fix: implement short-term fixes for lock contention and add tests

Makefile

@@ -19,4 +19,4 @@ build-in-docker:
 
 check-cuda-hook-consistency:
 	python3 hack/check_cuda_hook_consistency.py
-.PHONY: check-cuda-hook-consistency
+.PHONY: check-cuda-hook-consistency

src/utils.c

@@ -13,9 +13,10 @@
 #include "multiprocess/multiprocess_memory_limit.h"
 
 const char* unified_lock="/tmp/vgpulock/lock";
-static int lock_fd = -1;
+
 extern size_t context_size;
 extern int cuda_to_nvml_map_array[CUDA_DEVICE_MAX_COUNT];
+static int unified_lock_fd = -1;
 
 // 0 unified_lock lock success
 // -1 unified_lock lock fail

test/CMakeLists.txt

@@ -36,6 +36,12 @@ foreach(TEST_SCRIPT ${TEST_SCRIPTS})
     list(APPEND TEST_TARGET_NAMES_LIST ${TEST_TARGET_NAME})
     if (TEST_TARGET_NAME STREQUAL "test_postinit_owner_death")
         target_link_libraries(${TEST_TARGET_NAME} -lrt -lpthread)
+    elseif(TEST_TARGET_NAME STREQUAL "bench_lock")
+        # Tell CMake to compile utils.c together with bench_lock.c
+        target_sources(${TEST_TARGET_NAME} PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/../src/utils.c)
+        # Add the include directory so it can find the headers
+        target_include_directories(${TEST_TARGET_NAME} PRIVATE ${CMAKE_CURRENT_SOURCE_DIR}/../src)
+        target_link_libraries(${TEST_TARGET_NAME} -lrt -lpthread)
     else()
         target_link_libraries(${TEST_TARGET_NAME} -lrt -lpthread
             -lnvidia-ml -lcuda -lcudart -L${CUDA_HOME}/lib64)

test/test_alloc_concurrent.cu

@@ -0,0 +1,45 @@
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <sys/wait.h>
+#include <cuda_runtime.h>
+
+#define NUM_WORKERS 4
+#define ITERATIONS 10
+
+void worker_task(int worker_id) {
+    void *ptr = NULL;
+    for (int i = 0; i < ITERATIONS; i++) {
+        // Allocate 1MB to trigger HAMi-core's memory interception
+        cudaError_t err = cudaMalloc(&ptr, 1024 * 1024);
+        if (err != cudaSuccess) {
+            printf("Worker %d failed allocation: %s\n", worker_id, cudaGetErrorString(err));
+            exit(1);
+        }
+        cudaFree(ptr);
+    }
+    printf("Worker %d completed successfully.\n", worker_id);
+    exit(0);
+}
+
+int main() {
+    printf("Starting %d concurrent workers...\n", NUM_WORKERS);
+
+    for (int i = 0; i < NUM_WORKERS; i++) {
+        pid_t pid = fork();
+        if (pid == 0) {
+            // Child process
+            worker_task(i);
+        } else if (pid < 0) {
+            printf("Fork failed.\n");
+            return 1;
+        }
+    }
+
+    // Parent waits for all children to finish
+    int status;
+    while (wait(&status) > 0);
+
+    printf("All workers finished.\n");
+    return 0;
+}