The following problems are known to exist as of:
94Dec10 version 1.34
    CORRECTION TO 1.33a ---> Marcio Saito
    Marcio found a serious bug : If more than 1
    card are installed only the first one works.
    The others only receive. FIXED !
    A small bug in the setup program INST_CYCLOM
    prevented the users of kernels previous to 1.1.68
    to apply patches. FIXED !
    INST_CYCLOM lacks of a good recovery routine. Currently
    no auto recovery is implemented although a backup routine
    is implemented and works flawlessly !
    steffen@laguna.schiele-ct.de

94Dec01 version 1.33-C:
    CORRECTION TO 1.33 ---> PATCHES ONLY !!!
    Patches of interrupt.h and main.c didn't fit !
    Have corrected the patches supplied by Randolph.
    Now all patches will fit into kernels from 1.1.68 to 1.1.70
    and hopefully up. 
    Please report any "failed HUNK" problems to 
    steffen@laguna-schiele-ct.de

94Dec01 version 1.33:
    Don't know of the direct high speed works.
    The BogoMip problem persists--gotta talk to Linus.

94Nov28 version 1.32 (still):
    Documentation in earlier versions of the README file was
    incorrect with respect to the application of the patch
    files.  The patche files have been corrected.

94Nov23 version 1.32:
    The BogoMip problem persists.

94Nov20 version 1.31:
    Patches from several contributers have
    corrected or reduced the effects of these problems in
    the 1.28 version.

    Specifically the DTR, tcsh, and chars-in-buffer problems
    have been addressed.

    The BogoMip problem persists.

94Nov7 version 1.28:

    DTR is being mis-handled.  It drops during a login
    session (when the port is being used for dial-in
    access).   The current work-around is to tell the
    modem to ignore DTR changes or do ignore changes
    that occur over some small interval.

    For some reason, tcsh fails.  We suspect some
    condition is being mis-reported, but don't yet
    know which.

    There's an ugly hack inside that reports "chars_in_buffer"
    to be zero at all times.  When it told the truth, sessions
    are unceremoniously disconnected.

    CTS/RTS is not thoroughly tested, but is suspected
    to work much better.

    Works for 1.1.50, should work for 1.1.30+ to 1.1.59.
    Please report your successes!

    The BogoMip calculation has been moved earlier in the kernel.
    Because it enables interrupts a bit earlier than before, you'll
    get a message "irq.c:bad bottom half entry" that is benign.  (The
    move was to allow the driver to use the udelay() function with
    real values for loops_per_second.  Can someone develop a good fix
    to prevent the error message?)

