(Emacs: -*- indented-text -*-)

			 A list of known bugs
		Copyright (c) 1996 - 1999 Ulrich Windl
		 <Ulrich.Windl@rz.uni-regensburg.de>
		     $Date: 1999/04/21 20:23:29 $

- A long-lasting bug in the Linux kernel is the incorrect updates to
  the CMOS clock, especially during daylight saving time.  Since
  kernel 2.0.32/2.2.0 a kernel warning is printed at least. It's not
  quite obvious how to fix it without knowing the hardware very well
  (it seems).

- Occasionally the serial driver only detected a pulse once every ten
  seconds in Linux 2.0.  In Linux 2.2 every 10th pulse was lost of an
  Rx character was ready at the same time.  Currently I have no idea
  why this happens.

- Related to the problem above is the fact that the serial driver
  produces significantly less jitter upon the second open operation.
  I suspect it's the line discipline stuff, but I have no good idea
  where to start.

- Architectures other than i386 are possibly still incomplete, but
  should be rather straight forward (files <arch>/kernel/time.c,
  include/asm-<arch>/{ioctls.h,termios.h}).

- The line discipline code should be revised.

- For multiple CPUs the cycle counters may vary significantly.  When
  changing CPUs from interrupt to interrupt, the time offset may jump
  accordingly, probably even backwards!

- People with good time references have reported high variation in
  time stamps (up to 3 ms).  Unfortunately, this is not repeatable.
  Try to find the cause of that effect.

- Some Linux extensions within ``adjtimex()'' (which happens to be
  used as ``ntp_adjtime()'') collided with bits newly introduced by
  the nanokernel.  Therefore the bits had to be moved, breaking binary
  compatibility for the following flags and routines:
  ``ADJ_OFFSET_SINGLESHOT'' (``ADJ_ADJTIME'' now), ``ADJ_TICK'',
  ``ADJ_TICKADJ'', adjtime().  The latter two collide with
  ``STA_NANO'' and ``STA_CLK''.
