fix: rmmod unloads nvidia before dependent nvidia-peermem

[andrewwhitecdw]

This PR addresses the following issue in rhel8/nvidia-driver: rmmod unloads nvidia before dependent nvidia-peermem.

Changes

• rhel8/nvidia-driver: rmmod unloads nvidia before dependent nvidia-peermem.

Details

--- a/rhel8/nvidia-driver
+++ b/rhel8/nvidia-driver
@@ -1,14 +1,14 @@
-    if [ -f /sys/module/nvidia_uvm/refcnt ]; then
-        nvidia_uvm_refs=$(< /sys/module/nvidia_uvm/refcnt)
-        rmmod_args+=("nvidia-uvm")
-        ((++nvidia_deps))
-    fi
-    if [ -f /sys/module/nvidia/refcnt ]; then
-        nvidia_refs=$(< /sys/module/nvidia/refcnt)
-        rmmod_args+=("nvidia")
-    fi
-    if [ -f /sys/module/nvidia_peermem/refcnt ]; then
-        nvidia_peermem_refs=$(< /sys/module/nvidia_peermem/refcnt)
-        rmmod_args+=("nvidia-peermem")
-        ((++nvidia_deps))
-    fi
+    if [ -f /sys/module/nvidia_uvm/refcnt ]; then
+        nvidia_uvm_refs=$(< /sys/module/nvidia_uvm/refcnt)
+        rmmod_args+=("nvidia-uvm")
+        ((++nvidia_deps))
+    fi
+    if [ -f /sys/module/nvidia_peermem/refcnt ]; then
+        nvidia_peermem_refs=$(< /sys/module/nvidia_peermem/refcnt)
+        rmmod_args+=("nvidia-peermem")
+        ((++nvidia_deps))
+    fi
+    if [ -f /sys/module/nvidia/refcnt ]; then
+        nvidia_refs=$(< /sys/module/nvidia/refcnt)
+        rmmod_args+=("nvidia")
+    fi

Tests

• rhel8/tests/test_unload_order.sh

--- /dev/null
+++ b/rhel8/tests/test_unload_order.sh
@@ -0,0 +1,20 @@
+#!/bin/bash
+set -eu
+
+script_dir=$(dirname "$(readlink -f \"$0\")")
+driver_file=\"$script_dir/../nvidia-driver\"
+
+peermem_line=$(grep -n 'rmmod_args+=(\"nvidia-peermem\")' \"$driver_file\" | head -n1 | cut -d: -f1 || true)
+nvidia_line=$(grep -n 'rmmod_args+=(\"nvidia\")' \"$driver_file\" | head -n1 | cut -d: -f1 || true)
+
+if [[ -z \"$peermem_line\" ]] || [[ -z \"$nvidia_line\" ]]; then
+    echo \"FAIL: could not locate rmmod_args entries\"
+    exit 1
+fi
+
+if [[ \"$peermem_line\" -gt \"$nvidia_line\" ]]; then
+    echo \"FAIL: nvidia-peermem is added after nvidia in rmmod_args\"
+    exit 1
+fi
+
+echo \"PASS\"

Contributor guidelines

Per this repo's CONTRIBUTING.md:

• All commits are signed off (Signed-off-by trailer, DCO).

[copy-pr-bot]

This pull request requires additional validation before any workflows can run on NVIDIA's runners.

Pull request vetters can view their responsibilities here.

Contributors can view more details about this message here.