NetWare 4.x MPS14.PSM not detecting SMP on Vbox

[michaln]

Did you try with MS-DOS 6.22? Or booting with F5?

In the meantime I tried again with NetWare 6.0 and 6.5; both detect SMP just fine, but are horribly unstable.

[michaln]

This isn't really about NW 4.x but...

Recently I tried NW 6.0 again. With the latest not-quite-VirtualBox 5.1 code, it's actually stable with 2 CPUs. Or at least stable enough to boot up, run for a while, and shut down cleanly.

NW 6.5 is detects 2 CPUs as well but is highly unstable. For whatever reason, NW 6.0 uses MPS tables while NW 6.5 uses ACPI; at the moment I don't know if that makes any difference. The reason why 6.5 is super unstable is rather odd. Both NW 6.0 and 6.5 use NMI IPIs (Inter-Processor Interrupts) which is not unheard of, although NetWare seems to use NMIs a lot more than other OSes. Anyway, somehow sometimes(?) NW 6.5 decides that the NMI was not really an IPI but rather a system board parity error, and shuts down. I have so far not been able to find out why NW 6.5 would think so.

Hmm, guess what... with MPS 1.4 instead of ACPI, NW 6.5 is actually pretty stable too with 2 VCPUs. I should poke around some more, maybe this will give me some ideas.

[michaln]

It did give me ideas, problem sorted out. It turned out to be APICDRV.PSM doing very unusual things because of what should have been a harmless bug in it. VirtualBox 5.1 should be able to run NetWare 6.5 SMP. No more "parity errors" causing abends.

[Kumba]

It did give me ideas, problem sorted out. It turned out to be APICDRV.PSM doing very unusual things because of what should have been a harmless bug in it. VirtualBox 5.1 should be able to run NetWare 6.5 SMP. No more "parity errors" causing abends.

Oh, nice! I had a reply typed up earlier, but got distracted trying to get FreeBSD 10.3 to run with Vbox and MATE....which didn't go the way I wanted. Then tried Solaris 8 x86, which crashes if you have VT-x enabled. I think I found a way around it, but that's for later. Need Sol8 for its s5/s10 slice support so I can hack up my Banyan Vines VDI image and see if I can get it to boot farther.

Anyways, what's the specific commit that fixes NW65? Kinda curious. I'm on 5.0.24, but let me know when a stable 5.1 beta is available and I can give my NW65 install a go.

[michaln]

[quote="Kumba"Then tried Solaris 8 x86, which crashes if you have VT-x enabled.[/quote]
Certain versions of Solaris are veeery picky about the CPU and don't like newer Intel processors. Usually there are workarounds. Do you happen to have the panic details?

Anyways, what's the specific commit that fixes NW65? Kinda curious. I'm on 5.0.24, but let me know when a stable 5.1 beta is available and I can give my NW65 install a go.

Not committed yet, I'll let you know when it is. I doubt the fix will go into 5.0 BTW.

I could not find anything about NetWare SMP on other hypervisors. Years ago, VMware didn't support SMP with NetWare 6.x, don't know if it changed. The performance indeed isn't brilliant because NW 6.x doesn't halt the CPU when idle. So lots of wasted cycles.

[Kumba]

Certain versions of Solaris are veeery picky about the CPU and don't like newer Intel processors. Usually there are workarounds. Do you happen to have the panic details?

Yup, you're right. I booted an ISO I found on Google that was SunOS 5.8 Generic_108529-05, which has the issue. It triggers not only a genunix panic and bad trap fault, but also crashes the VM with a NULL pointer deref in Windows. That said, I found another ISO on an open FTP site on Google, and that booted SunOS 5.8 Generic_108529-11, which loads the second part of the installer fine. We'll see about it actually finishing the install, though. Screenshots attached, but I'll go ahead and open a ticket, as this is on 5.0.24 and that NULL deref, though caught by the application, is probably something you guys want to look at.

Solaris 8 GPF in install in Vbox 5.0.24, w/ VT-x

solaris8-vbox-vtx-panic-20160704.png (32.97 KiB) Viewed 13762 times

NULL deref error dialog in Windows from Solaris 8 panic

solaris8-vbox-vtx-null-20160704.png (15.9 KiB) Viewed 13762 times

I doubt the fix will go into 5.0 BTW.

