Advance the Performance Analysis on RISC-V
summary
Ported LKP (Linux Kernel Performance) to RISC‑V and adds representative workloads and background monitors. Uses perf with QEMU/SBI patches to expose PMU events, plus ftrace and eBPF for observability, and outputs CSV/JSON for postanalysis. Comparative runs on multiple workloads including STREAM and Dhrystone show STREAM performing very poorly on the RISC‑V board, with low IPC and a memory‑bandwidth bottleneck; an application case spends >70% CPU in memcpy. Tracing finds 4 ms wake delays caused by an HR timer being disabled.
AI-written summary · about
chapters (15)
- 0:00Intro
- 0:17Outline
- 0:39Problem Statement
- 1:50Targeted Scenarios
- 3:00Framework-Goal
- 3:43The Chosen Framework-LKP
- 5:12Workloads - Goal
- 7:30Performance Tools
- 8:22The "perf" Tool
- 10:59Performance Comparison
- 12:03Performance Metrics - STREAM
- 13:35Hotspot Profiling-LZ4
- 14:44Tracing - Stress-NG Semaphore
- 15:55Next Step Work
- 17:14Summary
related talks
22mEnhancing RISC-V Performance Analysis Infrastructure
16mPerf feature improvements in RISCV
41mLinux Systems Performance
45mPerf tools updates and beyond
18mScheduler Tracing With ftrace + eBPF
37mAdvanced Performance Profiling Using Perf Tools
36mObservability and performance analysis with BPF
16mProcessor Trace in a Holistic World