add portable cache irritators stress-ng coverage

Runner/config/pkg_command_map.conf

@@ -83,6 +83,18 @@ apt:gst-launch-1.0=gstreamer1.0-tools
 apt:gst-inspect-1.0=gstreamer1.0-tools
 apt:efivar=efivar
 
+# ---------------------------------------------------------------------------
+# Cache irritators userspace workload.
+#
+# Yocto remains image-managed. Debian, Ubuntu, and CentOS direct execution can
+# recover stress-ng through the native package provider. Runtime dependencies
+# are resolved by apt or dnf/yum as part of the same package transaction.
+# ---------------------------------------------------------------------------
+
+debian:package-set:cache-irritators=stress-ng
+ubuntu:package-set:cache-irritators=stress-ng
+centos:package-set:cache-irritators=stress-ng
+
 # ---------------------------------------------------------------------------
 # AI/ML TFLite and QNN package mappings.
 #

Runner/suites/Kernel/Stress/Cache_Irritators/Cache_Irritators.yaml

@@ -0,0 +1,25 @@
+metadata:
+  name: Cache_Irritators
+  format: "Lava-Test Test Definition 1.0"
+  description: "Exercise whole-cache, per-level cache, and TLB shootdown workloads with stress-ng"
+  os:
+    - linux
+  scope:
+    - functional
+    - performance
+
+params:
+  MODE: "nominal"
+  WORKERS: "1"
+  OPERATIONS: "auto"
+  TIMEOUT: "120"
+  CACHE_OPTIONS: "auto"
+  CACHE_SIZE: "auto"
+  CACHE_WAYS: "auto"
+
+run:
+  steps:
+    - REPO_PATH=$PWD
+    - cd Runner/suites/Kernel/Stress/Cache_Irritators
+    - ./run.sh --mode "${MODE}" --workers "${WORKERS}" --operations "${OPERATIONS}" --timeout "${TIMEOUT}" --cache-options "${CACHE_OPTIONS}" --cache-size "${CACHE_SIZE}" --cache-ways "${CACHE_WAYS}" || true
+    - $REPO_PATH/Runner/utils/send-to-lava.sh Cache_Irritators.res

