#asahi-dev · 2026-06-05 08:31
jannauI suspect there is a single gpio pin which bypasses the usb hub and allows usb device mode. unfortunately I have no good idea how to find it. likely on the smc pin controller#asahi-dev · 2026-06-05 08:28
StanfordAppliedCyber[m]thanks for the advice here! curse the horrible usb on this thing#asahi-dev · 2026-06-05 08:28
StanfordAppliedCyber[m]we're making significant progress through userspace#asahi-dev · 2026-06-05 08:27
StanfordAppliedCyber[m]i've been talking with yuka about kisd, will try, it's not a priority for now because i'm still in the 'make it work' phase (albeit definitely in the final 25%, if not more)#asahi-dev · 2026-06-05 08:27
jannausee https://github.com/AsahiLinux/kisd/issues/11 for macos usage#asahi-dev · 2026-06-05 08:27
jannaufor the debugusb troubles you could test https://github.com/AsahiLinux/kisd which should give you at least 10mbps#asahi-dev · 2026-06-05 08:26
StanfordAppliedCyber[m]and presumably everyone else who ever boots m1n1 hypervisor on the neo will do the normal "stream the kernelcache" thing#asahi-dev · 2026-06-05 08:26
StanfordAppliedCyber[m]it's just that i want to test the path i upstream#asahi-dev · 2026-06-05 08:26
StanfordAppliedCyber[m]i'm likely going to continue using my cursed setup because it still just works, adding nvme support will be had because there are nvme changes and for obvious reasons i can't contribute those#asahi-dev · 2026-06-05 08:25
jannauStanfordAppliedCyber[m]: you could create a fat partition and load the kernelcache from there using m1n1's nvme support (not sure if it will work out of the box on a18 pro / macOS 26). that's a less cursed setup than bundling the kernelcache#asahi-dev · 2026-06-05 07:58
StanfordAppliedCyber[m]have other people gotten debugusb to work on neo?#asahi-dev · 2026-06-05 07:57
StanfordAppliedCyber[m]that would be correct#asahi-dev · 2026-06-05 07:46
JamesCalligeros[m]Oh yeah :/#asahi-dev · 2026-06-05 07:43
chaos_princessi think stanford is on neo with it's weird usb topology#asahi-dev · 2026-06-05 07:41
JamesCalligeros[m]But i thought we sorted that out already#asahi-dev · 2026-06-05 07:40
JamesCalligeros[m]cdc acm should work once usb over spmi works#asahi-dev · 2026-06-05 07:38
StanfordAppliedCyber[m]right now i'm doing an even uglier hack than what i mentioned above#asahi-dev · 2026-06-05 07:38
StanfordAppliedCyber[m]okay, and yuka said she got this to work so i'll definitely need to do the same before i upstream this#asahi-dev · 2026-06-05 07:37
chaos_princesscustom dongle is uart, and going that way would take over a minute to send over the kcache#asahi-dev · 2026-06-05 07:37
chaos_princessso, runs at usb 2.0 speeds, not uart speeds#asahi-dev · 2026-06-05 07:37
chaos_princessno, it is uart protocol, but over usb. #asahi-dev · 2026-06-05 07:36
StanfordAppliedCyber[m]is cdc acm that custom hardware dongle that the macvdmtool readme talks about?#asahi-dev · 2026-06-05 07:36
StanfordAppliedCyber[m]oh, fair enough then#asahi-dev · 2026-06-05 07:35
chaos_princessall other machines have cdc acm that is much, much faster than uart. so, yes, 3-4 seconds is a good approximation#asahi-dev · 2026-06-05 07:34
StanfordAppliedCyber[m]if the kernelcache changes rarely (if this is wrong then correct me) then not having to send it over the wire for every test seems like the obvious decision#asahi-dev · 2026-06-05 07:33
StanfordAppliedCyber[m]wdym you want to chainload m1n1 itself? you just overwrite the first 2mb of m1n1#asahi-dev · 2026-06-05 07:33
StanfordAppliedCyber[m]my debugusb is having issues so i've been running over a 115200 uart#asahi-dev · 2026-06-05 07:33
JamesCalligeros[m]Also what if you want to chainload m1n1 itself#asahi-dev · 2026-06-05 07:33
StanfordAppliedCyber[m]it takes 2-3 seconds? i thought it takes 50+ in a best case#asahi-dev · 2026-06-05 07:32
StanfordAppliedCyber[m]does it change often though? for me it's only changed once#asahi-dev · 2026-06-05 07:32
JamesCalligeros[m]It takes like 2-3 seconds to stream a kcache who cares#asahi-dev · 2026-06-05 07:32
JamesCalligeros[m]Because thats a pain in the arse, no one wants to go through 1tr every time it needs to change#asahi-dev · 2026-06-05 07:32
StanfordAppliedCyber[m]this makes kernelcache iterations longer, but those are rare#asahi-dev · 2026-06-05 07:32
StanfordAppliedCyber[m]and then you just boot the kernelcache at some fixed offset#asahi-dev · 2026-06-05 07:31
StanfordAppliedCyber[m]as opposed to, say, building a raw boot image where the first 2mb are reserved for m1n1, you then bless a 1tr object containing the kernelcache, then chainload only sends 2mb#asahi-dev · 2026-06-05 07:31
StanfordAppliedCyber[m]--> you have to stream a 120mb kernel cache for every experiment#asahi-dev · 2026-06-05 07:31
StanfordAppliedCyber[m]like as i understand it (i've never actaully done it, and only vaguely looked at the code) you specify a path to a kernel cache, you stream that over uart, and then that boots under the hypervisor#asahi-dev · 2026-06-05 07:30
StanfordAppliedCyber[m]can i ask why the current chainload.py design exists as is, with a streamed kernelcache over uart?#asahi-dev · 2026-06-05 02:05
nicolas17sven: oh I didn't realize it was google login only, fair enough ^^#asahi-dev · 2026-06-05 00:38
StanfordAppliedCyber[m]as far as i can tell it's at least one sysctl and a bunch of random commpage bits#asahi-dev · 2026-06-05 00:38
StanfordAppliedCyber[m]there are a lot of knobs to turn off all the apple proprietary stuff#asahi-dev · 2026-06-04 19:35
amarioguythe actually awful part is the lack of ARM PMU support unironically, because it means that any instance of those reads if Apple doesn't add those MSRs in future cores will fault, and must be changed#asahi-dev · 2026-06-04 19:33
amarioguyfwiw, i'm not entirely sure you *need* the cursed parts of ACPI, all you actually need is a pseudo-DT in ACPI format, and have drivers carry the rest#asahi-dev · 2026-06-04 19:32
amarioguy> Bare metal arm windows relies on acpi and a bunch of other awful shit, its not happening#asahi-dev · 2026-06-04 19:29
amarioguyoh wtf just read again, hacr's gone?? ughhh#asahi-dev · 2026-06-04 19:25
amarioguy(on m4/m5, because of SPTM troubles)#asahi-dev · 2026-06-04 19:25
amarioguyreading asahi backlog about windows guests....man all paths really are leading back to "kernel patching is the sustainable path" despite everything i've ever read saying no...#asahi-dev · 2026-06-04 19:12
svenjust had to create it to send it to someone i know#asahi-dev · 2026-06-04 19:12
sventhat's it essentially :D#asahi-dev · 2026-06-04 19:12
svenFB21831949 / Write access to sys_reg(3, 5, 15, 1, 3) / APL_INTENABLE_EL2 faults in non-Apple-entitled boot mode