Bpf/optimized usdt ci

.github/actions/veristat_baseline_compare/action.yml

@@ -0,0 +1,49 @@
+name: 'run-veristat'
+description: 'Run veristat benchmark'
+inputs:
+  veristat_output:
+    description: 'Veristat output filepath'
+    required: true
+  baseline_name:
+    description: 'Veristat baseline cache name'
+    required: true
+runs:
+  using: "composite"
+  steps:
+    - uses: actions/upload-artifact@v4
+      with:
+        name: ${{ inputs.baseline_name }}
+        if-no-files-found: error
+        path: ${{ github.workspace }}/${{ inputs.veristat_output }}
+
+    # For pull request:
+    # - get baseline log from cache
+    # - compare it to current run
+    - if: ${{ github.event_name == 'pull_request' }}
+      uses: actions/cache/restore@v4
+      with:
+        key: ${{ inputs.baseline_name }}
+        restore-keys: |
+          ${{ inputs.baseline_name }}-
+        path: '${{ github.workspace }}/${{ inputs.baseline_name }}'
+
+    - if: ${{ github.event_name == 'pull_request' }}
+      name: Show veristat comparison
+      shell: bash
+      run: ./.github/scripts/compare-veristat-results.sh
+      env:
+        BASELINE_PATH: ${{ github.workspace }}/${{ inputs.baseline_name }}
+        VERISTAT_OUTPUT: ${{ inputs.veristat_output }}
+
+    # For push: just put baseline log to cache
+    - if: ${{ github.event_name == 'push' }}
+      shell: bash
+      run: |
+        mv "${{ github.workspace }}/${{ inputs.veristat_output }}" \
+           "${{ github.workspace }}/${{ inputs.baseline_name }}"
+
+    - if: ${{ github.event_name == 'push' }}
+      uses: actions/cache/save@v4
+      with:
+        key: ${{ inputs.baseline_name }}-${{ github.run_id }}
+        path: '${{ github.workspace }}/${{ inputs.baseline_name }}'

.github/scripts/compare-veristat-results.sh

@@ -0,0 +1,18 @@
+#!/bin/bash
+
+if [[ ! -f "${BASELINE_PATH}" ]]; then
+    echo "# No ${BASELINE_PATH} available" >> "${GITHUB_STEP_SUMMARY}"
+
+    echo "No ${BASELINE_PATH} available"
+    echo "Printing veristat results"
+    cat "${VERISTAT_OUTPUT}"
+
+    exit
+fi
+
+selftests/bpf/veristat \
+    --output-format csv \
+    --emit file,prog,verdict,states \
+    --compare "${BASELINE_PATH}" "${VERISTAT_OUTPUT}" > compare.csv
+
+python3 ./.github/scripts/veristat_compare.py compare.csv

.github/scripts/download-gcc-bpf.sh

@@ -0,0 +1,30 @@
+#!/bin/bash
+
+set -euo pipefail
+
+GCC_BPF_RELEASE_GH_REPO=$1
+INSTALL_DIR=$(realpath $2)
+
+cd /tmp
+
+tag=$(gh release list -L 1 -R ${GCC_BPF_RELEASE_GH_REPO} --json tagName -q .[].tagName)
+if [[ -z "$tag" ]]; then
+    echo "Could not find latest GCC BPF release at ${GCC_BPF_RELEASE_GH_REPO}"
+    exit 1
+fi
+
+url="https://github.com/${GCC_BPF_RELEASE_GH_REPO}/releases/download/${tag}/${tag}.tar.zst"
+echo "Downloading $url"
+wget -q "$url"
+
+tarball=${tag}.tar.zst
+dir=$(tar tf $tarball | head -1 || true)
+
+echo "Extracting $tarball ..."
+tar -I zstd -xf $tarball && rm -f $tarball
+
+rm -rf $INSTALL_DIR
+mv -v $dir $INSTALL_DIR
+
+cd -
+

