fix(profiler): recover arm64 native stacks across sampling signals

.gitlab-ci.yml

@@ -1,3 +1,5 @@
+# Copyright 2026, Datadog, Inc.
+
 image: alpine:3.22.1@sha256:4bcff63911fcb4448bd4fdacec207030997caf25e9bea4045fa6c8c44de311d1
 
 variables:
@@ -183,6 +185,13 @@ generate-reliability-child-pipeline:
     - if: '$JDK_VERSION != null || $DEBUG_LEVEL != null || $HASH != null || $DOWNSTREAM != null'
       when: never
     - when: on_success
+  # Replaces the default before_script: its CANCELLED guard would exit before
+  # generated-reliability.yml is written, and run-reliability-tests (a bridge
+  # job, so not skipped by that guard) then fails to create its child pipeline
+  # from the missing artifact. Merge-queue branches (mq-working-branch-*) have
+  # no open PR, so prepare.sh always sets CANCELLED=true there. RUN_RELIABILITY
+  # is never set when CANCELLED is, so the noop child pipeline is generated.
+  before_script: []
   script:
     - |
       if [ "${RUN_RELIABILITY:-}" = "true" ]; then

.gitlab/benchmarks/.gitlab-ci.yml

@@ -1,3 +1,5 @@
+# Copyright 2026, Datadog, Inc.
+
 variables:
   DD_OCTO_STS_IMAGE: registry.ddbuild.io/images/dd-octo-sts-ci-base:2025.06-1
 
@@ -28,6 +30,11 @@ variables:
       when: never
     - if: '$CI_PIPELINE_SOURCE == "merge_request_event"'
       when: never
+    # Merge-queue branches have no open PR, so prepare.sh cancels the pipeline
+    # and deploy-artifact publishes nothing to benchmark; the BP run would
+    # only delay the queue (run-benchmarks uses strategy: depend).
+    - if: '$CI_COMMIT_BRANCH =~ /^mq-working-branch-/'
+      when: never
 
 # Decide whether benchmarks run for this branch and generate the child
 # pipeline YAML accordingly.

AGENTS.md

@@ -1,5 +1,7 @@
 <!-- Copyright 2026, Datadog, Inc. -->
 
+<!-- Copyright 2026, Datadog, Inc. -->
+
 <!-- CLAUDE BOOTSTRAP INSTRUCTIONS
 If you are Claude and reading this file, check if CLAUDE.md exists in the same directory.
 If CLAUDE.md does not exist or only contains the redirect + Compact Instructions below, that
@@ -86,7 +88,9 @@ and `:ddprof-lib:benchmarks`), `ddprof-test` (Java integration tests), `ddprof-t
 ## Build & Test Quick Reference
 
