Re: nvidia-304xx is broken

On Thu, Feb 04, 2016 at 06:55:11PM -0700, Robin Laing wrote: > On 31/01/16 16:29, Ed Greshko wrote: >>> I wrote the inquiry below to the fedora users maillinglist >>> but there was no response except a suggestion to ask here, >>> so this is it. >> >> I recently had a problem with the nvidia-340xx. I would only get a blank screen with no X >> being started. I did not check the logs I just did what has worked for me in the past. >> >> 1. dnf erase *nvidia* >> 2. reboot (You could either boot to gui using nouveau or multi-user mode) >> 3. dnf install akmod-nvidia-340xx (In your case akmod-nvidia-304xx >> 4. reboot >> >> >>> >>> Since writing the text below, another Fedora 23 >>> kernel-4.3.3-303.fc23.x86_64 made an appearance, along with a new >>> kmod-nvidia-304xx-4.3.3-303.fc23.x86_64-304.131-1.fc23.x86_64 >>> but they exhibit the identical fault. >>> >>> ****** >>> The latest two kernels and the nvidia-304xx module are incompatible. >>> That is, the nvidia.ko module built by akmod-nvidia-304xx won't load. >>> That means that my three machines with older nvidia video cards (one >>> built into the mobo) are stuck running the old >>> >>>> kernel-4.2.8-300.fc23.x86_64 >>> kernel-4.3.3-300.fc23.x86_64 >>> kernel-4.3.3-301.fc23.x86_64 >>> >>> In the past, incompatibilities were resolved in about a week. >>> This time, it's nearly a month. No new package of akmod-nvidia-304xx >>> or kmod-nvidia-304xx has shown up at the >>> rpmfusion-nonfree-updates[-testing] repo. >>> >>> The akmod-nvidia-340xx module on another machine that needs it runs fine. >>> >>> Does anyone have a clue when this older version of the nvidia module >>> might be fixed? >>> >>> Here are the error messages with the newer kernel: >>> >>> Console error message when trying to run 'startxfce': >>> >>> (==) Using system config directory "/usr/share/X11/xorg.conf.d" >>> modprobe: ERROR: could not insert 'nvidia': Unknown symbol in module, >>> or unknown parameter (see dmesg) >>> >>> >>> From dmesg: >>> >>> ... >>> [ 12.563315] nvidia: module license 'NVIDIA' taints kernel. >>> [ 12.563325] Disabling lock debugging due to kernel taint >>> [ 12.573867] nvidia: Unknown symbol mtrr_del (err 0) >>> [ 12.573977] nvidia: Unknown symbol mtrr_add (err 0) >>> ... >>> [ 31.162308] nvidia: Unknown symbol mtrr_del (err 0) >>> [ 31.162397] nvidia: Unknown symbol mtrr_add (err 0) >>> ... >>> [ 81.656643] nvidia: Unknown symbol mtrr_del (err 0) >>> [ 81.656736] nvidia: Unknown symbol mtrr_add (err 0) >>> >>> >> > > I tried akmods for 340 on F23 and it doesn't work either. It won't > compile. > > Testing from a terminal using akmods --force to rebuild them, I found that > there is an error in the kernel source that won't work. > > I will start a different thread about it. > > Robin Robin: I'm glad to see (from your other thread) that you were able to fix your nvidia-340xx setup. As I said above, my experience with that version has been completely satisfactory. It is only the nvidia-304xx module that is, and remains, broken. Perhaps your experience demonstrates why fresh installs should be preferred over upgrades. Today my brother's machine in Italy that I maintain via an ipsec tunnel also failed with the nvidia-304xx package and a dnf update. To my surprise I found his video card is a GeForce 605, which purports to be new enough to be supported by the mainstream nvidia driver. Sure enough, when I installed the akmod-nvidia-358.16-1.fc22.x86_64, it worked. So my experiments confirm that akmod-nvidia-358.16-1.fc22.x86_64 - works akmod-nvidia-340xx-340.96-1.fc23.x86_64 - works akmod-nvidia-304xx-304.131-1.fc23.x86_64 - fails