SIGBUS on ( double * )

[Jeffrey Walton]

>> Yes, that's not legal C/C++. Its undefined behavior.
>
> how do you mean?
> I get SIGBUS also for things like:
> tempx = ( double* )ptr;
Its not legal C/C++.

> I run my code on ARMv7 after building with:
That's interesting that you could induce that on an ARMv7. I was
trying to do it for the longest time (for use as a test case, see
http://stackoverflow.com/questions/15233717/linux-x86-x64-turn-off-misaligned-data-fixups).

Jeff

On Wed, Oct 23, 2013 at 2:00 PM, Marco Bernasocchi
<marco at bernawebdesign.ch> wrote:
> On 10/23/2013 07:45 PM, Jeffrey Walton wrote:
>>>
>>> I often get SIGBUS when my code (qgis.org) hits stuff like this:
>>>
>>> GEOSCoordSeq_getX( cs, j, ( double * )&mGeometry[position] );
>>
>> Yes, that's not legal C/C++. Its undefined behavior.
>
> how do you mean?
> I get SIGBUS also for things like:
> tempx = ( double* )ptr;
>
> and replacing with the following does the job:
> memcpy( &tempx, ptr, sizeof( tempx ) );
>
>>
>>> double a;
>>> memcpy( a, &mGeometry[position])
>>> GEOSCoordSeq_getX( cs, j, &a )
>>>
>>> then all works correctly,
>>
>> Yep. And if the buffers overlap, you need to `memmov` it.
>>
>>> could this be because of a
>>> compiler setting?
>>
>> No, its the QT devs doing something they should not be doing. Newer
>> ARMs are OK with unaligned data, while older ARMs are not. They appear
>> to be testing on newer ARMs only (ARMv6 or ARMv7, IIRC).
>
> I run my code on ARMv7 after building with:
> for debug:
> CFLAGS: -DANDROID -Wno-psabi -std=c99 -g -O0 -marm -finline-limit=300
> -fno-strict-aliasing -fno-omit-frame-pointer -DDEBUG -D_DEBUG -march=armv7-a
> -mfloat-abi=softfp -mfpu=vfp
> CXXFLAGS: -DANDROID -Wno-psabi -std=gnu++0x -g -O0 -marm -finline-limit=300
> -fno-strict-aliasing -fno-omit-frame-pointer -DDEBUG -D_DEBUG
> LDFLAGS:  -Wl,--fix-cortex-a8
>
> for release:
> CFLAGS: -DANDROID -Wno-psabi -std=c99 -mthumb -O0 -march=armv7-a
> -mfloat-abi=softfp -mfpu=vfp
> CXXFLAGS:-DANDROID -Wno-psabi -std=gnu++0x -mthumb -O0 -march=armv7-a
> -mfloat-abi=softfp -mfpu=vfp
> LDFLAGS: -Wl,--fix-cortex-a8
>
>
>>
>> I believe -Wstrict-aliasing or -Wcast-align will catch it at compile
>> time. See http://gcc.gnu.org/onlinedocs/gcc/Warning-Options.html and
>>
>> https://www.gnu.org/software/gsl/manual/html_node/GCC-warning-options-for-numerical-programs.html#GCC-warning-options-for-numerical-programs.
>>
> thanks a lot
>>
>> Jeff
>>
>> On Wed, Oct 23, 2013 at 1:12 PM, Marco Bernasocchi
>> <marco at bernawebdesign.ch> wrote:
>>>
>>> Hi all, I often get SIGBUS when my code (qgis.org) hits stuff like this:
>>>
>>> GEOSCoordSeq_getX( cs, j, ( double * )&mGeometry[position] );
>>>
>>> if I change the call to use memcpy like
>>>
>>> double a;
>>> memcpy( a, &mGeometry[position])
>>> GEOSCoordSeq_getX( cs, j, &a )
>>>
>>> then all works correctly, the issue is that I've a lot of those calls and
>>> If
>>> I'm not wrong in the past it used to work). could this be because of a
>>> compiler setting? I'm using necessitas Qt sdk with ndk r8b1 compiling
>>> both
>>> with -mthumb and without has the same issue.