for running several NetBSD ports properly on the emulator. PR pkg/41742
Discussed with and okay'ed by wiz@.
Upstream changes:
The changes between release 0.4.6.6 and 0.4.7 are:
* Motorola 88K emulation is now complete enough to run a guest operating
system: OpenBSD/mvme88k 4.4.
A big thank you goes to Miod Vallat, who sent me physical 88100
documentation, via snail mail. This has been invaluable when implementing
the 88K emulation mode.
The changes between release 0.4.7 and 0.4.7.1 are:
* Applying a man page hypenation patch from Jonathan Wiltshire.
* Minor documentation errors fixed.
* Some more missing M88K instructions added.
* NetBSD/hpcmips can now be installed and run using serial console (i.e.
without -X), in addition to using the framebuffer.
The changes between release 0.4.7.1 and 0.4.7.2 are:
* NetBSD 5.0, OpenBSD 4.5, and HelenOS 0.4.0 have been released, and the
guest OS page has been updated to reflect this. (Not all versions and
architectures work; the versions listed on the guest OS page are the latest
ones that have been tested.)
* A bug has been fixed which prevented configuration files from being used.
Patches pulled from upstream trunk:
patch-aa:
Applying a patch for setting the ARCBIOS OSLOADPARTITION environment
variable in a more correct way.
http://gxemul.svn.sourceforge.net/viewvc/gxemul?view=rev&revision=5473
This makes NetBSD/arc work on PICA emulation with VGA console.
(more proper fix than the previous patch-aa)
patch-ab:
Better fix for the Cobalt board ID number.
http://gxemul.svn.sourceforge.net/viewvc/gxemul?view=rev&revision=5485
(fixes in the previous patch-ab has been integrated in the latest release)
patch-ac:
Applying a patch which makes NetBSD/hpcmips work well with the clock again.
http://gxemul.svn.sourceforge.net/viewvc/gxemul?view=rev&revision=5472
pkgsrc changes:
- update MASTER_SITES and HOMEPAGE (now they are in sourceforge)
- update PLIST (add share/doc/gxemul/unsupported.html)
Changes:
- Applying a patch from Kiyotaka Inaba which causes an ARM UND
exception to be generated by the instruction word 0xe6000011, which
is needed by gdb when running inside the emulator. Also enabling
ARM UND exceptions for ARM floating point instructions (needed by
both Linux and NetBSD/arm).
- Fix for a free() bug in bootblock_iso9660.c. Updating the
OpenBSD/landisk installation instructions from 4.2 to 4.4, after
verifying that it works.
Due to small issue with gxemul (PCI registers values of piix
controller are not saved having been written) FreeBSD in
gxemul panics after detecting IDE devices. Apply this patch to
fix it...
Bump PKGREVISION.
Changes since 0.4.6.1:
* Applying a patch from Oleksandr Tymoshenko for the wdc device, fixing the
order of 16-bit words within 32-bit words in the identify struct, and the
error register is not cleared on reads any more.
* The disk device used in the test machines can now access disk images
larger than 4 GBs when emulating 32-bit architectures (as suggested by
Julian Elischer).
* For ARM emulation, a special case for PC-relative load which wasn't
implemented before has now been implemented. (The problem could be
triggered by attempting large builds using NetBSD's pkgsrc inside e.g.
NetBSD/cats.)
* The ARM 'bkpt' instruction has been implemented (patch from Jiri Svoboda).
The changes between release 0.4.6 and 0.4.6.1 are mostly focused on documentation
updates, to reflect new releases of guest operating systems:
* OpenBSD 4.2 has been released. OpenBSD/landisk 4.2 works in GXemul.
* NetBSD 4.0 has been released. The installation instructions in GXemul's
documentation have been updated to reflect this.
o In most cases this simply means that NetBSD 4.0 should work,
in addition to NetBSD 3.1.
Some specifics worth mentioning are:
o For NetBSD/pmax, the change from 3.1 to 4.0 means that X Windows is now
working again out-of-the-box. (Simply run startx as root to start X.)
o For NetBSD/arc, the supported version has jumped from 1.6.2 to 4.0.
o NetBSD/landisk has been released, and works in GXemul.
o NetBSD/prep is unfortunately still at NetBSD 2.1.
o In some emulation modes, such as evbmips, hpcmips, and perhaps some more,
NetBSD 4.0 uses the clock in a way which is not yet implemented in the
emulator. (NetBSD 4.0 works, but NetBSD 3.1 worked better.)
o (I have not tried some of the "heavier" emulation setups, such as
NetBSD/sgimips.)
