Merge pull request qualcomm-linux#145 from smuppand/stress-tools

Add stress-ng and stressapptest validation scripts

Author: vnarapar

Runner/suites/Kernel/Stress/Stress-ng/README.md

@@ -0,0 +1,308 @@
+Scheduler/Stress Validation — stress-ng Runner
+
+This README explains how to use the stress-ng–based validation script (run.sh) we wrote to exercise CPU, memory, I/O, and scheduler paths on embedded Linux systems (Yocto, Debian/Ubuntu, RT & non-RT kernels, NUMA/non-NUMA). It also covers how to get stress-ng onto your target (cross-compile or sideload).
+
+---
+
+What this test does
+
+Launches stress-ng stressors sized to the current machine (online CPUs, RAM, and free disk) so we don’t overcommit tiny embedded boards.
+
+Affines worker threads to every online CPU to make scheduler regressions obvious.
+
+Applies fail criteria (max latency, OOM, I/O errors, stressor non-zero exits); returns non-zero exit code on failure for CI.
+
+Saves a short summary and optional detailed logs; runs a dmesg scan via your functestlib.sh.
+
+---
+
+Requirements
+
+stress-ng binary on the target
+
+Standard tools: awk, grep, sed, cut, tr, sleep, date, head, getconf
+
+(Optional) taskset, numactl for CPU pinning/NUMA; dd for I/O prechecks
+
+Your test framework’s init_env and functestlib.sh (already handled by run.sh)
+
+The runner reuses helpers from your existing functestlib.sh:
+
+check_dependencies
+
+find_test_case_by_name
+
+log_info, log_warn, log_pass, log_fail, log_skip, log_error
+
+scan_dmesg_errors
+
+---
+
+Getting stress-ng
+
+Project: https://github.com/ColinIanKing/stress-ng
+
+A) Native install (Debian/Ubuntu)
+
+sudo apt-get update
+sudo apt-get install -y stress-ng
+
+B) Cross-compile (Yocto)
+
+Add to your image or build it as an SDK tool:
+
+In your layer, ensure stress-ng is available (meta-openembedded has a recipe in meta-oe on many branches).
+
+Add to image:
+
+IMAGE_INSTALL:append = " stress-ng"
+
+Rebuild image / SDK:
+
+bitbake core-image-minimal
+
+C) Cross-compile (generic cmake/make)
+
+On your host:
+
+git clone https://github.com/ColinIanKing/stress-ng.git
+cd stress-ng
+make CROSS_COMPILE=aarch64-linux-gnu-  # or your triplet
+# artifact is src/stress-ng
+
+Copy the binary to your target (see “Sideload” below).
+
+D) Android / BusyBox targets (sideload)
+
+Push a statically linked stress-ng:
+
+adb push stress-ng /usr/local/bin/
+adb shell chmod 755 /usr/local/bin/stress-ng
+
+Or with SSH:
+
+scp stress-ng root@TARGET:/usr/local/bin/
+ssh root@TARGET chmod 755 /usr/local/bin/stress-ng
+
+---
+
+run.sh quick start
+
+From the test case directory (the script finds its own path via find_test_case_by_name):
+
+./run.sh
+
+By default, it:
+
+Detects online CPUs, total RAM, and free disk.
+
+Picks safe defaults: worker threads == online CPUs, memory workers sized to a small percentage of RAM, I/O workers sized to free space.
+
+Runs for a sane duration (e.g., 5–10 minutes configurable).
+
+Fails on stressor non-zero exit, OOM, major I/O error, or dmesg anomalies.
+
+## Usage
+
+```
+Usage: ./run.sh [--p1 <sec>] [--p2 <sec>] [--mem-frac <pct>] [--disk-frac <pct>]
+                [--cpu-list <list>] [--temp-limit <degC>] [--stressng "<args>"]
+                [--repeat <N>] [--help]
+```
+
+### Options
+
+| Option              | Description |
+|---------------------|-------------|
+| `--p1 <sec>`        | Phase 1 duration in seconds (default: 60) |
+| `--p2 <sec>`        | Phase 2 duration in seconds (default: 60) |
+| `--mem-frac <pct>`  | Percentage of total memory per worker (default: 15) |
+| `--disk-frac <pct>` | Percentage of free disk space per worker (default: 5) |
+| `--cpu-list <list>` | Comma-separated list or range of CPUs to stress |
+| `--temp-limit <degC>` | Maximum temperature threshold |
+| `--stressng "<args>"` | Additional arguments passed to stress-ng |
+| `--repeat <N>`      | Repeat the entire test sequence N times (default: 1) |
+| `--help`            | Show this help message and exit |
+
+> Exact flags may differ slightly depending on your final script; the examples below assume the version we discussed (auto-sizing, affinity, fail criteria, reuse of functestlib.sh).
+
+---
+
+Example invocations
+
+1) Quick CPU & memory smoke (auto sizing, 5 min)
+
+./run.sh --duration 300 --stressors cpu,vm
+
+2) Full platform shake (CPU+VM+I/O; pinned per-CPU)
