site_archiver@lists.apple.com Delivered-To: darwin-kernel@lists.apple.com

I have tried to use a mutex (which is what an IOLock is as far as I can see) and no luck. It sometimes works, sometimes it doesn't... It doesn't look like it is the good way to detect a thread termination.

While I agree with you about that, I think that perhaps you've got another problem. The filesystem I ported from Linux created what I would charitably call an "assload" of threads. I used mutexes to wait for the threads to finish up before the module was unloaded. I can unload my module just fine. And I don't hear a bunch of people complaining about this problem; if it was a serious problem then you would think it would come up a lot (or at least it would come up at least _once_ by someone else).

From what I remember you never actually _verified_ that this was the actual problem with a debugger; it was just your best guess. You also never posted a crash log or anything else. I'm willing to believe that your module is different enough that it might be triggering something that doesn't come up for most people, but I think the onus is on you at this point to prove that this is really the core issue (if my memory is wrong, then my apologies).

And you know what? I took a longer look at this (using the kernel source that anyone can download); it turns out that I was slightly wrong. While there isn't an exported function that waits for a thread to exit (there is an internal one called thread_wait()), you have a possible option. If you look at the places where thread_wait() is used, you've got thread_terminate() and thread_suspend(). Unfortunately, you can't use thread_terminate() on kernel threads other than yourself, so that leaves thread_suspend(). If you try to suspend a thread that has stopped, thread_suspend() will return KERN_TERMINATED. You don't actually want the thread to be suspended, so what you could do is a loop where you do a thread_suspend() and a thread_resume() until thread_suspend() (or thread_resume()) returns KERN_TERMINATED. I would caution that I'm not 100% sure this will work properly; this depends on a thread being marked as not active, and I haven't convinced myself that happens when a thread function simply exits. But that doesn't work, then I _think_ this should work: In your waitee thread: - Set your mutex to indicate that you're done with whatever you need to be done with. - Call thread_terminate() on yourself. In your waiter thread: - Wait for the above mutex - Do the aforementioned thread_suspend()/thread_resume() loop until you get a KERN_TERMINATED return code (or an error). --Ken _______________________________________________ Do not post admin requests to the list. They will be ignored. Darwin-kernel mailing list (Darwin-kernel@lists.apple.com) Help/Unsubscribe/Update your Subscription: http://lists.apple.com/mailman/options/darwin-kernel/site_archiver%40lists.a... This email sent to site_archiver@lists.apple.com