The changes between release 0.4.5.1 and 0.4.6 include, among other things:
o) NetBSD/pmppc can now run in the emulator (with root-on-nfs), on
an emulated Artesyn PM/PPC board.
o) An instruction combination has been implemented for the idle loop
used by NetBSD/arm (cats, netwinder, and iq80321). In plain English,
this means that if the guest OS inside the emulator is not using
any CPU, the emulator should not use much CPU on the host either.
o) Some minor SuperH emulation speed improvements.
o) General code cleanup: Non-working (skeleton) emulation modes have
been removed, to make it easier to maintain the source code in
the long run, and many unused/legacy constructs have been removed.
The changes between release 0.4.5 and 0.4.5.1 include, among other things:
o) OpenBSD/landisk has now had its first release (4.1, 2007-05-01).
Landisk emulation is now stable enough to allow OpenBSD/landisk
to be installed onto a disk image, and run from it.
o) Finally found and fixed an old bug in the address to symbol name
lookup mechanism, which caused some symbols to be missed. Debug
output with -t or -i should now show all symbols.
o) Initial support for "disk overlays" has been implemented. This
enables e.g. simple roll-back of emulated disk contents to a
previous state.
o) Dyntrans bug fixes; code translations on physical addresses that
were offset a multiple of 128 MB from each other could either
cause weird bugs, or translation leaks (leading to unnecessary
dyntrans cache overflows).
o) Some cleanup: The GDB debugging stub support, some dummy machine
modes (TS7200, Walnut, PB1000, and Meshcube), and some dummy or
experimental CPUs (RCA180x and Transputer) have been removed, to
make the emulator slightly more maintainable.
o) SuperH (SH4) emulation is now somewhat more stable.
x) GXemul 0.4.4: A NetBSD/dreamcast Live CD works.
x) GXemul 0.4.4.1: A Linux/dreamcast Live CD also works.
o) PowerPC "G4" emulation is now stable enough to let NetBSD/macppc
run from a disk image. (Installing actually worked before, but the
NetBSD/macppc GENERIC kernel uses AltiVec instructions which were
not implemented correctly.)
o) The PICA-61 (arc) and i80321 (evbarm) emulation modes now have their
timers fixed at 100 Hz. A hardcoded speed like this is very ugly, but
it is at least better than before (when the timer wasn't really
running at any specific speed at all).
Changes between release 0.4.3 and 0.4.4 include, among other things:
o) The interrupt subsystem has been redesigned. This means two things:
x) Internal code cleanup, which makes the whole emulator more
maintainable. Instead of using magically encoded integers
for interrupts, strings are now used. These strings are in
the form of "paths", so that devices and busses can more
easily be connected to other busses, devices, or CPUs.
x) Some machine types which happened to work in release 0.4.3,
but were not listed in the documentation as working, may
have stopped working now. As always, the documentation should
indicate the combinations of machine modes and guest OSes that
are supposed to work.
o) SuperH (SH4) emulation is now somewhat more stable, enough to let a
NetBSD/dreamcast Live CD be usable.
o) PowerPC "G4" emulation is now stable enough to let NetBSD/macppc
run from a disk image. (Installing actually worked before, but the
NetBSD/macppc GENERIC kernel uses AltiVec instructions which were
not implemented correctly.)
o) The PICA-61 (arc) and i80321 (evbarm) emulation modes now have their
timers fixed at 100 Hz. A hardcoded speed like this is very ugly, but
it is at least better than before (when the timer wasn't really
running at any specific speed at all).
INSTALLATION_DIRS, as well as all occurrences of ${PREFIX}/man with
${PREFIX}/${PKGMANDIR}.
Fixes PR 35265, although I did not use the patch provided therein.
Changes between release 0.4.2 and 0.4.3 include, among other things:
o) SuperH (SH4) emulation is now stable enough to let a NetBSD/dreamcast
GENERIC_MD (ramdisk) kernel reach userland.
o) There is now a simple framework for letting emulated clocks, as seen
by guest operating systems, run at the same speed as the host clock.
