AMDGPU Linux 6.3 Addition To Help With Optimized Buffer Placement

On Friday AMD sent out another round of feature patches for new kernel graphics driver material they have readied in advance of the Linux 6.3 kernel cycle.

In prior weeks AMD sent in patches to DRM-Next for better dealing with missing firmware / unsupported GPUs and other "new stuff" for that next version of the Linux kernel. On Friday was another round of new AMDGPU/AMDKFD material to queue in DRM-Next until the Linux 6.3 merge window opens around mid-February.

The latest round of patches include fixes for Secure Display handling, DCN 3.2 display code fixes, SR-IOV fixes, HDCP fixes, RAS updates, fixes for GC 11.0 found with RDNA3 GPUs, and other updates.

Notable on the feature side is adding PCI Express information to the AMDGPU INFO ioctl. The PCI Express generation and number of lanes is to be exposed via the information ioctl to user-space.

The PCIe details are being added to this ioctl as from user-space the AMD Radeon Mesa drivers are looking at relying on this information for making improved determinations around optimized buffer placement. There is this pending merge request to Mesa that would allow making use of the PCIe information to be exposed by AMDGPU on Linux 6.3+.

This is also useful for troubleshooting any PCIe problems. In writing the kernel patch, well known open-source AMD developer Marek Olšák commented that he's been running a Navi 21 GPU at PCIe Gen 1 speeds for two years and recently was his first time noticing that obvious bottleneck.

More details on this week's AMDGPU feature patches slated for Linux 6.3 via this pull request. Linux 6.3 stable in turn should arrive around April.