Currently the Linux kernel's "perf" performance monitoring subsystem has a limit on 2,048 CPU cores for its CPU map that is set by the MAX_NR_CPUS value. But that's becoming not enough in today's high core count era that patches are looking to raise it to a 4,096 CPU core limit by default.

HPE and Google engineers are hitting the 2,048 CPU core limit set by MAX_NR_CPUS for the perf CPU map. So along with some other improvements a pending patch series would raise the limit to 4,096 CPU cores.

While we are seeing high core count server CPUs these days from AMD, Intel, and Ampere, among others, in the case of exceeding 2,048 CPUs with this perf issue being encountered is reportedly being experienced on a 32 socket Xeon 6 "Granite Rapids" server that is hitting 2,048+ logical CPUs.

"Prompted by Kyle Meyer's report of the MAX_NR_CPUS value being too small, initiate some clean up of its use. Kyle's patch is at the head of the series. The additional patches hide MAX_NR_CPUS as exposed from cpumap.h, reduce its use by removing perf_cpu_map__read, and try to better size the temporary CPU array in perf_cpu_map__new.

...

Systems have surpassed 2048 CPUs. Increase MAX_NR_CPUS to 4096."

[1]This patch series is going into perf-tools-next and thus raising this limit should be in place for the upcoming Linux 6.14 kernel cycle.



[1] https://lore.kernel.org/lkml/20241206044035.1062032-1-irogers@google.com/