wadrasil wrote on 2020-06-06, 01:44:

As far as I know you would need to compile qemu for OSX on the hosts itself or another Mac. However qemu is supported on that pl […]
Show full quote

As far as I know you would need to compile qemu for OSX on the hosts itself or another Mac. However qemu is supported on that platform. I asked the author earlier about what is needed for this patch on other platforms and opengl was the main requirement.

I would try building from source with 3dfx patch and see if it works.

wiki.qemu.org/Host/Mac
Has more details.

Oh, many thanks - that's a project for another weekend then. This weekend I have other ideas 😉

It did not work because it complains about a pkg-config link, even though I added pkg-config through homebrew ?

This link discusses that error, not sure if it will help but does have some instructions for resolving that issue:

https://stackoverflow.com/questions/46812712/ … ready-installed

export PKG_CONFIG=/path/to/pkg-config

Good luck, I hope that helps.

wadrasil wrote on 2020-06-08, 12:15:

This link discusses that error, not sure if it will help but does have some instructions for resolving that issue: […]
Show full quote

This link discusses that error, not sure if it will help but does have some instructions for resolving that issue:

https://stackoverflow.com/questions/46812712/ … ready-installed

export PKG_CONFIG=/path/to/pkg-config

Good luck, I hope that helps.

Actually it didn't. But thanks anyway. Here's the error:

ERROR: sizeof(size_t) doesn't match GLIB_SIZEOF_SIZE_T.
You probably need to set PKG_CONFIG_LIBDIR
to point to the right pkg-config files for your
build target

when I do a 'which pkg-config', mac terminal returns /usr/local/bin/pkg-config. However what it wants is apparently the libraries. (Sorry, I don't know much about compiling stuff).

also, "echo $PKG_CONFIG_LIBDIR" returns empty.

Looks like you need to set $PKG_CONFIG_LIBDIR as well.

Also could be gcc version. Or qemu version. Try compiling an older version 4.x and see if that gives the same error.

wadrasil wrote on 2020-06-13, 07:02:

Looks like you need to set $PKG_CONFIG_LIBDIR as well.

Also could be gcc version. Or qemu version. Try compiling an older version 4.x and see if that gives the same error.

I was using 4.1.1 and 4.2.0 when I had this error. To set $PKG_CONFIG_LIBDIR, I need to know where they are (the libraries) and I don’t know.

i guess it's one of these
/usr/bin:/bin:/usr/sbin:/sbin:/usr/local/bin:/usr/X11/bin
i guess it may be not the user one but the general one:
/bin

but i guess it wants /lib not /bin

Any ideas? I am still not finding these libraries to compile on my mac. I wanted to try it out with qemu 4.2.0 so I could show to the UTM developer that it could be possible.

EDIT: I give up - I stop at a point where it complains about requiring glib 2.48 and gthread 2.0 even though I have glib 2.64 installed from homebrew.

2020-06-12: New QEMU installers (5.0.0).
https://qemu.weilnetz.de/w64/

alberthamik wrote on 2019-09-10, 07:19:

I'm rather curious, realizing the potential of this project, if it would be possible to add this pass-through feature to QEMU's PowerPC platform, specifically it's Mac99 emulation for Mac OS 9.X, which the voodoo cards technically worked on, but usually had to be run with a specific Macintosh release of voodoo card, or you could flash the GPU firmware. However, I'm unaware if Macintosh actually supported glide per se, or if it's support was mostly with just the GPU accelerating OpenGL or something. I've been glancing into other things if it falls into the latter, such as PCem's voodoo emulation.

No firmware flash was required to use a Voodoo2 (and maybe Voodoo1) card in a mac. You could install most reference design PC voodoo2 cards in a PCI based Mac using 3dfx's reference drivers for mac. 3dfx did release a mac firmware you can flash onto their voodoo3-2000 and voodoo3-3000 PCI/AGP cards which was necessary to use them as a full 2D/3D graphics card. I used a 3dfxVoodoo3-2000 for years. I have never seen the original Unreal game look as good as it did using the glide driver. There were some mac specific Voodoo1 cards such as the VillageTronic Voodoo1. The only mac-specific difference in those was that it had a built in mac to vga adapter for the video passthough. I also had one of those.

So with 3dfx's Mac/PPC reference drives installed in the MacOS9 PPC VM we should be able to do PCI passthough to a standard PCI Voodoo2 card just like if the same card was installed in a mac. (or a QEMU side voodooHardware Emulator. ) Is that possible? I'd like to understand more about achieving this goal.

