Intel Linux graphics driver engineers continue to be very busy enabling the Xe Direct Rendering Manager that is becoming the default kernel graphics driver beginning with Xe2 Lunar Lake and Battlemage hardware (it currently works as an experimental option with existing Intel graphics hardware going back to Tigerlake). The latest work coming out of Intel is their latest push on enabling GPU Shared Virtual Memory (SVM) support.

GPU Shared Virtual Memory is important for seamlessly sharing memory between the system's CPU and GPU for better performance and increased programming flexibility. Tacking SVM support onto the Xe driver has been a lengthy journey and out today are the latest "request for comments" (RFC) patches on this front.

Today's patches get the GPU SVM layer built for the Xe driver and exposing a new user-space API around the Shared Virtual Memory handling. The code in its initial form is working for GPU page faults for system allocations, runtime allocations, migration to/from vRAM, and unified eviction handling.

Intel engineers have successfully tested this GPU SVM code across Lunar Lake, Battlemage, and Ponte Vecchio hardware. Still to be tackled is multi-GPU support, user pointer "userptr" handling with GPU SVM, and madvise and prefetch IOCTL handling.

Those interested in this GPU Shared Virtual Memory work for the modern Intel Xe kernel graphics driver can check out [1]these 28 RFC patches for all the details.



[1] https://lore.kernel.org/dri-devel/20240828024901.2582335-1-matthew.brost@intel.com/