basilisk-devel — Basilisk II development-related discussion

Re: [B2-devel] JIT on x86 problem

[Gwenole B. <gb...@di...>]

Hi,

Sorry, I don't know why this happens for you. It could be a 
miscompilation, would you mind telling me which compiler do you use? 
For reference, I am using gcc 3.2.2 on x86 and gcc 3.3.2 on amd64.

> This is my config file:
> ========
> frameskip 8

Ouch, video refreshes won't be nice looking at. Since people told me 
they found video stuff slow in SheepShaver, I always set that to 0 
(which turns out to be 1 in SheepShaver), thus a 30 Hz resolution. My 
eyes don't suffer from that and benchmarks show twice the speed as a 
native 8100/80 for video.

> ignoresegv true

Please avoid this. You could get hangs instead of a real SIGSEGV and 
you wouldn't know where SheepShaver possibly crashed.

> I suspect that using of ./configure script from the older release of
> SheepShaver may cause this problem. Bot how can I generate it myself if
> every attempt to ./autogen.sh ends with:
> ========
> ./configure: line 6541: syntax error near unexpected token `1.2.0,'
> ./configure: line 6541: `AM_PATH_GTK(1.2.0,'
> ========

Make sure you have gtk+-1.2 autoconf macros, e.g. 
/usr/share/aclocal/gtk.m4.

> Would you be so kind to prepare a new tarball snapshot?

That was the plan to make it this week but there is another issue I 
want to check first.

Bye,
Gwenole.

Re: [B2-devel] JIT on x86 problem

[Tomasz J. <to...@je...>]

> On Thu, 13 May 2004, Tomasz Jerzykowski wrote:
>
> > As soon as Gwenole informed about JIT speed improvements checked in
> > (yesterday), I downloaded the latest CVS code and compiled it.
Compilation
> > succeeded and when I ran it, configuration window appeared. But as soon
as I
> > started emulation, the application just terminated.
>
> Please make clean, make. Or at least rm -f obj/sheepshaver_glue.cpp
> obj/ppc-*.o. Otherwise objects layout is different thus causing strange
> things. ;-)
>
> I know, I should fix Makefile deps... Hmm, or probably people should run
> make dep first?
>

<Robocop_voice>
Thank you for your cooperation.
</Robocop_voice>

Didn't help. This is the ending of 'strace ./SheepShaver' log:

==========
clone(child_stack=0, flags=CLONE_CHILD_CLEARTID|CLONE_CHILD_SETTID|SIGCHLD,
child_tidptr=0xf70010c8) = 12098
waitpid(12098, Reading ROM file...
WARNING: Cannot open /dev/fd1u1440 (No such device or address)
PowerPC CPU emulator by Gwenole Beauchesne
[WIFSIGNALED(s) && WTERMSIG(s) == SIGSEGV], 0) = 12098
--- SIGCHLD (Child exited) @ 0 (0) ---
exit_group(-1) = ?
========


This is my config file:
========
disk /root/sheepshaver/hd
extfs /root/sheepshaver
windowmodes 3
screenmodes 0
seriala /dev/ttyS0
serialb /dev/ttyS1
rom mac.rom
bootdrive 0
bootdriver 0
ramsize 67108864
frameskip 8
gfxaccel true
nocdrom false
nonet false
nosound true
nogui false
noclipconversion false
ignoresegv true
jit true
keyboardtype 5
keycodes false
mousewheelmode 1
mousewheellines 3
dsp /dev/dsp
mixer /dev/mixer
ignoresegv true
========

I suspect that using of ./configure script from the older release of
SheepShaver may cause this problem. Bot how can I generate it myself if
every attempt to ./autogen.sh ends with:
========
./configure: line 6541: syntax error near unexpected token `1.2.0,'
./configure: line 6541: `AM_PATH_GTK(1.2.0,'
========

I have gtk ang glib devel packages installed (otherwise I wouldn't be able
to compile it at all). 'aclocal-1.4' trick doesn't work.Would you be so kind
to prepare a new tarball snapshot? I would be glad to try to compile it
myself.

Jerzu

Re: [B2-devel] JIT on x86 problem

[Gwenole B. <gbe...@ma...>]

On Thu, 13 May 2004, Tomasz Jerzykowski wrote:

> As soon as Gwenole informed about JIT speed improvements checked in
> (yesterday), I downloaded the latest CVS code and compiled it. Compilation
> succeeded and when I ran it, configuration window appeared. But as soon as I
> started emulation, the application just terminated.

