Diagnosing Deadlocks in Win32 Program

Question

What are the steps and techniques to debug an apparent hang due to a deadlock in a Win32 production process. I heard that WinDbg can be used for this purpose but could you please provide clear hints on how this can be accomplished?

Answer 1

What language/IDE are you using?

In .Net you can view the threads of an application: Debug->Windows->Threads or Ctrl+Alt+H

Answer 2

This post should get you started on the various options..Check the posts tagged with Debugging..

Another useful article on debugging deadlocks..

Answer 3

Debugging deadlocks can be tricky. I usually do some kind of logging and see where the log stops. I either log to a file or to the debug console using OutputDebugString().

Answer 4

Debugging a true deadlock is actually kind of easy, if you have access to the source and a memory dump (or live debugging session).

All you do is look at the threads, and find the ones that are waiting on some kind of shared resource (for example hung waiting in WaitForSingleObject). Generally speaking from there it is a matter of figuring out which two or more threads have locked each other up, and then you just have to figure out which one broke the lock heirarchy.

If you can't easily figure out which threads are locked up, use the method shown in this post here to trace the lock chain for each thread. When you get into a loop, the threads in the loop are the ones that are deadlocked.

Answer 5

The best thing is to start by adding logging statements. Generally I would recommend only around the shared resources that are deadlocking but also adding them in general might point to situations or areas of code you weren't expecting. The much publicized stackoverflow.com database issue actually turned out to be log4net! The stackoverflow team never suspected log4net, and only by examining logging (ironically) showed this. I would initially forgo any complicated tools e.g., WinDgb since using them is not very intuitive IMHO.

Answer 6

If you are very lazy, you can install Application Verifier, then add you module and select just "locks" from the basic test. then you can run your application under any debugger.
if a critical section deadlock happens you with find the reason right away.