llcstat [-h] [-c SAMPLE_PERIOD] [duration]
llcstat instruments CPU cache references and cache misses system-side, and summarizes them by PID and CPU. These events have different meanings on different architecture. For x86-64, they mean misses and references to LLC. This can be useful to locate and debug performance issues caused by cache hit rate.
This works by sampling corresponding events defined in uapi/linux/perf_event.h, namely PERF_COUNT_HW_CACHE_REFERENCES and PERF_COUNT_HW_CACHE_MISSES, using BPF perf event tracing. Upon each sampled event, the attached BPF program records the PID and CPU ID on which the event happened, and stores it in table.
This makes use of a Linux 4.9 feature (BPF_PROG_TYPE_PERF_EVENT).
Since this uses BPF, only the root user can use this tool.
CONFIG_BPF and bcc.
Print usage message.
- -c SAMPLE_PERIOD
Sample one in this many cache reference and cache miss events.
Duration to trace, in seconds.
- Sample one in 100 events, trace for 20 seconds:
# llcstat -c 100 20
Number of cache reference events
Number of cache miss events
Cache hit ratio
This is from bcc.
Also look in the bcc distribution for a companion _examples.txt file containing example usage, output, and commentary for this tool.
Unstable - in development.
- Perf can be used as a generic event counter tool. An example for LLC:
# perf top -e cache-misses -e cache-references -a -ns pid,cpu,comm