kern/72979: unkillable process(es) stuck in `STOP' state

[David Xu]

The following reply was made to PR kern/72979; it has been noted by GNATS.

From: David Xu <davidxu at freebsd.org>
To: Ken Smith <kensmith at cse.Buffalo.EDU>
Cc: Mikhail Teterin <mi+mx at aldan.algebra.com>,
	Michael Nottebrock <michaelnottebrock at gmx.net>, re at freebsd.org,
	davidxu at t2t2.com, davidxu at viatech.com.cn,
	freebsd-gnats-submit at freebsd.org
Subject: Re: kern/72979: unkillable process(es) stuck in `STOP' state
Date: Thu, 28 Oct 2004 07:19:57 +0800

 Ken Smith wrote:
 
 >On Tue, Oct 26, 2004 at 12:07:47PM -0400, Mikhail Teterin wrote:
 >  
 >
 >>=On Tue, Oct 26, 2004 at 07:11:54PM +0800, David Xu wrote:
 >>=That looks both promising (given symptoms) and low-risk.
 >>=
 >>=Mikhail et. al., will you have any problems testing that?
 >>
 >>This problem was not easy to reproduce to begin with :-(
 >>
 >>I don't even know, how it can be positively claimed gone. Shall we merge 
 >>David's fix into RELENG_5 and hope for the best?
 >>
 >> -mi
 >>    
 >>
 >
 >Has anyone had time to test David's patch?  If yes have you had
 >any problems since?
 >
 >I should know later today if the other person who had been reporting
 >problems has had them stop.  He reported last night he was still able
 >to reproduce the problem but he hasn't had any "luck" doing it today.
 >
 >David, my only real question I guess is whether perhaps	P_STOPPED_SINGLE
 >should be added to the flags that are being turned off by your patch.
 >If I'm following the code right what had made P_STOPPED_TRACE special
 >was it's one of the three signals lumped into P_STOPPED and the code
 >in do_tdsignal() looks for that, as well as I'm sure the schedulers...
 >
 >  
 >
 P_STOPPED_SINGLE needn't be removed here, in fact, the flag should 
 already be
 turned off at this point, if the flag is still there, then there must be a
 bug, thread_single(SINGLE_EXIT) should turn it off if it returns or thread
 will exit in the function.
 P_STOPPED_TRACE is cleared by debugger via PT_DETACH in sys_process.c, but
 if debugger dies without chance to call ptrace(PT_DETACH), then exit1() must
 clear it.
 
 David Xu