See http://www.freebsd.org/~julian/ for pointers to diffs or other ways of getting the sources..
At this stage it is as good as I can get it right now. It builds world. It runs teh sampel threaded program, it even does both at the same time.. Matt has run it on an SMP machine and there are patches for various architectures (stubbed) to allow them to keep working (as long as they don't try threading :-). I would like to commit this as soon as I have feedback from the usual suspects. I do have the following issues on my list but I'd like ot commit BEFORE fixing them: 1/ psignal has been rewritten, but needs to be re-re-written. It works but I'd like any signal gurus to feel free to hack at it after it's committed. 2/ Ptrace works but once again, Anyone whoe knows and loves ptrace could probably do with rewriting parts of it in teh light of threads once it is committed. (I can gdb normal programs but threaded programs are not likely to cooperate. 3/ I have some cleanups to do in the generation of upcalls. there is still un-needed information being stored for theupcall and it can probably be simplified. 4/ The definition of a 'suspended' process has been changed. It should be changed again. I have some code to keep a track of which threads are currently in userland and which are in the kernel, with a counter in teh process so you can instantly tell how many threads are where. This will sim-lify some of the suspend code. 5/ Matt and I decided that we need to change the behaviour of 'BOUND' threads in a KSE process so that their KSEs can be lent to other threads in the same group that need to complete kernel operations.. BOUND threads are those that for some reason are not operating in threaded mode (e.g. they were in the UTS or some other critical code). The change is that BOUND threads can lend out their KSEs (scheduler spots) to other threads as long as those threads do not return to user land. (At least not on that KSE). 6/ code cleanup in the code that aborts sleeps and condvar waits. There are also a lot of debugging aids that I will delete BEFORE checkin. Please ignore them when reviewing. (e.g. KASSERTS in kern_switch.c) IN ADDITION TO THIS, (damned caps lock) there are still code paths that can not possibly succeed in a multi threaded world. What milestone 3 gives is enough of the most common codepaths cleaned up so that a very simple threaded program can run, be suspended, restarted and exit. There is still no ability to run threads on multiple processors as there is still only one KSE (schedulable entity) per process. Having said all this I think it is ready to commit. I will try respond to immidiate concerns people have but since we will have a working system after the commit, I'd like to do as much stuff after the commit as possible to allow others to test and comment. Next week I will have to spend a lot of time on work matters so I'd like to get this in in 24 hours from now, and have the weekend to handle problems that arrise. Estimated commit time: (assuming current works at that time) will be around 1AM GMT June 29 that will be 6PM here in San Francisco. Julian p.s. I will put the test programs in the tools tree soemwhere.. To Unsubscribe: send mail to [EMAIL PROTECTED] with "unsubscribe freebsd-current" in the body of the message
