#asahi-dev · 2023-01-10 14:37
sven Device tree node: /sys/firmware/devicetree/base/soc/phy@503000000/pcie@530000000/pci@0,0#asahi-dev · 2023-01-10 14:37
sven0001:00:00.0 PCI bridge: Apple Inc. Device 1010 (prog-if 00 [Normal decode])#asahi-dev · 2023-01-09 14:27
svenhah, I think I can access some of the thunderbolt pciec regs. I guess the rest just needs some additional init on the pciec side before they stop reading as 0xffffffff#asahi-dev · 2023-01-09 08:52
svenwouldn’t surprise me if it does the same for the thunderbolt iommus as well#asahi-dev · 2023-01-09 08:51
svenat least for the non-thunderbolt iommus macOS kernel panics on any fault #asahi-dev · 2023-01-08 14:34
svenso now I can at least read the thunderbolt pcie MMIO, it's just all 0xffffffff but at least it doesn't throw exceptions anymore#asahi-dev · 2023-01-08 14:33
svenduh, i forgot that driver needs custom handling once they have more than one power domain#asahi-dev · 2023-01-07 12:30
svenusb also doesn't work *sigh*#asahi-dev · 2023-01-07 12:30
svenNHI seems to be up but I still can't access the pcie MMIO :/#asahi-dev · 2023-01-07 12:30
svenhates thunderbolt#asahi-dev · 2023-01-06 17:00
sveniirc the tty node disappears a moment later #asahi-dev · 2023-01-06 16:59
sventhat looks similar to what I get when I try to suspend #asahi-dev · 2023-01-06 16:54
svencould be, I only have this m1 here #asahi-dev · 2023-01-06 16:53
svenlike, Python exception dies because the serial port disappears #asahi-dev · 2023-01-06 16:53
svenmy m1 mini reliably dies on suspend fwiw#asahi-dev · 2023-01-06 16:49
svenhaven’t ever seen the hv survive suspend/resume though #asahi-dev · 2023-01-06 16:47
svenyeah, I’ve even seen the watchdog write just before reboot occasionally #asahi-dev · 2023-01-06 15:28
sveniirc the pmgr reset is really just an interrupt for the asc so that missing shmem thing probably makes sense :(#asahi-dev · 2023-01-04 16:13
svenand i'm really trying to avoid writing a tracer so far to see how apple deviates from the standard there *sigh* #asahi-dev · 2023-01-04 16:12
svenspeaking of thunderbolt, I think I bring up at least one of those NHI rx/txrings now but then something gets stuck#asahi-dev · 2023-01-04 11:20
sventhey'll test the patch next and reply with tested-by if that quirk fixes it again#asahi-dev · 2023-01-04 11:20
svenfrom the t2 folks: for 6.2.0-rc3-00010-g69b41ac87e4a nvme nvme0: Identify NS List failed (status=0xb) in dmesg and only /dev/nvme0 exists#asahi-dev · 2023-01-04 09:58
svenlet me poke the t2 people to see if they can test some 6.2-rc#asahi-dev · 2023-01-04 09:56
svenyeah, no access to any t2 mac here either#asahi-dev · 2023-01-04 09:54
svenmarcan: nvme nvme0: Identify NS List failed (status=0xb) seems to present in the T2 logs as well (e.g. https://bbs.archlinux.org/viewtopic.php?id=281389). they probably also need that NVME_QUIRK_IDENTIFY_CNS quirk#asahi-dev · 2023-01-03 11:36
svenbah... and the acio driver is back to resetting the soc *sigh*#asahi-dev · 2023-01-01 17:34
svenkettenis: all the thunderbolt tunables are 32bit masks as well so I’ll stick to the current atcphy format #asahi-dev · 2023-01-01 15:48
sventhunderbolt-applesoc 501f00000.nhi: RTKit: IOP power state transition: 0x0 -> 0x20 <-- I think I was at that point before sometime last year but I'm still gonna count this as progress :)#asahi-dev · 2022-12-28 14:41
svenbut you can’t write drivers from that #asahi-dev · 2022-12-28 14:41
svensure, it’s also in the security whitepaper#asahi-dev · 2022-12-28 12:48
svenand that entanglement seems to change between boots So you can’t grab keys and reuse them later #asahi-dev · 2022-12-28 12:48
svenso essentially macOS asks sep to get the volume key, sep returns that key in an entangled form and just feeds that into ans #asahi-dev · 2022-12-28 12:47
svenI think you can set a flag somewhere and it’ll just take that key or set another flag and it’ll assume that key is “sep entangled” #asahi-dev · 2022-12-28 12:47
svenwas a bit bored and didn’t feel like thunderbolt so I looked a bit into nvme/FileVault. looks like there are aes key and IV fields for aes-xts in the tcb struct #asahi-dev · 2022-12-26 09:59
svenit very much sounds like a software limit only #asahi-dev · 2022-12-26 09:59
sveniirc we already allocate above vm-size for pcie and possibly dwc3#asahi-dev · 2022-12-26 09:50
svenlooks like that thing requires a different iommu for each allocation type, you shouldn’t need that #asahi-dev · 2022-12-22 12:21
svenbcrumb: marcan has to review and merge it. And usually we rebase PRs instead of merging back main #asahi-dev · 2022-12-22 12:20
sventhanks :)#asahi-dev · 2022-12-21 13:56
sven*feels#asahi-dev · 2022-12-21 13:55
sveni can hack it for now by just making everything child nodes of atcphy and probing/unprobing them I guess but that feel wrong#asahi-dev · 2022-12-21 13:54
sven(the pmgr clock thing is also weird: apparently it can only be turned on and it needs a ~0.5s delay or so to actually work and there's no status register. it can only be turned off by resetting atcphy afaict)#asahi-dev · 2022-12-21 13:52
svenright now the dart driver tries to access its MMIO inside probe to figure out some basic information about the dart (pagesize etc.)#asahi-dev · 2022-12-21 13:52
svenhow do we represent this? acio only (and its mbox) would be doable by just making sure the driver doesn't do anything before a tbt cable is connected and atcphy is up but acio-dart is going to be a bit more tricky#asahi-dev · 2022-12-21 13:51
svento turn that PMGR thing on atcphy has to be up and running. if atcphy is reset or powered off everything dies and some SoC-internal reset is triggered rather quickly#asahi-dev · 2022-12-21 13:51
svenurhg, so thunderbolt: there's acio and acio-dart, both need a special clock/power/whatever thing inside PMGR to be turned on to make their MMIO accessible and not SError.#asahi-dev · 2022-12-21 11:29
svenand the atcphy kext even fails if it sees any non-32bit tunable :D#asahi-dev · 2022-12-21 11:28
svenkettenis: I looked at the thunderbolt tunables again, and they also all seem to be 32bit masks. we can also just keep the current format then#asahi-dev · 2022-12-18 17:58
sveni thought they had a 64bit one somewhere in there but didn’t find that one anymore when I checked a few days ago