qcacmn: Include PID in SM history

Add PID name where SM events and transition
happened to help debug issues where the info
about which task started a certain event to
happen are crucial to rootcause the problem.

Change-Id: I208b46c6d0ec725ab0d9713bf2dd562c40719a24
CRs-Fixed: 3624811

qdf/inc/qdf_types.h

@@ -1606,6 +1606,14 @@ enum qdf_suspend_type {
  * of order packet counter values
  * @QDF_MGMT_RX_REO_ZERO_DURATION_PKT: Reception of management packet with zero
  * packet duration
+ * @QDF_VDEV_ACTIVE_SER_CONNECT_TIMEOUT: Active connect cmd in serialization
+ * timed out.
+ * @QDF_VDEV_ACTIVE_SER_DISCONNECT_TIMEOUT: Active disconnect cmd in
+ * serialization timed out.
+ * @QDF_VDEV_ACTIVE_SER_REASSOC_TIMEOUT:  Active reassoc cmd in serialization
+ * timed out.
+ * @QDF_VDEV_ACTIVE_SER_LINK_SWITCH_TIMEOUT: Active link switch cmd in
+ * serialization timed out.
  */
 enum qdf_hang_reason {
 	QDF_REASON_UNSPECIFIED,
@@ -1646,6 +1654,10 @@ enum qdf_hang_reason {
 	QDF_MGMT_RX_REO_INCONSISTENT_SNAPSHOT,
 	QDF_MGMT_RX_REO_OUT_OF_ORDER_PKT,
 	QDF_MGMT_RX_REO_ZERO_DURATION_PKT,
+	QDF_VDEV_ACTIVE_SER_CONNECT_TIMEOUT,
+	QDF_VDEV_ACTIVE_SER_DISCONNECT_TIMEOUT,
+	QDF_VDEV_ACTIVE_SER_REASSOC_TIMEOUT,
+	QDF_VDEV_ACTIVE_SER_LINK_SWITCH_TIMEOUT,
 };
 
 /**

umac/cmn_services/sm_engine/inc/wlan_sm_engine_dbg.h

@@ -51,7 +51,12 @@
 #define sm_engine_nofl_debug(params...) \
 	QDF_TRACE_DEBUG_NO_FL(QDF_MODULE_ID_SM_ENGINE, params)
 
+#ifdef CONN_MGR_ADV_FEATURE
+#define WLAN_SM_ENGINE_HISTORY_SIZE 15
+#else
 #define WLAN_SM_ENGINE_HISTORY_SIZE 50
+#endif /* CONN_MGR_ADV_FEATURE */
+
 struct wlan_sm;
 /**
  * enum wlan_sm_trace_type - history element type
@@ -64,7 +69,7 @@ enum wlan_sm_trace_type {
 };
 
 #ifdef SM_ENG_HIST_ENABLE
-
+#define WLAN_SM_PID_MAX_LEN 7
 /**
  * struct wlan_sm_history_info - history element structure
  * @trace_type:      history element type
@@ -72,6 +77,7 @@ enum wlan_sm_trace_type {
  * @initial_state:   Current state (state/sub-state)
  * @final_state:     New state
  * @time:            Timestamp
+ * @pid_name:        Name of task (truncated to WLAN_SM_PID_MAX_LEN bytes)
  */
 struct wlan_sm_history_info {
 	enum wlan_sm_trace_type trace_type;
@@ -79,6 +85,7 @@ struct wlan_sm_history_info {
 	uint8_t initial_state;
 	uint8_t final_state;
 	uint64_t time;
+	char pid_name[WLAN_SM_PID_MAX_LEN];
 };
 
 /**

umac/cmn_services/sm_engine/src/wlan_sm_engine_dbg.c

@@ -18,6 +18,7 @@
 /**
  * DOC: Implements general SM debug framework
  */
+#include <qdf_threads.h>
 #include <wlan_sm_engine.h>
 #include <wlan_sm_engine_dbg.h>
 
