16:32 TimurTabi: Why is there no "depends on CPU_LITTLE_ENDIAN" under "config DRM_NOUVEAU"? It's obvious that Nouveau doesn't support big-endian CPUs because it accesses little-endian firmware structures without any le32_to_cpu() calls. Shouldn't we protect against that in the Kconfig?
16:45 ajax: i thought we'd had it working on powermac at one point
16:56 TimurTabi: Hmmmmm
16:57 TimurTabi: Is there a vbios on powerpc?
16:57 TimurTabi: powermac, I mean.
17:12 RSpliet: It's EFI-provided I think
17:15 RSpliet: Or devicetree, or whatever it was. Anyway, there has to be a VBIOS to contain OEM-specific parameters and init sequences. Obvs it wouldn't need x86 init code
17:21 RSpliet: But this was in the pre-Cuda era, pre-Falcon firmware and the likes. I suspect that some code paths have not been extensively vetted.
17:23 TimurTabi: Hmm... I may be mistaken, then. Looking at the vbios code, it's endian safe. Well, the new GSP code isn't endian-safe, so I guess I'll have to wait for Ben to come back to ask about that.
17:26 RSpliet: I think you're generally right in expecting dragons in Power/BE support. The Power world is bigger than ancient Macs and IMHO still relevant given OpenPower. But also very niche, I don't think it's received much love at all beyond making an ancient Mac work in ancient times
18:03 ajax: iirc the powermac nvidia cards had an openfirmware "bios" code chunk for posting, but the data tables it worked from were in the same format as x86
18:04 ajax: whether that included endianness as stored in the rom... shrug
19:14 airlied: also nobody plugs a modern gpu into a BE system
19:14 airlied: gpus were one of the reasons power went LE
19:50 Lyude: there are currently some issues with powermacs and nouveau i'm aware of
19:50 Lyude: I don't have the hardware to reproduce but know some folks who do
19:51 Lyude: (if this would help anyone solve this issues let me know)
20:27 agd5f: plus we dropped all of the HW byte swappers several generations ago
20:53 TimurTabi: What about IBM P9 systems?
20:54 TimurTabi: RM supports them only in little-endian mode. Is that the standard? I would think that they normally run in big-endian mode.
20:54 airlied: normally in LE
20:56 TimurTabi: So be it, I guess. I was thinking it would be cool if Nouveau could run GSP on BE systems, but I guess no one really cares.
20:56 Lyude: tbf I think most of the cards running on those systems would be pre-GSP correct?
20:57 TimurTabi: Yes, although I don't know why.
20:58 TimurTabi: I know normal RM is broken on powerpc kernels newer than 5.9 because of some GPL thing.
20:58 TimurTabi: But I'd be surprised if that's the reason.
20:58 airlied: I think that is nvlink related
20:58 Lyude: some gpl thing? I certainly don't remember this
20:59 TimurTabi: Ok, I hope I'm not confused again. Let me dig it up. There was GPL_ONLY symbol that is needed by RM only on ppc kernels.
21:00 Lyude: if it's from the kernel syms that makes sense, if it's from nouveau.ko though I'm pretty sure our intent is for that driver to be MIT compatible
21:00 TimurTabi: I'm talking only about proprietary RM
21:00 Lyude: ahhh, gotcha