Intel's Mesa Drivers Begin Preparing For The New Xe Kernel Driver
One of the open-source Intel Linux graphics driver milestones we have to look forward to this year is the introduction of the new "Xe" kernel graphics driver to effectively succeed the existing "i915" Direct Rendering Manager driver for recent generations of Intel graphics. More prep code was merged this week to Mesa's Intel "ANV" Vulkan driver in preparing to be able to make use of that new kernel mode driver once its upstreamed into the Linux kernel.
The in-development Intel Xe kernel graphics driver will support Gen12/Xe Graphics and newer -- starting with Tigerlake integrated graphics and up through current DG2/Alchemist discrete graphics cards plus future Intel integrated and discrete graphics processors. The Xe kernel driver will allow them to focus on modernizing their kernel driver code-base, avoid the baggage of all the generations of Intel integrated graphics back to the Intel 915 chipset days, and make other optimizations and improvements moving forward with just having to worry about modern Intel graphics hardware.
Intel's user-space driver code has to be updated accordingly to be able to make use of the Xe kernel driver while also retaining support for the i915 kernel driver. In January some Mesa code began landing to abstract out the interfaces for supporting both the Xe and i915 KMDs.
Merged today to the Intel Vulkan driver was the basic kernel mode driver back-end infrastructure. This is part -- and a step toward -- the larger effort for allowing full Intel Xe kernel driver support by the Mesa drivers
This merge to Mesa 23.1 sets up the KMD infrastructure so the Vulkan driver can interface with both the i915 and Xe drivers. See this MR for the larger tracking effort of supporting the Intel Xe driver with Mesa.
The Intel Xe kernel driver remains under active development and hasn't yet been upstreamed into the mainline kernel. It wasn't submitted to DRM-Next yet either for v6.3, so not until v6.4 or later will it be merged. When it comes land it will likely be treated as experimental/opt-in for some time as well until it's proven itself and without any regressions for end-users on modern Intel graphics hardware. In any event this should be a very exciting year on the open-source Intel Linux graphics front.