The alternative would be to write macOS9 Glide shim drivers that pass on the glide commands to the natively running QEMU glide translater. There are a couple other projects that may have helpful information to do this. One was a shim that translated glide to opengl on the mac so that you could run glide games without a 3dfx card . The other did the opposite and used MESA to provide an implementation to run openGL games on my Voodoo card.

😀

Bruninho wrote on 2020-07-11, 20:36:

EDIT: I give up - I stop at a point where it complains about requiring glib 2.48 and gthread 2.0 even though I have glib 2.64 installed from homebrew.

what version of gthread have you got?

robertmo wrote on 2020-08-15, 17:42:

Bruninho wrote on 2020-07-11, 20:36:

EDIT: I give up - I stop at a point where it complains about requiring glib 2.48 and gthread 2.0 even though I have glib 2.64 installed from homebrew.

what version of gthread have you got?

Hi, I'm sorry, I don't remember. I've removed it all when I gave up on it. I'm currently using the latest qemu (5.1.0) from the homebrew.

Anyway, macOS qemu hvf acceleration doesn't work that well out of the box even with original qemu, at least for Windows 7 guests it doesn't work well. I always get a BSOD if I try to use it.

Does this 3dfx glide passthrough require anything specific from the host architecture? If no, it would be a great idea to contact the UTM developer and try to convince him to implement it (UTM is a wrapper for qemu), so 3dfx games on Windows 9x guests and on iPad Pro hosts could benefit from it. Although UTM does not have hvf acceleration too, like the macs have.

I already have a Win 9x guest being able to play Grand Prix 3, FIFA 98 and Flight Simulator 98 there, albeit with a small performance/graphics penalty. I can see potential for much more performance.

Hey, very interesting topic. Since I moved to Ubuntu 20.04 I'm still looking for a way to play old Windows games (especially these, for Glide or OpenGl). I read this guide, but I'm not familiar with applying this patch and compile programs under Linux. I've installed QEMU, but from a repository, so I don't know, how I can apply this patch now…

The authors GitHub has instruction for compiling this and applying the patch, there's a link on the first page. For ubuntu you need to install build essential and build-dep. Then you can try apt-get install build-dep qemu-system-i386 or x86_64. Then follow build instructions.

If you are familiar with docker you can follow the setup instructions from qemu's build page and use docker for the build. (You need to setup your own container unless you are building from the current qemu source tree and that won't work for the 3dfx patch as far as I can tell due to login access permissions for qemus repo)

There is no need to install dependencies other than gcc and make to your system when using docker. This also works for compiling from Linux for windows using qemus documentation for w32 builds.

It's a weekend project, but it's worth it. If you have a lot of free ram use mod RD and build to a ram drive. 3-4 gb should be enough.

Also you can't easily uninstall packages built from source, so it's good enough to run it from the build folder after the make completes and alias qemu to the location the executable.

If you can make a package that's better as it can be uninstalled. I'm not in a position to provide completed qemu builds but can help with compilation questions.

If there's any interest in how to use docker for cross builds I can post my notes. I did use an alpine Linux VM and ran docker from within that and everything compiled and exported without issues. My work VPN blocks docker but not kvm....

wadrasil wrote on 2020-08-21, 04:36:

My work VPN blocks docker but not kvm....

Might this workaround work for you? https://github.com/boot2docker/boot2docker/is … mment-247804033

😀

Hey, robertmo, big thanks! I'm really new in Linux world, but I want to learn as much as possible. This system is fascinating. 😀

Any chance of an update for QEMU 5.1.0 ?

I do want to report that Rloews and xmsdisk work for virtual memory with kvm enabled in qemu, I was not able to enable this with on separate drive / hdd-image. This uses compatibility mode paging but did work for a program that required swap space (Mech commander). This reboots without error with kvm enabled (missing ddll's). Rloews 32 bit and 64 bit ram disks work and you can assign 1GB plus for 98se and 4GB+ for ram drives within qemu. I did also figure out shortcuts for restart and shutdown commands work when using vbemp driver and or kvm in 98se(when normal commands fail / FS check on reboot). Also vhost-user-gpu works for the vbemp driver for win98se. This uses a separate process for the GPU and gets really good results on games that work with the driver. I have not tried the 3dfx patch so far with this so far.

