Sonnet 7200 PCI programming

[trixster]

Indeed

[trixster]

Inhibiting the dma space? Dunno tbh

[Timtheloon]

Originally Posted by trixster

Indeed

it took me forever to work out the second number in the setcachemode lines, for ages I was looking at the second number in the line on showconfig going how is that different from that one and then I realised it wanted the memory size not the end address of the memory address. don't know why I kept ignoring the word "size" before the second number

This is the webpage I got that highlighted line from its all the way at the bottom of the page

https://github.com/Sakura-IT/SonnetA...ki/Good-Config

[trixster]

I forgot we were talking about bytes and not kilobytes, and couldnt work in base 10

Yeah, it’s a range

[trixster]

I think that’s just an example config from one of grelbfark’s machines, so it’s probably not applicable to anyone else’s even if what he’s doing there is/was relevant.

Looks to me like he’s inhibiting part of the voodoo’s memory

[Timtheloon]

Originally Posted by trixster

I think that’s just an example config from one of grelbfark’s machines, so it’s probably not applicable to anyone else’s even if what he’s doing there is/was relevant.

Looks to me like he’s inhibiting part of the voodoo’s memory

I can see what it is now he has this line in his Showconfig

Node type $A, Attributes $2005 (FAST), at $64D00000-$64EFFFFF (2.0 meg)

[trixster]

That’s the dma space then I think.

[grelbfarlk]

Honestly I'm not really sure what I was doing there. The point was that once upon a year it was recommended to set the PCI space as CacheInhibit. You do that basically by taking your PPC Memory range from prmscan or pciinfo and setting it in MMU-Configuration.
Like in that config:
Pci Device 7:
VendorID: $1057
VendorName: Motorola
DeviceID: $4
DeviceName: MPC107 PCI Bridge/Memory Controller for PPC
ClassCode: $e0001
RevisionID: $12
InterruptPin: 4
SubsystemVendorID: $0
SubsystemID: $0
MemSpace0: $70000000 - $7bffffff
.......................^^^^........^^^^^ (size 196083712 this was 192MB on a Sonnet)
SetCacheMode from 0x70000000 size 0x0BB00000 CacheInhibit

Which is all anti-disrelevant, as you can see in Muscan, that it was already Cache Inhibited:
0x60000000 - 0x7FFFFFFF CacheInhibit I/O space
I am assuming later versions of mmu.library or maybe with Hedeon's libraries already set it to CacheInhibit, as I have a couple systems with just a blank or default MMU-Configuration.

So the point is run MuScan and see what it says.

[Hedeon]

Originally Posted by Timtheloon

Oh yes now I understand so would that line be

SetCacheMode from 0x68410000 size 0x16AF0000 CacheInhibit NonSerial Imprecise ;(just under 363mb of usable ppc memory)

This is incorrect. the 68K also needs to access stuff outside of the PPC memory that is in the memory list (the 2005 entry) for communications.

So just do either the whole Mediator/Prometheus range (so either 0x40000000 or 0x60000000 with size 0x20000000) or look at prmscan/pciinfo and and take the range of the PPC card. (in your case 0x68000000 size 0x18000000).

The first is prefered as adding any PCI card will mess up the address ranges and thus your MMU config. I can only imagine using the second option when you have a PCI card in there that needs copyback memory for speed (maybe a radeon used for memory, however Prometheus does not support that anyway). Not sure if it is usefull for Spider. You still need to invalidate the PCI DMA buffer range for the 68K every time the 68K needs to read it as PCI DMA in itself bypasses the 68K and therefore its caches. Or just make the range cache inhibited anyway.

EDIT:

I think the mmu library sets Z3 space as cache inhibited by default, as grelbfarlk says.

EDIT2:

Also, the PPC cards have a second PCI address range where config registers are. These need to be cache inhibited too. So option 1 is the only option to go really.

[Timtheloon]

Originally Posted by Hedeon

So just do either the whole Mediator/Prometheus range (so either 0x40000000 or 0x60000000 with size 0x20000000) or look at prmscan/pciinfo and and take the range of the PPC card. (in your case 0x68000000 size 0x18000000).

