修正io线程死锁检测不工作的问题。v4.6.50-20230905

3 files changed, 19 insertions, 32 deletions

diff --git a/app/src/marsio.c b/app/src/marsio.c
index 0ea11fe..72b48cf 100644
--- a/app/src/marsio.c
+++ b/app/src/marsio.c
@@ -564,10 +564,9 @@ static void mp_cache_init_for_each_mp(struct rte_mempool * mp, void * arg)
     }
 }
 
-static int mpapp_mp_cache_init(struct mr_instance * instance)
+static void mpapp_mp_cache_init(struct mr_instance * instance)
 {
     rte_mempool_walk(mp_cache_init_for_each_mp, (void *)instance);
-    return 0;
 }
 
 /* 注册应用 */
@@ -685,17 +684,6 @@ static int __ctrlplane_conn_close_handler(struct ctrlmsg_handler * ct_hand, stru
     exit(EXIT_FAILURE);
 }
 
-static void __open_device_unposion(struct vdev_instance * vdi)
-{
-    MR_ASAN_UNPOISON_MEMORY_REGION(vdi, sizeof(struct vdev_instance));
-    MR_ASAN_UNPOISON_MEMORY_REGION(vdi->vdev, sizeof(struct vdev));
-
-    vnode_mirror_unpoison_cons(vdi->vnode_rx_cons);
-    vnode_mirror_unpoison_prod(vdi->vnode_tx_prod);
-    vnode_mirror_unpoison_prod(vdi->vnode_ftx_prod);
-    vnode_mirror_unpoison_prod(vdi->vnode_ltx_prod);
-}
-
 #if 0
 static int __open_device_response_handler(struct ctrlmsg_handler * ct_hand,
     struct ctrlmsg_conn * ct_conn, struct ctrl_msg_header * msg, void * arg)
@@ -1038,12 +1026,7 @@ int marsio_init(struct mr_instance * instance, const char * appsym)
     }
 
     /* local mp cache */
-    if (mpapp_mp_cache_init(instance) != RT_SUCCESS)
-    {
-        MR_ERROR("Local mempool cache initialization failed.");
-        goto err;
-    }
-
+    mpapp_mp_cache_init(instance);
     pthread_t pid_ctrlplane_thread;
     int ret = pthread_create(&pid_ctrlplane_thread, NULL, mrapp_ctrlplane_thread, instance);
     if (ret < 0)
diff --git a/service/src/core.c b/service/src/core.c
index 3b1ecae..dc6318e 100644
--- a/service/src/core.c
+++ b/service/src/core.c
@@ -191,9 +191,10 @@ int sc_check_spinlock_loop(struct sc_main * sc_main)
 
 void sc_keepalive_failure_handler(void * data, const int id_core)
 {
-    MR_ERROR("Thread %d is dead, MRZCPD service exit. \n", id_core);
-    exit(EXIT_FAILURE);
-    return;
+    MR_ERROR("Thread %d is dead, exit. \n", id_core);
+
+    /* print the stack trace */
+    abort();
 }
 
 int sc_check_memleak_loop(struct sc_main * sc_main)
@@ -1228,25 +1229,26 @@ int main(int argc, char * argv[])
     if (sc->en_spinlock_check)
     {
         sc->keepalive = rte_keepalive_create(sc_keepalive_failure_handler, sc);
+        if (sc->keepalive == NULL)
+        {
+            MR_ERROR("Create spinlock checker handler failed. ");
+            ret = EXIT_FAILURE;
+            goto quit;
+        }
     }
 
-    if (sc->en_spinlock_check && sc->keepalive == NULL)
-    {
-        MR_ERROR("Create spinlock checker handler failed. ");
-        ret = EXIT_FAILURE;
-        goto quit;
-    }
-
-    http_serv_init(sc);
-
     unsigned int lcore_id_iter = 0;
     RTE_LCORE_FOREACH(lcore_id_iter)
     {
         if (sc->keepalive)
+        {
             rte_keepalive_register_core(sc->keepalive, (int)lcore_id_iter);
-        MR_INFO("Keepalive register for thread %d successfully.", lcore_id_iter);
+            MR_INFO("Keepalive register for thread %d successfully.", lcore_id_iter);
+        }
     }
 
+    http_serv_init(sc);
+
     /* 恢复CPU亲和性设置为EAL后的线程绑定参数 */
     ret = pthread_setaffinity_np(pthread_self(), sizeof(cpu_set_t), &sc->cpu_set_after_eal);
     MR_VERIFY_2(ret >= 0, "Cannot set init thread affinity: %s", strerror(errno));
diff --git a/service/src/node.c b/service/src/node.c
index dba70be..a38a1fa 100644
--- a/service/src/node.c
+++ b/service/src/node.c
@@ -168,6 +168,8 @@ int32_t node_manager_pkt_graph_service_entry(void * args)
         rte_graph_walk(graph_object);
     }
 
+    /* mark the lcore is alive */
+    rte_keepalive_mark_alive(sc->keepalive);
     return 0;
 }