current: X session won't start with nvidia driver and kernel 5.4.7

Hi there,

I installed the nvidia kernel driver on my box with Slackware current (kernel 5.4.7) manually, i.e. using the .run file. I installed the legacy 390.132 version from https://www.nvidia.com/Download/driv...x/153832/en-us because I have a GeForce 750M, which is considered legacy. During the installation I didn't get any errors and I allowed nvidia-xconfig to create an xorg.conf file.

When i run startx I get this error:

[ 149.549] (EE) No devices detected.
[ 149.549] (EE)
Fatal server error:
[ 149.549] (EE) no screens found(

I have attached boch /var/log/Xorg.0.log and /etc/X11/xorg.conf

Thanks

Perhaps, the 340.108 driver that was released on 23 December 2019? It also supports the 700M series.

Originally Posted by cwizardone Perhaps, the 340.108 driver that was released on 23 December 2019? It also supports the 700M series.

Thanks. I installed 340.108 and I got the same error. The only suspicious thing I see in /var/log/nvidia-installer.log is

-> Error.
-> Unable to build the Unified Memory kernel module.
-> The Unified Memory kernel module failed to build. Driver installation will pr
oceed without this module.

I attach the new /var/log/Xorg.0.log and /etc/X11/xorg.conf, and I pasted /var/log/nvidia-installer.log for NVIDIA-Linux-x86_64-340.108.run in https://justpaste.it/2p7gt

Thanks.

The nVidia-uvm kernel module version 340.108 will build if you apply this patch from Arch Linux:

  1 diff --git a/uvm/nvidia_uvm_lite.c b/uvm/nvidia_uvm_lite.c
  2 index 6943e7c..c5d232c 100644
  3 --- a/uvm/nvidia_uvm_lite.c
  4 +++ b/uvm/nvidia_uvm_lite.c
  5 @@ -820,7 +820,13 @@ done:
  6  }
  7
  8  #if defined(NV_VM_OPERATIONS_STRUCT_HAS_FAULT)
  9 +#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 17, 0)
 10 +vm_fault_t _fault(struct vm_fault *vmf)
 11 +#elif LINUX_VERSION_CODE >= KERNEL_VERSION(4, 11, 0)
 12 +int _fault(struct vm_fault *vmf)
 13 +#else
 14  int _fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 15 +#endif
 16  {
 17  #if defined(NV_VM_FAULT_HAS_ADDRESS)
 18      unsigned long vaddr = vmf->address;
 19 @@ -828,8 +834,15 @@ int _fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 20      unsigned long vaddr = (unsigned long)vmf->virtual_address;
 21  #endif
 22      struct page *page = NULL;
 23 +#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 17, 0)
 24 +    vm_fault_t retval;
 25 +#else
 26      int retval;
 27
 28 +#endif
 29 +#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 11, 0)
 30 +   struct vm_area_struct *vma = vmf->vma;
 31 +#endif
 32      retval = _fault_common(vma, vaddr, &page, vmf->flags);
 33
 34      vmf->page = page;
 35 @@ -868,7 +881,13 @@ static struct vm_operations_struct uvmlite_vma_ops =
 36  // it's dealing with anonymous mapping (see handle_pte_fault).
 37  //
 38  #if defined(NV_VM_OPERATIONS_STRUCT_HAS_FAULT)
 39 +#if LINUX_VERSION_CODE >= KERNEL_VERSION(4, 17, 0)
 40 +vm_fault_t _sigbus_fault(struct vm_fault *vmf)
 41 +#elif LINUX_VERSION_CODE >= KERNEL_VERSION(4, 11, 0)
 42 +int _sigbus_fault(struct vm_fault *vmf)
 43 +#else
 44  int _sigbus_fault(struct vm_area_struct *vma, struct vm_fault *vmf)
 45 +#endif
 46  {
 47      vmf->page = NULL;
 48      return VM_FAULT_SIGBUS;

I'm still on kernel 4.19.y so I haven't tested this running kernel 5.4.y.
I'm using the nvidia-legacy340-kernel.SlackBuild modified like this by ponce then the above patch shouldn't be needed:

55 (cd kernel || exit 1 56 make SYSSRC=$KERNELPATH module || exit 1 57 cd uvm 58 sed -i "s|-O2$|-O2 -Wno-error=incompatible-pointer-types|" Makefile 59 make SYSSRC=$KERNELPATH module || exit 1 60 )

Please note that nVidia 340.xxx driver doesn't work with libglvnd (requires 390.xxx or later).
Mats

Originally Posted by mats_b_tegner The nVidia-uvm kernel module version 340.108 will build if you apply this patch from Arch Linux:

thank you for your reply and sorry for the delayed reply.

I have posted this problem on the nvidia form to see if someone can help me with that.

Please note that nVidia 340.xxx driver doesn't work with libglvnd (requires 390.xxx or later). Mats

Do I have to remove libglvnd or the 340.xxx driver will simply not use it?

Thanks.

Originally Posted by gnw thank you for your reply and sorry for the delayed reply. I have posted this problem on the nvidia form to see if someone can help me with that. Do I have to remove libglvnd or the 340.xxx driver will simply not use it? Thanks.

Yes, I had to remove/blacklist libglvnd and blacklist/recompile mesa without libglvnd support since the 340.xxx driver series doesn't support libglvnd. Otherwise, the X-server will crash on startup.

Based on this, I reinstalled the nvidia 390.132 driver but this time I didn't run nvidia-xconfing. startx successfully starts the graphics environment but without using the nvidia driver (nvida-settings shows that the driver is not being used).

Based on info from the Arch wiki and /usr/doc/NVIDIA_GLX-1.0 I created a xorg.conf file manually and now I am getting and error related to prime synchronisation not being available. That bug has been reported but not yet fixed, although someone posted a patch that reportedly works for nvidia 440.44 and not 390.132. I will see if I can do a similar patch myself.

xorg.conf:

Section "ServerLayout"
	Identifier "layout"
	Screen 0 "nvidia"
	inactive "intel"
EndSection

Section "Device"
	Identifier "nvidia"
	Driver "nvidia"
	BusID "01:00:0"
EndSection

Section "Screen"
	Identifier "nvidia"
	Device "nvidia"
	option "AllowEmptyInitialConfiguration"
EndSection

Section "Device"
	Identifier "intel"
	Driver "modesetting"
EndSection

Section "Screen"
	Identifier "intel"
	Device "intel"
EndSection

~/.xinitrc:

xrandr --setprovideroutputsource modesetting NVIDIA-0
xrandr --auto

Thanks.