regression of waitid

[Wenji Huang]

Wenji Huang wrote:
> Hi,
> 
> When I tried the latest utrace patch to 2.6.25-rc6, found there is one 
> regression about waitid.
> It passed in upstream kernel, but hang in utrace-patched kernel. The 
> test case was old one and carried
> in previous kernel test. Roland once fixed the problem in 
> http://kernel.org/pub/linux/kernel/v2.6/ChangeLog-2.6.23.8.
> 
> It seems the second wait makes the kernel enter into infinite loop.

I did some analysis of outrace kernel according to this test case.

Parent's waitid is in such call tree:
	do_wait -> do_wait_thread -> ptrace_do_wait
In fact, wait_consider_task is called first, but return 0, so need to do 
ptrace_do_wait.

Piece of ptrace_do_wait code:
--------------------------------------------------------
		exit_code = xchg(&p->exit_code, 0);
		if (exit_code & PTRACE_TRAPPED_MASK)
			goto found;
                 /************/
		if (p->state == TASK_STOPPED)
			goto found;
---------------------------------------------------------

First waittid:
     child's state == TASK_TRACED, exit_code !=0, goto found, return 
child's pid

Second waitid:
     child's state == TASK_TRACED, exit_code ==0, continue next parts, 
return 0.
That means ptrace_do_wait blocking. Only Ctrl+C to make it exit.

But the kernel doesn't hang, just parent's do_wait is blocked. It's 
different from upstream kernel.
I'm not sure which behavior is correct.

Best regards,
Wenji