I've also found this which I've started to read:
https://www.filibeto.org/products-n-so ... /docs/pdf/806-1379-10.pdf

That (IMO) seems to do a slightly better job of explaining things.

I have to say though that how you get from this (qemu monitor output)


To this (OpenFirmware device properties)


is still largely a complete mystery. I'm hoping that somewhere in this FCode documentation there will be some clue about decoding those entries.

@balaton

Thanks! - that does look helpful.

Last night I opted for sleep. I did find some bits and pieces earlier related to Mac, but along a very similar fashion to this script so I have a couple of examples now. I will update if there is anything to report from my end.

@balaton

I spent the last hour or so looking at all that. Like you say, it's pretty cryptic. I think I at least have a clue where to start, and I've at least worked out how to create the file and so on.

Don't hold your breath as I feel way out of my depth here but I'm at least trying.

I'm going to head to the sofa and my macbook air to at least see about decoding the "assigned-addresses" from the properties. I realise that on a booting emulated machine they won't be 64 bit but the noise of an uninitialised GFX card is getting to me. If I get to the point of having some forth that can decode the properties then I can come back to this machine.

Incidentally the only way I could stop QEMU from choking on the card BIOS was to remove it from the vfio-pci device with romfile="". I have no idea if OS4 needs it later or not.

My relavent lines look like this at the moment:

That gets me to a working firmware prompt and indeed a working OS4 if I want.

@balaton

I do, yeah (RX580). I'll have another proper look at it all tonight after work referncing what you've just posted.

@balaton

Quote:

This is back to hardware again. I'd be happy to investigate this, I even know someone who is somewhat of a Forth protagonist (right word?) and would probably be keen to point me in the right direction but my hands are tied without hardware.

I guess I could look blind, but I'd prefer to spend my time looking into something that I have the hardware for. Definitely appreciate how it must feel right now though :)

@balaton

Looking at the datasheet of the ICH10R IO controller chip it's connected to, it looks like it's a fully independent PCI Bus. But it's a moot point anyway if it can't be passed through.

I didn't know the difference between VT-d and VT-x.

So back in it's box and back on the shelf it goes then. I won't get rid of it as it's the last system I have here with Floppy and IDE (aside from classic Amiga) but it's not going to be any help here.

I'm not convinced that a system supporting VT-d and providing a dedicated PCI bus will exist (outside of proprietery hardware where you may as well buy a NG system). The two technologies are just at different points in time.

@geennaam

Playing around is pretty much all I do - once things work they get boring

Are we sure that the bridge isn't causing the issues with the 9250 under OS4? I suppose we are given that it works fine in Morphos. Although I would say that jumping to a bit of a conclusion.

Not that a 9250 is really an ideal end-game card anyway. Would be better if the more modern cards worked and you have them covered anyway.

@geennaam

Looking at it, the 9250 is about as modern as I can get without getting into PCIe territory which given the consistency of eBay listings is a minefield (saying they're PCI when they're clearly AGP or PCIe and so on).

I'll keep an eye out for one and see if I can get one for next to nothing. Right now it looks like about £40 which is fine if I knew the rest of the system was OK with passthrough and all the other variables. I'll keep looking.

It would be good if I could repeat your tests but without the bridge to rule it out.

[EDIT] Doesn't look like buying one is likely to be an option since the world appars to have latched onto them being Amiga Mediator compatible. Typical. I will keep an eye out though.

@geennaam

no. 2: 9250, is that going through a bridge?

I've been looking at my older system that's here and am very tempted to start to build it up on the workbench later. It is from around 2009, uses an i7 950 CPU (so not 4700 or 6700 as I had previously guessed) but it still supports Intel VT-x. The BIOS has a single setting. Virtualisation enabled or not. But importantly it has a PCI slot, IDE and Floopy so I think it's a board I should keep anyway (nearly sold it a few times over the years).

If I'm going to test PCI passthrough without a bridge then I need to buy a GPU, like a £20 one from eBay or something, but frankly I'm getting completely lost as to all the options. Hans blog states Radeon HD does not work on Pegasos2 but not only is that page from around 2012 but that also doesn't take into account Qemu and I think we've proved that OS4 see's it OK haven't we?

I'd be happy with a decent WB resolution (1080?) and 2D acceleration for the moment, I'm not desperately concerned about 3D, I have my modern systems for that.

I don't know if me doing that would gain any insight for anyone whether it works or not for me?

I don't think this info is really going to add anything new, and because I am passing through an RX580, I think the issues are already known, but just I don't recall seeing anyone mention an "Unhandled 32 bit data prefix" error from Qemu yet.

I also don't think I can expect anything from OS4 anyway with this card without spending 80Gbp (sorry, wrong keyboard layout!) on the full enhancer software to get the Radeon RX drivers.

Anyway, that log, just in case it helps.

