Prev: 17th annual TCL/TK conference & call for papers
Next: showing progress when doing copy
From: Harry Huang on 18 Mar 2010 18:01
I installed TCL/TK into a hardware. But When I run a TCL program
(calling tcl libray API) on the hardward, the program crashed in the
TclpAlloc function (malloc function) which is defined as :

Char * TclpAlloc(unsigned int numBytes) {
return (char*) malloc(numBytes);
}

The error reported is "Alignment fault at adress xxxxx ...". I checked
numBytes is 1024. Anyone have ideas about this ?
From: Alexandre Ferrieux on 18 Mar 2010 18:13
On Mar 18, 11:01 pm, Harry Huang <hhuan...(a)gmail.com> wrote:
> I installed TCL/TK into a hardware. But When I run a TCL program
> (calling tcl libray API) on the hardward, the program crashed in the
> TclpAlloc function (malloc function) which is defined as :
>
> Char * TclpAlloc(unsigned int numBytes) {
>      return (char*) malloc(numBytes);
>
> }
>
> The error reported is "Alignment fault at adress xxxxx ...". I checked
> numBytes is 1024. Anyone have ideas about this ?

Most likely heap corruption, done *before* that call. A tool like
valgrind, if it is an option on your hardware, might help. Another
method, maybe more practical in your case, is to use 'configure --
enable-symbols=mem'. This will instrument internal allocations, and
the script-level function [memory validate on] will activate a mode
that verifies validity of the whole heap on each malloc/free.
Extremely slow but may help.

-Alex
From: Harry Huang on 18 Mar 2010 21:56
On 3ÔÂ18ÈÕ, ÏÂÎç6ʱ13·Ö, Alexandre Ferrieux <alexandre.ferri...(a)gmail.com>
wrote:
> On Mar 18, 11:01 pm, Harry Huang <hhuan...(a)gmail.com> wrote:
>
> > I installed TCL/TK into a hardware. But When I run a TCL program
> > (calling tcl libray API) on the hardward, the program crashed in the
> > TclpAlloc function (malloc function) which is defined as :
>
> > Char * TclpAlloc(unsigned int numBytes) {
> > return (char*) malloc(numBytes);
>
> > }
>
> > The error reported is "Alignment fault at adress xxxxx ...". I checked
> > numBytes is 1024. Anyone have ideas about this ?
>
> Most likely heap corruption, done *before* that call. A tool like
> valgrind, if it is an option on your hardware, might help. Another
> method, maybe more practical in your case, is to use 'configure --
> enable-symbols=mem'. This will instrument internal allocations, and
> the script-level function [memory validate on] will activate a mode
> that verifies validity of the whole heap on each malloc/free.
> Extremely slow but may help.
>
> -Alex

The interesting thing is it is passed if I used: "return (char*)
malloc(1024);". So use a positive integer will be Ok, but use
"numBytes" will be failed. Is it possible a "casting" issue ? I
remember the call to call "TclpAlloc" is something like "
"char * s = TclpAlloc(sizeof(structure) * a )" where a is a 'int'
From: Larry W. Virden on 19 Mar 2010 11:49
On Mar 18, 6:01 pm, Harry Huang <hhuan...(a)gmail.com> wrote:
> the program crashed in the
> TclpAlloc function (malloc function) which is defined as :
>
> Char * TclpAlloc(unsigned int numBytes) {
>      return (char*) malloc(numBytes);
>
> }
>
> The error reported is "Alignment fault at adress xxxxx ...". I checked
> numBytes is 1024. Anyone have ideas about this ?

That's seems a bit peculiar. Normally, a library built for a specific
platform should be able to malloc memory that is aligned properly.

What I have seen is someone malloc some memory, then later go to use
the memory in an operation that required a more stringent alignment
than malloc by default gave.

For instance, someone might have a system which required a certain
data structure, or floating numbers, or an int, etc. to reside on a
particular word boundary.

This was C code, not Tcl code...

The C code would then perform a malloc, take the returned address, and
adjust the address to be on the alignment boundary required for the
data type being used.

Messy.

 | 
Pages: 1
Prev: 17th annual TCL/TK conference & call for papers
Next: showing progress when doing copy