+
+./run.sh --duration 600 --stressors cpu,vm,io --logs
+
+3) Limit footprint on small RAM systems
+
+./run.sh --duration 180 --stressors vm --mem-pct 5
+
+4) Pin workers to a subset of CPUs
+
+./run.sh --cpu-list 0-3 --duration 240 --stressors cpu
+
+5) Exercise only I/O with conservative disk usage
+
+./run.sh --stressors io --disk-pct 3 --duration 120
+
+6) Mixed with latency guardrail (if cyclic path is enabled)
+
+./run.sh --stressors cpu,vm --max-latency-us 500 --duration 300
+
+7) Run with default phases, repeated 3 times
+
+./run.sh --repeat 3
+
+8) Run on specific CPUs with temperature limit
+
+./run.sh --cpu-list 0-3 --temp-limit 80
+
+9) Run memory-intensive workload for 90 seconds per phase
+
+./run.sh --mem-frac 30 --p1 90 --p2 90
+
+10) Run stress-ng with a custom workload twice
+
+./run.sh --repeat 2 --stressng "--cpu 4 --timeout 30 --verify"
+
+---
+
+What the script checks/fails on
+
+stressor exit codes (any non-zero → FAIL)
+
+Killed by OOM or ENOMEM patterns in stress-ng output → FAIL
+
+I/O failures (EIO, read/write errors) → FAIL
+
+dmesg anomalies via scan_dmesg_errors → WARN/FAIL as configured
+
+(Optional) latency threshold if you also run a small cyclic step
+
+Exit code:
+
+0 = PASS (no failures, at least one stressor ran)
+
+1 = FAIL (functional failure or threshold exceeded)
+
+2 = SKIP (dependencies missing)
+
+Artifacts:
+
+stress-ng-summary.log (always)
+
+stress-ng-*.log files (with --logs)
+
+*.res result file for your harness
+
+---
+
+Sizing & affinity logic (how it stays safe)
+
+CPU workers: ≤ online CPUs (default: one worker per online CPU)
+
+Memory workers: uses a small percentage of total RAM (cap per worker), adjustable via --mem-pct
+
+I/O workers: uses a small percentage of free disk (cap per worker), adjustable via --disk-pct
+
+Affinity: default is on (each worker pinned to a specific online CPU); disable with --no-affine
+
+NUMA: if numactl exists, the script prefers local node binding where appropriate; otherwise it simply CPU-affines.
+
+---
+
+Building stress-ng into your products
+
+Yocto (image integration)
+
+Add to your image recipe or local.conf:
+
+IMAGE_INSTALL:append = " stress-ng"
+
+Rebuild and flash your image.
+
+Debian/Ubuntu rootfs
+
+Bake into your rootfs recipe or install at first boot with a provisioning script:
+
+apt-get update && apt-get install -y stress-ng
+
+Sideload in CI
+
+For CI smoke on development hardware:
+
+scp stress-ng root@TARGET:/usr/local/bin/
+ssh root@TARGET chmod 755 /usr/local/bin/stress-ng
+
+---
+
+Cross-compiling notes & tips
+
+On ARM64 build hosts with Linaro/GCC toolchains:
+
+make CROSS_COMPILE=aarch64-linux-gnu-
+file src/stress-ng   # confirm aarch64 ELF
+
+Prefer static if your target is minimal:
+
+make static
+
+Validate dependencies: run src/stress-ng --version on the host and then on the target after copy.
+
+---
+
+Troubleshooting
+
+“stress-ng: command not found”
+Not on PATH. Install natively, or place it in /usr/local/bin and chmod +x.
+
+Out-of-memory or system lockups
+Lower --mem-pct, shorten --duration, drop io on small/flash media.
+
+I/O errors / read-only filesystems
+Switch to a writable mount (e.g., /tmp) or adjust --disk-pct down to 1–2%.
+
+High kernel latency on PREEMPT_RT
+Start with CPU-only tests, then introduce memory/I/O slowly; use --max-latency-us to gate.
+
+BusyBox environments
+Ensure the script’s dependencies exist (the runner checks and SKIPs otherwise). You can pre-install missed tools or adjust the stress mix.
+
+---
+
+Security & safety
+
+This script is destructive only in its I/O scratch area (e.g., under /tmp/stress-ng-io); it won’t touch other files.
+
+It will refuse to over-allocate RAM/disk beyond configured caps.
+
+Still, run on development hardware or staging boards when possible.
+
+---
+
+License
+
+The test runner: BSD-3-Clause-Clear (Qualcomm Technologies, Inc. and/or its subsidiaries).
+
+stress-ng is licensed upstream by its author; see its repository for details.
+
+---
+
+Appendix: Useful stress-ng commands (manual)
+
+See available stressors:
+
+stress-ng --class cpu --sequential 1 --metrics-brief --timeout 10
+
+Run with maximum stress on all classes (dangerous on small boards):
+
+stress-ng --aggressive --all 1 --timeout 60
+
+Only memory:
+
+stress-ng --vm 4 --vm-bytes 5% --vm-keep --timeout 120