MiSTer FPGA Forum

Hello all,

Following on from getting Linux working in this thread: viewtopic.php?f=13&t=4006&e=1, I have been continuing to look at getting Windows NT to boot on ao486. Today I finally got it to run!


Here are the problems I have had to solve to get it booted:

Problem: NT won't read the hard disks.
Solution: I found a driver which works here: http://alter.org.ua/en/soft/win/uni_ata/

Problem: NT won't read floppy disks.
Workaround: Not solved, but avoided. I have performed a "floppyless" installation of NT from DOS. This also required injecting the above hard disk driver into the setup files to avoid needing to provide it on a floppy during the setup process.

Problem: NT setup crashes with an "unable to initialize graphics" error
Solution: This is due to the ao486 BIOS, and happens in Bochs as well if you use this BIOS image there. I have compiled a new build of the BIOS from more recent source which avoids this error.

Problem: Newer BIOS images (see above) can't read hard disks on MiSTer/ao486
Solution: I have found that one of the "retired" fields in the IDE IDENITFY command is actually required by newer BIOS images (the current image works around this). I have added the missing field.

Problem: NT setup goes blank when selecting the video mode and does not recover
Solution: Just press cancel! It will save a 16-color 640*480 mode instead. Windows will complain about this again on first full boot, but just hit cancel again.


However, I have caused one new problem:

Problem: My new versions of the BIOS and IDE code do not work under DOS, so this is not a patch which can just be released and used by all MiSTer users. I am still working on solving this.

So... it's early days but it's running! So far I've caused one blue screen by trying to access the floppy drive (seriously, what's up with that?), and I need to find a better fix for BIOS/IDE so this can be released more broadly. More to come!

Alex

Wow! Finally a modern Domain Controller

ragnar wrote: ↑Tue Feb 15, 2022 6:09 pm Wow! Finally a modern Domain Controller

Well that just sounds like a challenge to get networking working too!

Nice! I made it run last year as well on the AO486, using the upgrade path from win 98 to NT, I even tried windows 2000 and XP! (spoiler XP is out of reach no matter what I tried) You can even have Msdos 8 running on NT. Still I feel Windows 98 is as far as it is worth to go.

Also the only way you will get native games running on NT properly is the upgrade path from windows 98, because it will then use the video drivers from windows 98 since there isn't any proper video driver for the ao486 core on NT. You can't install video driver directly from NT it won't work.

Neocaron wrote: ↑Tue Feb 15, 2022 7:43 pm ...You can even have Msdos 8 running on NT....

I'm confused what you mean by this or how upgrading from 98 would work without fixing some of the other issues I've listed. Is there any possibility you're talking about Windows ME rather than Windows NT?

alexoughton wrote: ↑Tue Feb 15, 2022 7:59 pm

Neocaron wrote: ↑Tue Feb 15, 2022 7:43 pm ...You can even have Msdos 8 running on NT....

I'm confused what you mean by this or how upgrading from 98 would work without fixing some of the other issues I've listed. Is there any possibility you're talking about Windows ME rather than Windows NT?

My brain did get confused, yoy are absolutely right but it doesn't change anything of what I said at the end. Because I played with both ME and NT and If you don't upgrade from windows 98, you won't get proper video driver support.
That was my point. Unless you found a work around for that, and in this case please share ^^

But yes, you will get proper graphics support if you install the vesa driver in windows 98 and then do an upgrade path to ME or NT from there.

Video driver work still to do. If I can't get an NT driver working, I will look at one from 98, as per your suggestion (as I know there were some NT-based drivers for 98, so the idea makes some sense).

I now have a version of this which doesn't require an IDE update, only the BIOS needs updating. This version seems to not break DOS!

Testing continues. WIP branch is here: https://github.com/alexoughton/ao486_Mi ... biosupdate

I think the biggest thing Sorgelig added to the custom video driver for Windows 98 was the ability to use the Linear Frame Buffer which allowed the higher res/color options. That's how I remember it anyway. I don't see the sourcecode for the driver he built but it is in the releases folder:

https://github.com/MiSTer-devel/ao486_M ... STERFB.INF
MiSTerFB.drv

New branch, as I've narrowed-down the specific change in the upstream Bochs BIOS which fixes the crash:

https://github.com/alexoughton/ao486_Mi ... test-touch

This is a change to the memory map which reports to the operating system which ranges are reserved for BIOS use. I will do some more testing tomorrow before submitting a Pull Request and then I'll give a write-up here on how to get NT installed with the third-party disk driver I linked to.

alexoughton wrote: ↑Tue Feb 15, 2022 11:07 pm New branch, as I've narrowed-down the specific change in the upstream Bochs BIOS which fixes the crash:

https://github.com/alexoughton/ao486_Mi ... test-touch

This is a change to the memory map which reports to the operating system which ranges are reserved for BIOS use. I will do some more testing tomorrow before submitting a Pull Request and then I'll give a write-up here on how to get NT installed with the third-party disk driver I linked to.

