WinUAE 3.6.0 beta series

[ross]

Originally Posted by Toni Wilen

- UAE boot ROM must 128k indirect
- MMU emulation must be enabled
- Use Thor's Mu680x0Libs.lha (from Aminet)

Ok, this is my setup but badly crashes
(WinUAE irresponsive, grey window, need to kill task)

Or use any emulated real world controller

Yes, this works

Best regards,
ross

[ross]

Originally Posted by mark_k

One suggested workaround for (virtual in this case ) hardware designers was to have longword-wide read/write registers not longword-aligned. So register at $xxxx2 instead of $xxxx0 for example.

And all the 32bit speed lost

[ross]

Sorry Toni for the multiple posts.. I re-read your response and thinked about.
When you write

"Use Thor's Mu680x0Libs"

do you mean setup a region at $Exxxxx (where is placed indirect FS ROM) with cache inhibited? (before Setpatch Data Cache activation in startup-sequence).

Thanks,
ross

[Toni Wilen]

Originally Posted by ross

Ok, this is my setup but badly crashes
(WinUAE irresponsive, grey window, need to kill task)

You know the only answer you get to this kind of replies is: "works for me"

Originally Posted by ross

Sorry Toni for the multiple posts.. I re-read your response and thinked about.
When you write

"Use Thor's Mu680x0Libs"

do you mean setup a region at $Exxxxx (where is placed indirect FS ROM) with cache inhibited? (before Setpatch Data Cache activation in startup-sequence).

It does it automatically. (MMU inhibit works with write-allocate. External cache inhibit (CI) pin does not prevent 68030 from caching writes in write-allocate mode. This is the stupid part)

Originally Posted by mark_k

One suggested workaround for (virtual in this case ) hardware designers was to have longword-wide read/write registers not longword-aligned. So register at $xxxx2 instead of $xxxx0 for example.

Why would I make it worse just to support some edge case?

EDIT: I'll probably add special case "really inhibit caching" flag to boot rom..

[ross]

Originally Posted by Toni Wilen

You know the only answer you get to this kind of replies is: "works for me"

Yes, even before writing it

It does it automatically.

Then I need to check my setup, i've lately updated Thor's package, maybe some mess..

(MMU inhibit works with write-allocate. External cache inhibit (CI) pin does not prevent 68030 from caching writes in write-allocate mode. This is the stupid part)

It's a feature!

Thanks for all,
ross

[sigma63]

Hello Toni,

there is a bug with the "Save Identity" Function:
When you select "Settings / CD & Hard drives / Add Hard Drive ... / Read Identity / Save" you get a file-select-dialogbox "Select geometry file".
The path in this dialogbox is preselected with the path where my .hdf- and .geo-files live. I select an existing file, say test.geo, and click on save. All seems well, but it turns out, the path and the selected filename were ignored! The file is saved where the winuaebootlog.txt lives with a filename constructed from drive manufacturer, name and serial and the selected file test.geo is unaltered.

[Toni Wilen]

Now indirect uae boot rom address space (upper 64k which has trap data area) is forced to not cache anything. Now it also works without MMU but data cache enabled.

Originally Posted by sigma63

there is a bug with the "Save Identity" Function

Fixed. I added automatic name generation but then forgot to test that manual still works

[sigma63]

Originally Posted by Toni Wilen

Fixed. I added automatic name generation but then forgot to test that manual still works

Thank you. But is it intentional that you overwrite an existing File? I would have expected that you somehow "append" it.

[Toni Wilen]

Originally Posted by sigma63

Thank you. But is it intentional that you overwrite an existing File? I would have expected that you somehow "append" it.

Yes. Automatic merging/appending gets too painful for this kind of rarely needed feature. It is easier to use text editor.

[sigma63]

Originally Posted by Toni Wilen

Yes. Automatic merging/appending gets too painful.

OK, all right.

[ross]

Originally Posted by Toni Wilen

Now indirect uae boot rom address space (upper 64k which has trap data area) is forced to not cache anything. Now it also works without MMU but data cache enabled.

Thanks, it works on my config without pre-MMU setup
(so I can enable/change/test MMU code in every situation when I boot from FS ROM)

Just for the record, I sorted out my previous problems: a Setpatch/MuLibs conflict (so I know how to replicate WinUAE 'bad' crash).

I was using OS3.9 with a BoingBag Setpatch (v44.38).
MuLibs on install declare that can patch Setpatch v44.xx editions but effectively v44.38 in not recognized/patched and for some reason 68030.library is not updated.
So I temporary copied Setpatch 43.7, updated MuLibs and restarted: a 'hybrid' 3.1 start and use the dummy 680x0.library that call the right needed library, but the system somehow works.
Now I re-copied the v44.38 setpatch and rebooted: the system works in OS3.9 mode and MMU works because v44.38 call directly 68030.library (only with a forced DataCache Burst mode enabled but can be easily disabled)..

