Fixing it

Now that we understand the issue here, how do we fix it? Seeing lockdep's report (Figure 7.9) and interpreting it, it's quite simple: (as mentioned) since the task structure spinlock named alloc_lock is already taken at the start of the do-while loop (via task_lock(t)), ensure that before calling the get_task_comm() routine (which internally takes and releases this same lock), you unlock it, then perform get_task_comm(), then lock it again.

The following screenshot (Figure 7.10) shows the difference (via the diff(1) utility) between the older buggy version (ch13/3_lockdep/buggy_thrdshow_eg/thrd_showall_buggy.c) and the newer fixed version of our code (ch13/3_lockdep/fixed_lockdep/thrd_showall_fixed.c):

Figure 7.10 – (Partial) screenshot showing the key part of the difference between the buggy and fixed versions of our demo thrdshow LKM

Great; another example follows – that of catching an AB-BA deadlock!