Re: [PATCH 0/3] nouveau: fix 3 null-ptr derefs

[Daniel Campos Ramos <Capitain_Jack@yahoo.com>]

Hi Jim,

I tested patches 1/3 and 2/3 on 7.3.0-rc1, with an RTX 3060 (GA106)
passed through to a VM (vfio-pci). Patch 3/3 does not apply on 7.3-rc1
(nvkm_gsp_gcx_ready() is not there yet), so it is untested here.

Without firmware (the whole nvidia/ga106 folder removed and the
initramfs rebuilt), nouveau reports acr, gr and sec2 "firmware
unavailable", so there is no GR engine:

- GETPARAM with NOUVEAU_GETPARAM_GRAPH_UNITS returns -ENODEV and logs
  "GETPARAM_GRAPH_UNITS: no gr engine or func"; the desktop session
  already reached that path at boot.
- DRM_IOCTL_NOUVEAU_GET_ZCULL_INFO returns -ENOTTY.

I also called both ioctls directly from a small test program on the
render node. No oops anywhere: KWin composites, NVK declines the device
and zink falls back to llvmpipe.

With the firmware restored, GSP-RM 570.144 loads, there is no oops, and
the card drives a Sony 3D TV in every HDMI 1.4 3D structure it declares,
at 12 bpc.

On your worry about firmware: the most "damaging" thing in all of this
testing was the TV getting stuck on a corrupted signal during an earlier
no-firmware run, solved by power-cycling it and feeding it a proper
signal. No GPU firmware is ever written, and the firmware files in the
VM are disposable.

The kernel logs, the test program and its output are here:
https://github.com/danielcamposramos/sony-bravia-linux/tree/main/docs/upstream/nouveau-null-v3-test-2026-10-03

Tested-by: Daniel Campos Ramos <Capitain_Jack@yahoo.com> # 1/3 and 2/3, GA106

The test was run with an AI assistant doing the legwork under my
direction; the results above are what I saw on the hardware.

Daniel