RE: NRF7002 using one virtual interface breaks a different one

SpeedCrash100 — Tue, 17 Sep 2024 08:01:36 GMT

Apparently, the original example driver has a flaw:

netdev_tx_t nrf_wifi_netdev_start_xmit(struct sk_buff *skb,
                       struct net_device *netdev)
{
    // Removed for brevity...
    fmac_dev_ctx = rpu_ctx_lnx->rpu_ctx;
    def_dev_ctx = wifi_dev_priv(fmac_dev_ctx);
    host_stats = &def_dev_ctx->host_stats;

    // Removed for brevity...

    if ((vif_ctx_lnx->num_tx_pkt - host_stats->total_tx_pkts) >=
        CONFIG_NRF700X_MAX_TX_PENDING_QLEN) {
        if (!netif_queue_stopped(netdev)) {
            netif_stop_queue(netdev);
        }
        schedule_work(&vif_ctx_lnx->ws_queue_monitor);
    }

    // Removed for brevity...

    vif_ctx_lnx->num_tx_pkt++;
    schedule_work(&vif_ctx_lnx->ws_data_tx);

out:
    return ret;
}

With second interface it seems that packet divided into parts and total_tx_pkts becomes bigger than num_tx_pkt the linux driver count. It creates integer underflow because both of them are unsigned and stops the queue forever.

As a workaround I will use the difference: `host_stats->total_tx_pkts - host_stats->total_tx_done_pkts`