 Configurations: `release`, `debug`, plus `asan`/`tsan`/`fuzzer` when the sanitizer runtime
-is found for the selected compiler (inactive on macOS and musl).
+is found for the selected compiler (inactive on macOS and musl), plus opt-in `coverage`
+with `-PenableCoverage` (`:ddprof-lib:coverageReport`; the C++ report covers only the
+`-O0 -DDEBUG` build — see [doc/building/CodeCoverage.md](doc/building/CodeCoverage.md)).
 
 ```bash
 ./.claude/commands/build-and-summarize buildRelease      # release library + JAR
@@ -125,10 +129,16 @@ debug symbols via `buildDebug`, memory errors via ASan, suppressions in
 Custom Kotlin-based native build plugin (`build-logic/`) instead of Gradle's
 `cpp-library`/`cpp-application` plugins; it invokes `clang++`/`g++` directly. Build
 configurations are discovered dynamically from `ConfigurationPresets.kt`, never hardcoded.
+Google Test entry-point archives precede their dependencies so architecture-gated
+test files also link on hosts where they register no cases.
 Two-JDK pattern: `JAVA_HOME` (build, JDK 17+, Gradle runs on 21 LTS) runs Gradle;
 `JAVA_TEST_HOME` runs tests against the target JDK; target bytecode is Java 8 via
 `--release 8`.
 
+The `ddprof-stresstest:chaosElements` outgoing configuration shares the chaos JAR
+with `ddprof-test` stack-correctness tests. See
+[TestingGuide.md](doc/building/TestingGuide.md#virtual-thread-remount-correctness).
+
 Details and troubleshooting:
 [build-logic/README.md](build-logic/README.md),
 [doc/building/NativeBuildPlugin.md](doc/building/NativeBuildPlugin.md) (task DSL, debug-symbol
@@ -148,6 +158,7 @@ before working on a subsystem.
 | String dictionary (arena chunks, rotation/clear protocols) | [doc/architecture/StringDictionary.md](doc/architecture/StringDictionary.md) |
 | RefCountGuard (lock-free RAII reader-drain before reclamation) | [doc/architecture/RefCountGuard.md](doc/architecture/RefCountGuard.md) |
 | TLS context (thread-local tracing context, distributed tracing) | [doc/architecture/TLSContext.md](doc/architecture/TLSContext.md) |
+| Native stack unwinding (DWARF CFA rules, arm64 temporary argument areas, JNI wrapper phases and LR/PLT, validated Linux signal-frame recovery) | [doc/architecture/JvmSupportLayer.md](doc/architecture/JvmSupportLayer.md) |
 | Profiling engines (CPU, wall-clock, allocation, native memory, liveness) | [doc/architecture/ProfilingEngines.md](doc/architecture/ProfilingEngines.md) |
 | JFR integration (FlightRecorder, metadata, constant pools, buffers) | [doc/architecture/JfrIntegration.md](doc/architecture/JfrIntegration.md) |
 | Live-heap reference chains (implementation, signal scheduling, design) | [doc/architecture/LiveHeapReferenceChainsImplementation.md](doc/architecture/LiveHeapReferenceChainsImplementation.md), [doc/architecture/ReferenceChainsSignalsExplained.md](doc/architecture/ReferenceChainsSignalsExplained.md), [doc/ReferenceChainsDesign.md](doc/ReferenceChainsDesign.md), [doc/ReferenceChainsCollectionSummary.md](doc/ReferenceChainsCollectionSummary.md) |
@@ -303,6 +314,10 @@ pointer-not-yet-stored. Never permanently advance an iteration cursor past NULL
 at the first NULL or track the last contiguous non-NULL entry so missing entries are
 retried.
 
+**Stack completeness.** A successful carrier crossing does not establish a complete
+virtual-thread stack. Preserve walker-reported truncation through JFR serialization;
+see [JvmSupportLayer.md](doc/architecture/JvmSupportLayer.md#continuation-completeness).
+
 **JVM abstraction.** Shared code must not refer to JVM-implementation-specific code directly,
 only through the abstraction files: `jvmSupport.{h,inline.h,cpp}`, `jvmThread.{h,cpp}`,
 `vmEntry.{h,cpp}`, `javaApi.cpp`. Implementation-specific code lives under `hotspot/`,

build-logic/conventions/src/main/kotlin/com/datadoghq/native/config/ConfigurationPresets.kt

@@ -1,3 +1,7 @@
+/*
+ * Copyright 2026, Datadog, Inc.
+ * SPDX-License-Identifier: Apache-2.0
+ */
 
 package com.datadoghq.native.config
 
@@ -54,6 +58,16 @@ object ConfigurationPresets {
         // reason: those configs instrument every memory access, so the measured
         // per-sample times say more about the sanitizer than about the sampler.
