Hold on, are you telling me installing Windows on a PPC Mac natively may actually be possible after all these years?
I started this with an iBook G3 Snow, building on the work I did to get NT running on the Wii. That works by basically bringing in an emulator (by hooking an exception handler) to do the endianness swapping, and ensuring all code gets patched on load to call into it if needed. It still doesn't correctly install, things break, but it does run further than one would expect (kernel needs to be patched in several places to handle big endian mode, but it works!)
But the earlier grackle+paddington systems can actually switch endianness (although something in Apple's Open Firmware implementation breaks, at least on Lombard), so none of that is needed.
Technically I could probably get away with switching MSR_LE on later uni-north systems too, I'd just need to handle that situation in drivers. (Basically, the way MSR_LE itself works is by XORing the low bits of the address for different kinds of accesses, such that given a little-endian PPC executable loaded into memory, blindly doing endianness swapping on it treating it as 64-bit data, puts everything in the correct place to run with MSR_LE on. Grackle's little endian mode implementation isn't even a full implementation (like some other memory controllers do), it just ensures a correct view of memory over the PCI bus, so that could be implemented in software if needed.)
Yes, it's NT4, and yes, it has almost no app support
Almost no app support?!
A version of the insignia softwindows emulator for PPC NT got dumped recently (with Motorola branding). Looking at it, it looks like a direct port, and the wx86 environment runs on top of it (so x86 applications can run). It emulates a 486, I want to see the performance there.
And of course, there's some PPC NT native applications. Excel for PPC NT got dumped some time ago, but Word is still undumped.
(And it has ntvdm support, through a version of insignia's emulator that emulates a 286. So it can run some DOS/Win16 applications too, which is more than modern NT can do out of the box.)
Maybe after I've done all the bare metal stuff I'll mess around with a port to OSX Blue Box, just so Red Box finally becomes a thing. (and hey, maybe I can learn some more stuff about OSX internals along the way).
the neat trick that seemed impossible
I personally want to know how far Apple themselves got getting PPC NT to run. There's some documentation from Apple themselves that refers to doing that; not just internal leaked documentation (the PMG3B&W bootrom ERS, which mentions PCI devices are moved to physical memory starting from 0x8080_0000 "to boot NT, just in case", implying that PPC NT was the final plan if everything else failed, but "Designing PCI Cards and Drivers for Power Macintosh Computers" mentions that "PCI-based Power Macintosh computers support little-endian addressing for several reasons [...] so that they can run operating systems (such as Windows NT) that require the underlying hardware to operate as if it were little-endian". This passage is in the February 1996 edition but was not removed from the March 1999 edition that adds a chapter about New World - given that the 1996 edition predates Gossamer entirely, it makes me wonder if (for example) Hammerhead has an unknown register bit that does some form of endianness swapping.
And if anyone wants to know "why 0x8080_0000"? NT and its bootloaders expects virtual address 0x8000_0000 to be mapped to physical address zero. veneer.exe maps exactly 8MB from start of memory at 0x8000_0000, which will fail if a PCI device is mapped there given Open Firmware's 1:1 physical->virtual mappings. Why exactly 8MB? Because PPC NT kernel init originally hardcoded a BAT mapping for exactly 8MB. This was fixed in NT 4 but means one has to take this into account and deal with it if loading something earlier for whatever reason.
forums.macrumors.com
