AmpereOne Core PMU Perf Events Added For Linux 6.6

Merged overnight were the perf subsystem tooling updates for the Linux 6.6 merge window that is set to close later on Sunday.

Among the perf tools updates this cycle is support for recording kernel data maps when using "perf record --data", removing the old and experimental BPF events support from the perf trace code, RISC-V mmap support in libperf, a new "uprobe" benchmark for perf bench, and a variety of fixes and other tooling updates. Perf tool startup should also be faster now by lazily reading PMU, JSON, and sysfs data.

When it comes to all the vendor event identification based on JSON data for each processor family, Intel has updated their Sapphire Rapids and Meteor Lake event data. IBM has also updated its POWER10 perf data. Meanwhile being introduced for the first time by Ampere Computing is having the perf event data for their new AmpereOne AArch64 server processors.

It's believed that there is good AmpereOne support already upstream into the mainline Linux kernel, the "ampere1" target has been available in the GCC and LLVM/Clang compilers for a while, and all the fundamentals are in place though this addition of the AmpereOne perf event data in Linux 6.6 shows there may be some late additions. I don't yet have access to any AmpereOne hardware for checking on the Linux (and BSD) compatibility or performance benchmarks yet.

Those curious about all of the perf tooling changes for Linux 6.6 can see this Git merge for the full change details.