Bug 2572845 - kernel 7.3 regression: with the same nvidia 620.70.06 driver, kernel 7.3.3 gives RTD3 "Not supported", 7.2.15 gave "Enabled (fine-grained)"
Summary: kernel 7.3 regression: with the same nvidia 620.70.06 driver, kernel 7.3.3 gi...
Keywords:
Status: NEW
Alias: None
Product: Fedora
Classification: Fedora
Component: kernel
Version: 44
Hardware: x86_64
OS: Linux
unspecified
high
Target Milestone: ---
Assignee: Justin M. Forbes
QA Contact: Fedora Extras Quality Assurance
URL:
Whiteboard:
Depends On:
Blocks:
TreeView+ depends on / blocked
 
Reported: 2026-11-12 21:58 UTC by William Bader
Modified: 2026-11-13 07:47 UTC (History)
15 users (show)

Fixed In Version:
Clone Of:
Environment:
Last Closed:
Type: ---
Embargoed:


Attachments (Terms of Use)
dmesg-nohost-7.3.3.txt (117.34 KB, text/plain)
2026-11-12 22:00 UTC, William Bader
no flags Details
nvidia-bug-report-7.3.3.log.gz (279.45 KB, application/gzip)
2026-11-12 22:01 UTC, William Bader
no flags Details
lspci-7.3.3.txt (1.72 KB, text/plain)
2026-11-12 22:02 UTC, William Bader
no flags Details
nvrm-debug-7.3.3.txt (238.22 KB, text/plain)
2026-11-12 22:04 UTC, William Bader
no flags Details
dmesg-nohost-7.2.15.txt (113.91 KB, text/plain)
2026-11-12 22:05 UTC, William Bader
no flags Details
nvidia-bug-report-7.2.15.log.gz (293.50 KB, application/gzip)
2026-11-12 22:05 UTC, William Bader
no flags Details
lspci-7.2.15.txt (1.72 KB, text/plain)
2026-11-12 22:06 UTC, William Bader
no flags Details
nvrm-debug-7.2.15.txt (246.22 KB, text/plain)
2026-11-12 22:07 UTC, William Bader
no flags Details

Description William Bader 2026-11-12 21:58:14 UTC
1. Please describe the problem:

Something changed between the 7.3 and 7.2 kernel to break the 620.70.06 nvidia driver. On 7.3, the driver compiles, but gets an error loading. The nvidia chip remains powered, which makes my laptop run hot, but it is not accessible, and X11 and chrome fall back to using the slower built-in Intel GPU.

2. What is the Version-Release number of the kernel:

7.3.3-200.fc44.x86_64

3. Did it work previously in Fedora? If so, what kernel version did the issue
   *first* appear?  Old kernels are available for download at
   https://koji.fedoraproject.org/koji/packageinfo?packageID=8 :

It worked on the final 7.2.15-200.fc44.x86_64 kernel, with both the 615 and 620 drivers. I built the 620 driver into the 7.2.15 to check whether the problem was the driver or the kernel.

4. Can you reproduce this issue? If so, please provide the steps to reproduce
   the issue below:

Booting with the 7.3.3 kernel makes the problem happen every time.
The problem happens with both cold and warm boots, including warm boots after shutting down 7.2.15, so it is something that the 7.3.3 kernel is doing, and not something random or uninitialized.

5. Does this problem occur with the latest Rawhide kernel? To install the
   Rawhide kernel, run ``sudo dnf install fedora-repos-rawhide`` followed by
   ``sudo dnf update --enablerepo=rawhide kernel``:


6. Are you running any modules that not shipped with directly Fedora's kernel?:

I have the nvidia driver.

7. Please attach the kernel logs. You can get the complete kernel log
   for a boot with ``journalctl --no-hostname -k > dmesg.txt``. If the
   issue occurred on a previous boot, use the journalctl ``-b`` flag.



