perf(mavlink): remove mutex from forward_message hot path

src/modules/mavlink/mavlink_main.cpp

@@ -513,7 +513,10 @@ Mavlink::forward_message(const mavlink_message_t *msg, Mavlink *self)
 		return;
 	}
 
-	LockGuard lg{mavlink_module_mutex};
+	// No mutex needed: instances can't be destroyed while receiver threads run
+	// (destroy_all_instances() stops all threads first). Pointer reads are
+	// naturally atomic on aligned architectures. pass_message() has its own
+	// per-instance _message_buffer_mutex for the ring buffer.
 
 	for (Mavlink *inst : mavlink_module_instances) {
 		if (inst && (inst != self) && (inst->get_forwarding_on())) {

test/mavsdk_tests/CMakeLists_bench.txt

@@ -0,0 +1,26 @@
+cmake_minimum_required(VERSION 3.5.1)
+
+project(mavlink_forwarding_bench CXX)
+
+set(CMAKE_CXX_STANDARD 17)
+set(CMAKE_CXX_STANDARD_REQUIRED ON)
+set(CMAKE_EXPORT_COMPILE_COMMANDS ON)
+
+find_package(MAVSDK REQUIRED)
+find_package(Threads REQUIRED)
+
+add_executable(mavlink_forwarding_bench
+    test_mavlink_forwarding_bench.cpp
+)
+
+target_link_libraries(mavlink_forwarding_bench
+    MAVSDK::mavsdk
+    ${CMAKE_THREAD_LIBS_INIT}
+)
+
+target_compile_options(mavlink_forwarding_bench
+    PRIVATE
+        -Wall
+        -Wextra
+        -Werror
+)

test/mavsdk_tests/run_forwarding_bench.sh

@@ -0,0 +1,117 @@
+#!/bin/bash
+#
+# MAVLink forwarding mutex benchmark runner.
+#
+# Builds the benchmark tool, starts PX4 SITL SIH, runs the benchmark,
+# then collects perf counter data.
+#
+# Usage:
+#   ./run_forwarding_bench.sh [--duration 60] [--rate 200] [--connections 2] [--label baseline]
+#
+# The --label flag tags the output files for before/after comparison.
+
+set -euo pipefail
+
+SCRIPT_DIR="$(cd "$(dirname "${BASH_SOURCE[0]}")" && pwd)"
+PX4_DIR="$(cd "$SCRIPT_DIR/../.." && pwd)"
+BUILD_DIR="$SCRIPT_DIR/build_bench"
+RESULTS_DIR="$PX4_DIR/bench_results"
+
+DURATION=60
+RATE=200
+CONNECTIONS=2
+LABEL="run"
+
+while [[ $# -gt 0 ]]; do
+    case "$1" in
+        --duration) DURATION="$2"; shift 2 ;;
+        --rate) RATE="$2"; shift 2 ;;
+        --connections) CONNECTIONS="$2"; shift 2 ;;
+        --label) LABEL="$2"; shift 2 ;;
+        --help|-h)
+            echo "Usage: $0 [--duration N] [--rate N] [--connections N] [--label NAME]"
+            exit 0
+            ;;
+        *) echo "Unknown arg: $1"; exit 1 ;;
+    esac
+done
+
+TIMESTAMP=$(date +%Y%m%d_%H%M%S)
+RESULT_PREFIX="${RESULTS_DIR}/${LABEL}_${TIMESTAMP}"
+
+mkdir -p "$RESULTS_DIR"
+
+echo "=== Step 1: Build PX4 SITL SIH ==="
+cd "$PX4_DIR"
+make px4_sitl_sih 2>&1 | tail -5
+
+echo ""
+echo "=== Step 2: Build benchmark tool ==="
+mkdir -p "$BUILD_DIR"
+cd "$BUILD_DIR"
+# Symlink the standalone CMakeLists and source into build dir
+ln -sf "$SCRIPT_DIR/CMakeLists_bench.txt" CMakeLists.txt
+ln -sf "$SCRIPT_DIR/test_mavlink_forwarding_bench.cpp" .
+cmake -DCMAKE_BUILD_TYPE=Release .
+make -j$(nproc 2>/dev/null || sysctl -n hw.ncpu)
+
+echo ""
+echo "=== Step 3: Start PX4 SITL SIH ==="
+cd "$PX4_DIR"
+
+# Start PX4 in background, capture PID
+PX4_SIM_MODEL=sihsim_quadx $PX4_DIR/build/px4_sitl_sih/bin/px4 \
+    -s $PX4_DIR/ROMFS/px4fmu_common/init.d-posix/rcS \
+    -w $PX4_DIR/build/px4_sitl_sih/rootfs \
+    $PX4_DIR/build/px4_sitl_sih/etc \
+    > "${RESULT_PREFIX}_px4.log" 2>&1 &
+PX4_PID=$!
+
+echo "PX4 started (PID=$PX4_PID), waiting 10s for boot..."
+sleep 10
+
+# Verify PX4 is running
+if ! kill -0 "$PX4_PID" 2>/dev/null; then
+    echo "ERROR: PX4 failed to start. Check ${RESULT_PREFIX}_px4.log"
+    exit 1
+fi
+
+echo ""
+echo "=== Step 4: Run benchmark ==="
+echo "  Duration:    ${DURATION}s"
+echo "  Rate:        ${RATE} Hz/connection"
+echo "  Connections: ${CONNECTIONS}"
+echo ""
+
+"$BUILD_DIR/mavlink_forwarding_bench" \
+    --duration "$DURATION" \
+    --rate "$RATE" \
+    --connections "$CONNECTIONS" \
+    --report "${RESULT_PREFIX}_sender.csv" \
+    2>&1 | tee "${RESULT_PREFIX}_bench.log"
+
+echo ""
+echo "=== Step 5: Perf counters ==="
+echo "The benchmark tool reads perf counters from the board automatically"
+echo "via the MAVLink shell protocol. Check the bench log above for BENCH lines."
+
+echo ""
+echo "=== Step 6: Stop PX4 ==="
+kill -INT "$PX4_PID" 2>/dev/null || true
+wait "$PX4_PID" 2>/dev/null || true
+
+# Extract bench lines from the benchmark log if available
+if grep -q "BENCH" "${RESULT_PREFIX}_bench.log" 2>/dev/null; then
+    echo "Perf counters:"
+    grep "BENCH" "${RESULT_PREFIX}_bench.log" | tee "${RESULT_PREFIX}_perf.txt"
+fi
+
+echo ""
+echo "=== Results saved ==="
+echo "  PX4 log:      ${RESULT_PREFIX}_px4.log"
+echo "  Bench log:    ${RESULT_PREFIX}_bench.log"
+echo "  Sender CSV:   ${RESULT_PREFIX}_sender.csv"
+echo "  Perf data:    ${RESULT_PREFIX}_perf.txt (if available)"
+echo ""
+echo "To compare runs:"
+echo "  diff <baseline>_perf.txt <after_fix>_perf.txt"