options.rst

nvidia kernel module options provided by vgpu unlock patcher

As the patching has been moved to runtime, there are new parameters that can be used to tune the patches applied.

The options can be specified using a .conf file in /etc/modprobe.d or put into linux kernel's command line configured in bootloader, for example:

• using /etc/modprobe.d/nv.conf: options nvidia vgpukvm=1
• or in kernel command line: nvidia.vgpukvm=1

There are three kinds of options:

1. provided by source level patches

   • vgpukvm and kmalimit these two options are provided by vgpu-kvm-optional-vgpu-v2.patch

     the kmalimit is rather for testing, so do not use that unless you know what are you doing

     vgpukvm allows to disable vgpu-kvm mode of host driver, basically switching it back to behavior of normal consumer desktop driver, particularly when using the "merged" variant of patched driver

     this may be used to create secondary boot entry within bootloader in order to boot without vgpu support, allowing easy "switching" of nvidia driver without reinstall of any files

   • nvprnfilter basic filter for nvrm logs, to be used with verbose nvrm logging for debugging, like for example: options nvidia NVreg_ResmanDebugLevel=0 nvprnfilter=1

2. provided by custom hooks into nvidia blob

   • cudahost this was used with older versions of the vgpu unlock patcher to experimentally enable cuda support on host with vgpu kvm merged driver - now this option defaults to 0 with not merged driver and to 1 with a merged one, so you can use it to disable cuda with merged driver in case of some problems
   • vupdevid this is mainly for testing override of pci devid, most of the time it does not need to be touched
   • klogtrace this can be used to enable tracing of nvidia blob code, may be useful for development or for bugs analysis for example when comparing traces of a working case with a not working case

3. provided by runtime binary patches of the blob

   • vup_vgpusig based on patch from mbuchel to disable vgpu config signature check
   • vup_kunlock kernel driver level method of vgpu unlock which does not need any ioctl hooks in userspace
   • vup_merged controls the patch to enable display output on host with merged driver
   • vup_qmode optionally switch off the qmode
   • vup_sunlock based on patch from LIL'pingu fixing xid 43 crashes when running vgpu with consumer cards (not needed when vup_kunlock is used)

Be careful when trying to tune the above options, usually the defaults work best already and they may rather be useful for analysis of some problems.