marcan changed the topic of #asahi to: Asahi Linux: porting Linux to Apple Silicon macs | General project discussion | GitHub: https://alx.sh/g | Wiki: https://alx.sh/w | Topics: #asahi-dev #asahi-re #asahi-gpu #asahi-stream #asahi-offtopic | Keep things on topic | Logs: https://alx.sh/l/asahi
loki_val has joined #asahi
crabbedhaloablut has quit [Ping timeout: 480 seconds]
yuyichao_ has joined #asahi
malvo has quit [Ping timeout: 480 seconds]
malvo has joined #asahi
phiologe has joined #asahi
PhilippvK has quit [Ping timeout: 480 seconds]
marvin24_ has joined #asahi
marvin24 has quit [Ping timeout: 480 seconds]
<phire>
Names of Firestorm-next/Icestorm-next are apparnetly Avalanche and Blizzard
nskl has joined #asahi
nsklaus_ has quit [Ping timeout: 480 seconds]
<chadmed>
is that the name of the cores in the next A series chip?
aleasto has joined #asahi
chadmed has quit [Remote host closed the connection]
chadmed has joined #asahi
<marcan>
and presumably M2
<marcan>
but I expect M1X will still be icestorm/firestorm
robinp has quit [Remote host closed the connection]
<tophevich[m]>
wow, double the streams double the fun :D
<alyssa>
🍿
<chadmed>
perfect timing for some comfy bedtime wind down viewing :)
<phire>
alyssa: is your talk live or pre-recorded?
<alyssa>
phire: live because executive dysfunction
<phire>
fair
<chadmed>
recording presentations sucks
<alyssa>
srsly
<chadmed>
i had to record a research presentation today because the conference has been made virtual and it took me over 2 hours to get out a 10 minute narrated slideshow of acceptable quality
<chadmed>
just evil stuff
<alyssa>
yep
<sven>
given that i usually only finish the slides (or the demo) maybe a few hours before the presentation i'm kinda sure i'd never be able to record one in time
<chadmed>
yeah the main problem is that a good slideshow is never truly done, and so the effort of having it recorded and prepared early means that its an extra mountain of work to go back and make last minute changes
<chadmed>
like already ive found problems with my recording but im not going to go back and fix it because it will take too much time and the audience is stuffy medical researchers who will probably not even notice
<chadmed>
but it annoys *me* and thats what matters at the end of the day
<sven>
yup
X-Scale has joined #asahi
<alyssa>
chadmed: oh shoot you're reminding me I need to update my slides right now
<chadmed>
alyssa: glad i could be of *some* use :D
<alyssa>
=)
mpaq has left #asahi [#asahi]
X-Scale` has quit [Ping timeout: 480 seconds]
aleasto has quit [Quit: Konversation terminated!]
aleasto has joined #asahi
aleasto has quit [Remote host closed the connection]
aleasto has joined #asahi
robinp has joined #asahi
<dnjmis[m]>
Don't worry alyssa, the drivers work we can hear you fine
<sven>
that's not new fwiw. it's just that SPSS command for the TI chip
<sven>
(possibly "system power state switch")
<alyssa>
sven: I do not envy trying to upstream that branch
<sven>
three commits of that branch are already almost upstream
<sven>
and it'll be two series. one for i2c and another one for tipd
<j_ey>
they were reviewed-by, this morning
<j_ey>
sven: can you try with pinctrl at some point, would be nice to know if there's any issues there
<sven>
and if you take a look at all those tipd commits you'll notice that they are small diffs
<sven>
j_ey: yup, i wanna see if i can do without that "irqs are overrated" commit
<alyssa>
sven: 👍
<j_ey>
sven: and about the power states, it sounds like the other chips do support SPSS, youre just saying 'false' to avoid changing other devices?
