build-prevail

作者: microsoft

獨立建置與測試 PREVAIL 驗證器(external/ebpf-verifier)。當被要求建置、測試或迭代 PREVAIL 驗證器、執行 YAML… 時,請使用此技能。

npx skills add https://github.com/microsoft/ebpf-for-windows --skill build-prevail

Build and Test PREVAIL Verifier

Build and test the PREVAIL eBPF verifier (external/ebpf-verifier) independently from the main eBPF for Windows solution.

When to Use

  • Working on verifier changes in external/ebpf-verifier/
  • Running or debugging YAML verification tests
  • Building run_yaml.exe or tests.exe for the verifier
  • Iterating on verifier fixes with faster feedback than full MSBuild

Prerequisites

The verifier's cmake build directory must exist. If external/ebpf-verifier/build/ is missing, run from the solution root:

.\scripts\initialize_ebpf_repo.ps1

This generates cmake projects for the verifier (and other submodules). You only need to do this once, or after resetting submodules.

Building Standalone

cd external\ebpf-verifier

# Build the YAML test runner (most common during development)
cmake --build build --config Release --target run_yaml

# Build the full Catch2 test binary
cmake --build build --config Release --target tests

# Clean build (rebuild everything from scratch)
cmake --build build --config Release --clean-first

Enabling Standalone Tests

When built as a submodule, prevail_ENABLE_TESTS defaults to OFF. To enable standalone tests:

cmake -B build -Dprevail_ENABLE_TESTS=ON
cmake --build build --config Release

Note: The initialize_ebpf_repo.ps1 script does NOT enable tests. You need to reconfigure with -Dprevail_ENABLE_TESTS=ON if you want to build tests.exe standalone.

Running YAML Tests

YAML test files are in external/ebpf-verifier/test-data/*.yaml. Each file is a test suite with individual tests separated by ---.

cd external\ebpf-verifier

# Run all tests in a suite
.\bin\run_yaml.exe test-data\loop.yaml

# Run tests matching a substring
.\bin\run_yaml.exe test-data\loop.yaml "while loop with"

# Run a specific test by exact name
.\bin\run_yaml.exe test-data\bitop.yaml "AND with 0xFF preserves relations when value fits"

Exit codes

  • 0 — all tests passed
  • 1 — one or more tests failed

Discovering postconditions for new tests

When writing new YAML tests, use placeholder postconditions and run the test to discover actual values:

post:
  - placeholder
messages:
  - placeholder

The failure output shows "Unexpected properties" (actual values) and "Unseen properties" (your placeholders). Copy the actual values into your test.

Available test suites

SuiteDescription
loop.yamlBounded loop verification (termination checking)
bitop.yamlBitwise operations (AND, OR, XOR)
movsx.yamlSign extension (MOVSX) operations
sext.yamlSign/zero extension relational tests
jump.yamlConditional jumps and branching
packet.yamlPacket access safety
pointer.yamlPointer arithmetic and safety
stack.yamlStack access verification
call.yamlHelper function calls
calllocal.yamlLocal (subprogram) calls
assign.yamlRegister assignment
add.yaml / subtract.yamlArithmetic operations
muldiv.yaml / sdivmod.yaml / udivmod.yamlMultiplication, division, modulo
shift.yamlShift operations
atomic.yamlAtomic operations
full64.yaml64-bit comparison operations
unsigned.yamlUnsigned comparison operations
unop.yamlUnary operations (neg, swap)
observe.yamlObservation/assertion tests
uninit.yamlUninitialized variable detection
map.yamlMap operations
parse.yamlYAML parsing tests
callx.yamlIndirect calls

Running the Full Catch2 Test Binary

cd external\ebpf-verifier

# Run all tests
.\bin\tests.exe

# Run with compact reporter
.\bin\tests.exe --reporter compact

# Abort on first failure
.\bin\tests.exe --abort --reporter compact

# Run specific test sections
.\bin\tests.exe "YAML suite: test-data/loop.yaml"

The tests.exe binary includes YAML tests, ELF verification tests, conformance tests, and unit tests.

