[lkml]   [1998]   [Apr]   [11]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
Messages in this thread
    SubjectRe: 2.1.95 lockups
    In article <>,
    Kevin Fenzi <> wrote:
    >[ I am cc'ing the author of might need a change for post
    > 2.1.95 kernels ]
    >This is likely un-related to the other lockups people have been
    >reporting, but I can reliably lock up my dual p166 machine under
    >2.1.95 by running xcdroast. ;(

    The SCSI ioctl's are _not_ currently protected by the SMP lock: we're
    still fixing some basic issues with normal IO (timeouts and bus resets).
    Any unprotected region that calls into code that expects to be protected
    is likely to get _very_ upset (because it will end up doing bad things
    to the lock that it thought was held).

    >2.1.94 works fine (well, I didn't burn a cd, but it came up fine).
    >I also have a report from a friend the same thing is happening on
    >their dual ppro 166.
    >No oops, no nothing, just a hard lockup. Nothing ever comes up so it's
    >likely when xcdroast is scanning for devices and whatnot.

    It probably works fine on UP, because the lock goes away completely on

    >I do have the generic scsi support compiled as a module, so this could
    >be a kmod problem.

    No, it's a generic SCSI problem right now. The sg code simply doesn't
    do the right thing wrt the io_request_lock, and what ends up happening
    is that something tries to lock the spinlock twice.

    (The code is _really_ broken: it tries to keep the spinlock across a
    sleep, for example).

    Right now my #1 priority is to make sure that the basic disk operations
    are safe. There's some work to be done for that still (timeouts are
    currently not getting the lock correctly, so if you have a SCSI command
    that gets lost for some reason the machine will lock on SMP). But it
    _looks_ like fixing that should be fairly simple, and after that I'll
    look into the sg.c code.


    To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
    the body of a message to

     \ /
      Last update: 2005-03-22 13:42    [W:0.023 / U:1.044 seconds]
    ©2003-2016 Jasper Spaans. hosted at Digital OceanAdvertise on this site