[zeromq-dev] ZMQ 4.2.0 stack overflow on Windows

[Robin Scher]

I am trying to update from 4.1.x to 4.2.0 for my project, and when I run it
linked to the 4.2.0 library, I get a stack overflow in a ZeroMQ background
thread when I bind a socket:



MyProgram.exe!__chkstk() Line 117

MyProgram.exe!std::map<unsigned
short,zmq::select_t::family_entry_t,std::less<unsigned
short>,std::allocator<std::pair<unsigned short const
,zmq::select_t::family_entry_t> > >::operator[](const unsigned short &
_Keyval) Line 191

MyProgram.exe!zmq::select_t::add_fd(unsigned __int64 fd_, zmq::i_poll_events
* events_) Line 77

MyProgram.exe!zmq::socket_base_t::start_reaping(zmq::select_t * poller_)
Line 1312

MyProgram.exe!zmq::reaper_t::process_reap(zmq::socket_base_t * socket_) Line
128

MyProgram.exe!zmq::object_t::process_command(zmq::command_t & cmd_) Line 135

MyProgram.exe!zmq::reaper_t::in_event() Line 98

MyProgram.exe!zmq::select_t::loop() Line 318

MyProgram.exe!zmq::select_t::worker_routine(void * arg_) Line 393

MyProgram.exe!thread_routine(void * arg_) Line 47

msvcr110d.dll!_callthreadstartex() Line 354

msvcr110d.dll!_threadstartex(void * ptd) Line 337

kernel32.dll!00007ffd243013d2() Unknown

ntdll.dll!00007ffd246054e4() Unknown



The executable has done nothing but create the context, create the socket,
set a few socket options, and then calls zmq_bind on it. The issue does not
happen in the main thread, and the main thread does not report any errors or
have any faults. No other client is running on this or any networked machine
that is trying to connect to the process yet.



I am using MSVC 2012. Stack size is at the default.



Any ideas of where I can look to resolve this? I would like to use the
current version of the library, if possible. Thank you,

-robin



Robin Scher | [phone] | <mailto:***@uberware.net>
***@uberware.net | <http://www.uberware.net/> www.uberware.net

[Luca Boccassi]

When you refer to main thread, are you by any chance creating a socket
in a thread and using it from another?

Post by Robin Scher
I am trying to update from 4.1.x to 4.2.0 for my project, and when I run it
linked to the 4.2.0 library, I get a stack overflow in a ZeroMQ background
MyProgram.exe!__chkstk() Line 117
MyProgram.exe!std::map<unsigned
short,zmq::select_t::family_entry_t,std::less<unsigned
short>,std::allocator<std::pair<unsigned short const
,zmq::select_t::family_entry_t> > >::operator[](const unsigned short &
_Keyval) Line 191
MyProgram.exe!zmq::select_t::add_fd(unsigned __int64 fd_, zmq::i_poll_events
* events_) Line 77
MyProgram.exe!zmq::socket_base_t::start_reaping(zmq::select_t * poller_)
Line 1312
MyProgram.exe!zmq::reaper_t::process_reap(zmq::socket_base_t * socket_) Line
128
MyProgram.exe!zmq::object_t::process_command(zmq::command_t & cmd_) Line 135
MyProgram.exe!zmq::reaper_t::in_event() Line 98
MyProgram.exe!zmq::select_t::loop() Line 318
MyProgram.exe!zmq::select_t::worker_routine(void * arg_) Line 393
MyProgram.exe!thread_routine(void * arg_) Line 47
msvcr110d.dll!_callthreadstartex() Line 354
msvcr110d.dll!_threadstartex(void * ptd) Line 337
kernel32.dll!00007ffd243013d2() Unknown
ntdll.dll!00007ffd246054e4() Unknown
The executable has done nothing but create the context, create the socket,
set a few socket options, and then calls zmq_bind on it. The issue does not
happen in the main thread, and the main thread does not report any errors or
have any faults. No other client is running on this or any networked machine
that is trying to connect to the process yet.
I am using MSVC 2012. Stack size is at the default.
Any ideas of where I can look to resolve this? I would like to use the
current version of the library, if possible. Thank you,
-robin
_______________________________________________
zeromq-dev mailing list
http://lists.zeromq.org/mailman/listinfo/zeromq-dev

