Software failures 68040 CPU on Amiga 4000 EC030

[meynaf]

But have you tried to load data from the hd to see if it comes up corrupt ?
I suspect it might just be the code cache not being properly cleared after loading programs.

[astremler]

Originally Posted by meynaf

But have you tried to load data from the hd to see if it comes up corrupt ?
I suspect it might just be the code cache not being properly cleared after loading programs.

Ok, also good one.. I will try this too later..

If this would be the case, what would be the solution to properly clear the code caches after loading programs?

[meynaf]

Originally Posted by astremler

If this would be the case, what would be the solution to properly clear the code caches after loading programs?

I don't know. Normally the OS should be doing that by itself. Perhaps this is a problem in 3.1.4 and so the easy solution is to return to regular 3.1 (you wrote in first post that it worked with 3.1).
But before getting a remedy, let's first see if the diagnostic is correct

[astremler]

Originally Posted by meynaf

I don't know. Normally the OS should be doing that by itself. Perhaps this is a problem in 3.1.4 and so the easy solution is to return to regular 3.1 (you wrote in first post that it worked with 3.1).
But before getting a remedy, let's first see if the diagnostic is correct

Right... this is what I did:

- I booted in 3.1.4 Workbench with caches disabled
- Opened the shell and went to DF0: where I have some programs on disk like Ed
- First I created a file on disk by: ed test
- I enabled the cache by 'cpu cache' while on DF0:
- I edited 'test' on disk again without problem
- Then open a file on hard disk from floppy: ed sys:s/Startup-Sequence

It gives me a 'checksum error on hard disk block ...'

Switching back to Kickstart 3.1 creates new problems as I described earlier. I cannot even install Workbench 3.1 so I don't see that as an option. I will check the Amiga PSU tomorrow..

[AmigaHope]

OK so to clarify -- system works fine with full data caches enabled, including copyback, as long as instruction cache is disabled? That's really weird.

Also to clarify, if MMU libraries are installed, and activated after boot with caches disabled, everything works fine? (other than some weird memory leak)

What happens if you boot using 3.1 hardware ROMs, but then include 3.1.4 maprom in your startup sequence first?

What is the version of your 3.1.4 hardware ROMs?

[astremler]

Originally Posted by AmigaHope

OK so to clarify -- system works fine with full data caches enabled, including copyback, as long as instruction cache is disabled? That's really weird.

Also to clarify, if MMU libraries are installed, and activated after boot with caches disabled, everything works fine? (other than some weird memory leak)

What happens if you boot using 3.1 hardware ROMs, but then include 3.1.4 maprom in your startup sequence first?

What is the version of your 3.1.4 hardware ROMs?

Yes, as long as caches are disabled in the early boot then everything works fine. With or without MMU libraries. Also i have tried with HD Toolbox 'Directory cache' and 'Direct scsi DMA' enabled/disabled, re-formatted and re-installed xxx times. Same result.

Once I put back the KS 3.1 roms the hard disk partitions becomes unreadable. With HDInstTools I can re-create them, but they never appear in the Workbench so I can not format them and get stuck there. Obviously I am using small partitions < 2 GB.

Also I tried to exchange the PSU but unfortunately the ATX power I have laying around here seems not good. I can not get into the boot screen. Normally boot gives a messed up screen and hangs. Even exchanging back to 68030 CPU same problem so this is not about the 68040. I will need to get another ATX power supply to be able to test.

I made screen captures of all the versions I am using which are displayed in SysInfo.

[astremler]

Originally Posted by Thomas Richter

The A4000 IDE is a rather harmless device, it doesn't do any DMA, just PIO, and there is nothing on the bus that would easily explain such an issue. For the 68040, the IDE registers look like any other chip registers, and the 68040.library should keep care about it not being cachable. However, is this a physical 3.1.4 ROM or the software/LoadModule solution? In the latter case, please ensure that the boot partition (the one the modules are on and LoadModule is on) must be below the 4GB barrier, or all hell may break loose. Also, if you run the MuTools based solution, an output of "MuScan" *after* SetPatch might be worth to look at. Probably post it here.

Thanks.. It is a physical 3.1.4 ROM. Not long ago I had some issues with my Amiga 4000 giving black screen when any hard disk was connected. No matter which hard disk I connected (I have several here with difference sizes). This was with the 68030 CPU inside. When removing the hard disk then I could boot into the Workbench by disk. But when connecting any hard disk it was just black screen at boot and bye bye.

First I thought the was a problem with the IDE port. I removed the IC's from the sockets of U-901 and R-902 (see attachments) and clean inside carefully. After cleaning and placing these IC's back in the sockets the whole problem disappeared like snow in front of the sun... I could boot again from hard disk and everything back to normal.. Until I replaced the 68030 with 68040..

I suspect here might be the problem somewhere although I am not sure what it can be. Some re-programming needed for these IC's because of 68040? I am not familiar with this. The best solution would be buying a FastATA card and completely by-passing the IDE on the mainboard by moving the IDE devices away, but still it's not 100% guaranteed.

I have attached a MuScan as well.

[hooverphonique]

Normally U-901/2 are not in sockets. Did the previous owner install a PIO2 mod for example? Could be some waitstate/latency issue when using 040 with cache on (it's much faster).
Could you try the original CBM PAL chips?

