"ble_gap_terminate" Occasionally Returns Unexpected HCI Error

With my application (on an ESP32), in some cases I have to call "ble_gap_terminate" from within the event callback of "ble_gap_adv_start", for the event "BLE_GAP_EVENT_CONNECT". The NimBLE documentation suggests that making this kind of call from a callback should work without issue. In some cases this call will return "BLE_HS_ENOTCONN," which is expected e.g. if the connection had immediately failed right after the connection event was raised. However, if I force this code path to run repeatedly, then about 1 out of every 2000 times it runs I instead get the HCI error "BLE_ERR_UNK_CONN_ID". This error probably isn't supposed to leak out like this in normal use.

Looking at the source for "ble_gap_terminate", it appears that the check for "BLE_HS_ENOTCONN" is implemented by a call to "ble_hs_conn_find," which is run before the HCI command is transmitted. There are several concurrency-related reasons why that check might pass, even though the later HCI command will fail. It'd probably be better to specifically check for the "BLE_ERR_UNK_CONN_ID" HCI error, and return "BLE_HS_ENOTCONN" in that case.

Start by tracing ble_gap_terminate, including its ble_hs_conn_find check and the HCI command path reached from the BLE_GAP_EVENT_CONNECT callback. Reproduce the repeated termination scenario if possible, then verify that BLE_ERR_UNK_CONN_ID is handled as BLE_HS_ENOTCONN without changing other HCI errors.

Written by the indexing model from the issue text.