Same hardware/software as in my other thread (LOQ 15IRX10, RTX 5050 Laptop, driver 595.91.07, BIOS graphics mode Dynamic).
Observed (timestamps relative to the AC change, from a sysfs/udev/journal trace, attached):
Unplug AC: ACPI power_supply change at 0 s; about 1.2 s later “nvidia 0000:01:00.0: Enabling HDA controller” and runtime status “resuming”; audio function 01:00.1 and DRM nodes removed; “[nvidia-drm] Removing device / Unloading driver”; PCI device 01:00.0 removed about 2 s after the AC change. Kernel also logs “pciehp: Slot(1): Card not present” and an acpi_os_execute_deferred workqueue warning.
Plug AC: PCI device re-added about 0.45 s after the AC change, nvidia-drm loads at about 0.8 s, DRM nodes appear at about 2.3 s, and the two SBIOS assertions follow.
With nvidia-powerd running at unplug time: “NVRM: Attempting to remove device 0000:01:00.0 with non-zero usage count!” and afterward nvidia-smi reports no devices until AC is reconnected.
The GPU is not absent after a battery boot (present, runtime-suspended); it disappears only on the AC-to-battery transition.
Questions: Is this the expected behavior for Dynamic mode on this machine, driven by firmware? Is the removal-with-nonzero-usage-count path safe? Should nvidia-powerd release the device on surprise removal?
Not tested: other BIOS modes, older drivers, Windows.
Attached: the trace logs, the journal excerpt around the “non-zero usage count” message, the nvidia-bug-report.log.gz.