fix(pve-nvidia-host): refuse early on Secure Boot without an enrolled DKMS key; record nh3-pve SB=on

persistent-memory.md

@@ -139,9 +139,12 @@ NH3 DNS are all fine. Open items:
 **nh3-ml1 (Prime, 2026-09-25 ~1410): IN PROGRESS.** CT 109 on nh3-pve at
 10.100.50.80 (VLAN 50, outside the UDM pool .150–.249), local-zfs, a second
 embed/rerank (TEI) backend on the new RTX. Playbooks are host-generic as of
-`bc278d4`. **Blocked:** the permission classifier refused the host driver install
+`bc278d4`. Prime cleared the driver install at ~1418. **It then failed: nh3-pve has
-(`pve-nvidia-host.yaml` on nh3-pve), and nothing host-side has run. Prime has to
+SECURE BOOT ON** (esh-pve has it off), and the unsigned DKMS module was refused.
-run it or allow it. After it: `gpu-lxc.yaml` with nh3 vars → embed-rerank with
+The installer rolled back. The playbook now refuses on this up front. Unblocking
+needs the console at boot (MokManager enroll, or SB off in the BIOS), which ties
+it to the IGFX/NanoKVM visit, unless Prime picks a VFIO VM. Waiting on his call.
+After it: `gpu-lxc.yaml` with nh3 vars → embed-rerank with
 `HOST_NAME/HOST_IP` → parity vs esh-ml1 → monitoring and DNS → gateway routing
 (Prime's call).
 

playbooks/pve-nvidia-host.yaml

@@ -33,6 +33,13 @@
 # module holds nothing and unloads cleanly. The unload step refuses if nouveau
 # has actually bound a device; then the blacklist plus a reboot is the only way.
 #
+# ⚠ SECURE BOOT: a DKMS-built module only loads under Secure Boot if its signing
+# key is enrolled as a MOK, and enrolling needs a person at the console at boot
+# (MokManager). esh-pve has Secure Boot OFF. nh3-pve has it ON (lockdown =
+# integrity), and the 2026-09-25 install failed there with "module failed to load".
+# The installer rolled itself back. The first step below now refuses early
+# instead of getting that far.
+#
 # Upgrading the driver: bump driver_version + driver_sha256 here AND
 # driver_version in playbooks/gpu-lxc.yaml, run this on every GPU host, then
 # gpu-lxc.yaml for every GPU LXC. Until both have run, CUDA in the container
@@ -50,6 +57,18 @@ vars:
   gpu_match: RTX 2000
 
 steps:
+  - name: Pre-flight — Secure Boot off, or the DKMS signing key already enrolled
+    shell: |
+      if mokutil --sb-state 2>/dev/null | grep -q 'SecureBoot enabled'; then
+        if [ -f /var/lib/dkms/mok.pub ] && mokutil --test-key /var/lib/dkms/mok.pub 2>&1 | grep -q 'already enrolled'; then
+          exit 0
+        fi
+        echo "Secure Boot is ON and no enrolled DKMS key: the nvidia module would be refused." >&2
+        echo "Needs the console at boot: disable Secure Boot in the BIOS, or enroll /var/lib/dkms/mok.pub via MokManager." >&2
+        exit 1
+      fi
+    changed_when: "false"
+
   # DKMS needs the headers for the RUNNING kernel (built now) and the series
   # meta-package (so each future kernel's headers arrive with it and DKMS
   # rebuilds on upgrade). nh3-pve had neither, nor dkms or a compiler.

servers/nh3-pve/README.md

@@ -90,6 +90,14 @@ not power back on by itself.
   Display = IGFX (or enable iGPU Multi-Monitor). Reaching the BIOS now needs a
   display on the RTX's mini-DP, for example the NanoKVM through a mini-DP→HDMI
   adapter. Verify with `lspci | grep 00:02.0` and `boot_vga` on `00:02.0`.
+- ⚠ **Secure Boot is ON here** (`mokutil --sb-state`, lockdown `integrity`;
+  enrolled MOK = the Proxmox Secure Boot CA). esh-pve, the same model, has it
+  OFF. Any DKMS-built module, the NVIDIA driver included, is refused until its
+  key is enrolled through MokManager at boot or Secure Boot is turned off in the
+  BIOS. Both need the console, and the console is blind (below). The 2026-09-25
+  NVIDIA install failed on this and rolled itself back. What it left, all
+  harmless: headers `6.8.12-11` plus the series meta, dkms and build-essential;
+  nouveau blacklisted and unloaded; the `.run` staged in `/root/nvidia`.
 - **OOB decision (Prime via Miranda, 2026-09-25): HOLD. nh3-pve stays console-blind
   until the next NH3 site visit.** No BIOS change and no reboot until then. The
   accepted risk: a boot without network means a site trip. **Target end state:**