x86 and x64 have no problems with unaligned read/write as far as I know --
which is why this hasn't come up. Can you open a ticket and continue this
discussion there? It should be fairly easy to do some conditional changes
based on CPU type.
Josh
On Mon, Jul 12, 2010 at 5:38 PM, Daniel Stonier <
d.stonier@gmail.com> wrote:
>
> This seems to be a bit of problem code with regards to the alignment
> issues:
>
> roscpp/src/libros/service_server_link.cpp
>
>
> *****************************************************************************************
>
> void ServiceServerLink::onResponseOkAndLength(const ConnectionPtr& conn,
> const boost::shared_array<uint8_t>& buffer, uint32_t size, bool success)
> {
> ROS_ASSERT(conn == connection_);
> ROS_ASSERT(size == 5);
>
> if (!success)
> return;
>
> uint8_t ok = buffer[0];
> uint32_t len = *((uint32_t*)(buffer.get() + 1));
>
>
> *****************************************************************************************
>
> That last line is where the signal is caught. I think the problem is, the
> arm core reads the uint32 values only at alignmed 4 byte boundaries - but
> that +1 generally shifts it off the 4 byte boundary and causes the signal.
>
> Do intel machines even have an alignment trap like this arm core does? I
> noticed my machine does not have /proc/cpu/alignment.
>
> Sorry, don't have a potential fix for it as I'm not familiar with this
> section of code yet.
>
> More to come,
> Regards,
> Daniel.
>
>
> On 12 July 2010 19:04, Daniel Stonier <d.stonier@gmail.com> wrote:
>
>>
>>
>> On 12 July 2010 17:24, René Wagner <rene.wagner@dfki.de> wrote:
>>
>>> On Fri, 2010-07-09 at 10:45 +0900, Daniel Stonier wrote:
>>> > - Patched (using the memcpy's rather than the reinterprets in the
>>> > SERIALIZER macro).
>>> > - In signal+warn mode (echo 5 > /proc/cpu/alignment)
>>> > - (float64&int32) client gives a bus error and kernel outputs an
>>> > alignment trap message, but servers are fine
>>> > - In default mode (echo 0 > /proc/cpu/alignment)
>>> > - (float64&int32) client and server function without hanging like
>>> before
>>> >
>>> > So...its working, but is this the right fix?
>>>
>>> You probably want to check where the unaligned memory access is
>>> happening. Can you run your float_client from within gdb? This should
>>> allow you to get a backtrace once the kernel has sent the signal (bus
>>> error) from the alignment trap.
>>>
>>> Cheers,
>>>
>>> Rene
>>>
>>> Getting some more meaningful backtraces now - catch bus errors for both
>> the float64's and the int32's in some cases, some with, some without that
>> memcopy patch above. I'll put together some results when I get back to work
>> in the morning.
>>
>>
>>
>>
>>> -
>>> ------------------------------------------------------------
>>> Dipl.-Inf. René Wagner Junior Researcher
>>> DFKI Bremen Enrique-Schmidt-Str. 5
>>> Safe and Secure Cognitive Systems D-28359 Bremen
>>>
>>> Phone: (+49) 421-218-64224 Fax: (+49) 421-218-98-64224
>>> Web: http://www.informatik.uni-bremen.de/agebv/en/ReneWagner
>>> --- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- -- --
>>> Deutsches Forschungszentrum für Künstliche Intelligenz GmbH
>>> Firmensitz: Trippstadter Strasse 122, D-67663 Kaiserslautern
>>>
>>> Geschäftsführung:
>>> Prof. Dr. Dr. h.c. mult. Wolfgang Wahlster (Vorsitzender)
>>> Dr. Walter Olthoff
>>> Vorsitzender des Aufsichtsrats:
>>> Prof. Dr. h.c. Hans A. Aukes
>>> Amtsgericht Kaiserslautern HRB 2313
>>> ------------------------------------------------------------
>>>
>>> _______________________________________________
>>> ros-users mailing list
>>> ros-users@code.ros.org
>>> https://code.ros.org/mailman/listinfo/ros-users
>>>
>>
>>
>>
>> --
>> Phone : +82-10-5400-3296 (010-5400-3296)
>> Home: http://snorriheim.dnsdojo.com/
>> Yujin Robot: http://www.yujinrobot.com/
>> Embedded Control Libraries:
>> http://snorriheim.dnsdojo.com/redmine/wiki/ecl
>>
>
>
>
> --
> Phone : +82-10-5400-3296 (010-5400-3296)
> Home: http://snorriheim.dnsdojo.com/
> Yujin Robot: http://www.yujinrobot.com/
> Embedded Control Libraries: http://snorriheim.dnsdojo.com/redmine/wiki/ecl
>
> _______________________________________________
> ros-users mailing list
> ros-users@code.ros.org
> https://code.ros.org/mailman/listinfo/ros-users
>
>
(text/plain)