native: segfault on heavy network usage in transceiver

[mehlis]

sending >100 packets per second causes a segfault in RIOT

in 9 of 10 cases it's:

Core was generated by `../projects/ccn-lite-client/bin/native/ccn-lite-client.elf grid5x5_b4 -t 4711 -'.
Program terminated with signal 11, Segmentation fault.
#0  0x08051cb1 in run () at transceiver.c:269
269                 response = send_packet(cmd->transceivers, cmd->data);
Missing separate debuginfos, use: debuginfo-install glibc-2.17-20.fc19.i686
(gdb) bt
#0  0x08051cb1 in run () at transceiver.c:269
#1  0x4a121ceb in makecontext () from /lib/libc.so.6
#2  0x08095444 in __end_stack ()
#3  0x080c33b4 in transceiver_stack ()
#4  0x080c33b8 in transceiver_stack ()
#5  0x080c33bc in transceiver_stack ()
#6  0x080c33c0 in transceiver_stack ()
#7  0x08095444 in __end_stack ()
#8  0x080bf560 in msg_buffer ()
#9  0x00000000 in ?? ()

in 1 of 10 cases it is:

Core was generated by `../projects/ccn-lite-client/bin/native/ccn-lite-client.elf grid5x5_c4 -t 4724 -'.
Program terminated with signal 11, Segmentation fault.
#0  receive_packet (type=0, pos=0 '\000') at transceiver.c:395
395             trans_p = NULL;
(gdb) bt
#0  receive_packet (type=0, pos=0 '\000') at transceiver.c:395
#1  0x4a121ceb in ?? ()
#2  0x0809645c in relay_stack ()
#3  0x080c43d4 in theRelay ()
#4  0x080c43d8 in theRelay ()
#5  0x0809645c in relay_stack ()
#6  0x080c0578 in transceiver_stack ()
#7  0x00000000 in ?? ()

This looks strange, because receive_packet with type argument=0 indicates it's a packet from CC1100 transceiver. So this might be a corrupted stack

[LudwigKnuepfer]

Another possible effect is in valgrind: process becomes unresponsive.

[LudwigKnuepfer]

Reportedly freezes are possible without a debugging environment as well. (that was unrelated)

[mehlis]

../RIOT/examples/ccn-lite-client/bin/native/ccn-lite-client.elf: _native_lpm_sleep: select(): Resource temporarily unavailable
../RIOT/examples/ccn-lite-client/bin/native/ccn-lite-client.elf: XXX: this should not have happened!

Core was generated by `../RIOT/examples/ccn-lite-client/bin/native/ccn-lite-client.elf grid5x5-ccn_e4'.
Program terminated with signal 11, Segmentation fault.
#0  0x080c1a6e in transceiver_stack ()
Missing separate debuginfos, use: debuginfo-install glibc-2.17-20.fc19.i686
(gdb) bt
#0  0x080c1a6e in transceiver_stack ()
#1  0x4a121c4b in setcontext () from /lib/libc.so.6
#2  0x0804bd43 in isr_cpu_switch_context_exit () at native_cpu.c:127
#3  0x0804be0a in cpu_switch_context_exit () at native_cpu.c:148
#4  0x0804b47f in native_irq_handler () at irq_cpu.c:281
#5  0x4a121ceb in makecontext () from /lib/libc.so.6
#6  0x00000000 in ?? ()

[mehlis]

Core was generated by `../RIOT/examples/ccn-lite-client/bin/native/ccn-lite-client.elf grid5x5-ccn_c3'.
Program terminated with signal 5, Trace/breakpoint trap.
#0  _native_sig_leave_tramp () at tramp.S:48
48      jmp *-4(%esp)
Missing separate debuginfos, use: debuginfo-install glibc-2.17-20.fc19.i686
(gdb) bt
#0  _native_sig_leave_tramp () at tramp.S:48
#1  0x080677cc in idle_stack ()
#2  0x4a121c4b in setcontext () from /lib/libc.so.6
#3  0x0804be52 in isr_thread_yield () at native_cpu.c:163
Backtrace stopped: previous frame inner to this frame (corrupt stack?)

[mehlis]

../RIOT/examples/ccn-lite-client/bin/native/ccn-lite-client.elf: _native_lpm_sleep: select(): Resource temporarily unavailable
../RIOT/examples/ccn-lite-client/bin/native/ccn-lite-client.elf: XXX: this should not have happened!

[LudwigKnuepfer]

Is this only happening when you use the ltc?

[LudwigKnuepfer]

@mehlis because then it is probably related to #495

[mehlis]

I'm not using the LTC since months

[OlegHahm]

Is it possible to create a testcase?

[LudwigKnuepfer]

It's nondeterministic, but somewhat reproducible with https:/LudwigOrtmann/riot-tools/tree/master/l2perf by sending packets really fast.

[cgundogan]

This issue is not apparent with the new network stack. It is possible to send pings with high throughput without crashing native (for the normal case). @OlegHahm what's your opinion on this? Should we close this?

[OlegHahm]

Yeah, let's close this. Even if there are problem with native and heavy network usage, this issue is about the transceiver module which has vanished.