Runner/suites/Kernel/Stress/Cache_Irritators/README.md

@@ -0,0 +1,124 @@
+# Cache irritators
+
+`Cache_Irritators` exercises CPU cache traffic and, when supported by the
+installed tool, TLB shootdowns using an image-provided `stress-ng` binary. It
+is intended for direct use on Yocto, Debian, Ubuntu, and CentOS targets.
+
+The source supplied for the original test is marked for internal use only and
+not for distribution. It also uses an ARM64 kernel module to issue privileged
+EL1 cache and TLB maintenance instructions. None of that restricted source is
+included or translated here. This suite is a public userspace replacement based
+on the stressors exposed by [stress-ng](https://github.com/ColinIanKing/stress-ng).
+
+The userspace replacement creates cache pressure over the complete detected
+cache topology and separately targets each runtime-discovered L1, L2, and L3
+level when `stress-ng` supports `--cache-level`. It also requests real
+kernel-mediated TLB shootdowns through mapping and protection changes.
+
+It does not claim to reproduce direct set/way cache maintenance. ARM64 cache
+set/way operations and direct TLBI instructions execute at EL1 and cannot be
+issued by a portable userspace test. Adding a runtime-built out-of-tree module
+would require matching kernel headers and toolchains on every target and would
+not be portable across the supported distributions.
+
+## Prerequisites and package policy
+
+- Yocto and other image-managed systems must provide `stress-ng` in the image.
+  The runner never invokes `opkg` or another package manager on these systems.
+- Debian and Ubuntu recover the `stress-ng` package with `apt` when it is
+  missing.
+- CentOS recovers the `stress-ng` package with `dnf` or `yum` when it is
+  missing.
+
+The package manager resolves the runtime dependencies of `stress-ng` in the
+same transaction. Package recovery uses the shared repository provider and the
+`cache-irritators` mappings in `Runner/config/pkg_command_map.conf`. No source
+is downloaded or compiled by this suite.
+
+Package recovery failure on Debian, Ubuntu, or CentOS is reported as `FAIL`.
+An image-managed or unsupported OS without `stress-ng` reports `SKIP` with the
+image prerequisite.
+
+The runner detects support from `stress-ng --help` instead of using distro or
+package-version gates:
+
+- `--cache` and `--cache-ops` are required for the primary workload.
+- `--cache-level` enables separate runtime-discovered L1, L2, and L3 workloads.
+- `--tlb-shootdown` and `--tlb-shootdown-ops` are optional. Their subcheck is
+  reported as `SKIP` when the installed build does not provide them.
+- Automatic cache tuning uses `--cache-enable-all` when advertised, matching
+  the recommendation emitted by the tested CentOS stress-ng 0.19.03 build.
+- Explicit cache tuning options, `--verify`, `--metrics-brief`, and the
+  stress-ng internal `--timeout` are used only when advertised.
+
+An outer managed watchdog always bounds each stressor.
+
+## Usage
+
+```text
+./run.sh [--nominal | --repeatability | --stress | --mode <mode>] [OPTIONS]
+
+Modes:
+  -n, --nominal          Run one operation unit per available workload (default)
+  -r, --repeatability    Run ten operation units per available workload
+  -s, --stress           Run one hundred operation units per available workload
+
+Options:
+  --workers <count>      stress-ng workers per supported stressor (default: 1)
+  --operations <count>   Override mode operation units (default: auto)
+  --timeout <seconds>    Maximum time per stressor (default: 120)
+  --cache-options <csv>  Cache flags: auto, none, enable-all, fence, flush,
+                         no-affinity, permute, prefetch (default: auto)
+  --cache-size <size>    Override stress-ng cache size, for example 4M
+  --cache-ways <count>   Limit the cache ways exercised
+  -h, --help             Show help and exit
+```
+
+Examples:
+
+```sh
+./run.sh
+./run.sh --repeatability
+./run.sh --stress --workers 2 --timeout 300
+./run.sh --mode nominal --cache-options prefetch,permute
+./run.sh --operations 25 --cache-size 4M --cache-ways 4
+```
+
+Only one mode may be selected. Worker and timeout values must be positive
+integers. `--operations` overrides the operation count selected by the mode.
+
+Custom cache flags are deliberately limited to cache-specific, non-destructive
+stress-ng options. The runner rejects unknown options and fails when an
+explicitly requested option is not advertised by the installed binary. Use
+`--cache-options none` to disable automatic cache tuning.
+
+## Yocto CI
+
+`Cache_Irritators.yaml` exposes the safe runner controls as LAVA parameters.
+Its default nominal run assumes `stress-ng` is already present in the Yocto
+image. The YAML does not enable package recovery.
+
+| YAML parameter | Runner option | Default |
+| --- | --- | --- |
+| `MODE` | `--mode` | `nominal` |
+| `WORKERS` | `--workers` | `1` |
+| `OPERATIONS` | `--operations` | `auto` |
+| `TIMEOUT` | `--timeout` | `120` |
+| `CACHE_OPTIONS` | `--cache-options` | `auto` |
+| `CACHE_SIZE` | `--cache-size` | `auto` |
+| `CACHE_WAYS` | `--cache-ways` | `auto` |
+
+## Results
+
+- `PASS`: an applicable stressor completed successfully, and no executed check
+  failed.
+- `FAIL`: an advertised stressor failed, timed out, reported an incomplete
+  workload, or relevant kernel errors were captured.
+- `SKIP`: an image-managed target lacks `stress-ng`, the required cache
+  stressor is unavailable, an optional TLB stressor is unavailable, or
+  kernel-log access is unavailable. A skipped optional subcheck does not
+  override a successful cache workload.
+
+Artifacts are retained in `logs_Cache_Irritators_<UTC timestamp>/` and include
+the stress-ng version and capability output, per-stressor logs, and the shared
+kernel-log scan files.