@@ -47,6 +48,8 @@ void wlan_sm_save_history(struct wlan_sm *sm,
 	p_memento->final_state = final_state;
 	p_memento->event_type = event_type;
 	p_memento->time = qdf_get_log_timestamp();
+	qdf_scnprintf(p_memento->pid_name, WLAN_SM_PID_MAX_LEN, "%.6s",
+		      qdf_get_current_comm());
 }
 
 void wlan_sm_history_init(struct wlan_sm *sm)
@@ -74,8 +77,8 @@ static void wlan_sm_print_history_entry(struct wlan_sm *sm,
 			return;
 
 		sm_engine_nofl_err(
-			"| 0x%016llx |%6d |%11d |%23s[%3d] |%19s[%2d] |%19s[%2d] |",
-			ent->time, i, ent->trace_type,
+			"| 0x%016llx |%6d |%6s |%11d |%23s[%3d] |%19s[%2d] |%19s[%2d] |",
+			ent->time, i, ent->pid_name, ent->trace_type,
 			event_name ? event_name : "UNKNOWN_EVENT",
 			ent->event_type,
 			sm->state_info[ent->initial_state].name,
@@ -84,8 +87,8 @@ static void wlan_sm_print_history_entry(struct wlan_sm *sm,
 			ent->final_state);
 	} else {
 		sm_engine_nofl_err(
-			"| 0x%016llx |%6d |%11d |%28d |%19s[%2d] |%19s[%2d] |",
-			ent->time, i, ent->trace_type,
+			"| 0x%016llx |%6d |%6s |%11d |%28d |%19s[%2d] |%19s[%2d] |",
+			ent->time, i, ent->pid_name, ent->trace_type,
 			ent->event_type,
 			sm->state_info[ent->initial_state].name,
 			ent->initial_state,
@@ -106,8 +109,8 @@ void wlan_sm_print_history(struct wlan_sm *sm)
 	 */
 	qdf_spin_lock_bh(&p_sm_history->sm_history_lock);
 
-	sm_engine_nofl_err("|%19s |%6s |%11s |%28s |%23s |%23s |", "Time",
-			   "Index", "Trace Type", "Event",
+	sm_engine_nofl_err("|%19s |%6s |%6s |%11s |%28s |%23s |%23s |", "Time",
+			   "Index", "PID", "Trace Type", "Event",
 			   "Initial State", "Final State");
 
 	for (i = 0; i < WLAN_SM_ENGINE_HISTORY_SIZE; i++) {
@@ -134,8 +137,8 @@ static void wlan_sm_print_fs_history_entry(struct wlan_sm *sm,
 			return;
 
 		qdf_debugfs_printf(
-			m, "| 0x%016llx |%6d |%11d |%23s[%3d] |%19s[%2d] |%19s[%2d] |\n",
-			ent->time, i, ent->trace_type,
+			m, "| 0x%016llx |%6d |%6s |%11d |%23s[%3d] |%19s[%2d] |%19s[%2d] |\n",
+			ent->time, i, ent->pid_name, ent->trace_type,
 			event_name ? event_name : "UNKNOWN_EVENT",
 			ent->event_type,
 			sm->state_info[ent->initial_state].name,
@@ -144,8 +147,8 @@ static void wlan_sm_print_fs_history_entry(struct wlan_sm *sm,
 			ent->final_state);
 	} else {
 		qdf_debugfs_printf(
-			m, "| 0x%016llx |%6d |%11d |%28d |%19s[%2d] |%19s[%2d] |\n",
-			ent->time, i, ent->trace_type,
+			m, "| 0x%016llx |%6d |%6s |%11d |%28d |%19s[%2d] |%19s[%2d] |\n",
+			ent->time, i, ent->pid_name, ent->trace_type,
 			ent->event_type,
 			sm->state_info[ent->initial_state].name,
 			ent->initial_state,
@@ -165,8 +168,8 @@ void wlan_sm_print_fs_history(struct wlan_sm *sm, qdf_debugfs_file_t m)
 	 * Save a pointer to next write location and increment pointer.
 	 */
 	qdf_spin_lock_bh(&p_sm_history->sm_history_lock);
-	qdf_debugfs_printf(m, "|%19s |%6s |%11s |%28s |%23s |%23s |\n", "Time",
-			   "Index", "Trace Type", "Event",
+	qdf_debugfs_printf(m, "|%19s |%6s |%6s |%11s |%28s |%23s |%23s |\n", "Time",
+			   "Index", "PID", "Trace Type", "Event",
 			   "Initial State", "Final State");
 