OK this is good to know, that was Matt3k showconfig I was working from so that would be his config

I will adjust mine to the mediator range and remove that 2mb line

[Hedeon]

Originally Posted by Timtheloon

OK this is good to know, that was Matt3k showconfig I was working from so that would be his config

I will adjust mine to the mediator range and remove that 2mb line

I did some heavy editing in the meantime, but I think you got it

[Timtheloon]

Well that seems to of fixed my Nightlong PPC installation would of normally crashed by now when in game

watch it crash has soon has I post this message

[matt3k]

Ok, my head hurts... (Hope Nighlong ran for you Timtheloon!)

I tried the supplied sample from Tim and it didn't work. Sorry to be sooooo think with this.

What can I do to help this along...

[Timtheloon]

Originally Posted by matt3k

Ok, my head hurts... (Hope Nighlong ran for you Timtheloon!)

I tried the supplied sample from Tim and it didn't work. Sorry to be sooooo think with this.

What can I do to help this along...

did you adjust the Prometheus range that Hedeon mentioned

[matt3k]

Tried it I think, Pciinfo seems to be for the mediator? Since it doesn't work and I can't finds prmscan?

I tried the ranges he specified and it still locks...

[grelbfarlk]

Just for lols, I played around with Musetcachemode on the Mediator.
Settings the two memory ranges for my Voodoo 62000000 64000000 to WRITETHROUGH, system still running. Setting them to Copyback would cause the system to freeze within a second or two.
Then for further lollercoasters, I set Writethrough and Copyback on the Ragnarok memory,
ADDRESS=70000000 SIZE=10000000.
GLQuake still running.
Then tried just setting musetcachemode ADDRESS=60000000 SIZE=20000000 WRITETHROUGH
It's not any faster, actually it is about 5 FPS slower in GLQuake or around 39 FPS.
Just to complete the circle I tried musetcachemode ADDRESS=60000000 SIZE=20000000 CACHEINHIBIT
GLQuake back to 44.5 FPS (or normal).

Next I tried setting Copyback on the IO space at 40000000, system locked up within a couple seconds.
Same for setting Writethrough to the IO space at 40000000, system locked up within a second or so.
So always IO space must be cache inhibited.

Originally Posted by matt3k

Tried it I think, Pciinfo seems to be for the mediator? Since it doesn't work and I can't finds prmscan?

I tried the ranges he specified and it still locks...

Post your muscan or at least run it.

[matt3k]

Here is my MUSCAN:
MuScan 46.1 (02.07.2016) © THOR

68040 MMU detected.
MMU page size is 0x1000 bytes.

Memory map:
0x00000000 - 0x001FFFFF CacheInhibit Imprecise NonSerial
0x00200000 - 0x00BBFFFF Blank
0x00BC0000 - 0x00BFFFFF CacheInhibit I/O space
0x00C00000 - 0x00D7FFFF Blank
0x00D80000 - 0x00DFFFFF CacheInhibit I/O space
0x00E00000 - 0x00EFFFFF Blank
0x00F00000 - 0x00F7FFFF CacheInhibit
0x00F80000 - 0x00FFFFFF ROM
0x01000000 - 0x06FFFFFF Blank
0x07000000 - 0x09FFFFFF CopyBack
0x0A000000 - 0x3FFFFFFF Blank
0x40000000 - 0x4000FFFF CacheInhibit I/O space
0x40010000 - 0x5FFFFFFF Blank
0x60000000 - 0x60FFFFFF CacheInhibit Imprecise NonSerial I/O space
0x61000000 - 0x7FFFFFFF CacheInhibit I/O space
0x80000000 - 0xFFFFFFFF Blank

[grelbfarlk]

Are you using the MMULIB 68040.library?

[matt3k]

Yes, I'm using the mmu 040 lib.

[torsti76]

I'm seeing very similar issues to matt3k, also with Firestorm, i.e. AA3000+, and a 128 MB Apocalypse:

Everything "basic" runs fine, memory is added correctly, all IO space and PPC RAM is set up as "CacheInhibit", Voxelspace is running (although at about half the pace of my 400 MHz CSPPC)

But as soon as I start W3D, serial terminal output shows 4 times the line:

Detected DSI hit outside of memory bounds at 00000004

Then graphics output freezes.

68k CPU seems still active though. Can do CTRL-A-A reset. Also when playing "Sulaco" demo by Encore the sound plays okay (via Paula DMA) but the screen stays grey. When the demo has finished, machine hangs completely.

Already tried rearranging cards in the slots, tested with another Voodoo, nothing helps.

Is the 128MB Harrier support somewhat broken perhaps?