> calling mark_bh() after freeing the buffers seems to substantially lessen > the condition (I assume this hastens some function which recaptures the > freed buffers?) Yes. However mark_bh(NET_BH) is done for each packet queued anyway so that doesnt make sense (netif_rx does it). It might by an indication of this being timing issues, however until we can get good databooks out of Intel who knows - : send the line "unsubscribe linux-net" in the body of a message to majordomo@vger.kernel.org