Do you mind uploading your compiled bios to releases in your branch? I would love to mess with it.

flynnsbit wrote: ↑Wed Feb 16, 2022 3:06 am Do you mind uploading your compiled bios to releases in your branch? I would love to mess with it.

Sure! Here's the most recent build: https://github.com/alexoughton/ao486_Mi ... 820-only-1

I've only had a little time to look into this today. So far I've found that I can improve stability significantly by installing Service Pack 3 (previously I was on SP1 because that's the ISO I have). This removes many of the crashes I have been seeing related to disk activity, even with the new driver. Floppy access still does not work. Upgrading to SP6a is the same as SP3.

I still have a persistent crash when performing certain activities caused by vga.dll. For example, attempting to open the "Properties" dialog of the C: drive will always cause a crash. I think my installation is still using SP1's vga.dll though, so I'm going to see about a fresh installation from an SP6 ISO. I'll write down my steps as I go with that and this will form the basis of the installation steps I publish. Note that I haven't made any progress with getting a proper Tseng ET4000 driver to load. This is all just to get stable 640*480 / 16-color video.

I've also removed a few lines of code from my branch as they don't seem necessary (only for machines with more than 4GB of RAM). Removing them has not changed anything, and I want the PR I submit to be as small as possible when I submit it.

OK, here are my instructions for getting NT running. The service pack slipstreaming is based off instructions found here: http://reboot.pro/topic/2383-windows-nt ... g-toolkit/

1. Download the tool at http://navozhdeniye.narod.ru/sskit.zip (note that the mscab.exe file mentioned in the forum post is already in this archive)

2. Make sure you have copies of the sp6a install files and the base NT i386 install files somewhere on disk. I am using c:\slipstream\sp6a and c:\slipstream\base as my examples. The output directory c:\slipstream\out should also exist.

3. Run: 4. Copy the contents of the "out" directory on top of the contents of the "base" directory. Windows will tell you there's about 600 files to be overwritten. Tell it to replace those files.

5. Download the uniata driver from here (tested with version 0.47b): http://alter.org.ua/en/soft/win/uni_ata/#uniata_beta

6. Extract the uniata driver rar file into c:\slipstream\base\uniata

7. Copy c:\slipstream\base\uniata\uniata.sys to c:\slipstream\base\uniata.sys

8. Edit c:\slipstream\base\dosnet.inf. There are two occurrences of the line "d1,atapi.sys". Add this line underneath both: 9. Edit c:\slipstream\base\layout.inf. Find the line "atapi.sys = 1,,18432,,,,_3,4,0,0" and add this line underneath: 10. Edit c:\slipstream\base\txtsetup.sif. Find the line "atapi.sys = 1,,,,,,_3,4,0,0" and add this line underneath: 11. Another edit to c:\slipstream\base\txtsetup.sif. Find the line "atapi = atapi.sys,4" and add this line underneath: 12. One more edit to c:\slipstream\base\txtsetup.sif. Find the line "atapi = "IDE CD-ROM (ATAPI 1.2)/PCI IDE Controller"" and add this line underneath: 13. Make an ISO of c:\slipstream\base. I used PowerISO for this, but you can use anything you like. No need to make it bootable or anything special.

14. Transfer your ISO to the MiSTer. You should also have my build of boot0.rom in place (or the official version if this has been updated with my patch by the time you read this).

15. You will also need a DOS boot disk with the "disk" and "format" tools, and it should have CD access set up (e.g. CD.SYS, MSCDEX.EXE). Make sure that your CD driver is not set up to call the device "CDROM" or "BANANA" (commonly used in boot disk images I have seen), as these will conflict with files which are being transferred. I call mine "ACD".

16. Boot ao486 with the DOS boot disk, prepared ISO, and blank HDD attached.

17. Use fdisk and format to set up the hard disk with a single FAT partition and reboot from the floppy drive again.

18. Go to the CD drive (e.g. D:) and run "winnt.exe /B". This will start a "floppyless" installation. Press enter to confirm the location of the setup files (the CD drive). The files will now take a while to transfer. Supposedly this will go faster if you use SMARTDRV.EXE on your DOS boot disk, but I haven't tried this.

19. When setup tells you the "MS-DOS based portion is complete", unmount the floppy image and press ENTER.

20. The system will reboot into the next stage of setup. Press ENTER to continue. Press ENTER again to confirm the mass storage devices and accept the license agreement.

21. Set the "Display" to "Standard VGA (640x480, 16 colors)". Then select "the above list matches my computer" and press ENTER.

