Messages in this thread | | | From | Roland McGrath <> | Subject | Re: [PATCH 16/16] ptrace: remove the extra wake_up_process() from ptrace_detach() | Date | Mon, 6 Dec 2010 16:10:43 -0800 (PST) |
| |
The plain wake_up_process was certainly wrong from the beginning.
We were conservative about changing it because of the difficulty of chasing all the corners where userland debuggers' behavior might be made to regress when it had been reliable in practice before (even if not always in theory, such as possible races that didn't bite in reality). The userland code has gone to many contortions to cope with how the kernel behaved in the past, whether or not that behavior ever made any good sense.
For that sort of reason, none of this stuff should change at all in a -stable kernel, nor late in a release cycle.
For new kernels, I think changing the behavior in the direction of something that can actually be described is OK as long as userland debugger maintainers like Jan agree to the new behavior and that the behavior really and truly does follow an articulated set of rules that the kernel and userland sides agree to.
Thanks, Roland
| |