows about 5W idle, which lines up with the 6W you saw on Windows.

So here's what I found:

The Problem:
The i915 kernel driver only creates hwmon power monitoring for discrete GPUs (DG1/DG2). For integrated GPUs like your UHD 770, it sets all the power registers to INVALID, so no hwmon files get created at all. That's why you're getting 0 power handles.

The specific code is in drivers/gpu/drm/i915/i915_hwmon.c - there's an if statement that checks for DG1 or DG2, and everything else gets blocked.

Workaround:
The RAPL uncore domain works on your system - that's what we just tested. It's at /sys/class/powercap/intel-rapl/intel-rapl:0/intel-rapl:0:1/energy_uj and it's tracking your iGPU power.

Problem is it needs root by default (security thing - CVE-2020-8694). You'd need to set up a udev rule to make it readable without sudo, which isn't ideal but it's what other tools like turbostat do. Also, not every iGPU platform has this domain available.

Long-term:
Really the kernel driver should just support iGPUs the same way it does for discrete cards. That would match what Windows does.

Anyway, that's what I found. Up to the Intel folks whether they want to add RAPL fallback support or fix the kernel driver.