22. Select to install on the single FAT partition which exists. Select to leave the current file system intact (i.e. do NOT convert to NTFS, as this has issues on the current ao486 IDE implementation).

23. Confirm the installation directory of \WINNT by pressing ENTER. Press ESC to skip the disk check (again, this has issues with ao486 IDE right now, and it's somewhat pointless for an emulated disk anyway).

24. During the file copy process I got an error transferring one file - oleacc.dll. I'm not sure why yet. Ignore it if it happens to you, and consider manually putting it into place from the install CD once finished.

25. Continue through the setup process, and then on to the graphical portion of setup. Set your setup options as desired (I use "custom" and select everything, but that's just me).

26. Do not select to create an emergency disk, as the floppy drive currently doesn't work under NT. Select "Do not connect this computer to a network at this time".

27. Finish setup by selecting your timezone and confirming 640x480 16-color mode. Click "restart computer" when prompted.

28. Log on to NT and go to Control Panel. Open "SCSI Adapters" and go to the "Drivers" tab.

29. Click "Add". Click "Have Disk". Enter the path "d:\uniata\nt4". Select "Universal ATA Driver (Win NT4) (Win2003)". Select "No" when prompted to reboot.

30. Highlight "IDE CD-ROM (Atapi 1.2)/Dual-Channel PCI IDE Controller". Click "Remove". Click "Yes". Reboot.

31. In Control Panel go to "Devices". Select "atapi" and click "Startup". Change to "Disabled" and confirm.

32. Run Service Pack 6a setup. This is necessary despite our file slipstreaming to complete all of the registry and security changes from the service pack.

Note that the vga.dll crash is still happening. I spent a little longer on this today. I managed to get a "universal" VBE driver up-and-running but it would only start in 320*200 256-color mode, which Windows was very upset about. That resolution is too low to be usable, but I was able to see that this solved the vga.dll driver crash issue. So if I can find a working driver, even if it only supports 640*480, it will be worth it as an alternative just to solve the crash. Then this thing will actually be pretty stable from what I can see.

PR has been raised... https://github.com/MiSTer-devel/ao486_MiSTer/pull/95

I'm happy to report that sorgelig and I have been discussing the changes. They got simplified to just a single number being updated and were merged!

The new BIOS version is now released: https://github.com/MiSTer-devel/ao486_M ... /boot0.rom


Update: Aaaand reverted again. Sorgelig realized this will actually conflict with the shared folder driver. Let me see if I can find something which works in both cases.

Update 2: Released again! Fixed MisterFS.

alexoughton wrote: ↑Fri Feb 18, 2022 8:25 pm I'm happy to report that sorgelig and I have been discussing the changes. They got simplified to just a single number being updated and were merged!

The new BIOS version is now released: https://github.com/MiSTer-devel/ao486_M ... /boot0.rom


Update: Aaaand reverted again. Sorgelig realized this will actually conflict with the shared folder driver. Let me see if I can find something which works in both cases.

Update 2: Released again! Fixed MisterFS.

I saw your interaction with Sorg on Github and I didn't want to muddy that thread but here are my notes from trying to get the LFB working with UNIVBE in DOS using the ET4000 Driver. There are a few games this would fix if LFB worked in DOS.

Here is what LFB will do if you actually get a picture in DOS, note that image is the boot logo on boot used in my Top 300 pack: It's like it is stuck in the buffer.

And here is what the actual logo.sys bootscreen looks like on boot: "it seems that LFB with VESA doesn't work at all, either a blank screen or that. Banked framebuffer mode works great on all resolutions. To test, UNIVBE Release 5.3 (anything below the latest SDD version will work as they removed all the PCI video cards) then you have to run UNIVBE, then enable both VBE 2.0 32-bit interface and Linear framebuffer support. Now one last bug, VBETEST.EXE using DPMI and will crash so you have to run it with dos32a or stub it with dos32a to get it to run. Now after that rabbit hole the video modes can all be tested including banked and Linear modes to see the issues"




The incredible Machines
and anything that I referenced as a black screen. I can't remember now what video mode it was trying to use. Depending on the resolution, my TV says 109Hz or 99Hz if I select LFB.

Verified fixed with Trident: the incredible machines1 and 2, troddlers, Oregon trail deluxe, chess master 3000 and a few others working by switching from the ET4000 bios to the Trident one. I put it here in the event that will help you with any of your work, testing, or progress: https://github.com/flynnsbit/scripts/bl ... nt.rom.bin

I could never get a single LFB mode above 240 to work in DOS no matter what version of UniVBE I used.

Lfb modes: Hopefully this gives you some breadcrumbs or shortcuts from some of the troubleshooting I've done in the past.

That’s some great info, thank you! I do suspect that there may be a BIOS change needed to get video working under NT so I’ll try with that one soon.