Please make clean, make. Or at least rm -f obj/sheepshaver_glue.cpp 
obj/ppc-*.o. Otherwise objects layout is different thus causing strange 
things. ;-)

I know, I should fix Makefile deps... Hmm, or probably people should run 
make dep first?

[B2-devel] JIT on x86 problem

[Tomasz J. <to...@je...>]

Hello.

As soon as Gwenole informed about JIT speed improvements checked in
(yesterday), I downloaded the latest CVS code and compiled it. Compilation
succeeded and when I ran it, configuration window appeared. But as soon as I
started emulation, the application just terminated. There was no error
message on the console, nor any message box. However, when I swiched JIT
off, it worked (booted 8.5 CD).

A couple of weeks ago I was compiling the latest tgz snapshot from February
and it worked fine, so I think I know how to configure SheepShaver.
This time however I might have screwed something, because I wasn't able to
generate ./configure script from autoconf (as many other people), so I
copied it from tgz snapshot. Another potential problem might be that I am
running FC 2 test 3 distribution.

Is there anything that you want to know in order to troubleshoot this
problem (strace or whatever)?

Jerzu

[B2-devel] Performance improvements

[Gwenole B. <gb...@di...>]

Hi,

I have committed a small improvement to the JIT, i.e. direct chaining of
basic blocks in the same page, which represent around 99.99% of cases for
"coin" benchmark from ssbench.

It's a clear win but I have not fully tested it yet. nbench benchmarks on
a P4 @ 1.8 GHz now yields a 7.3 slow down vs. native x86 execution. i.e.  
nbench.ppc under kpxrun vs. nbench.x86.

BTW, I noticed that sheepshaver_cpu::interrupt() is sometimes nested. This
should not happen since XLM_IRQ_NEST is normally checked in
HandleInterrupt() prior to invoking the real NanoKernel one. Atomic
operations are normally OK as they are derived from boehm's gc. I think I
could arrange the code for multiple interrupt handler stacks like in
native ppc mode, but I really think this should normally not happen...

This reminds me I apparently forgot to commit x86_64 spinlocks too.

Re: [B2-devel] SheepShaver status (2004/05/10)

[Christian B. <Chr...@un...>]

Hi!

On Mon, May 10, 2004 at 07:56:24PM +0200, Gwenole Beauchesne wrote:
> What is the advantage of cloning the Mac cursor into X11 by the way? 

It makes the cursor movement independent of the Mac window update rate
(which in the era of 66MHz machines and DMA-less PCI graphics cards was
usually around 5-10Hz...) leading to a much smoother user experience.

Bye,
Christian

-- 
  / Physics is an algorithm
\/ http://www.uni-mainz.de/~bauec002/

Re: [B2-devel] SheepShaver status (2004/05/10)

[Gwenole B. <gb...@di...>]

Hi,

> On Mon, May 10, 2004 at 06:54:57PM +0200, Gwenole Beauchesne wrote:
>> - TUN/TAP device support for networking.
>
> So this replaces the ethertap (which appears to be deprecated anyway)?

Actually, you can still use "ethertap" but it's deprecated and anyway 
it doesn't work for me with a 2.4 kernel. I refer to "TUN/TAP" device 
as per the use a "tun" device set in Ethernet mode thus equivalent to a 
tap.

This method is the only one working for me at home with my DSL 
connection. I am a complete newbie here and I understood how frustated 
users could get with a cable or DSL connection not working right away.

BTW, QEMU people have a user mode network stack through slirp. I tried 
it but I only got at most 5 KB/sec transfers in Basilisk II on PPC. I 
don't know yet wether it's a libslirp limitation or my glue not working 
correctly. I tried to check with an actual QEMU on ppc but it does not 
work with live boot CDs I gathered...

>> - I have disabled MacOS -> X11 cursor tracking as I noticed it
>> sometimes disappears with e.g. MS Internet Explorer or simply TextEdit
>> fields.
>
> Which is due to X11 not doing XOR when mask = 0 and data = 1, unlike 
> MacOS...

What is the advantage of cloning the Mac cursor into X11 by the way? 
Actually, I turned SheepShaver's behavior into Basilisk II's by not 
fully supporting cscSetHardwareCursor. If we really want that, I don't 
know how to fix it yet.

Bye,
Gwenole.