So far, the DECstation, MobilePro (hpcmips), NetWinder, CATS, Malta
(evbmips), Cobalt, Algor, Dreamcast, and testmips machine modes
use the new clock/timer framework.
o) Some changes to the way expressions are evaluated in the built-in
debugger, and some changes in command behaviour:
x) Expressions (including assignments) can now be arbitrarily
complex, using parentheses, and the following operators:
+ - * / % (modulo) ^ (xor) & (and) | (or)
x) Some internal emulator variables can now be read/written using
normal expressions. Examples of commands that did not work
earlier, but should work now:
print verbose
r5 = sp - arch_pagesize * 4
machine[0].statistics_enabled = 1
x) To force a name to be interpreted as a setting/register name,
a hash sign (#) is now used instead of the percentage sign (%).
(In the new expression evaluator, % means arithmetic modulo.)
x) The 'focus' command now also selects a cpu, in addition to
selecting machine and emul.
x) The 'reg' command only prints registers for one cpu now, not
all cpus in the currently focused machine.
o) The wdc (standard IDE controller) had a bug which prevented disk
images larger than 2 GB to work correctly. This has been fixed.
o) For MIPS emulation, some combinations of emulated processor + guest
operating system should now work better when idling (i.e. the host
should not run at 100% CPU):
x) For MIPS32/MIPS64 and RM5200, the 'wait' instruction should
now work more or less as expected.
x) For VR41xx (e.g. MobilePro) emulation, the standby instruction
should work like the 'wait' instruction.
x) For R3000 emulation, where there is no hardware wait instruction,
I've implemented "instruction combination" hacks for both
NetBSD/pmax and Debian/pmax, so that their cpu idle loops are
detected and treated almost as a wait instruction.
o) MIPS 64-bit address translation (X=1) was not fully working before;
TLB exception handling for xkseg and larger-than-2GB-userland should
now actually work. (Thanks to Juli Mallett and Carl van Schaik for
noticing these problems.)
o) The mouse cursor update routines in DECstation (LK201) emulation
previously used the fact that guest OSes set the _hardware_
cursor position. In order to support X Windows when emulating
modern versions of NetBSD/pmax, which don't set the hardware
position anymore, a workaround has been implemented which only
sends relative coordinates to the guest OS. This has two drawbacks:
1. Ultrix emulation with dual- and tripple-head emulation will
most likely feel very strange. It will still work, though.
2. Cursor movement feels "accelerated", because the emulator
sends unaccelerated movements to the guest OS, which then
accelerates them. This can however be compensated to some
degree by running 'xset m 1 0' in the guest OS.
Having weird accelerated mouse movement is better than having no
mouse support at all, so this change was necessary.
READ_TRACKINFO, sufficient to get a NetBSD/cats kernel mount a
cd-rom image. I still get ""cd_setblksize: trying to change bsize,
but no blk_desc" but it seems to be a benign warning.
Pkgrevision bumped to 1.
found on cats and netwinder. This gets a NetBSD/cats -current
install kernel booting again after the move to timecounters. Before
this change, the emulated kernel would loop during the tlp0
attachment, inside delay().
Note, I'm guessing that a free-running timer should not generate
an interrupt on wrap (I don't have the hardware documentation).
Patch sent upstream.
Changes between release 0.4.1 and 0.4.2 include, among other things:
o) The NetWinder emulation mode now works well enough to let
NetBSD/netwinder run from a disk image.
(NetBSD/netwinder has to be installed manually, though, because
there is no installation ramdisk kernel. The GXemul documentation
has an example of how to install NetBSD/netwinder using an emulated
NetBSD/pmax machine.)
o) Algorithmics P5064 emulation works well enough to let NetBSD/algor
run from a disk image. (Similar to NetBSD/netwinder, it has to
be installed manually, using another emulated machine.)
o) PCI configuration register writes can now be handled, which allows
NetBSD/Malta (evbmips) 3.0.1 and NetBSD/cobalt 3.0.1 to run from
PCI IDE harddisk images. (Previously, only NetBSD 2.1 worked for
the Malta and Cobalt emulation modes.)
o) Some performance increases:
The virtual translation table update routine has been simplified
to work the same way for 32-bit and 64-bit emulation, and this
apparently results in a speedup for all 64-bit modes.
For 32-bit MIPS, some more instruction combinations have been added.
