segfault when compiled with dpdk but run without dpdk

[rstory]

using this click config:

// test3.click
// This simple test demonstrates how RoundRobinSched works as a round robin
// scheduler.
// Run with 'click test3.click'
rr :: RoundRobinSched;
TimedSource(0.2) -> Queue(20) -> Print(q1) -> [0]rr;
TimedSource(0.5) -> Queue(20) -> Print(q2) -> [1]rr;
rr -> TimedSink(0.1);

and fastclick configured like so:

./configure \
  --prefix=/usr/local \
  --enable-multithread \
  --disable-linuxmodule \
  --enable-intel-cpu \
  --disable-batch \
  --enable-user-multithread \
  --verbose \
  --disable-dynamic-linking \
  --enable-poll \
  --enable-bound-port-transfer \
  --with-netmap=no \
  --enable-zerocopy \
  --enable-dpdk \
  --enable-dpdk-pool \
  --disable-dpdk-packet \
  RTE_SDK=/usr/local/share/dpdk RTE_TARGET=x86_64-native-linuxapp-gcc \
  CFLAGS='-g -O3 -I/usr/local/include/dpdk' \
  CXXFLAGS='-g -O3 -I/usr/local/include/dpdk -std=gnu++11' \
  --enable-local

from git commit 419abd7 (Wed Jul 5 10:59:15 2017 +0200).

and run click test3.click:
I get a segfault:

Program received signal SIGSEGV, Segmentation fault.
0x00000000007ab1c9 in DPDKDevice::get_mpool (socket_id=4294967295) at ../lib/dpdkdevice.cc:128
128         return _pktmbuf_pools[socket_id];
(gdb) bt
#0  0x00000000007ab1c9 in DPDKDevice::get_mpool (socket_id=4294967295) at ../lib/dpdkdevice.cc:128
#1  0x000000000075b936 in DPDKDevice::get_pkt (numa_node=<optimized out>)
    at ../include/click/dpdkdevice.hh:217
#2  DPDKDevice::get_pkt () at ../include/click/dpdkdevice.hh:229
#3  Packet::alloc_data (tailroom=0, length=2176, headroom=0, this=0xe15940) at ../lib/packet.cc:605
#4  WritablePacket::pool_data_allocate () at ../lib/packet.cc:401
#5  0x000000000075c1fd in WritablePacket::pool_allocate (tailroom=0, length=69, headroom=128)
    at ../lib/packet.cc:418
#6  Packet::make (headroom=128, data=0xe15dcc, length=69, tailroom=tailroom@entry=0) at ../lib/packet.cc:717
#7  0x00000000004397a6 in TimedSource::configure (this=0xe11b00, conf=..., errh=0x7fffffffe2e0)
    at ../elements/standard/timedsource.cc:51
#8  0x0000000000798935 in Router::initialize (this=this@entry=0xe15ac0, errh=errh@entry=0xe0da90)
    at ../lib/router.cc:1158
#9  0x000000000074eed9 in parse_configuration (text=..., text_is_expr=text_is_expr@entry=false,
    hotswap=hotswap@entry=false, errh=errh@entry=0xe0da90) at click.cc:400
#10 0x0000000000466b9b in main (argc=<optimized out>, argv=<optimized out>) at click.cc:730

if I configure without any dpdk arguments:

./configure \
  --prefix=/usr/local \
  --enable-multithread \
  --disable-linuxmodule \
  --enable-intel-cpu \
  --disable-batch \
  --enable-user-multithread \
  --verbose \
  --disable-dynamic-linking \
  --enable-poll \
  --enable-bound-port-transfer \
  --with-netmap=no \
  --enable-zerocopy \
  CFLAGS='-g -O3 ' \
  CXXFLAGS='-g -O3 -std=gnu++11' \
  --enable-local

It works great:

dl3n-1 $ click test3.click
q1:   69 | 52616e64 6f6d2062 756c6c73 68697420 696e2061 20706163
q1:   69 | 52616e64 6f6d2062 756c6c73 68697420 696e2061 20706163

[tbarbette]

Hi,
Indeed --enable-dpdk-pool will crash if no DPDK element is in the graph. I should initialize at least one memory pool no matter if there are devices or any DPDK element. For now, it's doing it only if a DPDKDevie or EnsureDPDKBuffer element is there.

The lightest temp solution for now would be to add a Idle -> EnsureDPDKBuffer -> Discard; as this will not require any physical device.

[rstory]

Yep, that works great. Thanks!

[tbarbette]

Finally solved by the last commit ...