From d2932dbcbe1e8d98754dc34620e122adf6f39c97 Mon Sep 17 00:00:00 2001
From: Lou MassKonFuzion Herard
 <6656037+masskonfuzion@users.noreply.github.com>
Date: Wed, 21 Aug 2019 01:58:13 -0500
Subject: [PATCH] Replace sleep BUS_RESCAN_WAIT_SEC with lspci-based GPU check;
 tweak MODULES_LOAD in config

---
 config/nvidia-xrun | 11 +++++++----
 nvidia-xrun        | 25 +++++++++++++++++++++++--
 2 files changed, 30 insertions(+), 6 deletions(-)

diff --git a/config/nvidia-xrun b/config/nvidia-xrun
index c13363e..a241bc1 100644
--- a/config/nvidia-xrun
+++ b/config/nvidia-xrun
@@ -19,11 +19,14 @@ CONTROLLER_BUS_ID=0000:00:01.0
 # Bus ID of the graphic card
 DEVICE_BUS_ID=0000:01:00.0
 
-# Seconds to wait before turning on the card after PCI devices rescan
-BUS_RESCAN_WAIT_SEC=1
-
 # Ordered list of modules to load before running the command
-MODULES_LOAD=(nvidia nvidia_uvm nvidia_modeset "nvidia_drm modeset=1")
+# >>>>> ORIGINAL
+#MODULES_LOAD=(nvidia nvidia_uvm nvidia_modeset "nvidia_drm modeset=1")
+
+# >>>>> MASSKONFUZION: "nvidia_drm modeset=1" seems to cause sporadic general
+# protection faults on my machine. At a minimum, perhaps the "modeset=1"
+# option can be documented as a parameter to tweak?
+MODULES_LOAD=(nvidia nvidia_uvm nvidia_modeset nvidia_drm)
 
 # Ordered list of modules to unload after the command exits
 MODULES_UNLOAD=(nvidia_drm nvidia_modeset nvidia_uvm nvidia)
diff --git a/nvidia-xrun b/nvidia-xrun
index 56d9855..ae5d602 100755
--- a/nvidia-xrun
+++ b/nvidia-xrun
@@ -41,8 +41,27 @@ function turn_on_gpu {
   if [[ ! -d /sys/bus/pci/devices/${DEVICE_BUS_ID} ]]; then
     echo 'Rescanning PCI devices'
     execute "sudo tee /sys/bus/pci/rescan <<<1"
-    echo "Waiting ${BUS_RESCAN_WAIT_SEC} second for rescan"
-    execute "sleep ${BUS_RESCAN_WAIT_SEC}"
+
+    # >>>>> MASSKONFUZION: This comment for review; feel free to delete:
+    # This change proposes to do away with BUS_RESCAN_WAIT_SEC, in favor of
+    # waiting for the NVIDIA GPU to appear in lspci. The code, as written,
+    # assumes the GPU will appear. But perhaps a more robust solution (not
+    # implemented as of yet) would be to fail if the GPU does not appear
+    # after a given timeout interval.
+    if [[ ${DRY_RUN} -eq 1 ]]; then
+        echo '>> Dry run. Waiting for NVIDIA video device'
+    else
+        echo 'Waiting for NVIDIA video device'
+        RESULT="1"
+        while [[ ${RESULT} -ne 0 ]]; do
+            echo -n '.'
+            lspci | grep -v Audio | grep NVIDIA > /dev/null
+            RESULT=$?
+            sleep 1
+        done
+        echo 'Woo-hoo! Got it :fist pump:'
+        lspci | grep -v Audio | grep NVIDIA
+    fi
   fi
 
   echo 'Turning the card on'
@@ -54,6 +73,7 @@ function load_modules {
   do
     echo "Loading module ${module}"
     execute "sudo modprobe ${module}"
+    sleep 1
   done
 }
 
@@ -62,6 +82,7 @@ function unload_modules {
   do
     echo "Unloading module ${module}"
     execute "sudo modprobe -r ${module}"
+    sleep 1
   done
 }