YAML Test Format

---
test-case: descriptive test name
options: ["termination"]    # optional; enables loop termination checking

pre: ["r1.type=number", "r1.svalue=[0, 100]", "r1.uvalue=r1.svalue"]

code:
  <start>: |
    r0 = 0
  <loop>: |
    r0 += 1
    if r1 > r0 goto <loop>
  <out>: |
    exit

post:
  - r0.type=number
  - r0.svalue=[1, 100]

messages: []    # expected verifier messages; omit for no messages

Key conventions

  • r prefix for 64-bit registers, w prefix for 32-bit
  • w2 = r2 generates a 32-bit self-MOV (SHL 32 + RSH 32 truncation pattern)
  • r2 &= 255 generates 64-bit AND with immediate
  • svalue = signed value, uvalue = unsigned value
  • r2.svalue=r1.svalue means relational constraint (r2 tracks r1)
  • r1.svalue=[0, 100] means interval [0, 100]
  • pc[N] refers to the loop counter at instruction N
  • post: must list ALL expected properties — unlisted ones cause "Unexpected properties" failure
  • messages: field is optional (defaults to empty)

Building via MSBuild (within the main solution)

When you need to build the verifier as part of the main solution (e.g., to build bpf2c which depends on it):

# From solution root
msbuild ebpf-for-windows.sln /m /p:Configuration=Debug /p:Platform=x64 /t:"tools\bpf2c" /v:q /nologo

The MSBuild target for the verifier library is libs\user\prevail (for Debug/Release configs). The ubpf_fuzzer\ebpfverifier target is a separate copy used only in FuzzerDebug configuration.

Submodule Notes

  • The verifier lives at external/ebpf-verifier/ as a git submodule
  • git stash in the parent repo does NOT affect the submodule working tree — stash separately in the submodule if needed
  • To reset to the committed pointer: git submodule update --init --recursive (from the parent repo root)
  • After resetting submodules, re-run .\scripts\initialize_ebpf_repo.ps1 to regenerate cmake projects

來自 microsoft 的更多技能

oss-growth
microsoft
開源增長駭客角色
agent-framework-azure-ai-py
microsoft
使用Microsoft Agent Framework Python SDK(agent-framework-azure-ai)构建Azure AI Foundry代理。适用于使用AzureAIAgentsProvider创建持久化代理、使用托管工具(代码解释器、文件搜索、网络搜索)、集成MCP服务器、管理对话线程或实现流式响应。涵盖函数工具、结构化输出和多工具代理。
development
airunway-aks-setup
microsoft
在AKS上設定AI Runway——從裸叢集到執行模型。涵蓋叢集驗證、控制器安裝、GPU評估、供應商設定及首次部署。時機:「設定AI Runway」、「上線AKS叢集」、「安裝AI Runway」、「airunway設定」、「部署模型至AKS」、「在AKS上進行GPU推論」、「在AKS上設定KAITO」、「在AKS上執行LLM」、「在AKS上使用vLLM」、「在AKS上設定模型服務」、「AI Runway控制器」。
devops
appinsights-instrumentation
microsoft
使用Azure Application Insights檢測Web應用程式的指南。提供遙測模式、SDK設定與組態參考。適用時機:如何檢測應用程式、App Insights SDK、遙測模式、什麼是App Insights、Application Insights指南、檢測範例、APM最佳實踐。
devops
applicationinsights-web-ts
microsoft
使用Application Insights JavaScript SDK(@microsoft/applicationinsights-web)為瀏覽器/Web應用程式進行檢測。適用於真實使用者監控(RUM)——頁面檢視、點擊、AJAX/fetch依賴、例外、自訂事件,以及與後端OpenTelemetry追蹤關聯的瀏覽器端GenAI代理追蹤。涵蓋SDK載入器指令碼與npm設定、框架擴充(React、React Native、Angular)、點擊分析、遙測初始化器,以及從瀏覽器發出的代理/工具/模型span的OTel GenAI語意慣例。
devops
azure-ai-anomalydetector-java
microsoft
使用適用於 Java 的 Azure AI 異常偵測器 SDK 建置異常偵測應用程式。在實作單變量/多變量異常偵測、時間序列分析或 AI 驅動監控時使用。
development
azure-ai-language-conversations-py
microsoft
使用 azure-ai-language-conversations Python SDK 實作對話語言理解(CLU)。當使用 ConversationAnalysisClient 分析對話意圖與實體、建置 NLP 功能,或將語言理解整合至應用程式時使用。
development
azure-ai-ml-py
microsoft
Azure Machine Learning SDK v2 for Python。用於機器學習工作區、作業、模型、資料集、計算資源與管線。 觸發詞:「azure-ai-ml」、「MLClient」、「workspace」、「model registry」、「training jobs」、「datasets」。
development