[lwip-users] Pbuf starvation

[Baptiste Chaboud-crousaz]

Hi all,

I use the lwIP stack without OS on a STM32F107 processor (ST micro). The processor Ethernet module is configured to assert an interrupt when a complete Ethernet Frame is received. Upon interrupt I call "ethernetif_input(struct netif *netif)" which the code is given below:

void ethernetif_input(struct netif *netif)
{
  struct ethernetif *ethernetif;
  struct eth_hdr *ethhdr;
  struct pbuf *p = NULL;

  ethernetif = netif->state;

  /* move received packet into a new pbuf */
  while((p = low_level_input(netif)) != NULL)
  { 
      /* points to packet payload, which starts with an Ethernet header */
      ethhdr = p->payload;

      switch (htons(ethhdr->type))
      {
          case ETHTYPE_ARP:  /* ARP packet */
              etharp_arp_input(netif, ethernetif->ethaddr, p);
              break;
          case ETHTYPE_IP:  /* IP packet */
              /* Update the ARP table */
              etharp_ip_input(netif, p);

              /* Skip the Ethernet header */
              pbuf_header(p, -((s16_t)sizeof(struct eth_hdr)));

              /* Pass the packet to the network layer.*/
              netif->input(p, netif);
              break;
          default:
              pbuf_free(p);
              p = NULL;
              break;
      }
  }
}

The function "struct pbuf * low_level_input(struct netif *netif)" is called until no more data is available. This function allocates a pbuf buffer to store the received data: pbuf_alloc(PBUF_RAW, len, PBUF_POOL);.

Between each call to the "struct pbuf * low_level_input(struct netif *netif)", the received data are processed according to their type: ARP or IP packet (for IP packet netif->input is "err_t ip_input(struct pbuf *p, struct netif *inp)").

As I can see once the packet is processed (except if it is a reassembled packet), the allocated pbuf ,in which the data are stored , is freed by a call to pbuf_free.

Nevertheless after a short period without error, the pbuf allocation failed because no more free space is available for a new allocation. I don't understand why!!! Is someone can help me to understand why the previously allocated pbuf are not correctly freed?

I must underline the fact that this problem doesn't occur when the network traffic is very low loaded. But when the network has to support a heavy loaded traffic, the problem appears rapidely.

Thanks for your help.
Best regards.

Baptiste