That's perfectly fine, I have no problem waiting for 5.1.x. finally be able yo play with my NW65 VM again, and start backing out some of my hacks, will be fun! :D

Btw, are you using the LsiLogic SCSI driver or the BusLogic driver for the NW65 VM you have? I never got far with the LSI, but the BusLogic driver and using an old HAM file from (I think) NW60 or NW51 worked quite well in NW65 under 1 VCPU.

I could not find anything about NetWare SMP on other hypervisors. Years ago, VMware didn't support SMP with NetWare 6.x, don't know if it changed. The performance indeed isn't brilliant because NW 6.x doesn't halt the CPU when idle. So lots of wasted cycles.

I tried VMWare one time on this machine, but there's a conflict somewhere with either my particular hardware or something else that would randomly BSoD my system. It was a known issue, as afew Google searches on the specific STOP code kept returning semi-related hits to the VMWare forums where users reported similar BSoDs, so once I removed VMWare, everything ran fine. Not a loss at all, as far as I am concerned. VBox works well, and is far more flexible. The configuration dialogs alone make changing most VM settings a snap, whereas VMWare's VM config is doing some non-standard GUI/widget stuff that's always been confusing. But I digress...

That said, NW is notorious for not idling the CPU. Ditto for MS-DOS. For both platforms, I have a CPU idling driver/TSR that solves that problem. Though, the types of systems those ran on weren't known for needing to idle a CPU anyways, back in the old days. Same goes for a lot of my SGI hardware...no idling CPUs there. When those machines aren't doing real work, they convert electricity to heat :)

[Kumba]

For the Solaris 8 bug, I created Ticket #15571.

[michaln]

Yup, you're right. I booted an ISO I found on Google that was SunOS 5.8 Generic_108529-05, which has the issue. It triggers not only a genunix panic and bad trap fault, but also crashes the VM with a NULL pointer deref in Windows.

Okay, the guest OS panic is a genuine Solaris bug, described in detail here. The VM process crash is something else of course and yes, that would be a VirtualBox bug.

Btw, are you using the LsiLogic SCSI driver or the BusLogic driver for the NW65 VM you have? I never got far with the LSI, but the BusLogic driver and using an old HAM file from (I think) NW60 or NW51 worked quite well in NW65 under 1 VCPU.

No, I'm just using IDE. For testing purposes I was really only interested in SMP. The LSI Logic SCSI emulation ought to work but I'm really not certain.

That said, NW is notorious for not idling the CPU. Ditto for MS-DOS. For both platforms, I have a CPU idling driver/TSR that solves that problem. Though, the types of systems those ran on weren't known for needing to idle a CPU anyways, back in the old days. Same goes for a lot of my SGI hardware...no idling CPUs there. When those machines aren't doing real work, they convert electricity to heat

Actually MS-DOS 6 and later comes with POWER.EXE which does idle the CPU. On IBM's DOS releases it's even installed by default. But yeah, without that, the CPU just spins and spins.

For NetWare, I expect it wouldn't be a problem at all in the olden days. And even with more modern systems, server hardware is designed to run under a load. But energy efficient it's not

[Kumba]

Okay, the guest OS panic is a genuine Solaris bug, described in detail here. The VM process crash is something else of course and yes, that would be a VirtualBox bug.

Okay, I added a comment to #15571 that the Ticket's title needs to be updated to reflect the real bug.

No, I'm just using IDE. For testing purposes I was really only interested in SMP. The LSI Logic SCSI emulation ought to work but I'm really not certain.

Okay, that's another one that will still be out there. I forget the actual bug...I think it was NW65 not recognizing the specific PCI vendor or device ID for the virtualized LsiLogic driver. I'll have to jog my memory after VBox 5.1.x is out.

Actually MS-DOS 6 and later comes with POWER.EXE which does idle the CPU. On IBM's DOS releases it's even installed by default. But yeah, without that, the CPU just spins and spins.

For NetWare, I expect it wouldn't be a problem at all in the olden days. And even with more modern systems, server hardware is designed to run under a load. But energy efficient it's not :)

Hmm, I did not know about POWER.EXE. I'm using DOSIDLE.EXE instead. I forget where I picked it up from, but it looks like it does the same job.

