Delete i2c cmd_mux semaphore more cleanly (v4.3 branch) (IDFGH-5061)

[makermelissa]

See adafruit#1 for more info.

[CLAassistant]

All committers have signed the CLA.

[dhalbert]

This change fixes an problem we have seen in CircuitPython between i2c_driver_delete() and esp_wifi_start(). It is not a hard crash, but in our case we lose the USB CDC connection, and things stop running: it may be a runaway set of interrupts or some other loop. The usual IDF threads that are running when the issue happens look the same as in the good case.

The essence of the problem is this sequence:

i2c_driver_delete(NULL);   // does nothing
i2c_param_config();
i2c_driver_install();
i2c_driver_delete();   // This is crucial.

esp_netif_init();               
esp_event_loop_create_default();
xEventGroupCreateStatic();
esp_event_handler_instance_register();
esp_event_handler_instance_register();
esp_wifi_init();
esp_wifi_set_mode();
esp_wifi_start();

I arrived at this change by divide and conquer, seeing which part of the i2c.c was causing the issue.Unfortunately esp_wifi_start(), which is where the problem seems to occur, is in the closed-source wifi stack.

The critical piece of the original i2c_driver_delete() code is:

xTakeSemaphore();
vSemaphoreDelete();

If the xTakeSemaphore() is completely removed, or if the xGiveSemaphore() is added, then we don't see the problem. I arrived at adding the xGiveSemaphore() by examining other semaphore cleanup code in the IDF.

I have tried to create a minimal example in ESP-IDF. I have one I, but the only difference I see immediately is that the DEBUG traces between the "good" and the "bad" cases. It would be great if someone with access could look inside esp_wifi_start() to see what the actual issue might be.

Other drivers (e.g., SPI, etc.) do not affect esp_wifi_start() in this way. As mentioned, there is more information in adafruit#1 and a longer discussion in adafruit/circuitpython#4387.

[Alvin1Zhang]

Thanks for your contribution.

[Alvin1Zhang]

Thanks for your contribution, and sorry for late reply, we have put the changes into our internal review queue, and the PR will be updated once the changes are available on GitHub. Thanks.

[L-KAYA]

Thanks for your contribution again! This PR has been merged into release/v4.3 (commit:2ac59cc8). Could you help to check whether the issue has been fixed.

[Alvin1Zhang]

Thanks for your contribution.