[lttng-dev] 32-bit application crash

[jamie explorer]

Hi All,
Here is more info.
I'm using 2.6 lttng-ust
I'm seeing this crash on 32-bit app's only when compiling with g++
compiler. Dont see this issue when compiling with gcc.
is there any known issue or fix for this.

Thanks,
Anand Neeli



/Jam

On Wed, Aug 5, 2015 at 10:43 PM, jamie explorer <jamieexplorer at gmail.com>
wrote:

> Hi All,
> i looked into this further, generated .c file with expanded macros for
> debugging
> Some of the function pointer are not populated correctly.
> I have attached the expanded source files which i used.
> please see the following GDB analysis:
>
>
> (gdb) bt
> #0  0x00000000 in ?? ()
> #1  0x080490a0 in __event_probe__sample_tracepoint___message
> (__tp_data=0x80512d0,
>     text=0x804ca64 "Hello World\n") at sample_tracepoint.c:8567
>
> #2  0x0804ba23 in __tracepoint_cb_sample_tracepoint___message (
>     text=0x804ca64 "Hello World\n")
>     at /build/home/jam/gentp/src/sample.c:4026
> #3  main (argc=1, argv=0xffffdd74)
>     at /build/home/jam/gentp/src/sample.c:4731
>
> ****bc_runtime->filter* --- is not populated correctly, because of which
> it crashes
> *(gdb) frame 1*
> #1  0x080490a0 in __event_probe__sample_tracepoint___message
> (__tp_data=0x80512d0,
>     text=0x804ca64 "Hello World\n") at sample_tracepoint.c:8567
> 8567     if (__builtin_expect(!!(*bc_runtime->filter*(bc_runtime,
> __stackvar.__filter_stack_data) & LTTNG_FILTER_RECORD_FLAG), 0))
>
>
> *(gdb) list*
> 8562     (__stackvar.__filter_stack_data, __tp_data,text);
> 8563     for (bc_runtime = ((__typeof__(*bc_runtime) *) ((char *)
> ((reinterpret_cast<__typeof__((&__event->bytecode_runtime_head)->next)>(tracepoint_dlopen.rcu_dereference_sym_bp((reinterpret_cast<void
> *>((&__event->bytecode_runtime_head)->next)))))) - (unsigned long)
> (&((__typeof__(*bc_runtime) *) 0)->node)));
> 8564     &bc_runtime->node != (&__event->bytecode_runtime_head);
> 8565     bc_runtime = ((__typeof__(*bc_runtime) *) ((char *)
> ((reinterpret_cast<__typeof__(bc_runtime->node.next)>(tracepoint_dlopen.rcu_dereference_sym_bp((reinterpret_cast<void
> *>(bc_runtime->node.next)))))) - (unsigned long)
> (&((__typeof__(*bc_runtime) *) 0)->node))))
> 8566     {
> *8567     if (__builtin_expect(!!(bc_runtime->filter(bc_runtime,
> __stackvar.__filter_stack_data) & LTTNG_FILTER_RECORD_FLAG), 0))*
> 8568      __filter_record = 1;
> 8569     } if (__builtin_expect(!!(!__filter_record), 1)) return;
> 8570     } __event_len =
> 8571
>
>
> *printing bc_runtime shows filter is 0*
> *(gdb) p *bc_runtime *
> $3 = {bc = 0x0,* filter = 0x0,* link_failed = 0, node = {next =
> 0x8051304,
>     prev = 0x8051304}}
>
>
> *also i see __event->bytecode_runtime_head->prev = 0x1 which may be wrong*
> *(gdb) p *__event*
> $4 = {id = 2, chan = 0xf6c047c0, enabled = 1,
>   desc = 0x804c700 <__event_desc___sample_tracepoint_message>,
> _deprecated1 = 0x0,
>   ctx = 0x0, instrumentation = LTTNG_UST_TRACEPOINT, u = {<No data
> fields>}, node = {
>     next = 0x8051344, prev = 0x0}, _deprecated2 = {next = 0x0, prev =
> 0x0},
>   _deprecated3 = 0x0, _deprecated4 = 0,
> *bytecode_runtime_head = {next = 0x8051304,     prev = 0x1}*,
> has_enablers_without_bytecode = 134552592, enablers_ref_head = {
>     next = 0x8051c10, prev = 0x0}, hlist = {next = 0xf6c00ea0, prev =
> 0x1},
>   registered = 89}
>
>
> I guess this structure is populated in tracepoint_register using urcu.
> Can anyone give me more pointers to debug this. Is this a known issue and
> is there any patch which i can use
>
>
> /Jam
>
>
>
> On Mon, Aug 3, 2015 at 4:19 PM, jamie explorer <jamieexplorer at gmail.com>
> wrote:
>
>> Hi,
>> i am trying to run 32-bit application and see crash at the the line where
>> "tracepoint" is used.
>> i am using sample application given in
>> lttng-ust-2.6.0/doc/examples/gen-tp/
>> Not able to understand why the problem is seen only with 32-bit apps? Can
>> experts here pls help me fix this.
>>
>> Have tried using ELF 32-bit app on Intel 80386 and PowerPC and see the
>> crash. Where as 64-bit applications have no problem.
>>
>> following is the code where it crashes
>>         for (i = 0; i < 100000; i++) {
>>                 tracepoint(sample_tracepoint, message,  "Hello World\n");
>>  <<<< Crash here
>>                 usleep(1);
>>         }
>>
>> GDB output:
>>
>> #0  0x00000000 in ?? ()
>> #1  0x10001e54 in __event_probe__sample_tracepoint___message (
>>     __tp_data=0x10015420, text=0x100034a4 "Hello World\n")
>>     at /build/home/jam/gentp/sample_tracepoint.h:13
>> #2  0x10000b9c in __tracepoint_cb_sample_tracepoint___message (
>>     text=0x100034a4 "Hello World\n")
>>     at //build/home/jam/gentp/sample_tracepoint.h:13
>> #3  main (argc=<optimized out>, argv=<optimized out>)
>>     at /build/home/jam/gentp/sample.c:115
>>
>>
>> i am running 32-bit version of lttng-sessiond
>> root at host:~# ps aux | grep lttng
>> root      4224  0.0  0.0  92212  1648 ?        Ssl  00:42   0:00
>> /usr/bin/lttng-sessiond --consumerd32-path
>> /usr/lib/lttng/libexec/lttng-consumerd --consumerd32-libdir /usr/lib/ -b
>> --no-kernel
>> root      4243  0.0  0.0  64212   956 ?        Sl   00:42   0:00
>> lttng-consumerd  -u --consumerd-cmd-sock
>> /var/run/lttng/ustconsumerd32/command --consumerd-err-sock
>> /var/run/lttng/ustconsumerd32/error --group tracing
>>
>>
>>
>> /Jam
>>
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.lttng.org/pipermail/lttng-dev/attachments/20150806/a5d182da/attachment.html>