Changes between release 0.4.0.1 and 0.4.1 include, among other things:
o) Some bugs in the dynamic translation core have been fixed, making
all emulation modes (especially the MIPS mode) more stable.
o) In the refactoring effort between 0.4.0 and 0.4.0.1, single-stepping
(and instruction tracing) of 64-bit programs was accidentally broken.
Single-stepping in 64-bit mode could result in strange exceptions.
This has been fixed in 0.4.1.
o) MIPS emulation performance has been improved somewhat:
R3000: After removing some buggy code (hints for physical page
translations), it was possible to remove the workaround for R3000
caches which was needed in 0.4.0.1 to make Linux and Ultrix run.
This gives an overall speedup for R2000/R3000.
For non-R3000, there have been some speedups as well. After fixing
reference count bugs for 64-bit addressing in the dyntrans system,
workarounds/hacks in the tlbwr/tlbwi instructions and in the ASID
change helper function could be removed.
o) A new -s command line option is now available, for dumping
raw runtime data/statistics on every instruction to a file.
Currently, the following kinds of data can be dumped:
1. the program counter (virtual address)
2. the physical address representation of the program counter
3. the internal dyntrans instruction call pointer, useful when
developing/optimizing the emulator
This data can then be analyzed by external tools.
(There was a -s command line option in previous releases of
the emulator, but it did not work as intended, and was not
usable from dyntrans emulation modes.)
The most important change between release 0.3.8 and 0.4.0.1 is:
o) The emulation of MIPS processors has been completely rewritten;
it now uses the same portable dynamic translation system as the
ARM and PowerPC emulation modes.
On Alpha and i386 hosts (and AMD64 hosts running in 32-bit mode),
GXemul previously used translation into native code. This release
will perform worse than 0.3.8 on those host architectures.
On all other hosts (including AMD64 running in native 64-bit mode),
0.4.0 is likely to be faster than 0.3.8, when emulating MIPS.
I think that in the long term, moving towards full portability like
this is a good idea.
(0.4.0 was a bit buggy and unstable; 0.4.0.1 is a quick-fix release.)
There have also been many other changes, including, but not limited to:
o) The "test machine" functionality is more well-defined than before,
and some tutorial-like demos have been added. These could be useful
e.g. in operating system construction courses.
o) NetBSD/sgimips 3.0 works now. This is most likely due to the rewrite
of the MIPS emulation mode. Previous releases of GXemul only worked
with NetBSD/sgimips 2.1.
o) I have begun implementing rudimentary support for GDB remote serial
protocol connections. This means that you can run e.g. the Data
Display Debugger, and connect it to a GXemul instance.
No advanced GDB functionality is working yet, but starting and
stopping the emulated machine and single-stepping should work.
The user-visible changes between release 0.3.7 and 0.3.8 include:
o) The IQ80321 (Xscale) machine mode is now working well enough to
run NetBSD/evbarm 2.1.
o) Faster framebuffer output in some situations.
User-visible change between release 0.3.6.2 and 0.3.7 include:
o) The experimental PowerPC emulation mode is now working well enough
to allow NetBSD/prep 2.1 to be installed and run inside the emulator.
It is not 100% stable, and it is not optimized for performance yet,
but hopefully enough for simple experiments.
o) I finally took the time to implement a DEC 21143 NIC; this brings
network connectivity to NetBSD/cats. (The userland "NAT"-like
networking layer is still a bit buggy, and does not work with
everything. However, NetBSD/cats can now be installed via ftp.)
o) CD-ROM images can now in some cases be detected as ATAPI CD-ROMs
instead of IDE harddisks. It works for at least NetBSD, OpenBSD,
and Linux on CATS, and NetBSD on hpcmips.
Internal (code related) changes include:
o) Cleanup of the PCI and ISA bus frameworks; in practice this means
that more code can be shared between different emulated machine
models than before, and that adding new machine types will become
easier.
o) Dyntrans updates; 32-bit PowerPC mostly, but also many performance
related updates for ARM.
The 0.3.6.1 release fixes some issues related to ARM emulation:
x) The emulator can now be compiled inside NetBSD/cats or OpenBSD/cats,
inside the emulator itself. (In 0.3.6, some bugs prevented this.)
x) Performance increase: A non-scientific but realistic test, measuring
the real-world time it takes to do a full NetBSD/cats installation,
seems to indicate that 0.3.6.1 can be twice as fast as 0.3.6 was.