 	for (i = 0; i < WLAN_SM_ENGINE_HISTORY_SIZE; i++) {

umac/mlme/connection_mgr/core/src/wlan_cm_connect.c

@@ -175,6 +175,7 @@ cm_ser_connect_cb(struct wlan_serialization_command *cmd,
 	QDF_STATUS status = QDF_STATUS_SUCCESS;
 	struct wlan_objmgr_vdev *vdev;
 	struct cnx_mgr *cm_ctx;
+	enum qdf_hang_reason hang_reason = QDF_VDEV_ACTIVE_SER_CONNECT_TIMEOUT;
 
 	if (!cmd) {
 		mlme_err("cmd is NULL, reason: %d", reason);
@@ -219,7 +220,7 @@ cm_ser_connect_cb(struct wlan_serialization_command *cmd,
 	case WLAN_SER_CB_ACTIVE_CMD_TIMEOUT:
 		mlme_err(CM_PREFIX_FMT "Active command timeout",
 			 CM_PREFIX_REF(wlan_vdev_get_id(vdev), cmd->cmd_id));
-		cm_trigger_panic_on_cmd_timeout(cm_ctx->vdev);
+		cm_trigger_panic_on_cmd_timeout(cm_ctx->vdev, hang_reason);
 		cm_connect_cmd_timeout(cm_ctx, cmd->cmd_id);
 		break;
 	case WLAN_SER_CB_RELEASE_MEM_CMD:

umac/mlme/connection_mgr/core/src/wlan_cm_disconnect.c

@@ -158,6 +158,8 @@ cm_ser_disconnect_cb(struct wlan_serialization_command *cmd,
 	QDF_STATUS status = QDF_STATUS_SUCCESS;
 	struct wlan_objmgr_vdev *vdev;
 	struct cnx_mgr *cm_ctx;
+	enum qdf_hang_reason hang_reason =
+				QDF_VDEV_ACTIVE_SER_DISCONNECT_TIMEOUT;
 
 	if (!cmd) {
 		mlme_err("cmd is NULL, reason: %d", reason);
@@ -192,7 +194,7 @@ cm_ser_disconnect_cb(struct wlan_serialization_command *cmd,
 	case WLAN_SER_CB_ACTIVE_CMD_TIMEOUT:
 		mlme_err(CM_PREFIX_FMT "Active command timeout",
 			 CM_PREFIX_REF(wlan_vdev_get_id(vdev), cmd->cmd_id));
-		cm_trigger_panic_on_cmd_timeout(cm_ctx->vdev);
+		cm_trigger_panic_on_cmd_timeout(cm_ctx->vdev, hang_reason);
 		cm_send_disconnect_resp(cm_ctx, cmd->cmd_id);
 		break;
 	case WLAN_SER_CB_RELEASE_MEM_CMD:

umac/mlme/connection_mgr/core/src/wlan_cm_host_roam.c

@@ -571,6 +571,8 @@ cm_ser_reassoc_cb(struct wlan_serialization_command *cmd,
 	QDF_STATUS status = QDF_STATUS_SUCCESS;
 	struct wlan_objmgr_vdev *vdev;
 	struct cnx_mgr *cm_ctx;
+	enum qdf_hang_reason hang_reason =
+				QDF_VDEV_ACTIVE_SER_REASSOC_TIMEOUT;
 
 	if (!cmd) {
 		mlme_err("cmd is NULL, reason: %d", reason);
@@ -614,7 +616,7 @@ cm_ser_reassoc_cb(struct wlan_serialization_command *cmd,
 	case WLAN_SER_CB_ACTIVE_CMD_TIMEOUT:
 		mlme_err(CM_PREFIX_FMT "Active command timeout",
 			 CM_PREFIX_REF(wlan_vdev_get_id(vdev), cmd->cmd_id));
-		cm_trigger_panic_on_cmd_timeout(cm_ctx->vdev);
+		cm_trigger_panic_on_cmd_timeout(cm_ctx->vdev, hang_reason);
 		cm_reassoc_cmd_timeout(cm_ctx, cmd->cmd_id);
 		break;
 	case WLAN_SER_CB_RELEASE_MEM_CMD:

