Thread: How can detect memory leakage?

Hi,

I wrote an application that use pthread. It has memory management problems.
It runs well without any problem but after some minutes, hours or day, is created Segmentation fault.

For detecting reason of segmentation fault, i use memwatch, a memory leakage detection tool. But when log file is created, it shows only function that cause problem. Beacuse of using some own functions(wroted by myself) for free specialized structures, the function cause problem, is always same function. I need to get stack trace so i can find where that it call the special function(that it free memory).

How can i do it?

I thank you for any other suggestions for memory leakage detection.

Originally Posted by pronetin

Hi,

I wrote an application that use pthread. It has memory management problems.
It runs well without any problem but after some minutes, hours or day, is created Segmentation fault.

For detecting reason of segmentation fault, i use memwatch, a memory leakage detection tool. But when log file is created, it shows only function that cause problem. Beacuse of using some own functions(wroted by myself) for free specialized structures, the function cause problem, is always same function. I need to get stack trace so i can find where that it call the special function(that it free memory).

How can i do it?

I thank you for any other suggestions for memory leakage detection.

Are you sure it's a memory leak and not a race condition? I consider that a lot more likely...

Originally Posted by EVOEx

Are you sure it's a memory leak and not a race condition? I consider that a lot more likely...

Even if be race condition(for example access to memory before initializing it), should be found by memory leakage detection tools, shouldn't?

Originally Posted by claudiu

I think you can give Valgrind a try:

Valgrind Home

I tried it before. It needs glibc 2.2 to 2.10 but in my system(ubuntu 10.04) installed 2.10-1. So it cannt be used.

Originally Posted by pronetin

I tried it before. It needs glibc 2.2 to 2.10 but in my system(ubuntu 10.04) installed 2.10-1. So it cannt be used.

Just use "apt-get install valgrind". Never had trouble with that.

Whether it should be detected by a memory leak... well, maybe, maybe not. That really depends on how advanced the application is ;-). But most proper ones probably will, yeah. Never used memwatch, though, but valgrind does it.
But it makes the way you should look at the function very different.

Originally Posted by pronetin

Even if be race condition(for example access to memory before initializing it), should be found by memory leakage detection tools, shouldn't?

Possibly... probably not. Accessing memory that is in your arena, but not allocated, isn't a memory leak. I believe he means something more along the lines of semaphores not being placed correctly, causing something to "go" with an invalid value as because there was no semaphore on the thread setting the value. It's very easy to get caught up with large multi-threaded applications.

Run it in gdb, when you get the SEGV signal, print out all your values, find out what's bad, exactly why it crashed, memory leakage typically won't cause a seg fault, a bad pointer will. Just run it, when it crashes, it'll tell you where, then start printing different values, find out what could possibly make that pointer (typically a pointer) invalid to make it crash there.

Just use "apt-get install valgrind". Never had trouble with that.

Ok. In Ubuntu's repository exist Valgrind. I tried to install Valgrind from source that showed me glibc compatibility error. When Valgrind is install from repositories, there isn't any problem.

Probably because it installs dependencies for you. E.g. it installed whatever version of glibc was required for it to work.

(Don't take my word for it, I'm a Windows person.)