[dpdk-dev] [Bug 116] Single-port, multi-core and multi-queue mode (open RSS), when configuring IP, may cause dpdk coredump

bugzilla at dpdk.org bugzilla at dpdk.org
Wed Dec 5 04:01:37 CET 2018


            Bug ID: 116
           Summary: Single-port, multi-core and multi-queue mode (open
                    RSS), when configuring IP, may cause dpdk coredump
           Product: DPDK
           Version: unspecified
          Hardware: x86
                OS: Linux
            Status: CONFIRMED
          Severity: critical
          Priority: Normal
         Component: ethdev
          Assignee: dev at dpdk.org
          Reporter: he.qiao17 at zte.com.cn
  Target Milestone: ---

Coredump stack information:
Thread 153 "lcore-slave-1" received signal SIGSEGV, Segmentation fault.
#0  0x00007ffff37b570d in ixgbe_rxq_rearm (rxq=0x7ffd4e5ed680) at
#1  0x00007ffff37b6740 in _recv_raw_pkts_vec (rxq=0x7ffd4e5ed680,
rx_pkts=0x926db0 <app+597232>, nb_pkts=32, split_packet=0x7fff227f5820 "")at
#2  0x00007ffff37b743b in ixgbe_recv_scattered_pkts_vec
(rx_queue=0x7ffd4e5ed680, rx_pkts=0x926db0 <app+597232>, nb_pkts=144) at
#3  0x0000000000515bd1 in rte_eth_rx_burst (port_id=0 '\000', queue_id=0,
rx_pkts=0x926db0 <app+597232>, nb_pkts=144) at rte_ethdev.h:2659
#4  0x000000000051c1b4 in app_lcore_io_rx

My dpdk IO rx/tx queue configuration is as follows:
 --rx (0,0,1),(0,1,9),(0,2,17) --tx (0,1)
one core corresponds to one queue, using multiple queues.

When configuring IP, function rte_kni_handle_request() can get the
RTE_KNI_REQ_CFG_NETWORK_IF message from the kernel. At this time,
rte_eth_dev_stop will be called. All the queue information on the current port
will be cleared in this process, but some queues are being used by other cores.

rte_eth_dev_stop() -> ixgbe_dev_stop() -> ixgbe_dev_clear_queues() ->

void __attribute__((cold))
ixgbe_dev_clear_queues(struct rte_eth_dev *dev)
        for (i = 0; i < dev->data->nb_rx_queues; i++) {
                struct ixgbe_rx_queue *rxq = dev->data->rx_queues[i];

static inline void
_ixgbe_rx_queue_release_mbufs_vec(struct ixgbe_rx_queue *rxq)
        /* set all entries to NULL */
        memset(rxq->sw_ring, 0, sizeof(rxq->sw_ring[0]) * rxq->nb_rx_desc);

The reason for coredump is known, but I do not know how to solve it gracefully

You are receiving this mail because:
You are the assignee for the bug.

More information about the dev mailing list