Reproducible: Always

Comment 1 William Bader 2026-11-12 22:00:32 UTC
Created attachment 2163451 [details]
dmesg-nohost-7.3.3.txt

Comment 2 William Bader 2026-11-12 22:01:29 UTC
Created attachment 2163452 [details]
nvidia-bug-report-7.3.3.log.gz

Comment 3 William Bader 2026-11-12 22:02:15 UTC
Created attachment 2163453 [details]
lspci-7.3.3.txt

Comment 4 William Bader 2026-11-12 22:04:19 UTC
Created attachment 2163454 [details]
nvrm-debug-7.3.3.txt

temporarily added NVreg_ResmanDebugLevel=0 to /etc/modprobe.d/nvidia-pm.conf, rebooted,  and ran

sudo dmesg > ~/nvrm-debug.txt
grep -iE 'rtd3|gc6|_PR3|d3cold|dynamic power' ~/nvrm-debug.txt

Comment 5 William Bader 2026-11-12 22:05:08 UTC
Created attachment 2163455 [details]
dmesg-nohost-7.2.15.txt

Comment 6 William Bader 2026-11-12 22:05:50 UTC
Created attachment 2163456 [details]
nvidia-bug-report-7.2.15.log.gz

Comment 7 William Bader 2026-11-12 22:06:29 UTC
Created attachment 2163457 [details]
lspci-7.2.15.txt

Comment 8 William Bader 2026-11-12 22:07:12 UTC
Created attachment 2163458 [details]
nvrm-debug-7.2.15.txt

Comment 9 William Bader 2026-11-12 22:16:21 UTC
Here is more information. Let me know if there is anything else that I can test.

Hardware: Lenovo ThinkPad T15p Gen 3
GPU: GA107M [GeForce RTX 3050 Mobile] 10de:25a2 at 0000:01:00.0, behind root port 0000:00:01.0
iGPU: Intel Alder Lake-P GT2 (i915), PRIME render offload
Driver: NVIDIA open kernel module 620.70.06 (RPM Fusion akmod-nvidia)
Module option: NVreg_DynamicPowerManagement=0x02
OS: Fedora 44

Results with the same driver build, same module options, same firmware; only the kernel differs:

Kernel 7.2.15-200.fc44.x86_64, driver 620.70.06:
  /proc/driver/nvidia/gpus/0000:01:00.0/power: Runtime D3 status: Enabled (fine-grained)
  /sys/bus/pci/devices/0000:01:00.0/power/runtime_status: suspended

Kernel 7.3.3-200.fc44.x86_64, driver 620.70.06:
  /proc/driver/nvidia/gpus/0000:01:00.0/power: Runtime D3 status: Not supported
  /sys/bus/pci/devices/0000:01:00.0/power/runtime_status: active

With NVreg_DynamicPowerManagement left at the default (3), 7.3.3 reports "Disabled by default" instead.

Already ruled out on 7.3.3:
- The root port has ACPI _PR3 (power_resources_D3hot present in sysfs). With NVreg_ResmanDebugLevel=0 there is no "Failed to evaluate _PR3 object" message, and the NVRM debug output is identical on both kernels apart from addresses and timestamps.
- d3cold_allowed=1 on both the GPU and the root port; power/control=auto on both.
- A test module that replicates the sysfs config file checks in nv_get_pci_sysfs_config() (nv-pci.c) passes on both 7.2.15 and 7.3.3 with identical results, so nv_dynamic_power_available() should return true on both.
- The nvidia.ko builds for 7.2.15 and 7.3.3 import the same kernel symbols (apart from renamed allocator internals), including the runtime PM, ACPI and sysfs functions, so conftest results match.
- No hypervisor flag in /proc/cpuinfo.
- No relevant ACPI/PCI/PM kernel config differences between the two Fedora kernels.

The SBIOS PlatformRequestHandler assertions ("failed to get platform power mode from SBIOS") appear on both kernels and also appeared with 615.78.08 when RTD3 worked, so they do not appear to be related.

The open driver code paths I could check behave the same on both kernels, so the "Not supported" decision appears to come from the closed RM code, based on something that changed in kernel 7.3.

Workaround: booting kernel 7.2.15 with the same 620.70.06 driver built with `sudo akmods --kernels 7.2.15-200.fc44.x86_64 --force`

Comment 10 ArekR 2026-11-13 00:30:37 UTC
looks same as:
https://github.com/NVIDIA/open-gpu-kernel-modules/issues/1507

it nothing can be done on kernel side.

Comment 11 William Bader 2026-11-13 07:47:23 UTC
(In reply to ArekR from comment #10)

Thanks, the fix on that page worked for me.


Note You need to log in before you can comment on or make changes to this bug.