Re: [patch] add arch_ptrace_stop() hook and use it on ia64

[David Mosberger <davidm@napali.hpl.hp.com>]

>>>>> On Tue, 10 May 2005 17:59:46 -0700, Roland McGrath <roland@redhat.com> said:

  Roland> (This leads to threads in TASK_TRACED with a pending
  Roland> SIGKILL, that cannot be killed with repeated kill -9s, and
  Roland> live until the tracer uses PTRACE_CONT, detaches, or dies.)

Ah, yes, I see now that SIGKILL is a rather nasty special-case.

  Roland> When I suggested this change for ia64 originally, I
  Roland> overlooked the need to handle blocking in writing to user
  Roland> memory.  I'd like to give a little more thought to the
  Roland> general case.  As long as only uninterruptible waits are
  Roland> provoked by an arch hook, then I think it is reasonably
  Roland> solvable.  I think that SIGKILL races are the only ones that
  Roland> can arise, and those can be addressed with some signal
  Roland> bookkeeping changes.  I'd like to give it a little more
  Roland> thought.  I expect it will wind up being some core changes
  Roland> that make it safe for an arch hook to drop and reacquire the
  Roland> siglock if it's doing something that won't always complete
  Roland> quickly, and then this will all happen before changing
  Roland> state, unlocking, and deciding to block (which won't be done
  Roland> if there was an intervening SIGKILL).

OK, that sounds good to me.  Please keep me posted.  It would be nice
to have this hook in place, since it does fix a bug on ia64 and makes
the code simpler.

Thanks,

	--david