Toni, so to replicate crash (tested): use OS3.9, use Setpatch v44.38, temporary rename 68030.library (so unused), reboot, the system crash (justly), click 'reboot' or 'suspend' and WinUAE get totally stuck and you need to remove process from task manager.
Or simply ignore this corner case and live happy

I surely appreciate this new auto-non-cache FS indirect ROM mode choice

Thanks!
ross

[Ian]

Originally Posted by Toni Wilen

- Disable low latency vsync if calibration returns impossibly large values (>1000Hz). Seems to happen with some recent NVidia drivers.

Does it just disable vsync or reverts to the old version?

I get a solid 50 fps, but the GUI doesn't say anything about vysnc being enabled.

[hyperlogos]

OS3.1, ROM3.1, A1200 style. AmiTCP 3.0b2. First boot, uaenet.device works great. Second boot, ping responses are doubled. Third boot, ping responses are tripled. Close window, start from scratch, everything is working fine. Latest winpcap release installed. Win7x64.

[Toni Wilen]

Originally Posted by Ian

Does it just disable vsync or reverts to the old version?

I get a solid 50 fps, but the GUI doesn't say anything about vysnc being enabled.

It disables vsync completely.

Originally Posted by hyperlogos

OS3.1, ROM3.1, A1200 style. AmiTCP 3.0b2. First boot, uaenet.device works great. Second boot, ping responses are doubled. Third boot, ping responses are tripled. Close window, start from scratch, everything is working fine. Latest winpcap release installed. Win7x64.

I don't think this is 3.6 beta series specific (see first post).. but does http://www.winuae.net/files/b/winuae.7z fix it?

[Ian]

Cheers for clearing it up Toni, must report that it seems to be working now with 16273 without an nvidia driver update, so maybe it was a windows problem more than a driver issue.

Going to brave 385.41, but 385.28 seems fine

Edit:

Still fine.

VSync calibrated: 50.001026Hz/1.0=50.001026Hz. MinV=1 MaxV=1079 Adj=3187 Units=78123 89.0%

[ross]

Hi Toni, a new interesting bug for you
I'm debugging some stupid code that make illegal access in 24bit memory.
I see with my eyes the illegal access but WinUAE debugger do not trigger (if you remember something similar reported in an old message but i was not able to replicate it, driving me crazy ).

Well, now i'm able.
Use my attached config (or other with z2 mem).
Open debugger: "w 0 800000", then "m 800000" and a magic 200000 mirroring appear at this location . Now "w 0" and no more mirroring.

"dm" is right:

00000000    2048K/1 =    2048K ID C32 Chip memory
00200000    6144K/1 =    6144K ID F32 Fast memory
00800000      64K/1 =      64K ID F32 Fast memory [D]
00810000    1984K/1 =    1984K ID F32 Fast memory

but every trigger in this zone no more works, redirected to mirror.

Regards,
ross

[Toni Wilen]

Probably fixed now but I don't think this had nothing to do with 3.6...

[ross]

Originally Posted by Toni Wilen

Probably fixed now but I don't think this had nothing to do with 3.6...

No, same behavior

(and sorry, i've not checked in previous version..)

[tested: the first not working version was 3.4]

Thanks,
ross

[Toni Wilen]

http://www.winuae.net/files/b/winuae_3600b3.7z
http://www.winuae.net/files/b/winuae64_3600b3.7z

Beta 3:

- 68030 data cache + indirect UAE Boot ROM will now always work, even without MMU. UAE Boot ROM address space now forces all caching off, even if write-allocate is enabled.
- 68030 hardware cache inhibit signal is now also emulated in non-MMU/EC modes, making data cache emulation usable in this mode.
- Identity save option ignored user selected file name and path.
- Read identity button didn't support ATAPI devices correctly, it only worked if ATAPI device appeared as SCSI or "RAID" device.
- Activate DTR and RTS when uaeserial.device is opened to match serial.device behavior.
- Serial port emulation to PC serial port reset DTR (and RTS state if RTS/CTS was unticked) when baud rate changed.
- Disassembler now also shows data in memory, makes step by step tracing less annoying, for example "CMP.W (A4)+ [2f09],D2". Not shown if address is not plain RAM or ROM.
- Assembler BSR fixed.
- uaenet.device didn't close low level ethernet handles when Amiga was reset, causing duplicate packets.
- 68020 CE statefiles created with 3.5 or older should load normally again.
- UAE controller hardfile custom unit number selection didn't work if unit numbers had gaps.
- Do not re-open main window when GUI on top option is changed.
- uaeserial.device crash, broken when OS4 compatible trap system was added.

[ross]

Mirror effect resolved

Regards,
ross