Maybe try the Trident Video bios and see if it makes a difference. There are maybe 10 or so bios files that actually function with the core from the VGA bios back from PCem rom pack here: https://github.com/86Box/roms/releases/tag/20220216 Not sure if they are "100% compatible" but they do function. Any/most of the 32KB ones will at least boot if set to boot1.rom

Don't know if that will help you or not.

Very helpful, thank you. I'm also experimenting with compiling my own BIOS with VBE support. It's larger than 32KB, but the system does see the whole ROM in memory when I apply it so it might potentially work eventually (not that it's currently working in VBE modes, but I'll keep at it).

I think I've run out of steam looking into the video issue. Compiling my own BIOS isn't going to go anywhere as the source code I have depends on very specific VBE emulation (which we don't have here). I've learned enough looking into this to understand that I don't know enough to make progress.

The driver I've being trying to make work (https://bearwindows.zcm.com.au/vbemp.htm#2) requires at least VBE 2 to be in any way usable, and the ET4000 BIOS only provides 1.2. I thought I was making some progress using the ET6000 BIOS because that provides VBE2. Many of the screen modes I tried worked fine under DOS using that BIOS, but nothing was working any better under NT. The official ET4000 driver provided with NT just freezes the system. So none of these are working.

The very best I've got so far using the above driver is a 640*480 256-color image which is cut-off after the first few lines. For that to work I have to use a registry override to point it at the 0xa0000 video buffer. If I were to guess, I'm seeing as many pixels fill the 64KB bank, and then the driver isn't properly handling the bank switching so the rest of the frame is blank.

The last thing I've been trying is looking into how/why Sorgelig's MisterFB.DRV works under Windows 95/98. Looking in a Hex editor it seems to be based on framebuf.drv from Windows 98. I was hoping I would discover it to be a simple Hex edit of the Microsoft original, but that doesn't seem to be the case (at least comparing it with any version of framebuf.drv I've been able to find). Does anyone know where this file came from? Is it compiled from source or a hex edit?

Is there a known address where the linear framebuffer resides? That's the information I was hoping to discover from MisterFB.DRV. I was hoping I could just try pointing the half-working driver at that address instead of 0xa0000 to see what happens.

Sorgelig wrote:

@Sorgelig
Do you have the source for the Windows98 framebuf.drv you built for AO486 available for @alexoughton to use? See above ^

The ET4000 driver probably crashes because there's no fully emulated ET4000 present to service it. That's also why I feel as if this approach is a bit backwards: fitting the software to an imperfect environment rather than improving the compatibility of the core itself. That's not to say I don't respect the effort or the achievement. I honestly wish I even knew where to start to code something like an Adaptec 1502 for this core..

I've attached the source of MiSTerFB driver.
I hope it's the latest version because i won't find where was my working environment.. Since SDK was too old to run on modern Windows, so i've compiled it in Win98 running on ao486.

I've implemented linear frame buffer in core but i didn't fully test it as i didn't make driver for that. Linear frame buffer should be at the end of available memory and it should be ~8MB..

Thank you!! That was all the info I needed to get full-color and resolution graphics working!

The driver I previously linked to works perfectly if you point it at the linear frame buffer address of 0xf800000. This also solves the weird vga.dll crashes I've been having (since I don't believe we're even using that DLL anymore).

Full instructions for installing the driver:

1. Download "STABLE Release version 2007.10.19" from https://bearwindows.zcm.com.au/vbemp.htm#2
2. Set up the following registry values before installing the driver:
3. Then go ahead and install the driver you downloaded above (specifically in the VBE20\NT4 directory).

4. On first reboot it will be running a strange resolution. You can go into display settings and set something more appropriate and a good color depth, and it will now work as expected.


One thing I did find a little strange is that I tried to take a screenshot and it was corrupted. Everything looks perfect on my HDMI output though.

A cellphone photo will have to do. Here's the most important picture to take of NT 4...

Congratulations!!

Thank you!

I just tried the Sound Blaster driver. Seems to work just fine! The only thing I'm aware of with lingering issues is floppy drive support. I'll try to do some digging in that area but it will probably involve too much low-level hardware debugging to be in my skillset. It's a shame too because floppy support would make the install process so much easier (no need for custom ISOs).

With NT working, does Windows 2000 maybe also work now using some of these same tricks?

Unfortunately not. I got it installed and running all the way to first graphical boot. It then becomes very unstable, with critical system processes quickly crashing. It's my understanding that it may require an FPU, and this is why it's crashing.

Yeah, I suspect so. Seems to be the same with Windows 98, although 98 just did not require it as much so you could get around it, and from what I can tell, Windows 98 even seems somewhat stable now. But Windows 2000 came later and I think it relies more on FPU as it was more common by then.