On Thu, Oct 25, 2018 at 09:51:38PM +0200, Alan Kayahan wrote: > Hello, > > I have 3 OVS bridges on the same host, connected to each other as > br1<->br2<->br3. br1 and br3 are connected to the docker container cA > via dpdkvhostuser port type (I know it is deprecated, the app works > this way only). The DPDK app running in cA generate packets, which > traverse bridges br1->br2->br3, then ends up back at the DPDK app. > This setup works fine. > > Now I am trying to put each OVS bridge into its respective docker > container. I connect the containers with veth pairs, then add the veth > ports to the bridges. Next, I add a dpdkvhostuser port named SRC to > br1, so far so good. The moment I add a dpdkvhostuser port named SNK > to br3, ovs-vswitchd services in br1's and br3's containers segfault. > Following are the backtraces from each, > > ------------------br1's container--------------- > > [Thread debugging using libthread_db enabled] > Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1". > Core was generated by `ovs-vswitchd > unix:/usr/local/var/run/openvswitch/db.sock -vconsole:emer -vsyslo'. > Program terminated with signal SIGSEGV, Segmentation fault. > #0 0x00005608fa0f321b in netdev_rxq_recv (rx=0x7ff13c34ee80, > batch=batch@entry=0x7ff1bbb4d890) at lib/netdev.c:702 > 702 retval = rx->netdev->netdev_class->rxq_recv(rx, batch); > [Current thread is 1 (Thread 0x7ff1bbb4e700 (LWP 376))] > (gdb) bt > #0 0x00005608fa0f321b in netdev_rxq_recv (rx=0x7ff13c34ee80, > batch=batch@entry=0x7ff1bbb4d890) at lib/netdev.c:702 > #1 0x00005608fa0cce65 in dp_netdev_process_rxq_port ( > pmd=pmd@entry=0x7ff1bbb4f010, rxq=0x5608fb651be0, port_no=1) > at lib/dpif-netdev.c:3279 > #2 0x00005608fa0cd296 in pmd_thread_main (f_=<optimized out>) > at lib/dpif-netdev.c:4145 > #3 0x00005608fa14a836 in ovsthread_wrapper (aux_=<optimized out>) > at lib/ovs-thread.c:348 > #4 0x00007ff1c52517fc in start_thread (arg=0x7ff1bbb4e700) > at pthread_create.c:465 > #5 0x00007ff1c4815b5f in clone () > at ../sysdeps/unix/sysv/linux/x86_64/clone.S:95 > > ------------------br3's container--------------- > > [Thread debugging using libthread_db enabled] > Using host libthread_db library "/lib/x86_64-linux-gnu/libthread_db.so.1". > Core was generated by `ovs-vswitchd > unix:/usr/local/var/run/openvswitch/db.sock -vconsole:emer -vsyslo'. > Program terminated with signal SIGSEGV, Segmentation fault. > #0 0x000055c517e3abcb in rte_mempool_free_memchunks () > [Current thread is 1 (Thread 0x7f202351f300 (LWP 647))] > (gdb) bt > #0 0x000055c517e3abcb in rte_mempool_free_memchunks () > #1 0x000055c517e3ad46 in rte_mempool_free.part () > #2 0x000055c518218b78 in dpdk_mp_free (mp=0x7f603fe66a00) > at lib/netdev-dpdk.c:599 > #3 0x000055c518218ff0 in dpdk_mp_free (mp=<optimized out>) > at lib/netdev-dpdk.c:593 > #4 netdev_dpdk_mempool_configure (dev=0x7f1f7ffeac00) at > lib/netdev-dpdk.c:629 > #5 0x000055c51821a98d in dpdk_vhost_reconfigure_helper (dev=0x7f1f7ffeac00) > at lib/netdev-dpdk.c:3599 > #6 0x000055c51821ac8b in netdev_dpdk_vhost_reconfigure > (netdev=0x7f1f7ffebcc0) > at lib/netdev-dpdk.c:3624 > #7 0x000055c51813fe6b in port_reconfigure (port=0x55c51a4522a0) > at lib/dpif-netdev.c:3341 > #8 reconfigure_datapath (dp=dp@entry=0x55c51a46efc0) at > lib/dpif-netdev.c:3822 > #9 0x000055c5181403e8 in do_add_port (dp=dp@entry=0x55c51a46efc0, > devname=devname@entry=0x55c51a456520 "SNK", > type=0x55c51834f7bd "dpdkvhostuser", port_no=port_no@entry=1) > at lib/dpif-netdev.c:1584 > #10 0x000055c51814059b in dpif_netdev_port_add (dpif=<optimized out>, > netdev=0x7f1f7ffebcc0, port_nop=0x7fffb4eef68c) at lib/dpif-netdev.c:1610 > #11 0x000055c5181469be in dpif_port_add (dpif=0x55c51a469350, > netdev=netdev@entry=0x7f1f7ffebcc0, > port_nop=port_nop@entry=0x7fffb4eef6ec) > at lib/dpif.c:579 > ---Type <return> to continue, or q <return> to quit--- > #12 0x000055c5180f9f28 in port_add (ofproto_=0x55c51a464ee0, > netdev=0x7f1f7ffebcc0) at ofproto/ofproto-dpif.c:3645 > #13 0x000055c5180ecafe in ofproto_port_add (ofproto=0x55c51a464ee0, > netdev=0x7f1f7ffebcc0, ofp_portp=ofp_portp@entry=0x7fffb4eef7e8) at > ofproto/ofproto.c:1999 > #14 0x000055c5180d97e6 in iface_do_create (errp=0x7fffb4eef7f8, > netdevp=0x7fffb4eef7f0, ofp_portp=0x7fffb4eef7e8, > iface_cfg=0x55c51a46d590, br=0x55c51a4415b0) > at vswitchd/bridge.c:1799 > #15 iface_create (port_cfg=0x55c51a46e210, iface_cfg=0x55c51a46d590, > br=0x55c51a4415b0) at vswitchd/bridge.c:1837 > #16 bridge_add_ports__ (br=br@entry=0x55c51a4415b0, > wanted_ports=wanted_ports@entry=0x55c51a441690, > with_requested_port=with_requested_port@entry=true) at > vswitchd/bridge.c:931 > #17 0x000055c5180db87a in bridge_add_ports > (wanted_ports=0x55c51a441690, br=0x55c51a4415b0) at > vswitchd/bridge.c:942 > #18 bridge_reconfigure (ovs_cfg=ovs_cfg@entry=0x55c51a46ea80) at > vswitchd/bridge.c:661 > #19 0x000055c5180df989 in bridge_run () at vswitchd/bridge.c:3016 > #20 0x000055c517dbc535 in main (argc=<optimized out>, argv=<optimized > out>) at vswitchd/ovs-vswitchd.c:120 > > Note that /dev/hugepages of the host is shared with all containers. I > have a feeling that br3 is overwriting the hugepage file of br1. Any > ideas?
It does look like some kind of bad pointer, since rx->netdev->netdev_class->rxq_recv shouldn't segfault. Is there a way you can rerun with Valgrind or Address Sanitizer? _______________________________________________ discuss mailing list disc...@openvswitch.org https://mail.openvswitch.org/mailman/listinfo/ovs-discuss