[dpdk-dev] [vhost] segment fault when virtio_user port init

[Yuanhan Liu]

Cc Jianfeng, who added the doc and wrote the virtio-user code.

On Mon, Mar 06, 2017 at 07:15:15PM +0800, wang.yong19 at zte.com.cn wrote:
> Following the description of "Virtio_user for Container Networking" in 
> 
> "how to guides", I ran this example in a VM created by qemu. I started 
> 
> a testpmd in the VM with a vhost-user port. The command is:
> 
> $(testpmd) -c 0x3 -n 4 --socket-mem 1024,1024 \
> 
> --vdev 'eth_vhost0,iface=/tmp/sock0' --no-pci -- -i
> 
> And then, I started a container instance with a virtio-user port. The 
> 
> command is:
> 
> docker run -i -t -v /tmp/sock0:/var/run/usvhost \
> 
> -v /dev/hugepages:/dev/hugepages \
> 
> dpdk-app-testpmd testpmd -c 0xc -n 4 -m 1024 --no-pci \
> 
> --vdev=virtio_user0,path=/var/run/usvhost \
> 
> -- -i --txqflags=0xf00 --disable-hw-vlan

Hmm, should not you add --file-prefix option to distinguish the huge
page file names while starting two DPDK apps in the same host?

	--yliu
> 
> 
> Then, a segment fault occured in the VM's testpmd. 
> 
> testpmd＞ VHOST_CONFIG: new vhost user connection is 15
> 
> Segmentation fault (core dumped)
> 
> 
> As a result, the container could not complete the initialization of the
> 
> virtio_user port. 
> 
> EAL: failed to initialize virtio_user0 device
> 
> PANIC in rte_eal_init():
> 
> Cannot init pmd devices
> 
> 6: [testpmd() [0x4497f9]]
> 
> 5: [/lib/x86_64-linux-gnu/libc.so.6(__libc_start_main+0xf5) 
> 
> [0x7f41c9b19ec5]]
> 
> 4: [testpmd(main+0x42) [0x448a32]]
> 
> 3: [testpmd(rte_eal_init+0xde2) [0x49e322]]
> 
> 2: [testpmd(__rte_panic+0xbe) [0x442a56]]
> 
> 1: [testpmd(rte_dump_stack+0x1a) [0x4a57ba]]
> 
> 
> I opened the "CONFIG_RTE_LIBRTE_VIRTIO_DEBUG_INIT" and  
> 
> "CONFIG_RTE_LIBRTE_VIRTIO_DEBUG_DRIVER" switch to try again. I got the 
> 
> following output in container.
> 
> PMD: vhost_user_sock(): VHOST_SET_OWNER
> 
> PMD: vhost_user_sock(): VHOST_GET_FEATURES
> 
> PMD: vhost_user_read(): Failed to recv msg hdr: -1 instead of 12.
> 
> PMD: vhost_user_sock(): Received msg failed: Connection reset by peer
> 
> PMD: virtio_user_dev_init(): get_features failed: Connection reset by 
> 
> peer
> 
> PMD: virtio_user_pmd_probe(): virtio_user_dev_init fails
> 
> 
> According to the output in above, I realized the reason was at VM side.
> 
> By adding some logs to the code, I found the segment fault was 
> 
> occured in the following code: vhost_user_server_new_connection()-＞
> 
> vhost_user_add_connection()-＞vhost_new_device()-＞rte_zmalloc()-＞
> 
> rte_zmalloc_socket()-＞rte_malloc_socket()-＞malloc_heap_alloc()-＞
> 
> malloc_elem_alloc()-＞elem_free_list_remove()-＞LIST_REMOVE();
> 
> 
> When there is a new vhost-user connection established, it need to 
> 
> malloc 528512 bytes memory(sizeof(struct virtio_net)) from heap. In my 
> 
> environment, malloc_heap_alloc() found the suitable element in heap. 
> 
> But when malloc_elem_alloc() called the elem_free_list_remove(), a 
> 
> segment fault occured in LIST_REMOVE().
> 
> 
> Would you please do me a favor to resolve this problem?
> 
> BTY, the VM used 1G hugepages and hugepage num was 4.
> 
> 
> 
> 
> 
>