<sven>
they support power states, i don't know if they support the SPSS command
<sven>
i'll write about that in cover letter
<sven>
if i can get someone to test with a regular SPSS chip that conditional can go away completely
<j_ey>
cool
<sven>
-SPP
<sven>
arg.. whatever. just get rid of the "SPSS" in that last message :D
<j_ey>
i understood :P
<j_ey>
and I like the use_int2/use_int2 thing
<sven>
yeah. i think the only behavior that's changed for regular TI chips is that the interrupts are explicitly enabled instead of just assuming they are
<sven>
and i can easily add another quirk if the maintainer would like to keep the original behavior there as well
Andalu30 has joined #asahi
<marcan>
all the SMP stuff is pushed, FWIW
<alyssa>
woot!
<alyssa>
Ooh sane boot time thank g-d
<alyssa>
marcan: very nice
<marcan>
CPU freq scaling doesn't work, but apparently the pcores boot pretty fast, or macos puts them there? (just not max)
<marcan>
need to figure out why it does that
<marcan>
it does work on bare metal
<marcan>
could be something I did wrong, or the code actively crippling itself when run as a guest
<alyssa>
i've been meaning to try to understand the corellium cpufreq code
<marcan>
I went through that in my stream at the beginning
<marcan>
before deciding to shave this yak
<alyssa>
👍
<j_ey>
remenants of the great clang-format.py "+ #TODO : turn into a real tracer"
<alyssa>
hah
<marcan>
derp
<marcan>
j_ey: meh, not worth fixing :p
<j_ey>
fix it when resolving the TODO :P
jbowen has joined #asahi
<tophevich[m]>
alyssa: Sadly I couldn't follow your talk live, thus only commenting now, but it was very informative and entertaining at the same time, nice job! Nice touch that you were able to go meta and show it from linux running on an m1 mac mini.
TheJollyRoger has quit [Remote host closed the connection]
aleasto has quit [Ping timeout: 480 seconds]
TheJollyRoger has joined #asahi
<rowang077[m]>
Is the talk online somewhere to watch? I couldn't find it
<rowang077[m]>
thx. It's 2 hours and 10 mins into the stream
<alyssa>
tophevich[m]: thank you!
<alyssa>
oh nice! didn't realize the stream was available immediately
<tophevich[m]>
alyssa: this is because it is still streaming, and you can go rewind in youtube livestreams. But AFAIK they can remove the stream afterwards (and sadly that happens for conferences too often). Hopefully the keep the stream vod, and in addition publish a vod for each talk seperately.
<alyssa>
sure, XDC publishes separate vods later. I guess the "all talks of the day" will be available immediately
<tophevich[m]>
I think we are in luck, I just found the ones from last year so they keep those aswell.
aleasto has joined #asahi
<alyssa>
yeah, got confirmation it'll be up tonight as a full video
<alyssa>
aside - I spent way too long on those slides 😋
<tophevich[m]>
well worth it (from the entertaining side of things)
<tophevich[m]>
Timesink where mostly the emojis I guess?
<tophevich[m]>
*were
<alyssa>
tophevich[m]: I mean
<alyssa>
...actually I made those before starting slides ;_p
<alyssa>
Apple Memoji
<alyssa>
seemed topical
<alyssa>
once I saw it had a witch hat I couldn't not
<tophevich[m]>
ahhh, clever
<alyssa>
most of the time was theming
<alyssa>
I'm used to using company templates as-is, this wasn't a company presentation though
<alyssa>
and didn't want to be Yet Another Beamer presentation
<jn>
ah, the joy of presentation theming
<tophevich[m]>
yeah, they were very unique and fitting without any overload
<alyssa>
❤️
<jn>
(side note: really cool to see people present *on* hardware, in a dual sense. both alyssa and Rob Clark did this today.)
* jn
made a presentation on the Wii U's ARM core once
<sven>
oh... what was that presentation about?
<jn>
said ARM core, and generally how to run code on the Wii U
<jn>
(based on other people's work)
<jn>
the slides were hand-drawn in gimp and RLE-compressed, the "next slide" button was the power button because GPIO is easy and USB is hard
<jn>
anyway, the point is: the topic was also the presentation device
<phire>
poor USB. Everyone neglects it
<j_ey>
not sven >:D
<alyssa>
jn: yo my m1 linux works better than my rk3399 linux the choice is obvious :_p
<sven>
:<
<jn>
alyssa: hehehe :)
<sven>
jn: was just curious because i did a lot of wii hacking and was also involved in the initial wiiu hack :)
<jn>
sven: my contributions to the wiiu scene are negligible. i wanted to run Linux on the PPC, but i gave up when i couldn't figure out how to start the PPC from the ARM, after using the kexec-like syscall in IOSu
<sven>
iirc you actually had to race the ppc to boot unsigned code. it's been a while.