Random: How hard would it be for VBox to emulate an ATI Mach64? Those video chips were embedded in quite a lot of old hardware, PCs and servers. I got Solaris 8 loaded, but its graphics selection is really limited, so I'm stuck in 640x480@16bpp. Sol8 has a driver for a Mach64, as do quite a few others, so if VBox could emulate that specific chip, it might make testing out older OSes a lot easier.

[michaln]

Okay, that's another one that will still be out there. I forget the actual bug...I think it was NW65 not recognizing the specific PCI vendor or device ID for the virtualized LsiLogic driver. I'll have to jog my memory after VBox 5.1.x is out.

I have a working NW 6.5 VM with LSI Logic on a different machine, so you should give it a try. Whether it's really better than the alternatives I don't know.

The NW 6.5 SMP fix is now committed but probably not yet visible externally.

Hmm, I did not know about POWER.EXE. I'm using DOSIDLE.EXE instead. I forget where I picked it up from, but it looks like it does the same job.

POWER.EXE was added in DOS 5.01 or 5.02 (I forget which) to support laptop hardware, which was just becoming more popular and started implementing power management. POWER.EXE works with APM if available, but can halt the CPU even without it. Quite handy.

Random: How hard would it be for VBox to emulate an ATI Mach64?

Hard enough that it probably won't happen. Yes, it would be great for old guests, but anything less than 15-20 years old or so can handle VBE. Basically a fair amount or work for a very small payoff, as cool as it would be.

But it reminds me that somewhere I have a PMI file that I think could be used with Solaris 8 to add better graphics modes for VirtualBox. What I haven't figured out was how to package it.

FYI there is a way to get VBE support into Solaris 8. I believe Sun had some patches, and there were 3rd party XFree86 packages.

[michaln]

The NetWare 6.5 SMP fix is revision 62018 in the VirtualBox OSE source code, now visible.

[Kumba]

The NetWare 6.5 SMP fix is revision 62018 in the VirtualBox OSE source code, now visible.

Tested 6 VCPUS out and NW6.5 boots fine :)

Still have to play with some other options, including the SCSI controller and it looks like multiple vcpus were not running in the idle loop, but other than that, it booted largely without complaint.

I did notice one quirk when the NW bootloader loaded up, seen in the below screenshot:

Vbox 5.1, NW65 Boot, Error 744

vbox51-nw65-boot-err744-20160714.png (24.94 KiB) Viewed 13708 times

That did not seem to impact the actual OS load, so whatever NW uses for a bootloader stub if you choose to not boot via DR-DOS is probably what's causing that.

[michaln]

The NetWare 6.5 SMP fix is revision 62018 in the VirtualBox OSE source code, now visible.

Tested 6 VCPUS out and NW6.5 boots fine

Good to hear, I only tried up to 4 VCPUs.

Still have to play with some other options, including the SCSI controller and it looks like multiple vcpus were not running in the idle loop, but other than that, it booted largely without complaint.

The idle loop is (as you know) generic, NetWare just doesn't halt. I saw that there are various solutions, not sure what's best. If you find something that works well, please let me know (BTW the VMware idling NLMs might well work.)

I also refreshed my memory re disks... IDE works, but does not use DMA for some reason. AHCI does not work because NetWare only recognizes a very small number of controllers. LSI Logic works out of the box in my experience but YMMV. And you said BusLogic works with their own drivers.

I did notice one quirk when the NW bootloader loaded up, seen in the below screenshot:

That did not seem to impact the actual OS load, so whatever NW uses for a bootloader stub if you choose to not boot via DR-DOS is probably what's causing that.

I have not seen that. Does it happen every time, and which NW version is it exactly? That's booting from disk, not CD-ROM, right?

[DrScriptt]

Need Sol8 for its s5/s10 slice support so I can hack up my Banyan Vines VDI image and see if I can get it to boot farther.

I want to learn more about S5 / S10 as I haven't run into it before hacking on Banyan Vines 8.5.

By the way, check out my recent post Banyan VINES 8.5 boots and runs in VB 5.1.36 for details on how I got Banyan Vines 8.5 to boot and run in VirtualBox. (I suspect the same technique could be used with physical hardware.)