Bluefield2 provides several performance counters via mlx_pmc driver under /sys/class/hwmon and Performance Monitoring Counters - NVIDIA Docs provides methods to program and access them. I would like to know if there is any way of accessing these performance counters in per-process granualarity. BTW, I have tried perf stat command, which can show per-process performance counters, but it does not have L3 cache information, while under /sys/class/hwmon, there are some L3 cache counters, but they are system-wide, not per-process. Thanks!
likely so far, linux kernel perf only support ARMV7 and no L3 counter.
I am not sure if ARMv8 PMU has L3 counter per process. that need ARMV8 A72 core manual. and also need kernel driver to implement.
According to Cortex A72 manual, there is no L3 cache-related counters for Cortex A72, which is why perf stat will not work if we want to view L3 cache-related counters. Then my question is if there is any method to access the counters under /sys/class/hwmon(the counters are provided by mlx_pmc driver and include L3 cache-related ones, PCIe, etc.) at per-core/per-process granualrity so we can access L3 cache counters even if perf stat does not provide them. I have tried to program and view these counters based on the instructions in Performance Monitoring Counters - NVIDIA Docs, but theses counters seem to be system-wide, is there any methods that I haven’t noticed to help me view them for each core/each process, or just simply no methods to do this? Thanks.
This topic was automatically closed 14 days after the last reply. New replies are no longer allowed.