#asahi-dev · 2025-04-16 13:26
chaos_princessok#asahi-dev · 2025-04-16 13:26
nickchanchaos_princess: there are always on support in upstream pmgr, it has to be there otherwise the kernel will explode hard on things like the ps_dcs1 (lpddr interfaces)#asahi-dev · 2025-04-16 11:42
chaos_princessWould it make sense to upstream? Do we have always-on support in upstream pmgr?#asahi-dev · 2025-04-16 11:21
jannaupushed as asahi-6.14.2-2 to github#asahi-dev · 2025-04-16 11:04
jannauyes, I have that patch. marking the domains as always on for j293/j493#asahi-dev · 2025-04-16 10:53
chaos_princessid say just add always-on on those and it should be fine#asahi-dev · 2025-04-16 10:53
jannaumipi seems to keep it's as long as ps_dispdfr_be/fe are powered on#asahi-dev · 2025-04-16 10:38
chaos_princessbut that will only allow powering down be/fe, mipi is more complicated#asahi-dev · 2025-04-16 10:38
chaos_princesswe can, just need to make a list of all the stuff that need to be preserved#asahi-dev · 2025-04-16 10:37
chadmeddumb question, can we not just use regcache/regmap to save/restore all registers during pm calls#asahi-dev · 2025-04-16 10:34
jannauI was wondering whether they were powered down due to the added pm callbacks sinc eI don't understand why this wasn't an issue in 6.13#asahi-dev · 2025-04-16 10:32
chaos_princessok, got it now#asahi-dev · 2025-04-16 10:31
jannauI'm getting the irq storm even without it and dispdfr_be and dispdfr_fe are powered of even without pm callbacks#asahi-dev · 2025-04-16 10:28
chaos_princessno, what is not clear is that you said that you added the pm callbacks locally, but how did other people get the irq storm then?#asahi-dev · 2025-04-16 10:27
jannauthe issue is the irq storm, probably caused by powering adp down#asahi-dev · 2025-04-16 10:22
chaos_princessi am not sure i follow, so what is the issue then?#asahi-dev · 2025-04-16 10:21
jannauin adp_drv. just locally for debugging#asahi-dev · 2025-04-16 10:20
chaos_princesswhere?#asahi-dev · 2025-04-16 10:20
jannauworks with always on. I guess it's now powering the domains down because I added pm callbacks#asahi-dev · 2025-04-16 10:17
chaos_princessyea, can't do that#asahi-dev · 2025-04-16 10:16
jannaubut yes, it looks to me like we're powering it down#asahi-dev · 2025-04-16 10:16
jannaulet me try with always-on on its power-domains#asahi-dev · 2025-04-16 10:15
chaos_princesslike, a kb+ worth of registers#asahi-dev · 2025-04-16 10:15
chaos_princessthere is a _ton_ of stuff that needs to be saved and restored for it to survive a power down#asahi-dev · 2025-04-16 10:15
chaos_princessno#asahi-dev · 2025-04-16 10:14
jannauADBE_BLEND_* seems to be messed up after resume as well but saving + restoring them doesn't help#asahi-dev · 2025-04-16 10:14
chaos_princessbecause we can't do that with current driver#asahi-dev · 2025-04-16 10:14
chaos_princessare we by any chance powering the ip block down#asahi-dev · 2025-04-16 10:12
jannausaving and restoring ADP_CTRL in adp_suspend/adp_resume prevents the irq storm but adp is broken after resume#asahi-dev · 2025-04-16 09:48
chaos_princessiirc "scanout in progress"#asahi-dev · 2025-04-16 09:48
jannauany idea what ADP_CTRL 0x2a13 is? tracing now and see that on some reads#asahi-dev · 2025-04-16 09:42
chaos_princessbut yes, there is a free run mode where it constantly scans out whatever is in the buffer, and writing ADP_CTRL_FIFO_ON turns it off and makes it only scan out once you write ADBE_FIFO_SYNC to the fifo#asahi-dev · 2025-04-16 09:39
chaos_princessmaybe? i do not remember exactly#asahi-dev · 2025-04-16 09:37
chaos_princessi think that means it is in "free-run" mode#asahi-dev · 2025-04-16 09:36
jannauADP_CTRL is 0x0412 at that point and writing 0 to it does't change anything#asahi-dev · 2025-04-16 09:32
jannauyes#asahi-dev · 2025-04-16 09:30
chaos_princessthe entire reg is 0?#asahi-dev · 2025-04-16 09:30
chaos_princessno, haven't seen that one#asahi-dev · 2025-04-16 09:28
jannauchaos_princess: have you seen the adp fe irq firing with ADP_INT_STATUS being consistently 0?#asahi-dev · 2025-04-16 07:08
svenand right now I can kinda reproduce it rather easily #asahi-dev · 2025-04-16 07:07
svenIt’s probably not bad for end users since m1n1 either boots Linux or fails but this could get really annoying with the hv #asahi-dev · 2025-04-16 07:06
svenwhich is why I’d really like to track this down #asahi-dev · 2025-04-16 07:06
svenyeah, same #asahi-dev · 2025-04-16 06:34
jannauI suspect a code issue which manifests only with specific compilers / optimisations#asahi-dev · 2025-04-16 06:27
svenI’m not sure this is a compiler issue tbh, but I can try an older clang this evening #asahi-dev · 2025-04-15 22:54
nicolas17if older clang works I volunteer to bisect :P#asahi-dev · 2025-04-15 22:52
nicolas17maybe try an older clang?#asahi-dev · 2025-04-15 20:20
svenassembly looks very different so this might also just be similar to adding printfs somewhere#asahi-dev · 2025-04-15 20:18
svenjannau: ugh.. looks like building with gcc on either a macos or a linux host doesn't run into the same issue#asahi-dev · 2025-04-15 19:12
jannaustill broken. I commented the whole pm_ops