[Robin Scher]

For testing purposes, no I am creating the context, creating the socket and binding the socket from a single thread (the main thread). There are other threads in my process, but none that interact with the ZeroMQ objects. I have not build a minimal sample to reproduce yet. Once this is resolved, my original code does pass the socket between threads, using a mutex to ensure that it is only accessed from a single thread, but I am not using that at the moment for testing.

The crash happens just after the main thread calls zmq_bind. There are no other uses of the socket object, for testing purposes, once it gets bound, the main thread just goes into an infinite loop that does nothing at all. No other thread is using any part of zmq for any purpose, and no other process is running that is trying to connect to this open socket or send any data to it.

Thanks for any other help anyone can provide.
-robin

Robin Scher | [phone] | ***@uberware.net | www.uberware.net

-----Original Message-----
From: zeromq-dev [mailto:zeromq-dev-***@lists.zeromq.org] On Behalf Of Luca Boccassi
Sent: Wednesday, November 30, 2016 3:42 AM
To: ZeroMQ development list <zeromq-***@lists.zeromq.org>
Subject: Re: [zeromq-dev] ZMQ 4.2.0 stack overflow on Windows

When you refer to main thread, are you by any chance creating a socket in a thread and using it from another?

Post by Robin Scher
I am trying to update from 4.1.x to 4.2.0 for my project, and when I
run it linked to the 4.2.0 library, I get a stack overflow in a ZeroMQ
MyProgram.exe!__chkstk() Line 117
MyProgram.exe!std::map<unsigned
short,zmq::select_t::family_entry_t,std::less<unsigned
short>,std::allocator<std::pair<unsigned short const
,zmq::select_t::family_entry_t> > >::operator[](const unsigned short &
_Keyval) Line 191
MyProgram.exe!zmq::select_t::add_fd(unsigned __int64 fd_,
zmq::i_poll_events
* events_) Line 77
MyProgram.exe!zmq::socket_base_t::start_reaping(zmq::select_t *
poller_) Line 1312
MyProgram.exe!zmq::reaper_t::process_reap(zmq::socket_base_t *
socket_) Line
128
MyProgram.exe!zmq::object_t::process_command(zmq::command_t & cmd_) Line 135
MyProgram.exe!zmq::reaper_t::in_event() Line 98
MyProgram.exe!zmq::select_t::loop() Line 318
MyProgram.exe!zmq::select_t::worker_routine(void * arg_) Line 393
MyProgram.exe!thread_routine(void * arg_) Line 47
msvcr110d.dll!_callthreadstartex() Line 354
msvcr110d.dll!_threadstartex(void * ptd) Line 337
kernel32.dll!00007ffd243013d2() Unknown
ntdll.dll!00007ffd246054e4() Unknown
The executable has done nothing but create the context, create the
socket, set a few socket options, and then calls zmq_bind on it. The
issue does not happen in the main thread, and the main thread does not
report any errors or have any faults. No other client is running on
this or any networked machine that is trying to connect to the process yet.
I am using MSVC 2012. Stack size is at the default.
Any ideas of where I can look to resolve this? I would like to use the
current version of the library, if possible. Thank you,
-robin
_______________________________________________
zeromq-dev mailing list
http://lists.zeromq.org/mailman/listinfo/zeromq-dev

[Osiris Pedroso]

You can enable Stop on All Exceptions on Dev2012 and run your process.
As soon as an exception happens, it will stop on that instructions and you
will be able to see the stack and environment variables for all frames,
which may help you find out what is wrong.

If you install the WinDBG debugger ("Debugging Tools For Windows"
<https://msdn.microsoft.com/en-us/library/windows/hardware/ff551063(v=vs.85).aspx>),
there is a tool called ADPLUS that can be used to script a debugging
session that attaches to your process and generate minidumps of the
exceptions it encounters (includes crash dumps). This is useful if you will
send the minidump to somebody else to analyze for you. With it comes a
adplus.doc file that explains how to use the tool.

