Driver Power State Failure (0x9F) on Windows Server 2022 VM using netkvm on KVM

[sunhwakang]

Describe the bug
We are encountering recurring network disconnection and DRIVER_POWER_STATE_FAILURE (BugCheck 0x9F) issues on a Windows Server 2022 VM running on KVM with netkvm.

Issue Summary:
2024-11-07 19:10 – First occurrence
2024-11-10 17:35 – Issue reoccurred
Symptoms observed during each incident:

• Remote access and network services became unreachable
• The network interface became unresponsive and could not be restarted
• After rebooting the OS, the system hung during startup and required an additional forced reboot to boot normally, and the crash dump referenced below was collected during this event

Screenshots

Additional context
Crash Dump Analysis
BugCheck: 0x9F (DRIVER_POWER_STATE_FAILURE)

• The crash dump includes multiple NDIS-related function calls.

Key stack functions:

• ndis!InvokeNetPnPEvent
• ndis!NdisDeliverNetPnPEventSynchronously
• ndis!NdisPnPNotifyBinding

Summary of findings:

• The Idle thread (PID 0) was holding a PnP lock while waiting for an NDIS event, which never completed
• The thread waited over 10 minutes in a Net PnP–related kernel call
• NdisPnPRemoveDevice indicates that the system attempted to remove or rebind a network device but the operation did not complete
• This strongly suggests that the NIC driver (Virtio-Net) or an NDIS component became unresponsive during a PnP event
• Reference: https://learn.microsoft.com/en-us/windows-hardware/drivers/debugger/bug-check-0x9f--driver-power-state-failure

Question
We found what appears to be a similar issue and a related patch:

• Related Issue:
  netkvm.sys causes DRIVER_POWER_STATE_FAILURE (0x9F) BSOD on repeated hibernation cycles (Windows 10 22H2 guest) #1394
• Related Commit:
  30c6bdf

Could you confirm whether this commit has already been included in the Virtio-Net driver version 0.1.285 (released around September 2025)?
If not, is this fix still under validation or planned for a future release?

[YanVugenfirer]

1. I don't think that the patch is included in the released version as we didn't get feedback on it
2. Are you enabling S4 or S3 in the VM? What's the use case for this?
3. Can you please upload the crash dump itself?

@ybendito , @heywji - tagging for awareness

[sunhwakang]

It appears that the S3 power state is not available in this environment, as it is not supported by the system firmware.
I will need to verify whether Hibernate (S4) is being used.
In the meantime, I am sharing the crash dump file for your analysis.
https://kr.object.ncloudstorage.com/forshare/MEMORY.DMP

[ybendito]

@sunhwakang Which hypervisor do you use? The feature set of the network adapter does not look like qemu typically suggests, so I suppose this is something different. What exactly?
From the dump (!powertriage) we can see that both S3 and S4 are supported. But the system is not in transition (may be yet).
The stack looks like there is an attempt to remove the device (nekvm adapter), but there are a lot of queued TX packets held that are not sent, so the adapter can't be disabled right now, the NDIS is pending. There is nothing common with issue #1394 (this issue is related to RX packets).
There should be some reason why these packets are not sent, need to check deeper. Do I understand correctly that TX queue size is limited by 32 entries and the indirect descriptors are not supported?