#asahi-dev · 2022-10-22 09:05
svenit’s possible to filter on the tp names of you just want to look at iomfb and I’m not sure you can easily have two trace systems in a single object #asahi-dev · 2022-10-22 09:04
svenI think we should call TRACE_SYSTEM appledcp or DCP instead of iomfb if you’re fine with that#asahi-dev · 2022-10-22 09:03
svenand finally resolve those tracepoint changes we both did :D#asahi-dev · 2022-10-22 09:02
sventhanks, will take a look later #asahi-dev · 2022-10-22 08:35
svenGuess I should try to rebase then #asahi-dev · 2022-10-22 08:35
svenah, nice#asahi-dev · 2022-10-22 08:26
svenThanks, I didn’t even notice that because I went back to atcphy when I didn’t feel like fixing that probe race #asahi-dev · 2022-10-21 14:33
svenand even t8112 seems to call back into the t8103 dp tunable magic#asahi-dev · 2022-10-21 14:30
svenlooks like the dp altmode tunable magic actually didn't change for t6000 phy, so this may just work there as well#asahi-dev · 2022-10-21 13:23
sventwo dcps + my ugly hacks seem to make everything just slow enough to hit it :)#asahi-dev · 2022-10-21 13:22
svenwe're just lucky that we never hit it with just dcp#asahi-dev · 2022-10-21 13:22
sventhat dcp_is_initialized issue is already present in the original driver I think#asahi-dev · 2022-10-21 13:16
svenso if you're unlucky enough the last devices probes, dcp_is_initialized is still false, then DCP sends that message and only then dcp_is_initialized becomes true but the display-subsystem probe is never called again#asahi-dev · 2022-10-21 13:15
svenbut dcp_is_initialized changes from false->true only after dcp_probe is already done#asahi-dev · 2022-10-21 13:15
svenEPROBE_DEFER will only try again as long as new devices are probed#asahi-dev · 2022-10-21 13:15
sventhat's also not a good idea#asahi-dev · 2022-10-21 13:15
sven return -EPROBE_DEFER;#asahi-dev · 2022-10-21 13:15
sven if (!dcp_is_initialized(dcp[nr_dcp]))#asahi-dev · 2022-10-21 13:09
sventhis entire thing sounds like a giant race condition right now tbh#asahi-dev · 2022-10-21 13:09
svenah, it's also a bit more complicated than that because dcp_is_initialized is only set after some messages have been sent over iomfb#asahi-dev · 2022-10-21 12:56
sventhat sounds like a proper fix, I don't do that here :D#asahi-dev · 2022-10-21 12:54
sveni can hack around it for now by just queueing the DPTX stuff (which takes a while) to a workqueue#asahi-dev · 2022-10-21 12:53
svenyeah#asahi-dev · 2022-10-21 12:51
svenif dcp_probe takes too long it can already get the hotplug cb and queue that work before apple_probe_per_dcp is done#asahi-dev · 2022-10-21 12:50
svenugh. i think i know what happens#asahi-dev · 2022-10-21 12:43
svenbut the link setup is also still broken in other ways afaict. it should send more ap calls than it does right now even before set_digital_out_mode is called#asahi-dev · 2022-10-21 12:42
svenjannau: fwiw, dcp->crtc is NULL inside dcp_delayed_vblank#asahi-dev · 2022-10-21 09:37
svenit'll also only work for link rate 0xa and one specific orientation of the cable#asahi-dev · 2022-10-21 09:35
svenit's not very stable yet and will also only work on t8103 for now#asahi-dev · 2022-10-21 08:48
svenlooks like I actually have to understand DCP now :(#asahi-dev · 2022-10-21 08:48
sven[ 9.486794] iomfb_callback: 271c00000.dcp: Callback D576 dcpep_cb_hotplug and then [ 9.491982] Unable to handle kernel NULL pointer dereference at virtual address 0000000000000558 dcp_delayed_vblank#asahi-dev · 2022-10-21 07:45
svenohh... duh.. because it has to ack the syslog message probably#asahi-dev · 2022-10-21 07:45
sventhough it's weird that it seems to break the tx side#asahi-dev · 2022-10-21 07:45
svenmarcan: ^-- looks like I can even break mailbox by just enabling lots of DCP debug spam :D#asahi-dev · 2022-10-21 07:45
sven[ 12.148215] apple-mailbox 271c08000.mbox: Try increasing MBOX_TX_QUEUE_LEN#asahi-dev · 2022-10-20 17:11
sventhough that may actually start to work once robin‘s bus iommu cleanup is done #asahi-dev · 2022-10-20 17:09
svenpractically that doesn’t work out because of a bunch of hacks along the way to io-pgtbl #asahi-dev · 2022-10-20 17:09
sventechnically the iommu api allows you to create a domain, create all mappings in there and only then attach it atomically #asahi-dev · 2022-10-20 14:58
sven[ 6.531518] dcp_dptxport_call: 271c00000.dcp: AP call 21 (deviceNotResponding) <-- and now ofc atcphy hates whatever I do again#asahi-dev · 2022-10-20 10:28
svenhides#asahi-dev · 2022-10-19 20:41
svenI remember testing that bus cleanup on top of 6.0 though #asahi-dev · 2022-10-19 20:41
svenhrm, i don’t think I ran 6.1 yet#asahi-dev · 2022-10-19 20:37
svenmaybe that combined with the bus_set_iommu? Would also explain why I never noticed it because I usually run upstream kernels which don’t have the usb nodes #asahi-dev · 2022-10-19 20:36
svenwell… USB has dual darts as well and it apparently also breaks sometimes #asahi-dev · 2022-10-19 20:35
sventhere was also that bus_set_iommu cleanup series in the core, maybe that interacts with dart in a strange way #asahi-dev · 2022-10-19 06:22
svenisn’t the wlan key stored in nvram and even propagated to the adt?#asahi-dev · 2022-10-19 05:24
svenmarcan: nope, never looked into where they are stored on macOS #asahi-dev · 2022-10-18 16:49
svennow let's see what happens if I connect a secondary screen with the linux driver#asahi-dev · 2022-10-18 16:48
sven[ 36.083214] apple-dcp 231c00000.dcp: RTKit: syslog message: entSource_rtkit.cpp:105: [AFK][AFKTimerEventSource:0xffffffff40fa85c8] worker <-- at least the AFK/EPIC stuff is finally working :)#asahi-dev · 2022-10-18 16:09
svenand it looks like I broke atcphy while trying to adjust it to support dp altmode :<