Opencores Emac Receive Buffer Memory Issue

[hibengu]

Hi,

I'm experiencing an issue with ESP32 emulation in QEMU where I get the error "E (104160) opencores.emac: no mem for receive buffer".

I have a setup as explained here (https://github.com/espressif/esp-toolch ... working.md). I am starting a docker container and set the network configurations. ESP emulations are assigned IP addresses without a problem from the DHCP server. I am training the a model on the devices and then devices are exchanging neural network parameters, and after a few rounds of training and communication, the EMAC buffer error occurs.

xTaskCreatePinnedToCore(tcp_server_task, "tcp_server", STACK_SIZE, (void*)AF_INET, TASK_PRIORITY, NULL, 0);

I am starting my tcpserver with the line (I had a problem with PSRAM which I fixed it pinning the task to a core). I tried to increase the CONFIG_EXAMPLE_ETHERNET_EMAC_TASK_STACK_SIZE value but regardless, I get the receive buffer error around same time which makes me think it doesn't make any difference. Below is my ethernet register function, I am also setting up a random Mac address at the start, if I keep the Mac value as default all of the devices have the same value and they are assigned to the same ip address.

Code: Select all

void register_ethernet(void)
{
    uint8_t random_mac[6];  // Example MAC address
    random_mac[0] = 0x02;  // Set the first byte as '02' (locally administered address)
    random_mac[1] = esp_random() & 0xFF;  // Random byte 1
    random_mac[2] = esp_random() & 0xFF;  // Random byte 2
    random_mac[3] = esp_random() & 0xFF;  // Random byte 3
    random_mac[4] = esp_random() & 0xFF;  // Random byte 4
    random_mac[5] = esp_random() & 0xFF;  // Random byte 5
    esp_err_t err = esp_base_mac_addr_set(random_mac);

    nvs_flash_init();

    ESP_ERROR_CHECK(esp_netif_init());
    ESP_ERROR_CHECK(esp_event_loop_create_default());
    esp_netif_config_t cfg = ESP_NETIF_DEFAULT_ETH();
    eth_netif = esp_netif_new(&cfg);

    eth_mac_config_t mac_config = ETH_MAC_DEFAULT_CONFIG();
    mac_config.rx_task_stack_size = CONFIG_EXAMPLE_ETHERNET_EMAC_TASK_STACK_SIZE;
    
    eth_phy_config_t phy_config = ETH_PHY_DEFAULT_CONFIG();
    esp_eth_mac_t *mac = esp_eth_mac_new_openeth(&mac_config);

    esp_eth_phy_t *phy = esp_eth_phy_new_dp83848(&phy_config);

    esp_eth_config_t config = ETH_DEFAULT_CONFIG(mac, phy);
    ESP_ERROR_CHECK(esp_eth_driver_install(&config, &eth_handle)); // THIS FAILS
    ESP_ERROR_CHECK(esp_netif_attach(eth_netif, esp_eth_new_netif_glue(eth_handle)));
    ESP_ERROR_CHECK(esp_event_handler_register(ETH_EVENT, ESP_EVENT_ANY_ID, &event_handler, NULL));
    ESP_ERROR_CHECK(esp_event_handler_register(IP_EVENT, IP_EVENT_ETH_GOT_IP, &event_handler, NULL));
    ESP_ERROR_CHECK(esp_eth_start(eth_handle));

}

I tried increasing the task stack size but it didn't change anything. I also checked the remaining stack size and never goes below around 4kbytes. I tried adding delay to my socket communication which also didn't fix the issue and regardless of the delay, it still crashes around the same time on each run which makes me think buffer is not cleared. Does anyone have an idea? thanks for the help in advance.