.github/scripts/matrix.py

@@ -0,0 +1,194 @@
+#!/usr/bin/env python3
+
+import os
+import dataclasses
+import json
+
+from enum import Enum
+from typing import Any, Dict, List, Final, Set, Union
+
+MANAGED_OWNER: Final[str] = "kernel-patches"
+MANAGED_REPOS: Final[Set[str]] = {
+    f"{MANAGED_OWNER}/bpf",
+    f"{MANAGED_OWNER}/vmtest",
+}
+
+DEFAULT_SELF_HOSTED_RUNNER_TAGS: Final[List[str]] = ["self-hosted", "docker-noble-main"]
+DEFAULT_RUNNER: Final[str] = "ubuntu-24.04"
+DEFAULT_LLVM_VERSION: Final[int] = 17
+
+
+class Arch(str, Enum):
+    """
+    CPU architecture supported by CI.
+    """
+
+    AARCH64 = "aarch64"
+    S390X = "s390x"
+    X86_64 = "x86_64"
+
+
+class Compiler(str, Enum):
+    GCC = "gcc"
+    LLVM = "llvm"
+
+
+@dataclasses.dataclass
+class Toolchain:
+    compiler: Compiler
+    # This is relevant ONLY for LLVM and should not be required for GCC
+    version: int
+
+    @property
+    def short_name(self) -> str:
+        return str(self.compiler.value)
+
+    @property
+    def full_name(self) -> str:
+        if self.compiler == Compiler.GCC:
+            return self.short_name
+
+        return f"{self.short_name}-{self.version}"
+
+    def to_dict(self) -> Dict[str, Union[str, int]]:
+        return {
+            "name": self.short_name,
+            "fullname": self.full_name,
+            "version": self.version,
+        }
+
+
+@dataclasses.dataclass
+class BuildConfig:
+    arch: Arch
+    toolchain: Toolchain
+    kernel: str = "LATEST"
+    run_veristat: bool = False
+    parallel_tests: bool = False
+    build_release: bool = False
+
+    @property
+    def runs_on(self) -> List[str]:
+        if is_managed_repo():
+            return DEFAULT_SELF_HOSTED_RUNNER_TAGS + [self.arch.value]
+        else:
+            return [DEFAULT_RUNNER]
+
+    @property
+    def build_runs_on(self) -> List[str]:
+        if is_managed_repo():
+            return ["codebuild"]
+        else:
+            return [DEFAULT_RUNNER]
+
+    @property
+    def tests(self) -> Dict[str, Any]:
+        tests_list = [
+            "test_progs",
+            "test_progs_parallel",
+            "test_progs_no_alu32",
+            "test_progs_no_alu32_parallel",
+            "test_verifier",
+        ]
+
+        if self.arch.value != "s390x":
+            tests_list.append("test_maps")
+
+        if self.toolchain.version >= 18:
+            tests_list.append("test_progs_cpuv4")
+
+        # if self.arch in [Arch.X86_64, Arch.AARCH64]:
+        #     tests_list.append("sched_ext")
+
+        # Don't run GCC BPF runner, because too many tests are failing
+        # See: https://lore.kernel.org/bpf/[email protected]/
+        # if self.arch == Arch.X86_64:
+        #    tests_list.append("test_progs-bpf_gcc")
+
+        if not self.parallel_tests:
+            tests_list = [test for test in tests_list if not test.endswith("parallel")]
+
+        return {"include": [generate_test_config(test) for test in tests_list]}
+
+    def to_dict(self) -> Dict[str, Any]:
+        return {
+            "arch": self.arch.value,
+            "toolchain": self.toolchain.to_dict(),
+            "kernel": self.kernel,
+            "run_veristat": self.run_veristat,
+            "parallel_tests": self.parallel_tests,
+            "build_release": self.build_release,
+            "runs_on": self.runs_on,
+            "tests": self.tests,
+            "build_runs_on": self.build_runs_on,
+        }
+
+
+def is_managed_repo() -> bool:
+    return (
+        os.environ["GITHUB_REPOSITORY_OWNER"] == MANAGED_OWNER
+        and os.environ["GITHUB_REPOSITORY"] in MANAGED_REPOS
+    )
+
+
+def set_output(name, value):
+    """Write an output variable to the GitHub output file."""
+    with open(os.getenv("GITHUB_OUTPUT"), "a", encoding="utf-8") as file:
+        file.write(f"{name}={value}\n")
+
+
+def generate_test_config(test: str) -> Dict[str, Union[str, int]]:
+    """Create the configuration for the provided test."""
+    is_parallel = test.endswith("_parallel")
+    config = {
+        "test": test,
+        "continue_on_error": is_parallel,
+        # While in experimental mode, parallel jobs may get stuck
+        # anywhere, including in user space where the kernel won't detect
+        # a problem and panic. We add a second layer of (smaller) timeouts
+        # here such that if we get stuck in a parallel run, we hit this
+        # timeout and fail without affecting the overall job success (as
+        # would be the case if we hit the job-wide timeout). For
+        # non-experimental jobs, 360 is the default which will be
+        # superseded by the overall workflow timeout (but we need to
+        # specify something).
+        "timeout_minutes": 30 if is_parallel else 360,
+    }
+    return config
+
+
+if __name__ == "__main__":
+    matrix = [
+        BuildConfig(
+            arch=Arch.X86_64,
+            toolchain=Toolchain(compiler=Compiler.GCC, version=DEFAULT_LLVM_VERSION),
+            run_veristat=True,
+            parallel_tests=True,
+        ),
+        BuildConfig(
+            arch=Arch.X86_64,
+            toolchain=Toolchain(compiler=Compiler.LLVM, version=DEFAULT_LLVM_VERSION),
+            build_release=True,
+        ),
+        BuildConfig(
+            arch=Arch.X86_64,
+            toolchain=Toolchain(compiler=Compiler.LLVM, version=18),
+            build_release=True,
+        ),
+        BuildConfig(
+            arch=Arch.AARCH64,
+            toolchain=Toolchain(compiler=Compiler.GCC, version=DEFAULT_LLVM_VERSION),
+        ),
+        BuildConfig(
+            arch=Arch.S390X,
+            toolchain=Toolchain(compiler=Compiler.GCC, version=DEFAULT_LLVM_VERSION),
+        ),
+    ]
+
+    # Outside of managed repositories only run on x86_64
+    if not is_managed_repo():
+        matrix = [config for config in matrix if config.arch == Arch.X86_64]
+
+    json_matrix = json.dumps({"include": [config.to_dict() for config in matrix]})
+    print(json.dumps(json.loads(json_matrix), indent=4))
+    set_output("build_matrix", json_matrix)

