#asahi · 2023-08-06 16:30
soreararmv8 has a ton of instructions with two destinations, largely approaching the "common code needs 2 bytes per elementary operation" problem from the other direction#asahi · 2022-03-24 10:12
sorear(i have to wonder about kexec in connection to the preboot DCP issues we have)#asahi · 2022-03-23 10:25
sorearyes#asahi · 2022-03-23 10:24
sorearthe m1n1 hypervisor is specialized for debugging/ reverse engineering apple silicon OSes (macOS and asahi itself); if you want to use a hypervisor for anything else you should use qemu/kvm#asahi · 2022-03-23 10:21
sorearm1n1 has several functions, one of which is an optional hypervisor which end users won't use#asahi · 2022-03-21 12:47
sorearswap the 3rd and 4th steps#asahi · 2022-03-20 11:32
sorearwhy does the allocator page size have to be the same as the kernel/hardware page size anyway? is chromium doing fun things with mprotect and SEGV handlers?#asahi · 2022-03-20 00:45
sorearI suspect you can run a 4K kernel under KVM#asahi · 2022-03-20 00:22
sorear(user, anyway)#asahi · 2022-03-20 00:22
sorearqemu tcg won't help you with mmap alignment requirements#asahi · 2022-03-19 07:36
sorearah. idk then#asahi · 2022-03-19 07:17
sorearthe former can be mixed in one process, the latter cannot (macOS supports a translation granule per process, linux presently doesn't)#asahi · 2022-03-19 07:17
soreardon't confuse hugepages (4K/2M/1G) with translation granules (4K/16K/64K)#asahi · 2022-03-18 10:14
sorearfeat_rng?#asahi · 2022-03-12 04:26
sorearrunning an emulated-16k kernel on M1 could provide interesting information (is the observed 16k speedup due to TLB/cache issues alone, or do kernel algorithms and fewer page faults also have a large impact?)#asahi · 2022-03-12 04:25
sorear"portable 16k-page kernels" wouldn't work, because the kernel needs to know the page table fanout at compile time and that's different between native and emulated 16k pages#asahi · 2021-12-26 10:05
sorearI'm interpreting the question as "does the Apple hardware allow people to play the silicon lottery or is the DVFS functionality hardware locked to only known safe voltage/freq combinations"#asahi · 2021-12-21 17:37
sorearguess that makes sense#asahi · 2021-12-21 17:33
sorearTIL there are cases where you need to build a modular kernel that don't come down to "extremely little ram" or "bootloader limitations". am curious if there is a list of other drivers that are limited in this way when compiled in#asahi · 2021-12-20 05:50
sorearmight be nice to have a giant warning on that saying when it was written (you can get that from the git history but still), that it will probably be outdated by the time the median reader sees it, what versions of macOS/m1n1/linux it applies to#asahi · 2021-12-17 23:22
sorearyou can also look at the “Logs:” link in the channel topic#asahi · 2021-12-17 23:21
sorearit’s working on the IRC side#asahi · 2021-12-15 11:59
sorearI just want to say that I love what you did with the color-coded support matrix#asahi · 2021-12-14 01:58
sorearthat's the size of framebuffer the bootloader gives you if it doesn't detect a display at all. iphone 5 native resolution#asahi · 2021-11-25 07:04
sorearremember that big.LITTLE has been a reality for Android for years and that's in the upstream kernel#asahi · 2021-11-09 16:32
sorearimagines you can do something in xkb#asahi · 2021-10-30 12:25
sorear.o( new channel for spoilers )#asahi · 2021-10-29 04:10
sorear64 byte cache blocks? does a LLC miss occupy 1 channel for 32 transfers or is it somehow strided?#asahi · 2021-10-22 04:18
sorear"no, I think the terms RISC/CISC should mostly stay in the 90s where it belongs." very this#asahi · 2021-10-20 07:01
soreari'm imagining that there's separate datapath cryptography in all 8 cores, plus PAC (which is likely distinct since it is connected to integer registers and not NEON), then there's the armv8 microcontroller cores they use which might also have PAC and/or simd cryptography. then we've got the ANS cryptography and whoever is responsible for HDCP#asahi · 2021-10-20 06:55
sorear're quite small these days and don't _need_ to be centralized#asahi · 2021-10-20 06:55
sorearit may be hard for anyone outside of apple's export control division to give you a complete list of cryptographic accelerators in the M1 - they#asahi · 2021-10-17 09:24
sorearincreasingly of the opinion ISA is a bad proxy for anything platform related#asahi · 2021-10-16 18:03
soreartil the history#asahi · 2021-10-16 18:00
sorearlooks like http://man.cat-v.org/plan_9/1/8c has a similar extension but with different syntax#asahi · 2021-10-16 17:56
sorearthat's a C99 designated initializer, not a plan9ism#asahi · 2021-10-15 07:22
sorearthe installer is in python so it needs python available in the macOS/1TR environment where it runs, if I've understood correctly. I don't think it matters whether there is also a copy of python in the installed distro, since the installer doesn't run there?#asahi · 2021-10-14 15:53
sorearwould m1x or m2 have aicv2?#asahi · 2021-10-13 19:03
sorearwhat happens if you try? synchronous exception?#asahi · 2021-10-10 03:09
sorearphire: avx is disabled by fuse in Pentium-branded parts for market segmentation reasons -> software can't rely on it -> baseline ISA patents expiring soon ... huge own goal by itnel if i've understood it correctly#asahi · 2021-10-10 03:08
sorearmarcan: presumably you've seen https://en.wikipedia.org/wiki/Amber_(processor_core)#asahi · 2021-10-08 22:49
sorear ha#asahi · 2021-10-04 16:04
sorearread the draft, nothing jumped out at me#asahi · 2021-10-03 00:57
soreari'd be concerned about machine safety if anything - are there any circumstances where the laptop is likely to overheat if m1n1 sets it to full power and the kernel hangs or panics before it has a chance to take over pstates?#asahi · 2021-09-29 13:18
sorearenjoying all the implicit trust on the window server, GUI libraries, keyboard and display drivers...#asahi · 2021-09-29 13:15
sorearhow is the SEP authenticating storagekitd#asahi · 2021-09-29 13:02
sorearyou can get raw disk access but you can't change the default boot partition without an entitlement?#asahi · 2021-09-29 12:17
sorear(and none of those three terms is completely standard)#asahi · 2021-09-29 12:16
sorearmobile socs always have integrated graphics, so there's no frequent need to talk about the "rendering engine, media codec, and display controller" as a group#asahi · 2021-09-29 12:15
sorearwell card would be physically inaccurate if you have integrated graphics, which is the situation here