DragonFly BSD
DragonFly kernel List (threaded) for 2005-11
[Date Prev][Date Next]  [Thread Prev][Thread Next]  [Date Index][Thread Index]

tsleep/wakeup patch #2 (Re: tsleep/wakeup)


From: Matthew Dillon <dillon@xxxxxxxxxxxxxxxxxxxx>
Date: Thu, 10 Nov 2005 15:27:22 -0800 (PST)

    Here's the second iteration of the tsleep/wakeup patch:

	fetch http://apollo.backplane.com/DFlyMisc/tsleep02.patch

    This is even more experimental, but I am finding that the 
    signal/sleep/wakeup interactions we inherited from FreeBSD are even
    more intertwined then I thought they were.  It's taking a lot to unwind
    the mess.

    Again, to test you want to create / destroy lots of processes as well
    as send lots of signals (^T, ^C, ^Z, etc).  Even try gdb'ing some 
    programs to test that the debugger still works :).

    On the bright side, this will make some of the future LWP work easier.
    I had to disentangle the process status (p_stat) from the sleeping vs
    stopped (SSLEEP vs SSTOP) status.  The process now has a P_STOPPED
    flag which is independant of whether the process has actually managed
    to stop yet or not.  SSTOP is no longer a valid p_stat state, but it
    is synthesized so programs like 'ps' can tell the difference between
    a stopped process and a sleeping process.

    As a consequence of the work, I also found a number of bugs... actually
    one very well known bug which is if tsleep() is called with PCATCH,
    a ^Z can sneak in there and cause the tsleep to not wake up until
    the process is continued again.  That little problem is now fixed.  I
    also fixed at least one ^Z / signal-parent race condition.

    In anycase, please test!  Clearly I won't be able to commit this on
    Friday like I hoped... perhaps monday, depending on how much more work
    I have to do on it.

						-Matt




[Date Prev][Date Next]  [Thread Prev][Thread Next]  [Date Index][Thread Index]