The most imporant user-visible change between release 0.3.5 and 0.3.6 is:
(X) The experimental ARM emulation mode is now working well enough
to install NetBSD/cats and OpenBSD/cats onto harddisk images.
The user-visible changes between release 0.3.4 and 0.3.5 are minor, and
can be summarized as follows:
o) Updates to the (old) binary translation subsystem, resulting
in minor speed improvements.
o) 64-bit MIPS dmult/dmultu has been fixed.
o) slt* instructions for 64-bit MIPS were incorrectly implemented
in the i386 backend. This has been fixed.
Changes:
x) When emulating a network of multiple machines, the emulated
machines can now be placed on different hosts.
x) NetBSD/evbmips can now be installed and run from a disk image.
(There is no INSTALL kernel for NetBSD/evbmips, so you need to
install using another OS, for example emulated NetBSD/pmax.)
x) NetBSD/sgimips can now be installed. Not onto a SCSI disk,
but the files can be exported via nfs from another emulated
machine. The sgimips machine can then netboot. (Read the
documentation for details.)
Several changes are involved since they are all interrelated. These
changes affect about 1000 files.
The first major change is rewriting bsd.builtin.mk as well as all of
the builtin.mk files to follow the new example in bsd.builtin.mk.
The loop to include all of the builtin.mk files needed by the package
is moved from bsd.builtin.mk and into bsd.buildlink3.mk. bsd.builtin.mk
is now included by each of the individual builtin.mk files and provides
some common logic for all of the builtin.mk files. Currently, this
includes the computation for whether the native or pkgsrc version of
the package is preferred. This causes USE_BUILTIN.* to be correctly
set when one builtin.mk file includes another.
The second major change is teach the builtin.mk files to consider
files under ${LOCALBASE} to be from pkgsrc-controlled packages. Most
of the builtin.mk files test for the presence of built-in software by
checking for the existence of certain files, e.g. <pthread.h>, and we
now assume that if that file is under ${LOCALBASE}, then it must be
from pkgsrc. This modification is a nod toward LOCALBASE=/usr. The
exceptions to this new check are the X11 distribution packages, which
are handled specially as noted below.
The third major change is providing builtin.mk and version.mk files
for each of the X11 distribution packages in pkgsrc. The builtin.mk
file can detect whether the native X11 distribution is the same as
the one provided by pkgsrc, and the version.mk file computes the
version of the X11 distribution package, whether it's built-in or not.
The fourth major change is that the buildlink3.mk files for X11 packages
that install parts which are part of X11 distribution packages, e.g.
Xpm, Xcursor, etc., now use imake to query the X11 distribution for
whether the software is already provided by the X11 distribution.
This is more accurate than grepping for a symbol name in the imake
config files. Using imake required sprinkling various builtin-imake.mk
helper files into pkgsrc directories. These files are used as input
to imake since imake can't use stdin for that purpose.
The fifth major change is in how packages note that they use X11.
Instead of setting USE_X11, package Makefiles should now include
x11.buildlink3.mk instead. This causes the X11 package buildlink3
and builtin logic to be executed at the correct place for buildlink3.mk
and builtin.mk files that previously set USE_X11, and fixes packages
that relied on buildlink3.mk files to implicitly note that X11 is
needed. Package buildlink3.mk should also include x11.buildlink3.mk
when linking against the package libraries requires also linking
against the X11 libraries. Where it was obvious, redundant inclusions
of x11.buildlink3.mk have been removed.
The most important/visible changes from 0.3.1 to 0.3.2 are:
x) NetBSD/cobalt can run from a harddisk image. (Installation
must be done using another OS though, for example NetBSD/pmax.)
x) Some minor fixes to make the binary translation system a bit
more stable.
The two most visible changes from version 0.3 to 0.3.1 are:
x) Name change (from mips64emul to GXemul).
x) NetBSD/hpcmips can now be installed and run from a harddisk
image on an emulated NEC MobilePro 770, 780, 800, or 880.
GXemul is a machine emulator. The initial goal was to write a simple
64-bit MIPS emulator for running multiprocessor experiments with a
microkernel, but the emulator can be used for many other things. While
some simulators only simulate a CPU, GXemul also simulates other
hardware components, making it possible to use the emulator to run
unmodified operating systems, such as NetBSD, OpenBSD, or Linux.
[previously known as mips64emul, it was renamed because now
supports more cpu archs than MIPS, as sparc, ppc...]