One would need the same WinDBG and open it with these commands:
CMD> windbg.exe -k minidump.dmp
windbg>.ecxr
windbg>kvn
windbg>dv
windbg>.frame 1
windbg>dv
windbg>.frame 2
windbg>dv
winsbg>q

These commands above would:
.ecxr - search in the saved minidump for the stack where the exception was
recorded.
kvn - show the stack at that time of the exception starting with a stack
frame number for each line
dv - show the local (stack) variables for the current frame
.frame n - changes the frame to nth line
q - quit the (minidump) debugging session
qd - if you ever debugging a live process that should not be killed when
the WinDBG exists, use qd to "detach and quit"

Post by Robin Scher
For testing purposes, no I am creating the context, creating the socket
and binding the socket from a single thread (the main thread). There are
other threads in my process, but none that interact with the ZeroMQ
objects. I have not build a minimal sample to reproduce yet. Once this is
resolved, my original code does pass the socket between threads, using a
mutex to ensure that it is only accessed from a single thread, but I am not
using that at the moment for testing.
The crash happens just after the main thread calls zmq_bind. There are no
other uses of the socket object, for testing purposes, once it gets bound,
the main thread just goes into an infinite loop that does nothing at all.
No other thread is using any part of zmq for any purpose, and no other
process is running that is trying to connect to this open socket or send
any data to it.
Thanks for any other help anyone can provide.
-robin
Robin Scher | [phone] <(213)%20448-0443> |
-----Original Message-----
Sent: Wednesday, November 30, 2016 3:42 AM
Subject: Re: [zeromq-dev] ZMQ 4.2.0 stack overflow on Windows
When you refer to main thread, are you by any chance creating a socket in
a thread and using it from another?

Post by Robin Scher
I am trying to update from 4.1.x to 4.2.0 for my project, and when I
run it linked to the 4.2.0 library, I get a stack overflow in a ZeroMQ
MyProgram.exe!__chkstk() Line 117
MyProgram.exe!std::map<unsigned
short,zmq::select_t::family_entry_t,std::less<unsigned
short>,std::allocator<std::pair<unsigned short const
,zmq::select_t::family_entry_t> > >::operator[](const unsigned short &
_Keyval) Line 191
MyProgram.exe!zmq::select_t::add_fd(unsigned __int64 fd_,
zmq::i_poll_events
* events_) Line 77
MyProgram.exe!zmq::socket_base_t::start_reaping(zmq::select_t *
poller_) Line 1312
MyProgram.exe!zmq::reaper_t::process_reap(zmq::socket_base_t * socket_) Line
128
MyProgram.exe!zmq::object_t::process_command(zmq::command_t & cmd_) Line 135
MyProgram.exe!zmq::reaper_t::in_event() Line 98
MyProgram.exe!zmq::select_t::loop() Line 318
MyProgram.exe!zmq::select_t::worker_routine(void * arg_) Line 393
MyProgram.exe!thread_routine(void * arg_) Line 47
msvcr110d.dll!_callthreadstartex() Line 354
msvcr110d.dll!_threadstartex(void * ptd) Line 337
kernel32.dll!00007ffd243013d2() Unknown
ntdll.dll!00007ffd246054e4() Unknown
The executable has done nothing but create the context, create the
socket, set a few socket options, and then calls zmq_bind on it. The
issue does not happen in the main thread, and the main thread does not
report any errors or have any faults. No other client is running on
this or any networked machine that is trying to connect to the process

yet.

Post by Robin Scher
I am using MSVC 2012. Stack size is at the default.
Any ideas of where I can look to resolve this? I would like to use the
current version of the library, if possible. Thank you,
-robin
_______________________________________________
zeromq-dev mailing list
http://lists.zeromq.org/mailman/listinfo/zeromq-dev

_______________________________________________
zeromq-dev mailing list
http://lists.zeromq.org/mailman/listinfo/zeromq-dev