56 Patches Volleyed For Improving Intel Linux Graphics Driver Scheduling
Longtime Intel open-source graphics driver developer Chris Wilson sent out a set of 56 patches today working to improve their kernel graphics driver's scheduling performance.
Among the 56 patches out today include implementing support for fair low-latency scheduling based in part on BFS/MuQSS kernel scheduler concepts. There is also new infrastructure for ring scheduling and other work helping with latency reductions and support going back to Gen6 graphics.
There is also a new knob for allowing administrators/users to specify whether the CPU should be prevented from entering higher C-states when the graphics driver is waiting on the CPU. This "dma_latency_ns" control via sysfs should help with reducing latency at the cost of keeping the CPU in higher performance states in some cases. The target DMA latency can be adjusted per-ring.
All 56 patches working on the i915 kernel driver scheduling improvements for now can be found on intel-gfx. This work is the latest on prior patches by Chris back during the summer but now revised for the latest state of the driver code-base. Hopefully this time around the patches are deemed in suitable shape for merging to an upcoming kernel cycle.
Among the 56 patches out today include implementing support for fair low-latency scheduling based in part on BFS/MuQSS kernel scheduler concepts. There is also new infrastructure for ring scheduling and other work helping with latency reductions and support going back to Gen6 graphics.
There is also a new knob for allowing administrators/users to specify whether the CPU should be prevented from entering higher C-states when the graphics driver is waiting on the CPU. This "dma_latency_ns" control via sysfs should help with reducing latency at the cost of keeping the CPU in higher performance states in some cases. The target DMA latency can be adjusted per-ring.
All 56 patches working on the i915 kernel driver scheduling improvements for now can be found on intel-gfx. This work is the latest on prior patches by Chris back during the summer but now revised for the latest state of the driver code-base. Hopefully this time around the patches are deemed in suitable shape for merging to an upcoming kernel cycle.
2 Comments