Using vhost user gpu in qemu:
/opt/qemu/vhost-user-gpu -s /run/user/1000/vg.sock & qemu-system-x86_64 -M q35 -chardev socket,id=vgpu,path=/run/user/1000/vg.sock -device vhost-user-gpu-pci,max_outputs=0,chardev=vgpu

building for linux example:
Assuming you have dsound.diff and sdlaudio_c.diff in /usr/local/src/ and the required packages to build qemu you can use this to build and patch on fedora for windows (adjust config as needed):

This assumes you have 4Gb od spare RAM to use for tempfs for the listed config, the end result will be in /usr/local/src/install and will be relocatable to opt or /usr/local/.. as needed. Make sure to change perms or setup udev rules (/dev/kvm) and use -L . to find the firmware when launching qemu afterwards . In linux it can be a pain to find and remove files built from source, so pointing to a single directory can be preferred. (still need to learn packaging)

Also assumes you have downloaded and installed the windows sdk and relocated files to be available within the specified folder.

Cross compile for windows on fedora, copy package list from qemus docker build files and adjust configure as needed. Feel free to use the provided one as an example, it does build & complete and work:
cd /usr/local/src/ ;
sudo mount -t tmpfs -o size=2048m tmpfs /usr/local/src/qemu-3dfx ;
sudo mount -t tmpfs -o size=1024m tmpfs /usr/local/src/install ;
sudo chown -R sw-maker:sw-maker qemu-3dfx ;
sudo chown -R sw-maker:sw-maker install ;
sudo chmod u+rwx qemu-3dfx ;
sudo chmod u+rwx install ;
git clone https://github.com/kjliew/qemu-3dfx.git ;
cd qemu-3dfx ;
curl -s https://download.qemu.org/qemu-4.2.1.tar.xz | tar -xJ -C . ;
cd qemu-4.2.1/ ;
rsync -r ../qemu-0/hw/3dfx ./hw/ ;
rsync -r ../qemu-1/hw/mesa ./hw/ ;
cd audio
cp /usr/local/src/dsound.diff . ;
cp /usr/local/src/sdlaudio_c.diff . ;
patch -i dsound.diff ;
patch -i sdlaudio_c.diff ;
cd .. ;
patch -p0 -i ../10-qemu411-mesa-glide.patch ;
mkdir ../build && cd ../build ;
../qemu-4.2.1/configure --cross-prefix=x86_64-w64-mingw32- --enable-guest-agent-msi --target-list="x86_64-softmmu,i386-softmmu" --enable-guest-agent --with-win-sdk=/usr/local/src/windows-kits/10 --mandir=/usr/local/src/install --datadir=/usr/local/src/install --docdir=/usr/local/src/install --firmwarepath=/usr/local/src/install --bindir=/usr/local/src/install --libdir=/usr/local/src/install --libexecdir=/usr/local/src/install --sysconfdir=/usr/local/src/install --localstatedir=/usr/local/src/install --prefix=/usr/local/src/install --enable-sdl-image --enable-iconv --disable-gtk --audio-drv-list=dsound,sdl --enable-docs --disable-membarrier --disable-pie --disable-capstone --with-coroutine=windows --disable-kvm --enable-whpx --disable-fdt && make install -j 4 ;
cd /usr/local/src/install ;
bash /home/sw-maker/mingw-copy-deps.sh /usr/x86_64-w64-mingw32/sys-root/mingw/bin/ qemu-system*.exe ;
tar -zcvf /home/sw-maker/qemu-3dfx/qemu-421-3dfx-no-gtk-QW64-"$(date "+%c").tar.gz" . ;
cd /usr/local/src/ ;
sudo umount install && sudo umount qemu-3dfx ;
echo done ;

Build for ubuntu/mint 19:

/usr/local/src/myqemu/qemu-3dfx/qemu-4.1.1/configure --target-list=x86_64-softmmu --enable-guest-agent --with-win-sdk=/usr/local/src/windows-kits/10 --enable-kvm --mandir=/dev/qemu --datadir=/dev/qemu --docdir=/dev/qemu --bindir=/dev/qemu --libdir=/dev/qemu --libexecdir=/dev/qemu --sysconfdir=/dev/qemu --localstatedir=/dev/qemu --prefix=/dev/qemu --with-coroutine=ucontext --enable-sdl-image --enable-iconv --disable-gtk --audio-drv-list=sdl,pa --disable-pie --disable-capstone --enable-opengl --disable-virglrenderer --iasl=/usr/bin/iasl --firmwarepath=/dev/qemu

-cross-prefix=x86_64-w64-mingw32- --enable-guest-agent-msi are for cross building from linux to windows and are not needed on linux. mingw-copy-deps.sh is on github, add the busybox command before the awk command if you have an awk version that does not work with the script (sometimes it works fine and you dont need to use busy box.

mingw-copy-deps.sh
https://gist.github.com/h0tw1r3/51084def72c9765a5d12

The first example is a script I made after going through it manually more than once to get it down pat, it goes pretty quickly on tempfs so I wanted to share the example how to. I used the advise on github for building/patching as the basis for the script.