umac/mlme/connection_mgr/core/src/wlan_cm_main_api.h

@@ -1007,10 +1007,12 @@ void cm_set_max_connect_attempts(struct wlan_objmgr_vdev *vdev,
 /**
  * cm_trigger_panic_on_cmd_timeout() - trigger panic on active command timeout
  * @vdev: vdev pointer
+ * @reason: Hang reason code
  *
  * Return: void
  */
-void cm_trigger_panic_on_cmd_timeout(struct wlan_objmgr_vdev *vdev);
+void cm_trigger_panic_on_cmd_timeout(struct wlan_objmgr_vdev *vdev,
+				     enum qdf_hang_reason reason);
 
 /**
  * cm_set_max_connect_timeout() - Set max connect timeout

umac/mlme/connection_mgr/core/src/wlan_cm_util.c

@@ -197,6 +197,24 @@ QDF_STATUS cm_set_key(struct cnx_mgr *cm_ctx, bool unicast,
 }
 #endif
 
+static void cm_dump_sm_history(struct wlan_objmgr_vdev *vdev)
+{
+	struct vdev_mlme_obj *vdev_mlme;
+	struct wlan_sm *vdev_sm;
+
+	vdev_mlme = wlan_objmgr_vdev_get_comp_private_obj(vdev,
+							  WLAN_UMAC_COMP_MLME);
+	if (!vdev_mlme)
+		return;
+
+	vdev_sm = vdev_mlme->sm_hdl;
+	if (!vdev_sm)
+		return;
+
+	wlan_sm_print_history(vdev_sm);
+	cm_sm_history_print(vdev);
+}
+
 #ifdef CONN_MGR_ADV_FEATURE
 void cm_store_wep_key(struct cnx_mgr *cm_ctx,
 		      struct wlan_cm_connect_crypto_info *crypto,
@@ -255,7 +273,8 @@ void cm_store_wep_key(struct cnx_mgr *cm_ctx,
 		   wep_keys->seq_len);
 }
 
-void cm_trigger_panic_on_cmd_timeout(struct wlan_objmgr_vdev *vdev)
+void cm_trigger_panic_on_cmd_timeout(struct wlan_objmgr_vdev *vdev,
+				     enum qdf_hang_reason reason)
 {
 	struct wlan_objmgr_psoc *psoc;
 
@@ -266,32 +285,15 @@ void cm_trigger_panic_on_cmd_timeout(struct wlan_objmgr_vdev *vdev)
 	if (qdf_is_recovering() || qdf_is_fw_down())
 		return;
 
-	qdf_trigger_self_recovery(psoc, QDF_ACTIVE_LIST_TIMEOUT);
+	cm_dump_sm_history(vdev);
+	qdf_trigger_self_recovery(psoc, reason);
 }
 
 #else
-void cm_trigger_panic_on_cmd_timeout(struct wlan_objmgr_vdev *vdev)
+void cm_trigger_panic_on_cmd_timeout(struct wlan_objmgr_vdev *vdev,
+				     enum qdf_hang_reason reason)
 {
-	struct vdev_mlme_obj *vdev_mlme = NULL;
-	struct wlan_sm *vdev_sm = NULL;
-
-	vdev_mlme = wlan_objmgr_vdev_get_comp_private_obj(
-			vdev,
-			WLAN_UMAC_COMP_MLME);
-	if (!vdev_mlme) {
-		mlme_err("VDEV MLME is null");
-		goto error;
-	}
-
-	vdev_sm = vdev_mlme->sm_hdl;
-	if (!vdev_sm) {
-		mlme_err("VDEV SM is null");
-		goto error;
-	}
-
-	wlan_sm_print_history(vdev_sm);
-	cm_sm_history_print(vdev);
-error:
+	cm_dump_sm_history(vdev);
 	QDF_ASSERT(0);
 }
 #endif

umac/mlme/connection_mgr/dispatcher/inc/wlan_cm_api.h

@@ -256,6 +256,16 @@ bool wlan_cm_is_link_switch_disconnect_resp(struct wlan_cm_discon_rsp *resp);
  */
 bool wlan_cm_is_link_switch_connect_resp(struct wlan_cm_connect_resp *resp);
 
