Guru on SCSI write (RAM help needed)

[ppieczul]

Thanks a lot for the link. I installed the other fixes on my 2630 board - resistor deletes. It did not change the situation dramatically.
The most stable (did not hang so far, installed WB3.9, more testing needed) configuration that I have now is:
- A2630 with 2MB, A2091 in leftmost slot, COLSP 4MB board, real HDD

Adding extra 2MB to 2630 slightly increases the chance of hang - things work most of the time.
Using scsi2sd makes the hang very likely.
Hardframe controller hangs generally regardless of the configuration.

I plan to stick to my config and play a bit with my Amiga.
Does it make sense to add bigram to A2630?

Thanks again
Pawel

 

[thebajaguy]

It's a complication of the situation, because it introduces 32-bit RAM that is not DMA-addressable, making DMA buffering and copy-up by the CPU needed when the target I/O address is in that space. It makes you a candidate for the GuruROM (for reasons explained on the pre-sale thread), but licking the instability issue you have with DMA is still more important.

Can you do me a favor and turn off the Data Cache with the CPU command? Do so right after SetPatch is called. Run with it that way for a little bit and let me know if the DMA problem goes away.

There is a bug in the 68030 CIIN functionality that can get fixed with the MuLibs 68030.library package (AmiNet) and an adjustment (patch) to the Setpatch command, for 68030's with MMUs, but the data cache turned off is a quick solution for testing. If the problem goes away, look into MuLibs package.

 

[ppieczul]

I inserted extra 2MB ram to 2630 which made it practically most of the time impossible to boot until GUI appears.
Then I added cpu nodatacache after setpatch (using workbench 3.1 from fdd, which worked fine). It did not change the situation. When eventually it would show GUI, it would hang afterwards.

What I noticed after all card insert/remove is that the problem may be worse or better (hang less frequently). I suppose it may be that the card has some issues with soldering. When I first got it and installed ram chips, memtest failed, because one of the pins on the ram chip was not connected even though it looked normally soldered on the surface. I had to add a small wire to make it work. The PCB looks like a three or even more layers with ground and power on the external layers, making it impossible to see where the pins are actually wired to on the inner layer. I think I may try to re-solder each of the pins or working with the schematic verify the connections. Both are lots of work. No other ideas. Other than maybe abandoning it and buying a working A4000
Thanks for your help.

 

[thebajaguy]

If you run a memory test tool against it, boot from a floppy with no startup-sequence, run setpatch, turn the data cache off (if on), and run the memory test tool. Leavign the data cache on can obscure some results. I'd let it run several loops. If no errors, it's not soldering.

One test tool you could use with DMA is my RSCP benchmark (run it after the memory check, no startup). It will raw-read from the target SCSI into either FastRAM or ChipRAM, and you have 16K, 128K, or 512K buffer targets.



For future ref: Note that CPU FastROM, if called from the startup, places it's image at the back end of the memory pool. OS 2.0+ merges all memory segments, even 16 and 32-bit, as one. I think there is a HEAD parameter which pushes it to the front (allocated memory range) in some versions (or may have been of the SetCPU days).

 

[ppieczul]

I ran memtest as you advised and it passes every time.
I also ran rscp benchmark and it read data very well. I get 800K-1600K scores. Unfortunately I don't know what I choose because the custom buttons with graphics on them are displayed as garbage (maybe this gives some hint as to my problem?)
I also noticed, that now with extra 2MB installed on A2630, but disabled with a jumper the system works much more stable than before the fixes I did to the board. But when I enable this 2MB, it crashed every time on Workbench boot. Memtest runs fine (from floppy) though.
I think I might limit my memory to 6MB and don't install these 2MB on 2630. As a second ram expansion I use colsp board. With 4MB enabled, it works fine, but when I set it to 6MB, it reports faulty board on boot, if combined with A2630 with 2MB ram.

 

[thebajaguy]

The RSCP tool defaults to MEMF_Public, 512K buffer. Rotates to 128K and 16K. The buffer type flips between Public (Fast normally) and Chip. That it's corrupted some of the image implies junk was loaded, be it from the media, across the SCSI/Amiga buses, or in memory (no way to tell at this point). Speeds are reasonable for your use. I was aiming to see if it hung anything.

Disable the A2630 and run RSCP - just to make sure nothing is corrupted in the copy you have on the drive, and so you see the options. This would rule out the copy of the file, and the data transfer across the SCSI bus.

Was any of the memory testing you did with just the 4MB A2630 memory and no 16-bit RAM?

I may be at the limit of my ability help. Things might be the point of trying different Amiga hardware or alternate cards, and that's not always an option these days.