.github/scripts/tests/test_veristat_compare.py

@@ -0,0 +1,75 @@
+#!/usr/bin/env python3
+
+import unittest
+from typing import Iterable, List
+
+from ..veristat_compare import parse_table, VeristatFields
+
+
+def gen_csv_table(records: Iterable[str]) -> List[str]:
+    return [
+        ",".join(VeristatFields.headers()),
+        *records,
+    ]
+
+
+class TestVeristatCompare(unittest.TestCase):
+    def test_parse_table_ignore_new_prog(self):
+        table = gen_csv_table(
+            [
+                "prog_file.bpf.o,prog_name,N/A,success,N/A,N/A,1,N/A",
+            ]
+        )
+        veristat_info = parse_table(table)
+        self.assertEqual(veristat_info.table, [])
+        self.assertFalse(veristat_info.changes)
+        self.assertFalse(veristat_info.new_failures)
+
+    def test_parse_table_ignore_removed_prog(self):
+        table = gen_csv_table(
+            [
+                "prog_file.bpf.o,prog_name,success,N/A,N/A,1,N/A,N/A",
+            ]
+        )
+        veristat_info = parse_table(table)
+        self.assertEqual(veristat_info.table, [])
+        self.assertFalse(veristat_info.changes)
+        self.assertFalse(veristat_info.new_failures)
+
+    def test_parse_table_new_failure(self):
+        table = gen_csv_table(
+            [
+                "prog_file.bpf.o,prog_name,success,failure,MISMATCH,1,1,+0 (+0.00%)",
+            ]
+        )
+        veristat_info = parse_table(table)
+        self.assertEqual(
+            veristat_info.table,
+            [["prog_file.bpf.o", "prog_name", "success -> failure (!!)", "+0.00 %"]],
+        )
+        self.assertTrue(veristat_info.changes)
+        self.assertTrue(veristat_info.new_failures)
+
+    def test_parse_table_new_changes(self):
+        table = gen_csv_table(
+            [
+                "prog_file.bpf.o,prog_name,failure,success,MISMATCH,0,0,+0 (+0.00%)",
+                "prog_file.bpf.o,prog_name_increase,failure,failure,MATCH,1,2,+1 (+100.00%)",
+                "prog_file.bpf.o,prog_name_decrease,success,success,MATCH,1,1,-1 (-100.00%)",
+            ]
+        )
+        veristat_info = parse_table(table)
+        self.assertEqual(
+            veristat_info.table,
+            [
+                ["prog_file.bpf.o", "prog_name", "failure -> success", "+0.00 %"],
+                ["prog_file.bpf.o", "prog_name_increase", "failure", "+100.00 %"],
+                ["prog_file.bpf.o", "prog_name_decrease", "success", "-100.00 %"],
+            ],
+        )
+        self.assertTrue(veristat_info.changes)
+        self.assertFalse(veristat_info.new_failures)
+
+
+if __name__ == "__main__":
+    unittest.main()

.github/scripts/tmpfsify-workspace.sh

@@ -0,0 +1,21 @@
+#!/bin/bash
+
+set -x -euo pipefail
+
+TMPFS_SIZE=20 # GB
+MEM_TOTAL=$(awk '/MemTotal/ {print int($2/1024)}' /proc/meminfo)
+
+# sanity check: total mem is at least double TMPFS_SIZE
+if [ $MEM_TOTAL -lt $(($TMPFS_SIZE*1024*2)) ]; then
+    echo "tmpfsify-workspace.sh: will not allocate tmpfs, total memory is too low (${MEM_TOTAL}MB)"
+    exit 0
+fi
+
+dir="$(basename "$GITHUB_WORKSPACE")"
+cd "$(dirname "$GITHUB_WORKSPACE")"
+mv "${dir}" "${dir}.backup"
+mkdir "${dir}"
+sudo mount -t tmpfs -o size=${TMPFS_SIZE}G tmpfs "${dir}"
+rsync -a "${dir}.backup/" "${dir}"
+cd -
+