#asahi-dev · 2023-01-13 06:52
jannaumarcan: sure, I'll do it tonight#asahi-dev · 2023-01-12 18:30
jannausven: starting at €24 https://geizhals.eu/?cat=nbdock&xf=5860_Thunderbolt+3 so I'm half doubting that it is thunderbolt#asahi-dev · 2023-01-10 14:52
jannauhttps://paste.debian.net/1266758/ works around that and unbreaks resume#asahi-dev · 2023-01-10 14:41
jannaunot sure if nvme-core or apple_nvme_reset_work() is to blame for this#asahi-dev · 2023-01-10 14:40
jannauon resume it tries to disable the shutdown controller and hangs on a mmio write#asahi-dev · 2023-01-10 14:39
jannauc76b8308e4c9148e44e0c7e086ab6d8b4bb10162 breaks apple-nvme resume since it prevents NVME_CC_ENABLE to be cleared on the shutdown path#asahi-dev · 2023-01-10 12:52
jannauare you sure iot's llvmpipe and not a different HW specific driver? is the ashi driver without uabi changes maybe leaking out of the arch mesa pkg?#asahi-dev · 2023-01-10 12:50
jannauI tested that since that combination apparently caused Xorg crashes on debian in #asahi-alt#asahi-dev · 2023-01-10 06:26
jannauthe inverse (mesa-asahi-edge with the non-edge kernel) was working at the end of december#asahi-dev · 2023-01-09 06:48
jannauso it is mostly an annoyance since we have to assume it's safe to start sending messages using shmem after the SetShmem message and we have to start it on our own instead of after receiving the initialized mbox message#asahi-dev · 2023-01-09 06:41
jannaumarcan: there's a misunderstanding. DCP updates its shared mem pointer on second SetShmem message. it just doesn't acknowledge the message like it does on the initial SetShmem message#asahi-dev · 2023-01-08 15:21
jannauneat, dcpep, returns the shmem address if SetShmem.flag != 4#asahi-dev · 2023-01-07 08:40
jannauat least amazon has a servo to provide virtual/physical prescene#asahi-dev · 2023-01-07 08:38
jannauin addition they would need to offer a method to hold the power button#asahi-dev · 2023-01-07 08:38
jannauit would require an ip kvm for display/keyboard/mouse access in the recovery os#asahi-dev · 2023-01-07 08:37
jannauunless specifically mentioned I would doubt that you can enter recovery os with any of the cloud offerings#asahi-dev · 2023-01-06 20:31
jannauhack to avoid the reset: overwrite the data for 'MBSE' in the smc tracer#asahi-dev · 2023-01-06 20:29
jannaulooks like macos simply doesn't shutdown DCP on restart/shutdown and DCP still issues callbacks and syslog messages#asahi-dev · 2023-01-06 16:58
jannauit looks like it dies while parsing dcp messages https://paste.debian.net/1266326/#asahi-dev · 2023-01-06 16:21
jannaudcp trace is incomplete due broken uart communication even with a single cpu core#asahi-dev · 2023-01-06 15:44
jannaudo we have code to intercept macos restart/shutdown in m1n1?#asahi-dev · 2023-01-06 15:38
jannauthe pmgr reset fixes the system endpoint re-initialization after IOP shutdown#asahi-dev · 2023-01-06 15:26
jannauyes#asahi-dev · 2023-01-06 15:25
jannauapple_rtkit_idle()#asahi-dev · 2023-01-06 15:25
jannaumarcan: apparently only before #asahi-dev · 2023-01-06 15:24
jannauthere's free_signal / stop_signal but both crash dcp#asahi-dev · 2023-01-06 14:32
jannaustill no ack for the initial SetSHM call though#asahi-dev · 2023-01-06 14:30
jannauDCP module reloading works! I must not call A500/set_dcp_power(0) on shutdown (or I suppose I have to call A500/set_dcp_power(1))#asahi-dev · 2023-01-05 15:09
jannaunot that we know of. the system endpoints are reset after shutdown and pmgr reset#asahi-dev · 2023-01-05 14:48
jannaudcp is strange, after IOP shutdown and pmgr reset it doesn't reply to the initial SetSHM call but I can communicate over the shared mem. everything except swapping surfaces works#asahi-dev · 2023-01-04 23:18
jannaubut that might be only my tree, still working on dcp/sleep on top of asahi-wip#asahi-dev · 2023-01-04 23:17
jannauhmm, looks like nvme resume is broken. "nvme nvme0: Device not ready; aborting reset, CSTS=0x9"#asahi-dev · 2023-01-04 23:15
jannauah, I misunderstood. I see now that it is completely broken. after resume it seems impossible to create new processes#asahi-dev · 2023-01-04 23:02
jannauthe keyboard (driver) doesn't yet work as wakeup source but that shouldn't have changed#asahi-dev · 2023-01-04 22:56
jannaumy understanding was that suspend had issues in general#asahi-dev · 2023-01-04 22:47
jannauchadmed: headphone audio? cs42l84 on j375 and j493 has annoying noise#asahi-dev · 2023-01-04 22:04
jannaucs42l83 headphone jack audio (j274) sounds fine with asahi-wip#asahi-dev · 2023-01-03 13:43
jannaubut responsible for the breakage is the reset as long as the power domain stays on#asahi-dev · 2023-01-03 13:43
jannauwell, the issue is that the dart is runtime suspended / resumed while the device is in use#asahi-dev · 2023-01-03 13:36
jannaups_apcie_gp_sys is always-on on t600x so it ought to be the dart reset#asahi-dev · 2023-01-03 13:28
jannauI suspect it resets the dart while the device is in use#asahi-dev · 2023-01-03 13:27
jannaupcie works initially, ethernet seems to randomly break when trasmitting data. I think the issue is the reset in the dart resume function#asahi-dev · 2023-01-03 13:13
jannauyes, the soc specific ones and arm-smmu (not arm-smmu-v3)#asahi-dev · 2023-01-03 12:51
jannauI wonder why this is not a problem on other systems. runtime pm in other iommu drivers looks close enough#asahi-dev · 2023-01-03 12:49
jannauugly workaround for now: add DL_FLAG_RPM_ACTIVE if dev_is_pci(dev) is true#asahi-dev · 2023-01-03 12:49
jannauthe issue might be that pm_runtime_enable() returns true for the pcie device in dart_probe_device(). sysfs claims that runtime pm is not used for them, but dart_probe_device() is too early to check the device runtime pm status anyway#asahi-dev · 2023-01-03 11:34
jannauI would have expected that the device link / runtime pm code keeps the supplier on in that case#asahi-dev · 2023-01-03 11:33
jannauhmm, I mistook the quirks for a generic suspend/resume handling so there is no runtime pm the PCIe drivers#asahi-dev · 2023-01-03 11:11
jannauit looked like it uses generic runtime pm callbacks if pci drivers don't have their own callbacks. I stopped looking after that#asahi-dev · 2023-01-03 10:45
jannaustill looks like the PCI's subsystem understanding of runtime pm is incompatible with powering the iommu down on suspend