Re: [B2-devel] SheepShaver status (2004/05/10)

[Christian B. <Chr...@un...>]

Hi!

On Mon, May 10, 2004 at 06:54:57PM +0200, Gwenole Beauchesne wrote:
> - TUN/TAP device support for networking.

So this replaces the ethertap (which appears to be deprecated anyway)?

> - I have disabled MacOS -> X11 cursor tracking as I noticed it 
> sometimes disappears with e.g. MS Internet Explorer or simply TextEdit 
> fields.

Which is due to X11 not doing XOR when mask = 0 and data = 1, unlike MacOS...

Bye,
Christian

-- 
  / Physics is an algorithm
\/ http://www.uni-mainz.de/~bauec002/

Re: [B2-devel] PearPC

[Gwenole B. <gb...@di...>]

Hi,

> Darwin 7 on PearPC on Windows.
> http://www006.upp.so-net.ne.jp/t_tanaka/images/ppc_darwin.jpg

Yes, that's nice but it doesn't seem to support MacOS Classic systems, 
does it? Anyhow, SheepShaver would still outperform in that case. ;-)

Once I finish latest SS changes, I will probably have a closer look at 
it.

Bye,
Gwenole.

[B2-devel] SheepShaver status (2004/05/10)

[Gwenole B. <gb...@di...>]

Hi,

New changes committed recently to CVS include:

- TUN/TAP device support for networking. TODO: the tunconfig script 
could be arranged so that a DHCP server is set up on the tunX interface 
thus enabling users to only configure MacOS with a DHCP client.

- Ethernet driver fixes for little-endian & 64-bit platforms. I have 
not tested the new code on x86/amd64 yet but it still works on PPC. I 
also introduced Mac_sysalloc() and Mac_sysfree() for allocating stuff 
in MacOS system heap zone, and therefore ensuring allocations under 
32-bit address space.

- I have disabled MacOS -> X11 cursor tracking as I noticed it 
sometimes disappears with e.g. MS Internet Explorer or simply TextEdit 
fields. However, you can still be re-enabled by setting 
Unix/video_x.cpp (mac_cursor_enabled) to true.

Bye,
Gwenole.

[B2-devel] PearPC

[Toshimitsu T. <t_t...@db...>]

Hi.

> PearPC is an architecture independent PowerPC platform emulator 
> capable of running most PowerPC operating systems.
http://pearpc.sourceforge.net/

Darwin 7 on PearPC on Windows.
http://www006.upp.so-net.ne.jp/t_tanaka/images/ppc_darwin.jpg
---
Toshimitsu Tanaka
t_t...@db...
http://homepage3.nifty.com/toshi3/index.html

Re: [B2-devel] TUN/TAP support

[Gwenole B. <gb...@di...>]

I wrote

> I have yet to little-endian & 64-bit fix networking code in 
> SheepShaver.

Some little-endian fixes are now in CVS, but I have still not tested it 
on x86, only minimal changes and reading of ether.cpp, ether_linux.cpp. 
As for 64-bit fixes, we need to replace the new/delete with proper 
32-bit memory allocation. An idea is to use NewPtrSysClear/DisposePtr, 
what do you think?

Bye,
Gwenole.

[B2-devel] TUN/TAP support

[Gwenole B. <gb...@di...>]

Hi,

I have committed code to enable TUN/TAP ("tun" module) support in both 
Basilisk II and SheepShaver. That's cool, I can even get Internet 
connection at home through an ethernet DSL modem. A possible extension 
to the tunconfig script (derived from MOL's) is to install a DHCP 
server on the tunX interface so that users only need to configure for a 
DHCP client on the Mac side.

I have yet to little-endian & 64-bit fix networking code in 
SheepShaver. TUN/TAP is also supposed to work on FreeBSD and Solaris 
but without adequate system I can't test and tunconfig may need some 
arrangements.

I am not a networking wizard, so if someone could proofread and make 
sure the following portion added to the README file is understandable, 
that would be fine. Thanks.

       3. Access the network through a "tuntap" interface.
          The "ethernet card description" must be set to "tun".

          TUN/TAP provides packet reception and transmission for user
          space programs.  It can be viewed as a simple Point-to-Point
          or Ethernet device, which instead of receiving packets from a
          physical media, receives them from user space program and
          instead of sending packets via physical media writes them to
          the user space program.

          A virtual network configuration script is required and the
          default is /usr/local/BasiliskII/tunconfig unless you specify
          a different file with the "etherconfig" item.

