#asahi-dev · 2026-07-30 18:12
sventhat's with hardware sprr/gxf though#asahi-dev · 2026-07-30 18:10
svenit's not even the macos27 device tree but random patches applied to the 12.x one until sptm was happy#asahi-dev · 2026-07-30 18:10
svenhttps://f.svpe.de/91ac7157b3b58a9b68611eef1208586284298ec300bdbc036b70b6a3f5cce4d2_xnu_sptm.txt hahahahaha, this had absolutely no right to get as far as it did :D#asahi-dev · 2026-07-30 16:07
svenah crap, i think my cute "let's pretend _EL1 is actually _GL1 and hope nothing in the guarded world tries to actually read/write the real _EL1 bank" doesn't seem to hold anymore for SPTM :(#asahi-dev · 2026-07-30 11:20
svenit's upstream now#asahi-dev · 2026-07-30 11:20
sven*look#asahi-dev · 2026-07-30 11:20
svenwe still do, looks for the dwc3 apple glue driver#asahi-dev · 2026-07-29 20:11
svenboth xnu and sptm code being read-only makes that kinda painless#asahi-dev · 2026-07-29 20:10
svenjust patch them to hvc then i guess :D#asahi-dev · 2026-07-29 20:10
svenugh#asahi-dev · 2026-07-29 20:09
svenhcr#asahi-dev · 2026-07-29 20:08
svenhrm, though tlbi might not an issue because i trap all of that already for the sprr/gxf emulation#asahi-dev · 2026-07-29 20:07
svenhm, also true#asahi-dev · 2026-07-29 20:07
sveni.e. if i have to take care of GL12/GL2 crap i may as well just handle the EL part the same#asahi-dev · 2026-07-29 20:05
svenanother reason against NV: they also use the _GL registers, so that would work on M2/M3 but i'd be stuck again on m4+#asahi-dev · 2026-07-29 20:04
svenyup :D#asahi-dev · 2026-07-29 20:01
svencan't be worse than fixing thunderbolt :P#asahi-dev · 2026-07-29 19:54
svenyeah :(#asahi-dev · 2026-07-29 19:49
sveni can probably just patch some _EL12 to just _EL1 and just anything that's truly _EL2-only to a hvc and try to emulate it#asahi-dev · 2026-07-29 19:48
sveni have another very cursed idea, but it's also going to be very annoying#asahi-dev · 2026-07-29 19:46
svenbah, this is going to very annoying. both sptm and xnu now seem to make the hard assumption that they run in EL2/GL2 :(#asahi-dev · 2026-07-29 15:09
sveni even tried all the cables i have#asahi-dev · 2026-07-29 15:09
svenyes :D#asahi-dev · 2026-07-29 13:04
svenshould probably actually install macos 27 at this point#asahi-dev · 2026-07-29 13:04
sven[cpu0] x0 = "%s: Couldn't find nvme-queue-entries"#asahi-dev · 2026-07-29 13:04
sven[cpu0] SPTM breakpoint: panic at 0xfffffff0270f395c#asahi-dev · 2026-07-29 10:47
svenonce that works we can move on to usb 3 tunneling (which is likely the easiest), then pcie tunneling (somewhat harder and won't work with very long chains due to the current linux hotplug code) and finally DP (that's gonna be annoying because DCP)#asahi-dev · 2026-07-29 10:46
svenit doesn't bringup pcie or dpd or anything else yet, but the first step is anyway to make it recognize the connection and setup tunnels#asahi-dev · 2026-07-29 10:46
svenunifiedgravity[m]: fwiw, https://github.com/AsahiLinux/linux/tree/sven/tbt-wip is my current (well... months old) work. it bringsup everything with a mmio trace similar to macos iirc but the co-processor then never sees the cable that's actually connected#asahi-dev · 2026-07-29 08:14
svenyou can also pass -S to run guest and you'll be dropped to a shell before the hv starts#asahi-dev · 2026-07-29 02:21
svenfair enough!#asahi-dev · 2026-07-28 22:26
svenbut that’s problems for later, first I need to actually support gl0 and then bring up SPTM and TXM in there #asahi-dev · 2026-07-28 22:25
svenyeah, and if we have to we can make safari or any other process that uses jit work #asahi-dev · 2026-07-28 22:25
svendunno what the gpu folks need to trace but even that shouldn’t be a problem #asahi-dev · 2026-07-28 22:24
svenyeah, I doubt it’ll be necessary #asahi-dev · 2026-07-28 22:23
svenif it comes to that we can probably patch that dylib to make the msr/mrs into a nop and just mal the jit index as rwx#asahi-dev · 2026-07-28 22:22
svenI also had to steal about a GB of RAM to have enough headroom to cache the shadow copies of the pagetables #asahi-dev · 2026-07-28 22:22
svenSafari with aprr_jit=0 in bootargs e.g. pegs a cpu core at 100% almost completely in the hv and doesn’t even start after 10 minutes #asahi-dev · 2026-07-28 22:21
svenIt’s not *great*, it’s just enough for tracing #asahi-dev · 2026-07-28 22:18
svenwe have a path that looks viable to fix that SPRR/GXF mess now at least #asahi-dev · 2026-07-28 22:16
svenI’ll be happy enough if they maybe unlock a few more regs (see above) and don’t regress them for every new soc release then #asahi-dev · 2026-07-28 22:11
svenespecially because they don’t seem to test the non-entitled boot mode and the approach seems to have been „just lock down everything that’s not in the arm arm“#asahi-dev · 2026-07-28 22:11
svenit’s a pity really #asahi-dev · 2026-07-28 22:08
svenyeah :(#asahi-dev · 2026-07-28 22:07
svenyeah, I don’t care all that much about TSO either. Would’ve been nice if it was an oversight and just lacking testing but apparently it was an intentional decision so *shrug*#asahi-dev · 2026-07-28 22:05
svenand if I could get a wish granted SPRR/gxf but well… :-D#asahi-dev · 2026-07-28 22:05
svenl2c_err thingy to be able to clear it#asahi-dev · 2026-07-28 22:03
svenTSO as well but I’ve heard that disabling that was very intentional #asahi-dev · 2026-07-28 22:03
sventhe core power off ones might be nice but we don’t even know how to correctly use them so kinda hard to ask for them