Hi, I am often getting overflow (OVF) packets from Intel PT on a small Goldmont Plus processor, and I'm wondering exactly what is overflowing. Is it some kind of on-chip buffer, or is it the RAM allocation that was configured by the operating system to receive the PT packets? It would be great to have a diagram that gives a precise overview of the dataflow stages for tracing on Linux via perf_event_open(). Here's my setup:
- Mini PC
- CPU model Gemini Lake
- 8GB RAM
- Fedora 29
- decoding via Intel libipt (pt_packet_decoder only)
I also get overflows on my Xeon and i7 machines when running larger benchmarks, such as the SPEC 2017 reference data sets. Is there anything I can do to improve the reliability of the tracing, or is it simply subject to overflow on occasion? Thanks for your help.
Byron