[astremler]

Originally Posted by hooverphonique

Normally U-901/2 are not in sockets. Did the previous owner install a PIO2 mod for example? Could be some waitstate/latency issue when using 040 with cache on (it's much faster).
Could you try the original CBM PAL chips?

I think so... I got it like this when I bought this Amiga from the previous owner. I don't have the original CBM PAL chips.

[hooverphonique]

Originally Posted by astremler

I think so... I got it like this when I bought this Amiga from the previous owner. I don't have the original CBM PAL chips.

Ok.. If you have the means to get a set of GALs programmed, the OEM jedecs are here: https://www.amigawiki.org/doku.php?i...s:pld_download

[alexh]

I'm sure that the PIO2 mod works fine with A3640@25MHz and A3660@50MHz.

Perhaps your A3640 is overclocked? Even then I think it works fine.

I think you should talk to Cosmos and maybe he can help you debug your PIO2 and/or supply some PIO0 GAL's.

http://leblogdecosmos.blogspot.com/2...00d-pio-2.html

[astremler]

Originally Posted by hooverphonique

Ok.. If you have the means to get a set of GALs programmed, the OEM jedecs are here: https://www.amigawiki.org/doku.php?i...s:pld_download

Thanks a lot... I have downloaded the original rom files for U901 and U902. I will burn them on new IC's and see if I can get the original state back first.

[astremler]

Originally Posted by alexh

I'm sure that the PIO2 mod works fine with A3640@25MHz and A3660@50MHz.

Perhaps your A3640 is overclocked? Even then I think it works fine.

I think you should talk to Cosmos and maybe he can help you debug your PIO2 and/or supply some PIO0 GAL's.

http://leblogdecosmos.blogspot.com/2...00d-pio-2.html

Thanks... But this guy is in France and I am based near Shenzhen in China. I can easily source the IC's here at the electronic market if I know the part numbers of the ICs, that's not a problem. Also for burning the roms is not a problem, because I know some Chinese on the electronics market here with the equipment who can do this. I just need to provide them the empty IC's with the files and tell them what to do.

I already downloaded the original rom files (Thanks for the link hooverphonique). First I will try to put it back in the original state and test again. I will also get an extra pair of IC's where I will burn the latest PIO2 files, also test these one. Maybe the IC's currently inside are faulty? At least I can just easily remove them without soldering. Let's see..

Regarding overclocking 3640, I am not sure. How can I see that? It doesn't look like it's over-clocked in SysInfo.

[hooverphonique]

Originally Posted by alexh

I'm sure that the PIO2 mod works fine with A3640@25MHz and A3660@50MHz.

Perhaps your A3640 is overclocked? Even then I think it works fine.

I have first-hand experience with PIO2 mods failing when overclocking the A3640 (which overclocks the whole system). The mod I've tried seems to work @ 25MHz though, but we don't know the fusing of the OP's U901/2, so I think it's a good idea to go back to standard to see if there's an issue there.

[hooverphonique]

Originally Posted by astremler

I already downloaded the original rom files (Thanks for the link hooverphonique).

No problem

Originally Posted by astremler

Regarding overclocking 3640, I am not sure. How can I see that? It doesn't look like it's over-clocked in SysInfo.

If it has a 50MHz oscillator, it isn't overclocked. Also, the delay line (for CLK90) should be connected at its 10ns tap @ 50MHz (just in case it was tampered with, but that should be easily visible). I'm not sure if CLK90 affects the IDE interface though.

[astremler]

Today I went to the electronic market in Shenzhen and managed to get those IC's. A kind Chinese helped me to burn the original Commodore U901/902 GAL files on those 2 IC's for free. Back home I replaced them and bingooo...

Everything works like a charm now! I am able to boot with cache enabled using 68040. I also got a new ATX PSU and checked, but there are no differences. I can still use the original Amiga PSU.

I have re-partitioned the harddisk, formatted and re-installed Workbench 3.1.4. After placing the 68040.library in the libs: folder I can boot from hard disk with cache enabled instantly. MuTools was not required to install (but I will install it anyway).

Conclusion: the problem were the PIO2 IC's U901/902 which the previous owner installed on the mainboard. These are working fine with the original 68030 CPU, but becomes problematic with caches when using a 68040 CPU. (Compatibility issues) The solution is to put the original Commodore U901/902 back and the problems are gone.

[gulliver]

I am glad you kept pushing for a solution until you solved it.

Well done!

[astremler]

Originally Posted by gulliver

I am glad you kept pushing for a solution until you solved it.

Well done!

Thanks

[AmigaHope]

Great! So glad your problem has been worked out. =) You should buy some beer for the person who helped you out by burning the GALs for ya. xD Shenzen can be a magical place (though I hear it's getting harder for foreigners living in China).

btw later if you do go the FastATA route be aware that it hits your CPU *hard* due to its PIO nature. If you get a DMA SCSI controller and find one of those rare SCSI-to-SD or SCSI-to-SATA adapters, it runs way smoother, especially if you ever get a CPU card with a SCSI controller.

Best solution is the (now impossible-to-find) Deneb running in DMA mode, let you use any USB 2.0 mass storage device in DMA mode. I wish someone would recreate the Deneb.