615 regression on GNOME/Wayland: every screen lock now sends an HDCP topology hotplug uevent for each TMDS connector (plus a drm_sysfs.c:497 WARN for DVI-D, which has no content-protection property). Combined with an older nvidia-drm leak when an EDID override applies to a disconnected connector, it runs the machine out of memory on the 4th lock of every boot. Full write-up, code references and a minimal sysfs reproducer: 615: every screen lock now sends HDCP hotplug uevents, which trigger an nvidia-drm EDID-override leak on disconnected connectors (kmalloc-128)
A second, smaller 615 regression: nvidia.ko’s module init now ends in wait_for_device_probe() (nv_pci_register_driver() → nv_pci_wait_for_probe_complete() in kernel-open/nvidia/nv-pci.c), which waits until every device in the system has finished probing, not just NVIDIA’s. For me it held the GPU behind a slow USB audio probe: the NVRM: loading line printed within 0.1 ms of snd-usb-audio finishing on all 8 boots with 615, ~6.5 s after the PCI probe instead of ~1.3 s on 610, so the login screen came up ~4 s later. Waiting only for NVIDIA’s own probes would avoid that.