Re: [Boost-users] Boost::Asio on ARMV5TE

Subject: Re: [Boost-users] Boost::Asio on ARMV5TE
From: Peter Meyer (residue_at_[hidden])
Date: 2011-02-20 12:26:04

Hi,

after some debugging and testing i track down the problem to epoll. 
I don't known if it is a wrong usage by boost::asio or a bug inside the kernel.

I've changed epoll_reactor.ipp with some debugging output and a memset.

226   void epoll_reactor::run(bool block, op_queue& ops)
...
239   epoll_event events[128];
240+  ::memset(events, 0XEE, sizeof(events));
241   int num_events = epoll_wait(epoll_fd_, events, 128, timeout);

In case there are more events then one reported by epoll_wait, only the first
epoll_event structure is filled with information. All following structures
includes garbage or 0xEE set by memset befor calling epoll_wait.
Debug output:
Count events: 2
----------------------------------------
EPOLL[0].data.ptr   = 0x1ce8b0
EPOLL[1].data.ptr   = 0xeeeeeeee

So a workaround is to set maxevents to one. Like this:
226  void epoll_reactor::run(bool block, op_queue& ops)
...
240  int num_events = epoll_wait(epoll_fd_, events, 1, timeout);

Is there a known bug about epoll on ARM? Or had boost::asio a bug?

With best regards
Peter


> -----Original Message-----
> From: Peter Meyer [mailto:r..._at_[hidden]] 
> Sent: Friday, February 11, 2011 7:57 AM
> To: boost-users_at_[hidden]
> Subject: [Boost-users] Boost::Asio on ARMV5TE
> 
> 
> Hi,
> 
> i've wrote a litte proxy using boost::asio (boost 1.45). Program runs
> well on Intel (Linux kernel 2.6.35.10 and gcc 4.5.1).
> Same code cross compiled for ARMV5TE (Linux 2.6.24.4 and GCC 4.1.2 or
> GCC 3.4.3) crashes always at the same point.
> Stacktrace looks like this
> 
> Program terminated with signal 11, Segmentation fault.
> #0  0x400e3e20 in ?? ()
> (gdb) backtrace 
> #0  0x400e3e20 in ?? ()
> #1  0x0001cfb0 in boost::asio::detail::posix_mutex::lock (this=0x7d) at
> /usr/armv5tel-redhat-linux-gnueabi/sys-root/usr/local/include/boost/asio
> /detail/posix_mutex.hpp:52
> #2  0x0001cfe6 in scoped_lock (this=0xbe8a98d0, m=@0x7d) at
> /usr/armv5tel-redhat-linux-gnueabi/sys-root/usr/local/include/boost/asio
> /detail/scoped_lock.hpp:36
> #3  0x00069bf4 in boost::asio::detail::epoll_reactor::run
> (this=0x191538, block=true, ops=@0xbe8a994c)
>     at
> /usr/armv5tel-redhat-linux-gnueabi/sys-root/usr/local/include/boost/asio
> /detail/impl/epoll_reactor.ipp:275
> #4  0x0006a06c in boost::asio::detail::task_io_service::do_one
> (this=0x1914e0, lock=@0xbe8a99d4, this_idle_thread=0xbe8a9990)
>     at
> /usr/armv5tel-redhat-linux-gnueabi/sys-root/usr/local/include/boost/asio
> /detail/impl/task_io_service.ipp:264
> #5  0x0006a33c in boost::asio::detail::task_io_service::run
> (this=0x1914e0, ec=@0xbe8a9a0c)
>     at
> /usr/armv5tel-redhat-linux-gnueabi/sys-root/usr/local/include/boost/asio
> /detail/impl/task_io_service.ipp:130
> #6  0x0006a426 in boost::asio::io_service::run (this=0xbe8a9bc4) at
> /usr/armv5tel-redhat-linux-gnueabi/sys-root/usr/local/include/boost/asio
> /impl/io_service.ipp:57
> #7  0x00073504 in proxy::Server::run (this=0xbe8a9bb4) at server.cpp:71
> #8  0x0004e2a4 in main (argc=3, argv=0xbe8a9da4) at proxy.cpp:111
> 
> It looks like the internal epoll_events data structure
> (descriptior_state) is corrupted.
> 
> Any advice is welcome. 
> 
> With best regards
> Peter
>
------------------------------------------------------------------------------------