+/**
+ * wlan_cm_trigger_panic_on_cmd_timeout() - Trigger recovery on CM command
+ * timeout.
+ * @vdev: VDEV object manager
+ * @reason: Hang reason code
+ *
+ * Return: void
+ */
+void wlan_cm_trigger_panic_on_cmd_timeout(struct wlan_objmgr_vdev *vdev,
+					  enum qdf_hang_reason reason);
 #ifdef WLAN_FEATURE_ROAM_OFFLOAD
 /**
  * wlan_cm_is_vdev_roam_started() - check if vdev is in roaming state and

umac/mlme/connection_mgr/dispatcher/src/wlan_cm_api.c

@@ -265,6 +265,12 @@ bool wlan_cm_is_link_switch_connect_resp(struct wlan_cm_connect_resp *resp)
 	return cm_is_link_switch_connect_resp(resp);
 }
 
+void wlan_cm_trigger_panic_on_cmd_timeout(struct wlan_objmgr_vdev *vdev,
+					  enum qdf_hang_reason reason)
+{
+	cm_trigger_panic_on_cmd_timeout(vdev, reason);
+}
+
 #ifdef WLAN_FEATURE_HOST_ROAM
 bool wlan_cm_get_active_reassoc_req(struct wlan_objmgr_vdev *vdev,
 				    struct wlan_cm_vdev_reassoc_req *req)

umac/mlo_mgr/src/wlan_mlo_mgr_link_switch.c

@@ -892,38 +892,17 @@ mlo_mgr_start_link_switch(struct wlan_objmgr_vdev *vdev,
 	return status;
 }
 
-/**
- * mlo_mgr_trigger_recovery_on_link_switch_timeout() - trigger panic on link
- * switch timeout
- * @vdev: vdev pointer
- *
- * Return: void
- */
-static void
-mlo_mgr_trigger_recovery_on_link_switch_timeout(struct wlan_objmgr_vdev *vdev)
-{
-	struct wlan_objmgr_psoc *psoc;
-
-	psoc = wlan_vdev_get_psoc(vdev);
-	if (!psoc)
-		return;
-
-	if (qdf_is_recovering() || qdf_is_fw_down())
-		return;
-
-	qdf_trigger_self_recovery(psoc, QDF_ACTIVE_LIST_TIMEOUT);
-}
-
 static QDF_STATUS
 mlo_mgr_ser_link_switch_cb(struct wlan_serialization_command *cmd,
-			   enum wlan_serialization_cb_reason reason)
+			   enum wlan_serialization_cb_reason cb_reason)
 {
 	struct wlan_objmgr_vdev *vdev;
 	QDF_STATUS status = QDF_STATUS_SUCCESS;
 	struct wlan_mlo_link_switch_req *req;
+	enum qdf_hang_reason reason = QDF_VDEV_ACTIVE_SER_LINK_SWITCH_TIMEOUT;
 
 	if (!cmd) {
-		mlo_err("cmd is NULL, reason: %d", reason);
+		mlo_err("cmd is NULL, reason: %d", cb_reason);
 		QDF_ASSERT(0);
 		return QDF_STATUS_E_NULL_VALUE;
 	}
@@ -931,7 +910,7 @@ mlo_mgr_ser_link_switch_cb(struct wlan_serialization_command *cmd,
 	vdev = cmd->vdev;
 	req = &vdev->mlo_dev_ctx->link_ctx->last_req;
 
-	switch (reason) {
+	switch (cb_reason) {
 	case WLAN_SER_CB_ACTIVATE_CMD:
 		status = mlo_mgr_start_link_switch(vdev, cmd);
 		if (QDF_IS_STATUS_ERROR(status)) {
@@ -947,7 +926,7 @@ mlo_mgr_ser_link_switch_cb(struct wlan_serialization_command *cmd,
 		break;
 	case WLAN_SER_CB_ACTIVE_CMD_TIMEOUT:
 		mlo_err("Link switch active cmd timeout");
-		mlo_mgr_trigger_recovery_on_link_switch_timeout(vdev);
+		wlan_cm_trigger_panic_on_cmd_timeout(vdev, reason);
 		break;
 	default:
 		QDF_ASSERT(0);