#
#asahi-dev · 2025-02-27 01:10
nickchananyways, metal doesn't support software rendering
#asahi-dev · 2025-02-27 01:10
nickchanand a bunch of stuff though not critical *only* support metal display output
#asahi-dev · 2025-02-27 01:07
nickchani think macOS can use opengl or metal display output but opengl on arm64 macos is nothing more than an opengl to metal translator
#asahi-dev · 2025-02-26 21:52
fl0_idmonkee even if it can use cpu acceleration, from in the past briefly testing x86_64 images of macos on linux vms, you really don't want that, because it will be really slow. Maybe kinda like me trying to run a win 10 x86_64 vm in qemu emulation mode instead of virtualization, cannot recommend...
#asahi-dev · 2025-02-26 21:05
j`eythe OS uses the Metal API.. that's what you'd need to have some implemtation of
#asahi-dev · 2025-02-26 21:04
j`eythere's a text-only mode I think
#asahi-dev · 2025-02-26 20:47
monkeedoes the osx kernel require a gpu or can it use cpu acceleration?
#asahi-dev · 2025-02-26 20:44
svenyeah, but the biggest challenge still is gpu para virtualization
#asahi-dev · 2025-02-26 20:40
monkeeapple has their own asm instructions, maybe not able to execute them can cause kernel panic?
#asahi-dev · 2025-02-26 20:40
j`eyit would require some Metal paravirt stuff, and i dont think anyone has looked at that
#asahi-dev · 2025-02-26 20:38
monkeehi. Is there any progress or known way to run osx in a vm with acceptable performance from apple silicon images? emulating arm
#asahi-dev · 2025-02-25 11:52
chadmedill see what mark thinks when i submit then. it seems to me that it's a consequence of the shared bus and therefore the mask is a description of the board/hardware but mark did question why we arent providing machine drivers a way to set it so i dunno...
#asahi-dev · 2025-02-25 11:40
jannauchadmed: no idea
#asahi-dev · 2025-02-25 10:07
chadmedim not 100% sure if the mask is a consequence of the physical bus arrangement or how the dai is set up by the machine driver
#asahi-dev · 2025-02-25 10:06
chadmedor keep it as a dt property
#asahi-dev · 2025-02-25 10:06
chadmedjannau: do you think that the tas2764 tdm slot mask should be set when the machine driver sets the dai's tdm slot
#asahi-dev · 2025-02-25 10:03
LarstiQnice!
#asahi-dev · 2025-02-25 08:50
chadmed(or a 70% drop in audio-related patches)
#asahi-dev · 2025-02-25 08:49
chadmedwhich is a drop from about 7% of the total patch set to 2%
#asahi-dev · 2025-02-25 08:49
chadmedcleaning up the audio bits branch and submitting the TI codec patches will get us from 114 audio-related patches to 35
#asahi-dev · 2025-02-25 06:47
chaos_princessnickchan: it is in fact my understanding that cpu voltages are locked and can't be controlled by the os. But, for obvious reasons, i can be wrong
#asahi-dev · 2025-02-25 06:39
chadmedwould be interesting if that is not the case
#asahi-dev · 2025-02-25 06:39
chadmednickchan: our understanding is that all of the regulators are controlled by pmgr or something in a way that is completely opaque to the rest of the platform. we just ask for one of the pstates and the platform does the rest with magic
#asahi-dev · 2025-02-25 02:11
nickchanm1n1 deals with it even, but in an inexplicable way according to my understanding
#asahi-dev · 2025-02-25 02:09
nickchan(and on m2 there's a change but a similar interface still exists)
#asahi-dev · 2025-02-25 02:08
nickchanand that interface still appear to exist in m1 looking at various source codes but i can't verify myself
#asahi-dev · 2025-02-25 02:07
nickchancause voltages and frequencies most definitely isn't locked on a7-a11
#asahi-dev · 2025-02-25 02:06
nickchanjust saying things I mean
#asahi-dev · 2025-02-25 02:06
nickchanchaos_princess: are voltages actually locked on m1 or are you actually saying things?
#asahi-dev · 2025-02-23 22:02
leioI guess I didn't use to see the kernel symbols that much in earlier times, assuming it's all good then. With the new sysprof working properly, I would recommend it for perf stuff ;)
#asahi-dev · 2025-02-23 21:59
leioso like if I do stuff that would keep 2 CPU cores busy (but the load is spread out to all with many threads), I have 30% of total accounted for that userspace app, 65% for do_idle/cpuidle_enter and rest on some other background things (and a few % other kernel stuff); that's quite... weird to read still :)
#asahi-dev · 2025-02-23 21:53
leioright, newer app version shows it like perf top but in a way I can track the tree sensibly: https://i.imgur.com/78nvxDS.png
#asahi-dev · 2025-02-23 16:41
leiohah, older kernel I was on didn't support `perf top` yet
#asahi-dev · 2025-02-23 16:27
leio(did not see those kind of things on amd64, so I guess just confused on top of the wrong symbol report under cpu_idle elsewhere)
#asahi-dev · 2025-02-23 16:26
leiokind of counter-intuitive to track perf results :)
#asahi-dev · 2025-02-23 16:24
jannaubut even if not you still would expect a high cpuidle_enter_state sampling if the cpu is mostly idle
#asahi-dev · 2025-02-23 16:22
jannauI assume the time the cpu is idle is counted under cpuidle_enter_state
#asahi-dev · 2025-02-23 16:22
svenno idea either, but my assumption would be that it's just sampling time the cpu is stuck in wfi or whatever
#asahi-dev · 2025-02-23 16:21
leiothe UI also showed cpu_idle, just children there wrong then somehow in my old version
#asahi-dev · 2025-02-23 16:21
leionot sure what the percentages signify in perf, hopefully not that half a CPU is busy going idle :)
#asahi-dev · 2025-02-23 16:20
leiook, I need to upgrade said GUI frontend quite a bit before can blame it too much
#asahi-dev · 2025-02-23 16:18
svenbut sart2_set_entry makes no sense there
#asahi-dev · 2025-02-23 16:18
svensounds like a bug in the GUI frontend tbh. never looked into cpuidle, but cpuidle_enter_state showing up might just mean that well, that cpu is idling a lot :D
#asahi-dev · 2025-02-23 16:17
leiono sart2_set_entry there though, with -g spam either
#asahi-dev · 2025-02-23 16:14
leioshows it as such there
#asahi-dev · 2025-02-23 16:14
leio 50.20% [kernel] [k] cpuidle_enter_state
#asahi-dev · 2025-02-23 16:13
leioI was measuring something else and this stood out instead; I'll reboot to old kernel in a bit too, once I can soon
#asahi-dev · 2025-02-23 16:13
leiomoment, need to install it, but you can take sysprof as a GUI frontend to perf and more
#asahi-dev · 2025-02-23 16:11
jannaudo you see sart2_set_entry in `perf top`
#asahi-dev · 2025-02-23 16:09
leioremains there after closing firefox too

plus ›