          This script requires you that "sudo" is properly configured
          so that "/sbin/ifconfig" and "/sbin/iptables" can be executed
          as root. Otherwise, you can still write a helper script which
          invokes your favorite program to enhance a user priviledges.
          e.g. in a KDE environment, kdesu can be used as follows:

            #!/bin/sh
            exec /usr/bin/kdesu -c /path/to/tunconfig $1 $2

[B2-devel] PPC CPU Emulation Question

[Mark H. <ma...@mm...>]

Hello all!

I wanted to ask a question about the PPC CPU emulator that is in the 
B2-Dev.  I grabbed the first release of the GameCube Emulator and was 
surprised at how fast it ran Zelda and Super Mario Sunshine.  I mean the 
graphics were messed up at times but I think the graphics emulation 
isn't as mature as the CPU emulation.

I know the GC uses an IBM PPC CPU @ 405 Mhz (Called Gekko).  I don't 
know how it compares to what it is in the B2 code but I just thought if 
this CPU emulator was better/faster/more complete/whatever, is this 
something that might allow B2 to do more?  I don't know if they are 
going to release the source code but if they have it done and its closer 
to say a G3 then I just thought it might offer more Mac emulation 
options down the road.  :-)

Just a thought, thanks,
Mark

Re: [B2-devel] Open Source Firmware?

[Gwenole B. <gb...@di...>]

Hi,

> I posted this before, but it did not show up in the archives. Sorry if
> this is a dupe.

Nope, the initial message did not go through.

> I would like to emulate a powerpc machine, but without licensing a ROM
> image.

You do need a PowerMac PCI ROM or a NewWorld ROM image, like e.g. the 
one from the retail 8.6 CD version (nwrom v1.4 IIRC) since it contains 
the 68k emulator, low-level service routines and the NanoKernel. You 
don't need OpenFirmware nor actual BootROM.

> My question is, are there any projects that aim to provide open source 
> firmware that works with SheepShaver?

Not that I am aware of. I once tried to merge the OF client interface 
from Mac-on-Linux but it does not provide any benefit over what is 
already done in augmenting the NameRegistry manually. And we don't have 
a full initialization process from the HardwareInit routine in the 
(uncompressed part of) NewWorld ROM. Rather, we directly jump to the 
NanoKernel entry point.

> Does OpenBIOS work with SheepShaver?

No, it doesn't.

Bye,
Gwenole.

[B2-devel] Open Source Firmware?

[Robbert H. <bas...@in...>]

I posted this before, but it did not show up in the archives. Sorry if
this is a dupe.

I would like to emulate a powerpc machine, but without licensing a ROM
image. My question is, are there any projects that aim to provide open source firmware that works with SheepShaver? Does OpenBIOS work with SheepShaver?

It shouldn't be that hard to write some boot firmware; we know how the code is interpreted, and what it does is pretty much up to the programmer, I'd say (unless you want to clone the behavior of some specific system).

May the Source be with you,

Bob

Re: [B2-devel] Native QuickDraw acceleration

[Christian B. <Chr...@un...>]

Hi!

On Thu, Apr 22, 2004 at 05:41:21PM +0200, Gwenole Beauchesne wrote:
> Apple really should release NQD specs to public now that Classic is
> obsoleted. That sort of magic numbers does not help to determine the
> underlying semantics.

Definitely. At least there is a supported interface for the acceleration.
The routines in ShapeShifter were based on complete guesswork (but the
NQD hooks are not entirely different from it).

Bye,
Christian

-- 
  / Physics is an algorithm
\/ http://www.uni-mainz.de/~bauec002/

Re: [B2-devel] Native QuickDraw acceleration

[Gwenole B. <gb...@di...>]

On Thu, 22 Apr 2004, Christian Bauer wrote:

> On Mon, Apr 19, 2004 at 01:18:53AM +0200, Gwenole Beauchesne wrote:
> > BTW, the following test should always yield true so I intend to remove 
> > it:
> > ((uint32 *)p)[0x15c >> 2] > 0
> 
> Uhmm, why would this always be true?

Ooops, forget about it, my mind read a ">=" with unsigned comparison. ;-)

Apple really should release NQD specs to public now that Classic is
obsoleted. That sort of magic numbers does not help to determine the
underlying semantics.

Bye,
Gwenole.

Re: [B2-devel] Native QuickDraw acceleration

[Christian B. <Chr...@un...>]

Hi!