<jn>
hmm
Andalu30 has quit [Ping timeout: 480 seconds]
<j_ey>
marcan: hv with smp works for me!
arahael has quit [Ping timeout: 480 seconds]
arahael has joined #asahi
skali has joined #asahi
<j_ey>
marcan: while you were streaming I mentioned about turning the clocks on/off for the pinctrl, but after trying I dont think we can do it, since the interrupt controller is the same hw, so if we turn it off, we dont get interrrupts for the gpios
<roxfan>
poll then?
<sven>
both the gpio and the pinctrl driver can call clk_enable/disable independently
<sven>
the clk framework does the required reference counting
<j_ey>
I need to look at the rockchip code a bit more, see if the GPIO side ever disable/enables the clocks
<marcan>
jn: something something ps4 ;)
<sven>
i'm not sure if doing it like that makes sense fwiw. i just know that the clock framework does all the reference counting internally :)
<j_ey>
you could enable the clock every time a gpio was set as an interrupt maybe? that would bump the reference counts..
<marcan>
it's not even a clock!
<j_ey>
s/clock/thingy/
<marcan>
let's not bikeshed over *gpio power management* until I figure out what to even do with power management ;)
<sven>
how's that coming along btw? :>
<j_ey>
:)
<marcan>
sven: I made the hv do smp
<marcan>
as part of wanting to work out the cpu pstate stuff :p
skali is now known as skali_
skali_ has quit [Quit: My MacBook has gone to sleep. ZZZzzz…]
<marcan>
unfortunately it doesn't actually control them right now, it seems; either I fail or it nerfs itself in EL1
<j_ey>
now I have 7 penguins on my screen!
<marcan>
hopefully not hard to figure out though
<marcan>
j_ey: ... 7? not 8?
<alyssa>
j_ey: macbook air? :-p
<marcan>
that's the *gpu*
<alyssa>
marcan: hypervisor steals one right?
<marcan>
nol
<marcan>
*no
<alyssa>
marcan: i know :-p
<alyssa>
and oh, nice
<sven>
did you check if the mini even cares about freq scaling?
<marcan>
it used to, for the watchdog, but I just made that optional
<sven>
maybe it's just a thing for the macbooks
<alyssa>
sven: I think so
<marcan>
sven: yes, in bare metal it works
<sven>
ah. weird.
<marcan>
the watchdog is only really useful for me anyway
<j_ey>
marcan: uh, yes, probably 8, I was thinking the wdt was still on
<marcan>
back when I was doing early bringup
<j_ey>
marcan: i didnt actually count them :P
<alyssa>
sven: like, I think some magic registers are needed just to get the performance cores to reach peak perf on the mini? i think
<marcan>
lol
<sven>
alyssa: yes
<alyssa>
sven: or maybe it's just wishful thinking to explain why mesa seems to build faster on macos than linux :-p
skali has joined #asahi
yuyichao_ has joined #asahi
<marcan>
at least macos has the pcores at close to peak though
<marcan>
not sure if that's a default or "something it does in lieu of actually making it dynamic"
<sven>
alyssa: i just thought that the mini maybe brings them up to 100% once and then leaves them along because it doesn't need to save power
<marcan>
like they're at 3GHz instead of the max 3.2
<sven>
*alone
<j_ey>
I switched back to 1 core, it boots a little faster in m1n1/hv
<marcan>
the ecores, otoh, are very underclocked
<marcan>
which explains the HV slowness until now
<alyssa>
sven: given the chips lineage I suspect anything mini specific is an after thought tbh
<sven>
sure, but that would be software only
<alyssa>
I know what I said ;-)
<sven>
anyway, doesn't matter since it's apparently not the case
<marcan>
j_ey: multicore is useful when you're actually *doing* stuff like macos is :p
<marcan>
linux takes a lot fewer instructions to boot ;)
<j_ey>
marcan: :-)
<marcan>
wait hold on
<marcan>
cpu scaling works now
<marcan>
what
<sven>
lol
<marcan>
nevermind lol, it's that I had the interval set to 1ms
<alyssa>
🍿
<marcan>
I'm guessing it clocks them up initially for some reason :p
<alyssa>
boot fasterr
<j_ey>
can I get a platform_device* from a device*?
