#asahi · 2022-03-12 12:30
marcanthe two matched woofers that work are within 1dB though#asahi · 2022-03-12 12:29
marcanpretty obvious in the video too#asahi · 2022-03-12 12:29
marcanchadmed, povik: https://photos.app.goo.gl/1C57sNwvmeVMpo4i7 :)#asahi · 2022-03-12 12:24
marcanoh yeah, the right tweeter is like 9 dB louder according to this#asahi · 2022-03-12 12:19
marcan"uninitialized memory", right#asahi · 2022-03-12 12:16
marcanlet me check with a mic#asahi · 2022-03-12 12:16
marcan(unless my ears are hosed#asahi · 2022-03-12 12:16
marcanbut yeah, the left tweeter does seem subtly quieter#asahi · 2022-03-12 12:16
marcanit's totally dead for me#asahi · 2022-03-12 12:16
marcanyup, that's the one I'm missing#asahi · 2022-03-12 12:15
marcanyeah, I do#asahi · 2022-03-12 12:15
marcan("side left" according to speaker-test)#asahi · 2022-03-12 12:15
marcan0 is dead#asahi · 2022-03-12 12:14
marcan6 and one is dead#asahi · 2022-03-12 12:14
marcanpovik: so I set everything to 50 and speaker-test only gives me 5 speakers, and it sounds like the left tweeter is quieter than the right one?#asahi · 2022-03-12 12:10
marcanlooks like we should change the name, there are some length issues :-)#asahi · 2022-03-12 12:10
marcan 0 [a ]: snd-soc-apple-m - MacBook Pro J314/6 integrated a#asahi · 2022-03-12 11:57
marcanwell, I want to test it :)#asahi · 2022-03-12 11:56
marcanI'll add that in for safety#asahi · 2022-03-12 11:55
marcanpovik: j313 doesn't have the speakers disabled?#asahi · 2022-03-12 11:41
marcanalso not relevant for anything but x86 emu#asahi · 2022-03-12 11:41
marcanthere's some other stuff about SSE flags#asahi · 2022-03-12 10:30
marcanyes, that's just the building for 4K alignment problem etc#asahi · 2022-03-12 10:28
marcanotherwise you'd have built it for arm64#asahi · 2022-03-12 10:28
marcanif you're running x86 code in the first place it's because it's not open source#asahi · 2022-03-12 10:28
marcanso no chance of recompiling those#asahi · 2022-03-12 10:28
marcanand yeah, for x86 the target is stuff like games#asahi · 2022-03-12 10:27
marcanthe emacs problem was assuming the build-time page size == the run-time page size#asahi · 2022-03-12 10:27
marcanwe still do, chromium/jemalloc/WebKit/Emacs and I think a few others were broken on asahi, some of those have landed fixes, some haven't#asahi · 2022-03-12 10:26
marcanalso emulators for other 4K native systems of course#asahi · 2022-03-12 10:26
marcanyeah, and because x86 applications can and do assume they can just mmap stuff at random 4K aligned vaddrs#asahi · 2022-03-12 10:26
marcanthat's most of it#asahi · 2022-03-12 10:26
marcanmmap() needs to work with 4K alignment, and 4K pages need to have independent protection status#asahi · 2022-03-12 10:25
marcanno, the 4K story is basically for x86 compat#asahi · 2022-03-12 10:24
marcanit'd complicate anonymous memory allocation though, how do you keep track of how much of a 16K page is actually in use and hand out the rest?#asahi · 2022-03-12 10:23
marcanin principle I want to say 4K processes working as a layer on top of a native-16K kernel should *work* in that you can just treat each 16K page as 4 4K pages that can be mapped independencly by 4K processes, though you'd need accounting for that kind of splitting#asahi · 2022-03-12 10:21
marcanbut then we still run into the IOMMU issues#asahi · 2022-03-12 10:21
marcanwith memory folios#asahi · 2022-03-12 10:21
marcanyeah, that would be easier#asahi · 2022-03-12 10:21
marcanOTOH 4K processes mapping chunks of 16K page cache pages isn't necessarily a problem#asahi · 2022-03-12 10:20
marcanthe point of 4K pages is so that mmap() on a 4K granularity actually works#asahi · 2022-03-12 10:20
marcan*purpose#asahi · 2022-03-12 10:20
marcanyeah, but that defeats the entire purposes#asahi · 2022-03-12 10:20
marcanand there's horrible corner cases like what happens when a 4K process tries to mmap memory that a 16K process has mapped, or worse, vice versa#asahi · 2022-03-12 10:19
marcanhowever, the weirdness will happen when the kernel has to access userspace pages, yes, since a lot of things will probably assume a 1:1 mapping between page sizes for kernel/userspace#asahi · 2022-03-12 10:19
marcan(the CPU does support mixing 4K/16K this way, simultaneously)#asahi · 2022-03-12 10:18
marcanarnd: there's two top level page tables, one for kernel and for userspace, and in principle you'd set only the user/lower one to 4K/16K and keep the kernel one at 16K always (that's how macOS does it)#asahi · 2022-03-12 08:00
marcansame chip, marketing says 380MB/s, so that sounds about right#asahi · 2022-03-12 08:00
marcanbluetail[m]: https://www.baslerweb.com/en/products/vision-components/pc-cards/usb-3-0-interface-card-pcie-fresco-fl1100-4hc-x4-4ports/#asahi · 2022-03-12 07:51
marcanUSB 3.0 performance can vary widely between controllers