Sonnet 7200 PCI programming

[Wrangler]

Originally Posted by spudje

Or would he have forgotten GND?...

Let's hope Chucky sees the joke!

[Magic]

Although I realize that a stable Sonnet system can be a moving target, I finally had my first night of success. Starting with Initppc many things work fine. I copied my Sys partition over to the Sonnet's memory (Ram disk) and ran some programs from it. I tried Cyberpi 100000 and it took 1 min 46 sec, right on time. I had good luck with Cybermand, Bogomips, Getinfo, PCIInfo, Voxelspace, Warptest, and Landscape. After trying for months it was fun to see it finally working.

I need to add a fan as the Voodoo gets HOT right next to the Sonnet. I am also working on building up a fresh partition.

[Hedeon]

Default RAM disk goes to the memory with the highest priority which is still cpu slot accelerator memory (if installed) and not sonnet. Most fast memory has priorities like 30 or 40. Sonnet has 1.

[Wrangler]

Well done Magic! It's worth the hassle to get it working!

[spudje]

My Sonnet is up for sale at Amibay, if anyone's interested.

[Wrangler]

Unlike everyone else it seems, I'm getting a storage access exception trying to run Cowcat's Q3 conversion on matter what I do. Snoopdos shows this happening as soon as powerpc.library is opened. Now I can turn on debugging of powerpc.library by setting the env debug to 3, but how do I capture the debug output? I can't find where it ends up. Thanks

[Hedeon]

Serial output. Or you can intercept it with sashimi.

[Wrangler]

Thanks Hedeon.

As far as I can tell, the storage address exceptions I am getting in Q3 occur when the data cache is being invalidated right at the start. The final call before the exception is logged as:

Process: quake3_PPC Function: SetCache r4,r5,r6,r7 = 0000000B,5202B240,02000000,89E6A989

Following this through, I believe this means the Q3 code attempts to invalidate the cache across 32MB of space via the routine .DCACHEINV. But this causes an exception as it works through the 32Mb when the virtual address it is trying to invalidate ($52C02000 in this case) does not map to a physical address. Is that possible? (My PPC assembly is -basic- to say the least.)

But even if it is, I'm not sure whether this means there is a hardware fault, a problem with the Q3 code or with the sonnetlib! Can you help?

See attached exception report.

[Hedeon]

You are right in your assumptions. However, I don't know why it errors. A DSI happens when the memory is not mapped, but seeing the other sonnets from people working and how yours is giving also other problems, I'm not sure if I can help.

Originally Posted by Wrangler

Thanks Hedeon.

As far as I can tell, the storage address exceptions I am getting in Q3 occur when the data cache is being invalidated right at the start. The final call before the exception is logged as:

Process: quake3_PPC Function: SetCache r4,r5,r6,r7 = 0000000B,5202B240,02000000,89E6A989

Following this through, I believe this means the Q3 code attempts to invalidate the cache across 32MB of space via the routine .DCACHEINV. But this causes an exception as it works through the 32Mb when the virtual address it is trying to invalidate ($52C02000 in this case) does not map to a physical address. Is that possible? (My PPC assembly is -basic- to say the least.)

But even if it is, I'm not sure whether this means there is a hardware fault, a problem with the Q3 code or with the sonnetlib! Can you help?

See attached exception report.

[Wrangler]

Thanks. The address that fails ($52c02000) is squarely in the memspace of the Sonnet so I am starting to think this is a hardware fault. Do you know of a memory checker that would work on the Sonnet memory? I'd like check this out more thoroughly...

[Hedeon]

Originally Posted by Wrangler

Thanks. The address that fails ($52c02000) is squarely in the memspace of the Sonnet so I am starting to think this is a hardware fault. Do you know of a memory checker that would work on the Sonnet memory? I'd like check this out more thoroughly...

Seeing that it works if the caches are off, I rather suspect cache memory to be flakey instead of the sonnet memory.

However, there is a program called PPC-MemTest from Christian Kummerow. I can't find it on Aminet, however. Maybe someone else has a link?

[jack-3d]

Originally Posted by Joe

I will contact Jack3D about visit with this issue.

