AMDKFD Has A Big Batch Of Improvements For The Mainline Linux Kernel

There hasn't been much to report on recently for Radeon's AMDKFD driver that serves as the kernel code for the Radeon GPU compute stack and part of the company's ROCm offering. AMDKFD work hasn't let up but has just been queuing for a while in the amd-kfd-staging Git branch and now there are a host of improvements to be mainlined.

Longtime AMD developer Felix Kuehling sent out a set of 27 patches on Sunday providing numerous improvements and fixes to AMDKFD that are now ready for the mainline kernel.

* Support for VegaM
* Support for systems with multiple PCI domains
* New SDMA queue type that's optimized for XGMI links
* SDMA MQD allocation changes to support future ASICs with more SDMA queues
* Fix for compute profile switching at process termination
* Fix for a circular lock dependency in MMU notifiers
* Fix for TLB flushing bug with XGMI enabled
* Fix for artificial GTT system memory limitation
* Trap handler updates

The Vega M support is notable for those with the Intel CPUs bearing Radeon graphics. The new SDMA queue type will be important moving forward with XGMI and EPYC 2 workstations/servers and the other optimizations and fixes are always welcome.

The patches are out there for testing. It remains to be seen if this work will still get pulled into DRM-Next for the upcoming Linux 5.2 cycle considering the period of new material to DRM-Next is largely over, but we'll see if it squeezes in otherwise is delayed until Linux 5.3. As well, come Linux 5.3 hopefully we'll see Navi support in tow.