[gst-embedded] oom-killer on gstreamer 0.10.20

[Stefan Kost]

yihe schrieb:
> Mike,
>  
>  It was understood. I want to know anyone in the mail list have good
> tools to debug this memory leakage issues. I was using the dmalloc. but
> after changing the configure file to use the arm crosscompile option,
> I got some werid errors about the dmalloc.h
>  /pro/devkit/arm/v5t_le/bin/arm_v5t_le-gcc -g -O2  -DHAVE_STDARG_H=1
> -DHAVE_STDLIB_H=1 -DHAVE_STRING_H=1 -DHAVE_UNISTD_H=1
> -DHAVE_SYS_MMAN_H=1 -DHAVE_SYS_TYPES_H=1 -DHAVE_W32API_WINBASE_H=0
> -DHAVE_W32API_WINDEF_H=0 -DHAVE_SYS_CYGWIN_H=0 -DHAVE_SIGNAL_H=1  -I.
> -I.  -c arg_check.c -o ./arg_check.o
> In file included from arg_check.c:40:
> dmalloc.h:460: error: parse error before "__extension__"
> dmalloc.h:460: error: `__len' undeclared here (not in a function)
> dmalloc.h:460: error: initializer element is not constant
> dmalloc.h:460: error: parse error before "if"
> dmalloc.h:460: error: conflicting types for '__retval'
> dmalloc.h:460: error: previous definition of '__retval' was here
> dmalloc.h:460: warning: data definition has no type or storage class
> dmalloc.h:460: error: parse error before '}' token
> 
> Regards,
>  
> Yihe

If you can compile the same code to x86, you can use valgrind with memcheck
(http://www.valgrind.org). There are some efforts for valgrind to support arm,
but they are quite new.
glibc also has some memory checking tools (see
http://www.buzztard.org/index.php/Debugging#LibC)

Stefan

>
> On Tue, Feb 17, 2009 at 12:57 PM, Michael Smith <msmith at xiph.org
> <mailto:msmith at xiph.org>> wrote:
> 
>     On Tue, Feb 17, 2009 at 10:49 AM, yihe <tihpa2004 at gmail.com
>     <mailto:tihpa2004 at gmail.com>> wrote:
>     > Mike,
>     >
>     >  I am a new hand to linux/gstreamer.
>     > The project is to let the system to decoder multples mepg2
>     avi.stream on the
>     > hardware and mux the decoded buffers to a single display buffer
>     and display
>     > out.
>     >
>     > our team has developed the decoder element and display element
>     based on the
>     > hardware platform.
>     > My job is to make a muxer element between these two
>     elements,besides that,
>     > filesrc and avimux elemetn were used in the pipeline.
>     >
>     > I guess somehow the project has some memory leakage. since the
>     oom-kiill did
>     > not occur at first running. it was only present after nth running
>     n,where n
>     > was decided by the number of input streams.
> 
>     Ok, that's entirely possible. We're not familiar with your system
>     though, and you're using a bunch of your own plugins - so we can't
>     really help you with any of that sort of thing.
> 
>     Good luck!
> 
>     Mike
>