Let me know if there is anything more useful I could do. Otherwise I'll continue reading up on network drivers.

OK, I'm out of time for now, need to go. I need to read through the thread again tonight. My suspicion is that this isn't going to work as QEMU spits out a crash on startup citing something about 32BIT addresses and an unimplemented OPCode.

In OS4 (I have sm501 device as well at the moment) it does show the PCI card as "Working" in HWInfo but I'm guessing that means little more than it knows that it's there.

This is my RX580 incidentally with 8GB of memory on board. Just giving it a try because it was there sat in a box doing nothing.

@geennaam

Ah, ok - same result though, only the GPU and it's embedded Audio are in group 30

@geennaam

if the group is the '0d' in '0d:00.0' then no. Apart from '0d:00.1' which is the HD Audio part of the GPU (also bound to virtio).

I specifically purchased this motherboard at the time based on it's ability to isolate and from memory I believe it puts every device in it's own group. This was pre-pandemic though so my memory of it isn't great.

@geennaam

No problem - I'll try the script but my Ubuntu is 22.04 and I don't desperately fancy reinstalling. Not really an issue though, I've done all of this several years ago on this machine, I just need to dig out my notes and remember how.

Incidentally, on this machine at least, it's possible in the BIOS to have virtualisation enabled (whatever AMD refer to VT-d as) but not have IOMMU isolation enabled. I'll figure it. But maybe not today as I need to go out soon.

[EDIT] Never mind - that was easy, the script worked and my secondary GPU is isolated to vfio-pci

I'm losing track of all the scenarios a bit here but as I mentioned before I have a motherboard with PCI and VT-d but I'm very doubtful that the BIOS will support the required isolation for passthrough as it definitely predates that really being something that end users were into. I may be wrong on the 6700K processor too, I think that machine in on my electronics workbench which means the one in the box is likely 4700K. Still has VT-d though, but even earlier.

On the plus side, I just realised that if I'm to put the spare GPU I have in my main gaming machine then I don't need the missing power cable as it already has one. I'm posting now with both GPUs installed and I know this motherboard has excellent isolation properties if I can remember how to enable it in the BIOS.

So when I get some time I can do some experiments and potentially add some logs etc., but I'll only do so if it looks like it adds value otherwise too much noise is not a good thing!

[EDIT] Forgot to mention that I also have a PCI to PCIe bridge to allow a PCIe card to be plugged into PCI. I was about to send it back as it's the wrong way round (I wanted to plug PCI into PCIe). I'm not sure I really see the point in me keeping it because IF the motherboard with PCI will allow isolation (doubtfull) then that board has PCIe. The only reason to put it into a bridge would be to test the bridge. Like I say, too many scenarios here, I'm losing track!

Lol! Yep. Anything associated with Amiga costs a premium.

One question - isn't Voodoo3 AGP? - never mind, there are PCI versions.

[edit] So what features are available aside from 32bit 1080p? Any acceleration?

@balaton

Understood - it's not a big problem for what I'm doing with OS4 at the moment. I can experiment to see if any other kinds of card work better for me. I also like the idea of looking at a driver.

Unfortunately that's not 100% the case because I'm already doing that.

Just so you know how I'm running on my Mac Mini (M1)

@balaton

You read my mind somewhat on the networking! So when I say it's problematic, the RTL device will just stop working randomly. I noted from your web page that after the OS updates you may need to downgrade back to the version on the CD but it still does it. I used bridged mode by running qemu as root. I've built Qemu from source so I'm running on master, though I'm guessing that doesn't make a huge difference.

My NAS at home has a bridged interface which was doing a good job of upsetting OS4 (Roadshow?) because the MAC address would randomnly flick between the two ports on the interface. I changed that bridge to just be in standby / failover mode instead and I haven't checked to see if that has helped. I could certainly see it being problematic if it decided to switch half way through a big transfer. If that's even something that it would do.

I did also spend time reading through various documentation and notes, some of which may well have been from you (what you said sounds familiar) and I have to say, it didn't all fly miles over my head and I did end up thinking I could possibly take a look. That was probably a week or two back? I haven't ruled out looking into it. From memory I think I forumlated a plan in my head to see if I could implement some sort of stub driver that at least spits out some logging and if I get that far go from there.

Apologies if this is de-railing the main topic here!

@geennaam
Sure, I'm aware. I already have OS4 fully up and running in Qemu using SM501 on a spare disk in my gaming PC and on my day-to-day Mac Mini. It runs very well mostly, better than under WinUAE or FS-UAE, but of course it runs like a snail, or not at all as soon as you hit it with something that needs real GPU capabilities.

I am indeed debating whether to follow along and experiment with GPU passthrough. When I first got my current gaming PC a few years back that's how I had it, a very slim linux boot and then either Windows or Linux via virtIO. The motherboard is extremely good at segmenting all the bits and allowing passthrough. Right now it's not setup that way.