The performance “perf” events changes for the Linux 7.0 kernel are continuing to prepare for next-generation Xeon Diamond Rapids processors as the successor to current Xeon 6 Granite Rapids.
There already has been a lot of kernel upstreaming in preparing for Diamond Rapids along with the other open-source enablement such as to the GCC and LLVM/Clang compilers. With Linux 7.0 the performance events integration is landing.
The performance events code adds the core performance monitoring unit (PMU) for Diamond Rapids to support its different counters and metrics capabilities. Diamond Rapids brings “a lot of changes” on the PMU side with a new OFF-MODULE RESPONSE “OMR” facility replacing the Off-Core Response “OCR” found on existing Xeon processors. There is also a new Processor Event Based Sampling “PEBS” data source encoding layout that needs to be adapted in the kernel plus enabling support for a new RDPMC user-disable feature.
The performance events code also brings a large patch series for uncore PMU support on Diamond Rapids. These patches deal with Diamond Rapids’ two integrated I/O and memory hub dies that are separate from the compute tile dies. Plus different discovery changes, several new performance monitoring types (SCA, HAMVF, D2D_ULA, UBR, PCIE4, CRS, CPC, ITC, OTC, CMS, and PCIE6), and other changes.
The Linux 7.0 changes also add some missing performance monitoring units for Panther Lake plus enabling support for Nova Lake too. The Intel C-State driver is also adapted to support the low-power Wildcat Lake SoCs too.
More details on these performance event changes for Linux 7.0 via this pull request that has already been merged to the Linux kernel Git.