Joe´s Sonnet 500/1 arrived weeks ago. Today I finally had time to insert it to my Amiga 4000 (CSMK1 040, 128MB, Voodoo4). Card was recognized in piiinfo, but initialization was not successful (problem with memory error). I removed the provided RAM module (64 MB) and exchanged with another one which I know it works with Sonnets and guess what? Yes it works now. Tested some benchmarks (Syspeed and GeardsPPC) and you have nice 500/1 Sonnet

I am just secretly thinking to send you back mine 400/512 by "mistake" ;oP

[grelbfarlk]

Originally Posted by jack-3d

Joe´s Sonnet 500/1 arrived weeks ago. Today I finally had time to insert it to my Amiga 4000 (CSMK1 040, 128MB, Voodoo4). Card was recognized in piiinfo, but initialization was not successful (problem with memory error). I removed the provided RAM module (64 MB) and exchanged with another one which I know it works with Sonnets and guess what? Yes it works now. Tested some benchmarks (Syspeed and GeardsPPC) and you have nice 500/1 Sonnet

I am just secretly thinking to send you back mine 400/512 by "mistake" ;oP

Good news. I wonder what the problem with the DIMM is. From the specs it should work. GM71C17400CJ-6 were the chips he said were on it which by the datasheet appear to be 5V 60ns 2k refresh.
Also this reminds me that PPC not responding will be triggered by invalid RAM setup.

[nexus]

PPCMEMTEST

http://nappe.acc.umu.se/mirror/archi...PC-Memtest.lha

http://members.upc.cz/ghh12/files/p5.lha

[grelbfarlk]

Originally Posted by nexus

PPCMEMTEST

http://nappe.acc.umu.se/mirror/archi...PC-Memtest.lha

http://members.upc.cz/ghh12/files/p5.lha

The first one is the one I remember. Just make sure you try to run this with a minimal startup sequence and not fully booted into workbench. And WTF did this disappear from main aminet?

[jack-3d]

Originally Posted by grelbfarlk

Good news. I wonder what the problem with the DIMM is. From the specs it should work. GM71C17400CJ-6 were the chips he said were on it which by the datasheet appear to be 5V 60ns 2k refresh.
Also this reminds me that PPC not responding will be triggered by invalid RAM setup.

Joe´s RAM (not working): Hyundai GM71C17400CJ6
Mine RAM (working): LGS GM71C16400CJ6

So I dont know, there is just one number difference, but could be also malfunction module, not sure.

[jack-3d]

I am thinking to upgrade my 400/512 to 400/1 and/or 500/1. I opened my passive cooler and I can see there is PPC750L-DB0G366 which is 366 MHz 750L overclocked to 400.

I know grelbfarlk already successfully upgraded 512 kB to 1 MB L2, can you please remember me where to get the chip(s) and if anythign else is needed.

Does anybody know what PPC chip is used on 500/1 sonnets? What about this one will that be working? https://www.ebay.com/itm/IBM-PPC750L...-/151374574697

[grelbfarlk]

Originally Posted by jack-3d

I am thinking to upgrade my 400/512 to 400/1 and/or 500/1. I opened my passive cooler and I can see there is PPC750L-DB0G366 which is 366 MHz 750L overclocked to 400.

I know grelbfarlk already successfully upgraded 512 kB to 1 MB L2, can you please remember me where to get the chip(s) and if anythign else is needed.

Does anybody know what PPC chip is used on 500/1 sonnets? What about this one will that be working? https://www.ebay.com/itm/IBM-PPC750L...-/151374574697

CY7C1347G-250AXC-ND were the 250MHz chips I used. You should be able to get them from Digikey or Mouser or any part supplier. And if you get it up to 500MHz you can use the 2:1 divider. I was hoping to use 1.5 divider with these at 400MHz for 266MHz but that extra 16MHz was too apparently too much.

I have to look up that CPU, as I recall you want a 2.2V version.

[nexus]

if you are gonna do all that might as well put a 7400/7410 G4 on it

[jack-3d]

Not sure if G4@500 is supported and so much boost comparing to G3@500 here.