#asahi-dev · 2026-06-05 15:27
svendo we know at which version they changes this? 26.0?#asahi-dev · 2026-06-05 15:26
svenespecially cause we don't know what they'll change next time#asahi-dev · 2026-06-05 15:26
svenagreed#asahi-dev · 2026-06-05 15:25
jannauwe could switch to new compatible and add unknown_ctrl and linear_sq_ctrl as reg entries or properties but I think that's even worse#asahi-dev · 2026-06-05 15:23
svendon't really care though, both options kinda suck i think#asahi-dev · 2026-06-05 15:22
sveni was thinking about adding the minimal fw verison where it changes to the compatible string#asahi-dev · 2026-06-05 15:21
jannauadd the firmware version to the compatible string#asahi-dev · 2026-06-05 15:20
jannauthe second option works until apple changes the firmware again#asahi-dev · 2026-06-05 15:19
svenor maybe just add a new compatible without the apple,firmware-version that only works for the new fw version, dunno what's better#asahi-dev · 2026-06-05 15:12
svenand patch the device trees to the new compatible on m1/m2 if we are running on a newer version *sigh*#asahi-dev · 2026-06-05 15:11
svenso i guess for linux we need to introduce a new compatible with a required apple,firmware-version property and use that for m3+#asahi-dev · 2026-06-05 14:59
svencmd.cdw12 is just a bug i think because i read the nvme spec wrong and the controller on m1/m2 didn't mind#asahi-dev · 2026-06-05 14:59
svenso really just two registers + SART needs a bigger allowed region apparently#asahi-dev · 2026-06-05 14:59
svenhttps://gist.github.com/Artingl/c5e7f9fbb0307f7c236e29e4f1fc1f8e those are the nvme differences#asahi-dev · 2026-06-05 14:08
yukaon the kernel side#asahi-dev · 2026-06-05 14:08
yukaI'll check which commits are required exactly#asahi-dev · 2026-06-05 14:07
yukahttps://github.com/AsahiLinux/tuxvdmtool/pull/9#asahi-dev · 2026-06-05 14:07
yukasven: it does work#asahi-dev · 2026-06-05 12:35
chadmedoh huh, we never did the per-opp power measurements for t602x#asahi-dev · 2026-06-05 09:01
yukabut indeed I have a /sys/kernel/debug/spmi-0/#asahi-dev · 2026-06-05 09:01
yukaneed to figure that out#asahi-dev · 2026-06-05 09:01
yukathe tipd-spmi probe also fails with -110 half of the time currently#asahi-dev · 2026-06-05 08:53
svenif that works maybe just send that as a RFC to the spmi maintianers with our usecase and see what they think#asahi-dev · 2026-06-05 08:52
svenlooks like that should be enough to directly talk to the tipd chip#asahi-dev · 2026-06-05 08:52
yukaI'll look into it#asahi-dev · 2026-06-05 08:52
yukaMaybe#asahi-dev · 2026-06-05 08:52
svenhah, https://cyberchaos.dev/yuka/linux/-/commit/88b7ece864c3a05342a10fa3b82260b18e3271ad is the interface you could use, isn't it? :D#asahi-dev · 2026-06-05 08:51
sventhere are also some old commit by jannau somewhere that directly add some sysfs or debugfs or whatever thing to tipd iirc#asahi-dev · 2026-06-05 08:51
yuka(https://cyberchaos.dev/yuka/linux/-/commits/tipd-spmi at least loads as a separate module on my system, and is able to turn on the 5V power)#asahi-dev · 2026-06-05 08:50
svenmight be a good idea to ask the spmi maintainers what they think about a generic i2c-style userspace interface#asahi-dev · 2026-06-05 08:50
yukaI'll try the sysfs route based on what I have#asahi-dev · 2026-06-05 08:50
sveneither that or create a /dev node like the i2c interface. but i guess sysfs might be easier#asahi-dev · 2026-06-05 08:48
yukaSo probably a sysfs file for sending VDMs and one for setting the local mode?#asahi-dev · 2026-06-05 08:48
svenif that's accepted upstream, yeah. otherwise add a userspace interface to tipd to send commands#asahi-dev · 2026-06-05 08:48
yuka*could#asahi-dev · 2026-06-05 08:48
yuka* for spmi#asahi-dev · 2026-06-05 08:48
yukawould of course add a generalized userspace interface spmi#asahi-dev · 2026-06-05 08:47
yukanot yet I guess#asahi-dev · 2026-06-05 08:47
sveni2c was convenient because you can just send commands from userspace, dunno if that's possible with spmi but i don't think so#asahi-dev · 2026-06-05 08:47
svenyes#asahi-dev · 2026-06-05 08:46
yukaFor spmi pd controllers we'll need a new interface to allow tuxvdmtool to send out VDMs, right?#asahi-dev · 2026-06-05 08:44
svenpresumably with a fixed firmware stored in ROM since i haven't seen anything in the ipsws#asahi-dev · 2026-06-05 08:44
sveni'm pretty sure that microcontroller is yet another core part of the main SoC fwiw. it also exposes SWD/JTAG via that debugusb protocol on development-fused devices#asahi-dev · 2026-06-05 08:39
yukaStanfordAppliedCyber: don't confuse cdc-acm and debugusb. For cdc-acm m1n1 directly drives the main usb phy itself and it appears like any usb-serial adapter on the other side. For debugusb it's a special apple thing and there's another microcontroller in the system doing usb to dockchannel uart internally.#asahi-dev · 2026-06-05 08:35
svenyeah, it's the two or so registers that moved to a new location then#asahi-dev · 2026-06-05 08:35
svenand as long as there are no further changes to the usb2 phy device mode should mostly just work if you manage to bypass the hub#asahi-dev · 2026-06-05 08:35
StanfordAppliedCyber[m]not sure, chaos princess told me we knew what the changes were#asahi-dev · 2026-06-05 08:34
sveniirc it was just that two registers moved to a new place#asahi-dev · 2026-06-05 08:34
svendidn't someone already reverse engineer the nvme changes for that newer firmware? or have they changed even more for the neo now?#asahi-dev · 2026-06-05 08:33
StanfordAppliedCyber[m]how would we actually use this? like just set some gpio pin --> usb device mode just works?