On Mon, Apr 19, 2004 at 01:18:53AM +0200, Gwenole Beauchesne wrote:
> BTW, the following test should always yield true so I intend to remove 
> it:
> ((uint32 *)p)[0x15c >> 2] > 0

Uhmm, why would this always be true?

Bye,
Christian

-- 
  / Physics is an algorithm
\/ http://www.uni-mainz.de/~bauec002/

Re: [B2-devel] SheepShaver updates

[Christian B. <Chr...@un...>]

Hi!

On Wed, Apr 14, 2004 at 12:46:02AM +0200, Gwenole Beauchesne wrote:
> BTW, if someone knows the Apple timing ID for 1152x768, I don't have 
> Universal Headers handy new enough to check.

Hm, mine aren't recent enough, either...

Bye,
Christian

-- 
  / Physics is an algorithm
\/ http://www.uni-mainz.de/~bauec002/

[B2-devel] Native QuickDraw acceleration

[Gwenole B. <gb...@di...>]

Hi,

I have just committed a few arrangements for Native QuickDraw (NQD) 
acceleration hooks under Linux/PPC. With extra changes, it will be easy 
to clean the code for little-endian and 64-bit platforms. So far, 
performance improved by approx. 2x in emulated mode with MacBench 5.0 
CopyBits/srcCopy benchmark.

BTW, the following test should always yield true so I intend to remove 
it:
((uint32 *)p)[0x15c >> 2] > 0

I don't know about other magic values either. I think we should also 
benefit from a CopyBits() acceleration but I'd want to get some NQD 
stuff done first.

Bye,
Gwenole.

Re: [B2-devel] SheepShaver updates

[Gwenole B. <gb...@di...>]

Hi,

> In native mode, we get stack corruption and we are jumping to garbage 
> when the mouse is moved in fullscreen mode. This is weird as we are 
> using an alternate stack in signal handlers.

Fixed in CVS by setting up another stack for next signal handler during 
ppc_interrupt(). I am not sure that MacOS normally handles nested 
ppc_interrupt() calls though.

This stack corruption problem reminds me a very similar problem when I 
was playing with 'PICT' copying through an offscreen GWorld, i.e. I got 
garbage value in return. Might worth trying again.

Bye,
Gwenole.

[B2-devel] SheepShaver performance

[Tomasz J. <to...@je...>]

Hello!

First of all I wanna say that SheepShaver is really a great piece of code. I
never imagined anyone would be willing/able to code a real PowerMac
emulator. Nonetheles, I have little performance problems with it.
On my Duron 1.3 it is pretty slow, but generally usable. However, sometimes
the emulator freezes for a few seconds. Host CPU consumption is still 100%,
but you can't do anything (within emulator, that is).
There are also some problems with mouse input (probably for the same reason)
- sometimes you need to click several times before a button is pressed down
or a folder opened.
It looks as if the mouse input thread (provided SheepShaver is written in
multithreaded manner) was starved by CPU emulation thread ot JIT thread... I
think so, because Basilisk II was multithreaded in Windows and you could
play with thread priorities with good results.

Jerzu

Re: [B2-devel] SheepShaver updates

[Gwenole B. <gb...@di...>]

> Got some time this weekend and committed a few changes to CVS:
> - Runtime depth switching now works, tested 1bpp -> 16bpp
> - Fullscreen now works in emulated PPC mode, only tested on 32-bit big 
> endian target

Forgot to mention:
- If XPRAM is not correctly set up, some initial default values are 
loaded. This prevents e.g. a long lag when booting with no initial 
~/.sheepshaver_nvram file.

[B2-devel] SheepShaver updates

[Gwenole B. <gb...@di...>]

Hi,

Got some time this weekend and committed a few changes to CVS:
- Runtime depth switching now works, tested 1bpp -> 16bpp
- Fullscreen now works in emulated PPC mode, only tested on 32-bit big 
endian target

In native mode, we get stack corruption and we are jumping to garbage 
when the mouse is moved in fullscreen mode. This is weird as we are 
using an alternate stack in signal handlers.

I also disabled VidMode extension by default as it's highly X server 
dependent and does not always work. e.g. on my PBG4, I get a black 
screen when switching to anything < 1024x768.

BTW, if someone knows the Apple timing ID for 1152x768, I don't have 
Universal Headers handy new enough to check.

Bye,
Gwenole.