upstream/0.3.3.2/BUGS

$Id: BUGS,v 1.33 2005/06/04 12:35:14 debug Exp $


Serious bugs:

  x)  An unknown math coprocessor bug. (Ultrix' dxclock sometimes looks
      weird.)

  x)  Enabling cache emulation (./configure --enable-caches) triggers bugs.

  x)  Compiling with certain versions of gcc and -O3 optimization on i386
      triggers bugs. I'm not sure whether they are in gcc or in GXemul.

  x)  Linux/DECstation (Debian) oopses extremely often unless -U is used
      at run-time. I'm not sure yet why it bugs out. With -U, the risk is
      lower, but not completely gone. _Maybe_ this is a bug in Linux. Why?
      Because the oops message contains things like ANSI escape codes and
      characters in registers (including the pc and return address register);
      this looks like a buffer overflow in the serial driver. (Another thing
      that gives weight to this theory is that the serial driver in Linux is
      still being developed.)  But this is just a guess.

  x)  bintrans bugs.  (For example, NetBSD/sgimips for IP22/24 hangs during
      boot with bintrans enabled.)


Less serious bugs / issues:

  o)  hardware device ticks are done at cycle specific intervals, not
      instruction intervals, so sometimes a fraction of a cycle can be
      "lost".

  o)  running Linux/DECstation 2.4.26 with no scsi disks attached causes
      a warning message to be printed by Linux.

  o)  UDP packets that are too large are not handled well by the Lance device.

  o)  Colors in X11 framebuffers on MacOS X hosts are wrong. (I'm not sure
      how to solve this; the code works on both little-endian (Alpha) and
      big-endian (UltraSPARC) X-servers...)

  o)  bintranslated 64-bit stuff checks to see if the top 32 bits are
      all zeroes or all ones, and then uses 32-bit tables and such. This
      is a bug. It should check the top 33 bits, not 32. (Alpha only, already
      fixed for i386?)

  o)  NetBSD/arc 2.0 uses the ASC controller in a way which GXemul cannot yet
      handle. (NetBSD 1.6.2 works ok.)

1	$Id: BUGS,v 1.33 2005/06/04 12:35:14 debug Exp $
2
3
4	Serious bugs:
5
6	x) An unknown math coprocessor bug. (Ultrix' dxclock sometimes looks
7	weird.)
8
9	x) Enabling cache emulation (./configure --enable-caches) triggers bugs.
10
11	x) Compiling with certain versions of gcc and -O3 optimization on i386
12	triggers bugs. I'm not sure whether they are in gcc or in GXemul.
13
14	x) Linux/DECstation (Debian) oopses extremely often unless -U is used
15	at run-time. I'm not sure yet why it bugs out. With -U, the risk is
16	lower, but not completely gone. _Maybe_ this is a bug in Linux. Why?
17	Because the oops message contains things like ANSI escape codes and
18	characters in registers (including the pc and return address register);
19	this looks like a buffer overflow in the serial driver. (Another thing
20	that gives weight to this theory is that the serial driver in Linux is
21	still being developed.) But this is just a guess.
22
23	x) bintrans bugs. (For example, NetBSD/sgimips for IP22/24 hangs during
24	boot with bintrans enabled.)
25
26
27	Less serious bugs / issues:
28
29	o) hardware device ticks are done at cycle specific intervals, not
30	instruction intervals, so sometimes a fraction of a cycle can be
31	"lost".
32
33	o) running Linux/DECstation 2.4.26 with no scsi disks attached causes
34	a warning message to be printed by Linux.
35
36	o) UDP packets that are too large are not handled well by the Lance device.
37
38	o) Colors in X11 framebuffers on MacOS X hosts are wrong. (I'm not sure
39	how to solve this; the code works on both little-endian (Alpha) and
40	big-endian (UltraSPARC) X-servers...)
41
42	o) bintranslated 64-bit stuff checks to see if the top 32 bits are
43	all zeroes or all ones, and then uses 32-bit tables and such. This
44	is a bug. It should check the top 33 bits, not 32. (Alpha only, already
45	fixed for i386?)
46
47	o) NetBSD/arc 2.0 uses the ASC controller in a way which GXemul cannot yet
48	handle. (NetBSD 1.6.2 works ok.)
49