Fix a segmentation fault bug when used with DPDK.

[boat0]

When DPDK is used we have:
lkl_hijack_netdev_create()
--> lkl_netdev_dpdk_create()
--> rte_eal_init()
--> rte_eal_intr_init()

rte_eal_intr_init() creates threads to use lkl_ops, but lkl_ops will only
be initialized in later call to lkl_start_kernel(). So LKL crashes with
SIGSEGV received.

Fix by switching calls to lkl_hijack_netdev_create()/lkl_start_kernel().

Signed-off-by: Xiaozhou Liu [email protected]

---

This change is 

[lkl-jenkins]

Can one of the admins verify this patch?

[thehajime]

Thanks for the patch !
I will give a review later (sorry for the late response).

btw, what dpdk version are you using in the case of crash ?

[boat0]

I am using the default dpdk version-17.02 that comes with lkl.

[thehajime]

@boat0 sorry for the late response.

lkl_hijack_netdev_create() before lkl_start_kernel() was meant to put kernel boot parameter (below) before lkl_start_kernel. looks like LKL can add virtio devices after lkl_running=1 (hungs in my local machine though), but the original code should work fine.

[    0.000000] Kernel command line:  virtio_mmio.device=268@0x1000000:1

I can reproduce segfault and found that the crash is not due to pthread_create in rte_eal_intr_init, but due to the commit 8f6b03d, which I did :(

commit 8f6b03d865d0fac79e9d9e011196e62df070bdce
Author: Hajime Tazaki <[email protected]>
Date:   Thu Feb 8 08:09:40 2018 +0900

    lkl: epoll_wait demultiplex between host and lkl

Applying the following patch seems to fix the issue, but I need to think more about what's the right way to do this.

diff --git a/tools/lkl/lib/hijack/hijack.c b/tools/lkl/lib/hijack/hijack.c
index 95bbcf7f549e..664f2018407d 100644
--- a/tools/lkl/lib/hijack/hijack.c
+++ b/tools/lkl/lib/hijack/hijack.c
@@ -421,6 +421,10 @@ int epoll_wait(int epfd, struct epoll_event *events,
        void *trv_val;
        int i, ret, ret_lkl, ret_host;
 
+       if (!lkl_running){
+               return host_epoll_wait(epfd, events, maxevents, timeout);
+       }
+
        ret = lkl_call(__lkl__NR_pipe, 1, l_pipe);
        if (ret == -1) {
                fprintf(stderr, "lkl pipe error(errno=%d)\n", errno);

Since I'm not yet able to run an app with dpdk, could you try this patch in your machine to see if it works or not ?

thanks.