Best Effort Thread Pinning to CPU_ID (#1096)

Author: waahm7

source/posix/thread.c

@@ -296,9 +296,9 @@ int aws_thread_launch(
             attr_return = pthread_attr_setaffinity_np(attributes_ptr, sizeof(cpuset), &cpuset);
 
             if (attr_return) {
-                AWS_LOGF_ERROR(
+                AWS_LOGF_WARN(
                     AWS_LS_COMMON_THREAD,
-                    "id=%p: pthread_attr_setaffinity_np() failed with %d.",
+                    "id=%p: pthread_attr_setaffinity_np() failed with %d. Continuing without cpu affinity",
                     (void *)thread,
                     attr_return);
                 goto cleanup;
@@ -382,7 +382,20 @@ int aws_thread_launch(
 
     if (attr_return) {
         s_thread_wrapper_destroy(wrapper);
-
+        if (options && options->cpu_id >= 0) {
+            /*
+             * `pthread_create` can fail with an `EINVAL` error or `EDEADLK` on freebasd if the `cpu_id` is
+             * restricted/invalid. Since the pinning to a particular `cpu_id` is supposed to be best-effort, try to
+             * launch a thread again without pinning to a specific cpu_id.
+             */
+            AWS_LOGF_INFO(
+                AWS_LS_COMMON_THREAD,
+                "id=%p: Attempting to launch the thread again without pinning to a cpu_id",
+                (void *)thread);
+            struct aws_thread_options new_options = *options;
+            new_options.cpu_id = -1;
+            return aws_thread_launch(thread, func, arg, &new_options);
+        }
         switch (attr_return) {
             case EINVAL:
                 return aws_raise_error(AWS_ERROR_THREAD_INVALID_SETTINGS);

tests/CMakeLists.txt

@@ -40,6 +40,7 @@ add_test_case(aws_load_error_strings_test)
 add_test_case(aws_assume_compiles_test)
 
 add_test_case(thread_creation_join_test)
+add_test_case(thread_creation_join_invalid_cpu_id_test)
 add_test_case(thread_atexit_test)
 add_test_case(test_managed_thread_join)
 add_test_case(test_managed_thread_join_timeout)

tests/thread_test.c

@@ -70,6 +70,40 @@ static int s_test_thread_creation_join_fn(struct aws_allocator *allocator, void
 
 AWS_TEST_CASE(thread_creation_join_test, s_test_thread_creation_join_fn)
 
+static int s_test_thread_creation_join_invalid_cpu_id_fn(struct aws_allocator *allocator, void *ctx) {
+    (void)ctx;
+    aws_common_library_init(allocator);
+    struct thread_test_data test_data = {.allocator = allocator};
+
+    struct aws_thread thread;
+    aws_thread_init(&thread, allocator);
+
+    struct aws_thread_options thread_options = *aws_default_thread_options();
+    /* invalid cpu_id. Ensure that the cpu_id is best-effort based. */
+    thread_options.cpu_id = 4096;
+
+    ASSERT_SUCCESS(
+        aws_thread_launch(&thread, s_thread_fn, (void *)&test_data, &thread_options), "thread creation failed");
+    ASSERT_INT_EQUALS(
+        AWS_THREAD_JOINABLE, aws_thread_get_detach_state(&thread), "thread state should have returned JOINABLE");
+    ASSERT_SUCCESS(aws_thread_join(&thread), "thread join failed");
+    ASSERT_TRUE(
+        aws_thread_thread_id_equal(test_data.thread_id, aws_thread_get_id(&thread)),
+        "get_thread_id should have returned the same id as the thread calling current_thread_id");
+    ASSERT_INT_EQUALS(
+        AWS_THREAD_JOIN_COMPLETED,
+        aws_thread_get_detach_state(&thread),
+        "thread state should have returned JOIN_COMPLETED");
+
+    aws_string_destroy(test_data.thread_name);
+    aws_thread_clean_up(&thread);
+    aws_common_library_clean_up();
+
+    return 0;
+}
+
+AWS_TEST_CASE(thread_creation_join_invalid_cpu_id_test, s_test_thread_creation_join_invalid_cpu_id_fn)
+
 static uint32_t s_atexit_call_count = 0;
 static void s_thread_atexit_fn(void *user_data) {
     (void)user_data;