<alyssa>
j_ey: uh, usually you go the other way around I think
<marcan>
looks good
<alyssa>
j_ey: but maybe to_platform_device
<marcan>
Package Power: 97 mW
<marcan>
I will say this thing has pretty impressive power management even on the mini ;)
<alyssa>
j_ey: yeah, to_platform_device
<alyssa>
marcan: nice!
<sven>
if you _know_ that device is a platform device to_platform_device will work
<j_ey>
alyssa: thanks!
<sven>
marcan: 97 mW. whoa.
<alyssa>
sven: I mean. my 2013 intel macbook air ran on like 4 W for the whole machine, and that was mostly the backlight
skali has quit [Ping timeout: 480 seconds]
<alyssa>
admittedly that was when idle after poking powertop for a few minutes
<mini>
my mini is running a few terminals, discord, a music player and two high res external displays (4k & 5k), and is still only using ~1.2W package power :) they're quite frugal
<mini>
(and would you believe this channel pings me a lot? :p)
<sven>
:D
<j_ey>
you pinged yourself!
<mini>
I need to mute it, but it's always interesting!
<alyssa>
i haven't done anything all day but the talk and it's 2pm
<alyssa>
uhm.
<psykose>
doing things is overrated
<sven>
gah... why is my builder server so slow today :/
<alessandrorzz[m]>
Are you creating the sw In c++?
<sven>
uh no
<j_ey>
sven: give it a good kick
<sven>
i'm sure they'll grant me access to their data center if i tell them i need to kick a server to make it go faster :D
<j_ey>
:D
skali has joined #asahi
<tophevich[m]>
sven: actual metal or virtual server on a badly balanced "enterprise cloud virtualization platform cluster" like vsphere?
<marcan>
btw, now that this works, did a bit of digging and I don't see macOS setting the PSINFO regs like corellium, nor touching that voltage bit thing
<marcan>
I should check on a laptop, but I'm getting the feeling whatever they were doing is obsolete/unnecessary... (bit of a theme with them)
<marcan>
all macos is really doing is changing the target pstate bits in the cluster control register
<marcan>
it does set the psinfo regs *once*
<marcan>
will have to run some tests with benchmarks from m1n1 to confirm that it works as intended
<j_ey>
marcan: considering a lot of this is probably from project sandcastle, I guess that makes sense
<marcan>
yeah, and that firmware changes and gets better and presumably automates more things
skali has quit [Ping timeout: 480 seconds]
<sven>
tophevich[m]: the latter, and i know it's slow because of that. i'm just venting :)
skali has joined #asahi
skali has quit [Ping timeout: 480 seconds]
erincandescent has quit [Remote host closed the connection]
erincandescent has joined #asahi
aleasto has quit [Quit: Konversation terminated!]
<tophevich[m]>
sven: I was asking because of the kicking picture you were painting, and now I see you kicking random racks all over in my mind :P
aleasto has joined #asahi
skali has joined #asahi
aleasto has quit [Remote host closed the connection]
skali has quit [Ping timeout: 480 seconds]
jbowen has quit [Ping timeout: 480 seconds]
jbowen has joined #asahi
<sven>
tophevich[m]: :D
skali has joined #asahi
skali has quit [Ping timeout: 480 seconds]
psykose has quit [Ping timeout: 480 seconds]
psykose has joined #asahi
skali has joined #asahi
skali has quit [Ping timeout: 480 seconds]
skali has joined #asahi
skali has quit [Ping timeout: 480 seconds]
jbowen has quit [Quit: leaving]
skali has joined #asahi
skali has quit [Ping timeout: 480 seconds]
<alyssa>
marcan: cpufreq came early in the corellium tree which means it's probably from sandcastle