Testing the "eth0 reports no resources" problems, it seems that there is a connection with linux's memory allocation scheme which exasperates the problem. The condition occurs with bridging when the device is first put into promiscuous mode and a large number of packets with an "unknown" destination address are fired at the machine. Even though the skbufs are freed immediately, the memory pool is quickly depleted as it seems the buffers are not put back into usable memory. calling mark_bh() after freeing the buffers seems to substantially lessen the condition (I assume this hastens some function which recaptures the freed buffers?) Calling mark_bh() is a bit kludgy and hopefully not the best solution...is there a direct function which could be called to make the freed buffers immediately available for reuse by the network drivers to handle these types of bursts? Clearly it is desirable to free the packets and make them available as soon as possible in this case without triggering a bogus network event. Dennis - : send the line "unsubscribe linux-net" in the body of a message to majordomo@vger.kernel.org
