[Bug 297298] x11/nvidia-kmod-580, x11/nvidia-driver-580, x11/linux-nvidia-libs-580, graphics/nvidia-drm-*-kmod-580: Update to 580.178.04

https://bugs.freebsd.org/bugzilla/show_bug.cgi?id=297298

            Bug ID: 297298
           Summary: x11/nvidia-kmod-580, x11/nvidia-driver-580,
                    x11/linux-nvidia-libs-580,
                    graphics/nvidia-drm-*-kmod-580: Update to 580.178.04
           Product: Ports & Packages
           Version: Latest
          Hardware: amd64
                OS: Any
            Status: New
          Severity: Affects Only Me
          Priority: ---
         Component: Individual Port(s)
          Assignee: [email protected]
          Reporter: [email protected]
                CC: [email protected]
             Flags: maintainer-feedback?([email protected])
          Assignee: [email protected]
 Attachment #273487 maintainer-approval?([email protected])
             Flags:
                CC: [email protected]

Created attachment 273487
  --> https://bugs.freebsd.org/bugzilla/attachment.cgi?id=273487&action=edit
Patch to upgrade to 580.178.04

Update Legacy version 580 series of drivers to 580.178.04.

In document below:
 
https://docs.nvidia.com/datacenter/tesla/pdf/NVIDIA_Data_Center_GPU_Driver_Release_Notes_580_v12.0.pdf

  1.2. Fixed Issues
    ‣ When FM deletes a partition, it doesn’t release MC teams, as a result MC
Team
    resources (i.e., MLIDs are leaked). We now delete all outstanding MC teams
    associated with a partition while de-activating it.
    ‣ Multibank failures have been identified in HBM3E fleet and the current
driver is
    better optimized to handle these failures by triggering channel repair upon
the 2nd
    uncorrectable error in a pseudochannel.
    ‣ On self-hosted systems, the WPR region is mapped in RM reserved memory,
while
    the CPU never explicitly accesses the mapping. The existence of the mapping
could
    lead to CPU speculative access and kernel panic. The fix removes WPR from
reserved
    region mapping and avoids CPU speculative access.
    ‣ GSP unload path is changed to use fast unload instead of full teardown
sequence,
    skipping channel stop/teardown code which may result in a GSP hang/crash.
    ‣ Fixed a bug that could cause Xid 32 errors and application crashes when
scaling via
    CUDA_SCALE_LAUNCH_QUEUES (2x or 4x) with CUDA graphs that have a large
number of
    chained kernel nodes.

-- 
You are receiving this mail because:
You are on the CC list for the bug.
You are the assignee for the bug.