leaf crash :-)

Matthew Dillon dillon at apollo.backplane.com
Tue Sep 2 10:51:15 PDT 2003


:
:Okay, should be fixed now.  Please give it a whirl.
:
:						Jeffrey

    All right, I have installed a new kernel on leaf and turned
    net.inet.tcp.limitedtransmit back on.  We'll see if leaf survives.

    I am rather gratified by the stability of the system.  When I first
    turned on leaf (the shell machine) a couple of minor gotchas came out
    of the woodwork, like this tcp issue and a null pointer dereference in
    NFS's signal checking, but once those were fixed it has been rock solid.

    I also didn't think I would be able to stabilize the slab allocator so
    quickly, but it has been rock solid.  There are still some minor issues
    with regards to low memory situations but that's pretty much it.

    --

    In anycase, now on to the next task, which is going to be considerably
    more difficult.  I am going to try to rewrite the VFS path cache and
    VOP_LOOKUP this week which should have the immediate effect of fixing
    the vnode directory locking's race-to-root issue that has been in FreeBSD
    for years.  It is a necessary precursor to being able to move the VOP
    ops to a message-passing format.

    --

    Oh, I noticed that on an MP system 'ps' rather hillariously only lists
    the pure kernel threads (the ones without an associated proc) connected
    to whatever cpu the sysctl was run on rather then listing the threads
    for all the cpus.  e.g. I run 'ps ax' and see idle_0, usched 0,
    and all the interrupt threads.  Then I run it again and I only see
    idle_1 and usched 1.  It's kinda funny, actually.  If someone wants to
    have a go at fixing this low-priority issue I'll be happy to review.

					-Matt
					Matthew Dillon 
					<dillon at xxxxxxxxxxxxx>





More information about the Kernel mailing list