         val samplerPerf = project.hasProperty("enableSamplerPerf")
+
+        // Opt-in gcov/llvm-cov instrumented build. Inactive by default -- only
+        // registered as an active configuration (and thus only gets its
+        // compile/link/assemble/gtest tasks created at all, see
+        // NativeBuildPlugin/GtestPlugin filtering on getActiveConfigurations)
+        // when -PenableCoverage is passed. This keeps coverage instrumentation
+        // completely out of the default build and out of CI runs that don't
+        // ask for it, mirroring how asan/tsan are gated off when unsupported.
+        val enableCoverage = project.hasProperty("enableCoverage")
+
         extension.buildConfigurations.apply {
             register("release") {
                 configureRelease(this, currentPlatform, currentArch, version)
@@ -74,6 +88,9 @@ object ConfigurationPresets {
             register("fuzzer") {
                 configureFuzzer(this, currentPlatform, currentArch, version, rootDir, compiler)
             }
+            register("coverage") {
+                configureCoverage(this, currentPlatform, currentArch, version, enableCoverage)
+            }
         }
 
         val activeConfigs = extension.getActiveConfigurations(currentPlatform, currentArch)
@@ -406,4 +423,31 @@ object ConfigurationPresets {
             }
         }
     }
+
+    /**
+     * gcov/llvm-cov-instrumented debug build (`--coverage` is understood by both
+     * gcc and clang and emits gcov-compatible .gcno/.gcda files, so one preset
+     * covers both toolchains/platforms without a separate clang source-based
+     * coverage format). Inactive unless the caller opts in via -PenableCoverage.
+     *
+     * `-fprofile-update=atomic`: the default counter updates are plain
+     * read-modify-writes, so the profiler's concurrent threads (samplers, signal
+     * handlers, JVM threads) lose and corrupt counts -- a Linux run produced
+     * negative branch counts (GCC bug 68080) that gcovr refuses to parse.
+     */
+    fun configureCoverage(
+        config: BuildConfiguration,
+        platform: Platform,
+        architecture: Architecture,
+        version: String,
+        active: Boolean
+    ) {
+        // Built on configureDebug so coverage measures the same binary testDebug runs
+        // (including -z,nodelete on Linux), only with the instrumentation added.
+        configureDebug(config, platform, architecture, version)
+        config.active.set(active)
+        config.publishable.set(false)
+        config.compilerArgs.addAll("--coverage", "-fprofile-update=atomic")
+        config.linkerArgs.add("--coverage")
+    }
 }

build-logic/conventions/src/main/kotlin/com/datadoghq/native/gtest/GtestPlugin.kt

@@ -76,7 +76,7 @@ class GtestPlugin : Plugin<Project> {
             }
 
             // Check if gtest is available
-            val hasGtest = checkGtestAvailable()
+            val hasGtest = PlatformUtils.isGtestAvailable()
             if (!hasGtest) {
                 project.logger.warn("WARNING: Google Test not found - skipping native tests")
             }
@@ -96,21 +96,6 @@ class GtestPlugin : Plugin<Project> {
         }
     }
 
-    private fun checkGtestAvailable(): Boolean {
-        // Check common gtest locations
-        val locations = when (PlatformUtils.currentPlatform) {
-            Platform.MACOS -> listOf(
-                "/opt/homebrew/opt/googletest",
-                "/usr/local/opt/googletest"
-            )
-            Platform.LINUX -> listOf(
-                "/usr/include/gtest",
-                "/usr/local/include/gtest"
-            )
-        }
-        return locations.any { File(it).exists() }
-    }
-
     private fun createBuildNativeLibsTask(project: Project, extension: GtestExtension, hasGtest: Boolean) {
         project.tasks.register("buildNativeLibs") {
             group = "build"

build-logic/conventions/src/main/kotlin/com/datadoghq/native/gtest/GtestTaskBuilder.kt

@@ -1,3 +1,4 @@
+// Copyright 2026, Datadog, Inc.
 
 package com.datadoghq.native.gtest
 
@@ -161,8 +162,8 @@ class GtestTaskBuilder(
                 Platform.LINUX -> { /* System paths */ }
             }
 
-            // Add gtest libraries
