Cellular connection fails on sequential connect

Description

mbed-cli: 1.8.3
toolchain: cmake_gcc_arm
gcc-arm-none-eabi: 7-2018-q2-update (7.3.1)
mbed-os: 5.11.1

I'm using the CellularContext as a cellular interface and am running into the issue where I don't get more than one connect working. The connect only works once when I reset the device.

I'm using a non-blocking connect call and am checking for NSAPI_STATUS_GLOBAL_UP to check whether I'm connected. After a connect, I wait for a couple of seconds before disconnecting. Then again waiting a couple of seconds before starting with the connect again. See the code snippet below for what I'm doing as test, where NetworkConnection is a simple class around CellularContext / NetworkInterface.

This is a simplified example from what I'm trying to do in my application

1. Do some measurements or whatever
2. Connect to the network
3. Send some stuff
4. Disconnect from the network
5. Go back to 1

See the attached log (0113.log) where I have tracing and AT commands enabled. Also see network_connection.txt for that class.

I don't understand why it would only do this section when I'm connecting for the second time. @AriParkkila any ideas?

Blocking calls don't make any difference.

2019-01-01 13:17:57 [16] Connecting... (submodules/mbed-system/source/main.cpp:334)

[INFO][CELL]: CellularContext set blocking 0

[INFO][CELL]: CellularContext connect

[DBG ][CELL]: checked operation 0 (0)


[DBG ][CELL]: returning

TRACE_DEBUG("Starting network connection test...");
    mbed_trace_init();

    wtg::connectivity::NetworkConnection network_connection;

    for (std::uint8_t i = 0; i < 50; i++) {
        TRACE_DEBUG("Connecting...");
        if (network_connection.Connect()) {
            TRACE_DEBUG("Connected to network: %s", network_connection.IpAddress().c_str());
        }
        TRACE_DEBUG("Done connecting! (%hhu)", network_connection.GetConnectionStatus());


        rtos::ThisThread::sleep_for(5000);


        TRACE_DEBUG("Disconnecting...");
        network_connection.Disconnect();
        TRACE_DEBUG("Done disconnecting! (%hhu)", network_connection.GetConnectionStatus());


        rtos::ThisThread::sleep_for(20000);
        PRINTF("\n\n\n\n\n");
    }

    TRACE_DEBUG("network connection test finished!");

Issue request type

[ ] Question
[ ] Enhancement
[x] Bug

[cmonr]

@ARMmbed/mbed-os-wan Please take a look.

[ciarmcom]

Internal Jira reference: https://jira.arm.com/browse/MBOCUSTRIA-890

An interesting thing here is that when I remove _interface->set_blocking(false); in the NetworkConnection (line 44). Every third network connection is able to make a connection. The latter is due to an incorrectly released semaphore (see this).

What I also noticed today is that when I disconnect the antenna from the u-blox SARA-G350 module, it cannot connect while the antenna is disconnected. That is what I'd expect in this case, but when I plug the antenna back in, it is still not able to make a connection. After I reboot my device it immediately is able to create the connection again.
Could it be that this is due to the state that the state machine is in and it is not equipped for the antenna being removed? Seems odd to me if it is, but could be the case.
How can I ensure to completely 'reset' the cellular interface, and via that also the internal state machine, so it wil start the cellular modem again with init and power states?

[mirelachirica]

I have looked a bit on this issue and at least one thing that seems to be problematic, but i need to check more, is that state machine is not reseted on disconnect, even if it should.
I tried to add following:
_device->cellular_callback(NSAPI_EVENT_CONNECTION_STATUS_CHANGE, NSAPI_STATUS_DISCONNECTED); in AT_CellularContext::disconnect() before call_network_cb(NSAPI_STATUS_DISCONNECTED);

Would you try if this is helping any of your cases?

[mirelachirica]

void CellularDevice::cellular_callback() is ready to reset state machine on disconnect:

if (ev == NSAPI_EVENT_CONNECTION_STATUS_CHANGE && ptr == NSAPI_STATUS_DISCONNECTED) {
                // we have been disconnected, reset state machine ...
                if (_state_machine) {
                    _state_machine->reset();
                }
}

, but cellular context is not notifying the device on disconnect. That is what I suggest to do above.

@mirelachirica Thanks for your response.
To be clear, do I need to add the _device->cellular_callback before this line or all call_network_cb(NSAPI_STATUS_DISCONNECTED); calls in AT_CellularContext.cpp?

If you clarify, I'll tests and post back my results.

[mirelachirica]

@vidavidorra I meant only in one place, in disconnect routine. Proper solution might require in all places but I propose to try in this one place first.

I didn't see a noticeable difference in behaviour, unfortunately. I do notice some extra logs (AT OoB) after my first disconnect call. See the log attached.
0124.log.
I've also tried adding the _device->cellular_callback before all call_network_cb(NSAPI_STATUS_DISCONNECTED); calls in AT_CellularContext.cpp, but again no noticeable difference.

@mirelachirica Also I really want to have the connection being set-up the same way it does the first time. I want to completely power down the modem and power it back on when I need to make a cellular connection again. How can I achieve this?

[mirelachirica]

The change I recommended earlier makes a difference only before commit 63fb35 where the call_network_cb(NSAPI_STATUS_DISCONNECTED); was moved under else branch.

When this commit is in use can you try to take call_network_cb(NSAPI_STATUS_DISCONNECTED); out of else and add the _device->cellular_callback before it? Does this change have any impact?

[mirelachirica]

@jarvte can you advice @vidavidorra on the following question:
"Also I really want to have the connection being set-up the same way it does the first time. I want to completely power down the modem and power it back on when I need to make a cellular connection again. How can I achieve this?"