#asahi · 2023-01-14 01:22
marcanback when I was first fixing idevicerestore for AS, the TCP proxy thing was broken so I ended up with a successful but not activated restore and then got the phone home prompt on first boot (which went into recovery)#asahi · 2023-01-14 01:22
marcanthe tickets are fetched by dedicated code in idevicerestore, while the activation stuff is just a TCP proxy thing#asahi · 2023-01-14 01:21
marcantpw_rules: you cannot restore without the internet, it needs to get the tickets for your machine since it always restores in full security mode (that's not just the activation)#asahi · 2023-01-13 05:24
marcannp, we also have logs (see topic)#asahi · 2023-01-13 05:14
marcan(it's what macOS does most of the time too, single X2RGB10 + A8 biplane layer with compression)#asahi · 2023-01-13 05:13
marcanwe should look into that first since it gives the best quality/bandwidth/etc ratio for a single plane and is what KDE ought to use by default if supported#asahi · 2023-01-13 05:13
marcanalso there's the whole story with multiplanar formats and framebuffer compression#asahi · 2023-01-13 05:12
marcanso unless we find a way to disable alpha (maybe with matrix stuff / some property?) for a given plane, we either have to drop support for XRGB8888 (will break some userspace) or only support it on some planes (not sure how practical that is)#asahi · 2023-01-13 05:11
marcanthere's also a very annoying quirk that the DCP firmware doesn't support XRGB8888 (i.e. alphaless standard RGB). we can emulate that for a single plane by treating it as ARGB8888 with premultiplied alpha, which is what we do now, but that doesn't work for overlay planes#asahi · 2023-01-13 05:10
marcanthere are 4 plane slots in the firmware API but not all can be used at once, and some seem to have special behavior on some platforms, so for now it's best to just keep the supported configurations simple#asahi · 2023-01-13 05:09
marcanwe can in theory support at least 2 or 3 depending on the platform, but we still have some things to figure out and make that practical#asahi · 2023-01-13 05:09
marcanthe DCP driver currently supports a single hardware plane#asahi · 2023-01-13 05:09
marcanthe GPU and display controller are independent#asahi · 2023-01-13 05:07
marcannot really worth spending time on software rendering when we have GPU drivers :)#asahi · 2023-01-13 05:07
marcanhonestly I'm not sure exactly how it all works out in the software rendering path and I don't really care to find out, I just know it's noticeably slower with DCP (particularly on Wayland)#asahi · 2023-01-13 05:07
marcansoftware rendering is always copying data on the CPU anyway, but I think specifically with the modesetting driver you end up doing full updates, while I think the simpledrm driver has better support for damage regions?#asahi · 2023-01-13 05:02
marcanbut none of that matters when rendering is done by the GPU#asahi · 2023-01-13 05:02
marcanit actually should hurt performance even more if it mapped buffers properly, I'm pretty sure what it does right now is technically not legal as far as the hardware goes (but it works anyway)#asahi · 2023-01-13 05:01
marcanthing is the dumb framebuffer driver is more optimized for software rendering platforms, and also doesn't have vsync which also helps performance when you are right on the edge of smoothness with software rendering#asahi · 2023-01-13 05:00
marcanwhich completely changes how display output works, from a dumb framebuffer to a proper modern atomic modesetting driver#asahi · 2023-01-13 05:00
marcanit goes together with the whole display controller driver#asahi · 2023-01-13 04:59
marcanso it will probably wait until we consider the GPU stuff stable enough to promote#asahi · 2023-01-13 04:58
marcanDCP is also probably getting to the point where we could promote it to stable, except for the part where it hurts performance without the GPU drivers, which is a bit of a conundrum (you gain brightness support but lose desktop perf)#asahi · 2023-01-13 04:57
marcanthat was only like a week or so#asahi · 2023-01-13 04:57
marcanthere was just a brief limbo period for edge where DCP didn't support M2 but we unconditionally disable that hack on the edge kernels, so M2 machines wound up with neither#asahi · 2023-01-13 04:57
marcanit is binary on m2 on non-edge kernels (and always has been)#asahi · 2023-01-13 04:54
marcan(actually it wasn't even binary on M2 for the brief period before that on edge, it was just outright always on, since we disable the binary brightness hack on edge)#asahi · 2023-01-13 04:53
marcani509vcb: it is, that was added shortly after the first announcement of edge#asahi · 2023-01-13 04:21
marcaneven today we still have people throwing around expectations that Asahi Linux will never support SEP/Touch ID/whatever because they expect it to be "locked down" in some vague way (it's not, nothing is locked down from third-party OSes except decrypting DRMed content)#asahi · 2023-01-13 04:20
marcanbut this kind of stuff happens all the time with Apple because people don't understand how these secure boot systems actually work and how companies actually implement these things and what is "security" vs "just undocumented"#asahi · 2023-01-13 04:19
marcanreality: Linux just needed a patch to its NVMe driver since Apple's hardware is slightly nonstandard#asahi · 2023-01-13 04:19
marcanone big FUD instance was when T2 Macs came out, and some YouTuber or something claimed they "lock out the HDD from being used by Linux" and started a whole cycle of fake news#asahi · 2023-01-13 04:18
marcanShadowJonathan: I'm not aware of any "evil checks" ever existing in Apple products, that's just a myth/misconception that keeps coming up whenever this third party OS stuff is discussed which is why I mentioned it#asahi · 2023-01-10 04:28
marcanbut you can happily play 4K60 videos on YouTube with CPU decoding, it'll just use more power than with hardware decoding#asahi · 2023-01-10 04:26
marcanshakes fist at AMD and Nvidia and Intel for convincing everyone that video decoding and display output are inherently GPU features just because they *chose* to design it and bundle it like that#asahi · 2023-01-10 04:25
marcanregalramp: video decoding has nothing to do with the GPU, and is not supported yet#asahi · 2023-01-06 15:31
marcan"why is my log scanner pulling in the porn detector?!"#asahi · 2023-01-06 15:31
marcanwhich was... quite important, given the hilarity of the google3 dependency tree#asahi · 2023-01-06 15:30
marcanChaosPrincess: google did reproducible builds which globally cached build artifacts IIRC, so everyone shared one giant ccache (except not just for C) effectively#asahi · 2023-01-06 15:29
marcanstuff just worked, fast#asahi · 2023-01-06 15:29
marcanit was provisioned enough#asahi · 2023-01-06 15:29
marcanaxboe: there were no build slots, and I don't recall ever waiting for anything#asahi · 2023-01-06 15:28
marcanaxboe: I don't know about chromium, but when I was there the infra build stuff was all very nicely optimized, distributed and cached#asahi · 2023-01-06 15:27
marcanchromium... lol.#asahi · 2023-01-06 15:27
marcaneven with the double-build stuff linux-asahi is quite tolerable to build#asahi · 2023-01-06 08:20
marcanlots of discussion for that in https://gitlab.freedesktop.org/upower/upower/-/merge_requests/49#asahi · 2023-01-06 08:19
marcanalso note that right now KDE messes with the charge limits directly. I would posit that this functionality really belongs in upower and the current state is a hack, so it might be worth pushing for it on the upower side first and then KDE can just provide the UI#asahi · 2023-01-06 08:16
marcanblazra: note that the KDE charge limit settings do not persist across reboots (there is a bug open for this)#asahi · 2023-01-06 08:13
marcanayke: yes, AP mode is known broken#asahi · 2023-01-05 05:24
marcanit's a computer, not an iPhone or a PS5 :p