-            lib("gtest", "gtest_main", "gmock", "gmock_main", "dl", "pthread", "m")
+            // Static archives must precede their dependencies, including when no tests are registered.
+            lib("gtest_main", "gmock_main", "gmock", "gtest", "dl", "pthread", "m")
             if (PlatformUtils.currentPlatform == Platform.LINUX) {
                 lib("rt")
             }

build-logic/conventions/src/main/kotlin/com/datadoghq/native/model/BuildConfiguration.kt

@@ -1,3 +1,7 @@
+/*
+ * Copyright 2026, Datadog, Inc.
+ * SPDX-License-Identifier: Apache-2.0
+ */
 
 package com.datadoghq.native.model
 
@@ -18,11 +22,19 @@ abstract class BuildConfiguration @Inject constructor(
     abstract val testJvmArgs: ListProperty<String>
     abstract val active: Property<Boolean>
 
+    /**
+     * Whether this configuration's jar is part of the published artifact set. An
+     * unpublishable configuration (e.g. the opt-in coverage build) is only packaged
+     * while it is active, for in-repo consumers such as the test projects.
+     */
+    abstract val publishable: Property<Boolean>
+
     override fun getName(): String = configName
 
     init {
         // Default to active unless overridden
         active.convention(true)
+        publishable.convention(true)
         testEnvironment.convention(emptyMap())
         testJvmArgs.convention(emptyList())
     }

build-logic/conventions/src/main/kotlin/com/datadoghq/native/util/PlatformUtils.kt

@@ -1,3 +1,7 @@
+/*
+ * Copyright 2026, Datadog, Inc.
+ * SPDX-License-Identifier: Apache-2.0
+ */
 
 package com.datadoghq.native.util
 
@@ -62,6 +66,24 @@ object PlatformUtils {
         )
     }
 
+    /**
+     * Google Test install locations probed by the gtest plugin (and by the coverage
+     * preflight check, since a missing gtest silently drops the gtest half of coverage).
+     */
+    val gtestLocations: List<String>
+        get() = when (currentPlatform) {
+            Platform.MACOS -> listOf(
+                "/opt/homebrew/opt/googletest",
+                "/usr/local/opt/googletest"
+            )
+            Platform.LINUX -> listOf(
+                "/usr/include/gtest",
+                "/usr/local/include/gtest"
+            )
+        }
+
+    fun isGtestAvailable(): Boolean = gtestLocations.any { File(it).exists() }
+
     /**
      * Check if a compiler is available and functional
      */

build-logic/conventions/src/main/kotlin/com/datadoghq/profiler/ProfilerTestPlugin.kt

@@ -1,3 +1,7 @@
+/*
+ * Copyright 2026, Datadog, Inc.
+ * SPDX-License-Identifier: Apache-2.0
+ */
 
 package com.datadoghq.profiler
 
@@ -78,6 +82,23 @@ import javax.inject.Inject
  */
 class ProfilerTestPlugin : Plugin<Project> {
 
+    companion object {
+        private const val ADDITIONAL_JVM_ARGS_EXTENSION = "additionalJvmArgs"
+
+        /**
+         * Extra JVM args a build script can contribute to a musl Exec test task (e.g.
+         * ddprof-test's JaCoCo agent for the coverage tasks, which Gradle only attaches to
+         * Test tasks). Get-or-create, since a build script's configureEach on the task can
+         * run before the plugin's own configuration of it.
+         */
+        @Suppress("UNCHECKED_CAST")
+        fun additionalJvmArgs(task: Exec): ListProperty<String> =
+            task.extensions.findByName(ADDITIONAL_JVM_ARGS_EXTENSION) as ListProperty<String>?
+                ?: task.project.objects.listProperty(String::class.java).also {
+                    task.extensions.add(ADDITIONAL_JVM_ARGS_EXTENSION, it)
+                }
+    }
+
     override fun apply(project: Project) {
         val extension = project.extensions.create(
             "profilerTest",
@@ -362,6 +383,9 @@ class ProfilerTestPlugin : Plugin<Project> {
                 }
                 allArgs.addAll(testConfig.extraJvmArgs)
                 allArgs.addAll(testConfig.configJvmArgs)
+                // Read here, not contributed via a doFirst of the build script's own: that
+                // would run before this one and its args would be overwritten.
+                allArgs.addAll(additionalJvmArgs(execTask).get())
 
                 // System properties
                 testConfig.systemProperties.forEach { (key, value) ->