#
#asahi-dev · 2026-06-07 09:45
svenand well... the atcphy struct would disappear, so that's probably not it because we'd just see a crash
#asahi-dev · 2026-06-07 09:45
svenhrm.. but the next probe would use a different mutex then as well and
#asahi-dev · 2026-06-07 09:45
svenmaybe some left-over state when atcphy probe fails?
#asahi-dev · 2026-06-07 09:44
jannauprobing tipd, dwc3 and atc at the same time is probably racy but I don't understand why that would easily reproducible on some devices and and at all on others
#asahi-dev · 2026-06-07 09:42
sventrue
#asahi-dev · 2026-06-07 09:42
jannauif something changed I would expect it to be reproducible and not racy
#asahi-dev · 2026-06-07 09:39
svenbut that would surprise me
#asahi-dev · 2026-06-07 09:39
svenbut maybe i overlooked something back then or something changed
#asahi-dev · 2026-06-07 09:39
sveni remember checking when i originally wrote the code
#asahi-dev · 2026-06-07 09:38
jannauye, atc->lock is locked while registering the devices. so the only issue could be that one of register calls tries to access the locked phy. it would be unexpected and I think it's unlikely to be racy either
#asahi-dev · 2026-06-07 09:35
svenbut i'm pretty sure i guard against that by taking the mutex before registering all those devices
#asahi-dev · 2026-06-07 09:33
sventhe only thing that could happen was that e.g. the mux was registered and would already trigger a callback to atc from a different thread while atc was busy registering the phys
#asahi-dev · 2026-06-07 09:32
svenmaybe something changed?
#asahi-dev · 2026-06-07 09:32
svenhrm... i thought i was careful to avoid that
#asahi-dev · 2026-06-07 09:29
jannausven: due you known if any of atc sub-device register functions might call back into atc and thus deadlock? I think a deadlock during probe could be catastrophic when other devices are probe as well. asking for https://github.com/AsahiLinux/linux/issues/465
#asahi-dev · 2026-06-07 09:04
jannauI guess a little less so due to the damage handling but unless the notebook is constantly used the powered backlight while the display is "off" will burn those savings
#asahi-dev · 2026-06-07 08:52
sventhough I guess that also happens with simple framebuffer
#asahi-dev · 2026-06-07 08:52
svenI just don’t wanna know how quick it burns through the battery; all cores were easily at 20-30% running llvmpipe
#asahi-dev · 2026-06-07 08:51
svengpu-less dcp was useable on m1 in December for me
#asahi-dev · 2026-06-07 08:43
jannaualso not how the driver works at the moment. only way I can image how it could work is that either the driver or dcp has a separate static framebuffer and dcp blits the damage into it
#asahi-dev · 2026-06-07 08:40
jannauI'm not even sure if that would fit into how compositors use kms with a hw driver
#asahi-dev · 2026-06-07 08:40
JamesCalligeros[m]macos never uses them either iirc
#asahi-dev · 2026-06-07 08:36
JamesCalligeros[m]Swap reqs have damage region support on paper but i couldnt make them work
#asahi-dev · 2026-06-07 08:11
jannauthat said I think the usability advantages of dcp are more important than performance
#asahi-dev · 2026-06-07 08:09
jannaui.e. simpledrm has an advantage when only a small part is updated. without overlay cursor movement would be the prime example
#asahi-dev · 2026-06-07 08:07
jannauthe main advantage simpldedrm has over dcp is that it supports damage while dcp always swaps complete framebuffers
#asahi-dev · 2026-06-07 08:05
jannauit won't be a slideshow and I don't remember display without gpu as unsable. under kwin it ishould have improved due to using an overlay plane for the cursor
#asahi-dev · 2026-06-06 23:38
JamesCalligeros[m]some people can't seem to tell though so i guess it depends
#asahi-dev · 2026-06-06 23:38
JamesCalligeros[m]it's not slideshow tier but it's pretty bad ime
#asahi-dev · 2026-06-06 23:27
yukaI hope enabling dcp without gpu will not turn it into a slideshow
#asahi-dev · 2026-06-06 23:24
yukaBecause the disp_fe does actually not turn off the backlight either
#asahi-dev · 2026-06-06 23:24
yukaOk after playing around with pmgr in m1n1 I concur with you that properly turning off the internal display needs dcp
#asahi-dev · 2026-06-06 23:03
yukaNo
#asahi-dev · 2026-06-06 23:03
JamesCalligeros[m]Is the thing marked as always on in the devicetree?
#asahi-dev · 2026-06-06 23:02
yukaStill invalid argument after unbinding the framebuffer device
#asahi-dev · 2026-06-06 22:57
chaos_princessright, but unless the driver declares support for runtime pm, turning off pm domains from under it is not a valid action
#asahi-dev · 2026-06-06 22:57
yukaBut I guess I can try over ssh while my compositor is not running...
#asahi-dev · 2026-06-06 22:57
yukaor my compositor will be unhappy
#asahi-dev · 2026-06-06 22:57
yukait is not supposed to go away
#asahi-dev · 2026-06-06 22:57
yukaThe panel still exists
#asahi-dev · 2026-06-06 22:56
chaos_princessmaybe unbind the drivers first?
#asahi-dev · 2026-06-06 22:56
yukaonly on or auto
#asahi-dev · 2026-06-06 22:55
yukaI am not allowed to write "off" to the control sysfs entry
#asahi-dev · 2026-06-06 22:54
yukaThis was also a general question regarding raw control of power domains from linux
#asahi-dev · 2026-06-06 22:53
JamesCalligeros[m]Than wasting time with simpledrm
#asahi-dev · 2026-06-06 22:53
JamesCalligeros[m]My point was more that the effort would be better spent on making dcp do that if it doesnt already and its possible
#asahi-dev · 2026-06-06 22:53
JamesCalligeros[m]I dont think anyones tested it
#asahi-dev · 2026-06-06 22:53
yukaor would the dcp driver disable that already?
#asahi-dev · 2026-06-06 22:52
yukaI think it's ps_disp_fe or sth
#asahi-dev · 2026-06-06 22:52
yukaAnd with dcp, the power domain should be left on at all times?

plus ›