uACPI - a portable and easy-to-integrate ACPI implementation

[gentle peak]

that sounds interesting

does ironclad have a website or where is this list located

[lofty dragon]

https://ironclad.nongnu.org

[gentle peak]

ah sick

[strong heath]

the list you gotta ask me

depends on whats there

[calm latch]

pmOS without UC stuff is faster on 5900X than on 13700 for some reason

[gentle peak]

probably the same locking stuff that proxima suffers from

[lofty dragon]

also #1006539285039697981

for ironclad talk

it also has a discord and a matrix but it has a very criminally underutilised channel here

let's not clobber this one

[strong heath]

Why does this have to happen, I have ACPI

[gentle peak]

seems to me like the first error cascaded into the other ones

[strong heath]

@fiery turtle do you happen to know where VDID actually is? I was checking it at offset 2 in the PCI config space, but it doesnt seem like enough

[gentle peak]

@lofty dragon if you're still available to test, i moved around some ordering stuff and suddenly the performance jumped from 7M to 8.5M so i figured it's worth testing on intel as well

[lofty dragon]

about 6.5/6.6

good improvement

[gentle peak]

that's a lot of improvement for simply moving save_disable_irq from before enable_preempt to after

sometimes i do not understand why the things i do improve performance

[lofty dragon]

lmao

[gentle peak]

this was supposed to just be cleanup

[lofty dragon]

the arcane art of optimisation

[fiery turtle]

What are you doing exactly

Vdid is just the vendor id field

Why do u even care btw

Just return Fs if you believe this device doesn't exist

Or trust the firmware and do what it tells you to do

[strong heath]

returning Fs is disguting

I am just checking VDID manually for it

[fiery turtle]

What was you workaround attempt

[gentle peak]

dealing with firmware is disgusting

[fiery turtle]

Wdym

[gentle peak]

offset 2 is the device id

[strong heath]

I am reading the PCI config space

[gentle peak]

offset 0 is vendor id

[strong heath]

checking for FFFFFFFF in the stuff

[fiery turtle]

Like in device open?

[strong heath]

yeah, in the search function

[fiery turtle]

And what do u do if its FFs

[strong heath]

I return the mockup

[fiery turtle]

That should work

[strong heath]

yeah I was just asking you the offset

because it wasnt 2

[gentle peak]

when does the mockup return Fs?

[strong heath]

turns out its 0x100

[fiery turtle]

Uhhh no

[strong heath]

what do you mean no

[frank canopy]

no its 0

[gentle peak]

the vendor id is at offset 0

[frank canopy]

the vendor id offset is always 0

[fiery turtle]

Vendor ID is the first field

[strong heath]

huh

well that does make it simpler

[frank canopy]

0x100 would imply youre accessing the wrong device i think

[fiery turtle]

BTW you dont have to care about which field, reading from any field of a non existent device will return Fs

[frank canopy]

for port io access that is

[gentle peak]

depends on whether you're using ecam or legacy io access

[frank canopy]

beat you by one message but yeah - its 0x100 for legacy and 0x1000 for ecam

also before i forget, @fiery turtle INFO (uacpi): successfully loaded 1 AML blob, 1708 ops in 1ms (avg 1504436/s) new number for imaginarium when i turn uncacheable to false like its supposed to have been lol

mightve already sent, im very forgetful today

but may as well get it in text since if i did before it was a screengrab

[fiery turtle]

Kvm off 7M is impossible

What are you measuring?

[fiery turtle]

Btw why is it supposed to be off

Oh nvm

Uncacheable false means cacheable

[frank canopy]

its a double negative which i think is what confused me into making the mistake in the first place lol

[fiery turtle]

Yeah ikr

[frank canopy]

500x speedup way bigger than i was expecting tho

[fiery turtle]

Wait I thought it was a legal iso from ci

[gentle peak]

apparently not

[fiery turtle]

Daaamn

Dreams shattered

[gentle peak]

CI is currently building a legal iso which i will test once it is done (usually takes ~20m)

[kindred beacon]

bruh

[fiery turtle]

Yeah thats surprising indeed

[slim panther]

you wanted to do it

[green valve]

My uacpi_kernel_get_nanoseconds_from_boot impl was calculating the current timestamp wrong, that does explain the >100M ops/s with KVM on lol

[fiery turtle]

indeed

[strong heath]

your kernel is just slow

100M is archievable

[green valve]

It's at around ~1.5M with KVM on now, but my notebook is pretty weak (mobile 8th gen i7), I'll also test it at home later (ryzen 7700X)

[fiery turtle]

thats not bad

[gentle peak]

@fiery turtle updated score with legal iso (test cpu Intel 13900KS)
hydrogen 225fc7ff228b70959b9b8a2a5a8cb2a7c0bcc9df (built by https://github.com/proxima-os/proxima/actions/runs/13269265857)

uacpi: info: successfully loaded 1 AML blob, 1705 ops in 0ms (avg 8468599/s)

[fiery turtle]

back to the old score ig?

[gentle peak]

slightly better and more consistent but yeah

[fiery turtle]

yeah

oh well

[gentle peak]

@lofty dragon ^ if you want to test on intel - i made a few final changes which should dramatically reduce the number of lock acquisitions

[lofty dragon]

7.1

🎉

[gentle peak]

still lower than i'd like considering many kernels are much faster on your cpu than mine but it's definitely a great improvement

[lofty dragon]

streaks should try this first though, on his faster 13900KS

my 12900 is slower

[gentle peak]

oh is it?

[lofty dragon]

of course

[gentle peak]

i thought 12900 was better than 5800x

ah you were saying in comparison to 13900ks

[strong heath]

sorry my qemu lost the ability to run with smm off

I cannot run your OS anymore

unlucky

[gentle peak]

this one doesn't break with smm

[lofty dragon]

it's 8.4

[gentle peak]

damn

on the 13900ks or?

[lofty dragon]

yes

[gentle peak]

also do you have an exact number so i can put it in the lb submission

[lofty dragon]

no @strong heath please send a screenshot

[mortal yoke]

Ill also post my thing here in case someone else wants to test

[gentle peak]

thank you!

[strong heath]

I wont

I have stuff to do

ill get to it later

[fiery turtle]

damn

streaks with nasa cpu

[gentle peak]

7.2M

[lofty dragon]

command line?

[gentle peak]

the one i used was qemu-system-x86_64 -accel kvm -cpu host,migratable=off -M q35,smm=off -cdrom image.iso

[lofty dragon]

6.5

you can extrapolate how fast it would be on streaks's CPU based on mine

without bothering him

and it will very very very likely be extremely accurate

[mortal yoke]

around the same as what I get then

[gentle peak]

assuming linear, 7.7M on 13900KS

actually the type of scaling doesn't apply here since it's only 2 datapoints anyway

[lofty dragon]

now i am double curious to double check the result but for getting the last one out of him i was literally screamt at so loud that my ears hurt so

[gentle peak]

so just "assuming the bottlenecks are the same as for proxima"

[mortal yoke]

idk why does the score vary so much tho, like if mint got 6.5 and I think I have also seen that myself (but with an even lower score like 5M or whatever)

[strong heath]

please change your font

[mortal yoke]

and sometimes I get something stupid like 10M lol

[strong heath]

idk who hurt you to do this but get a psychiatrist

[mortal yoke]

its a font that staticsaga sent at some point lol (iirc at least)

[gentle peak]

the font isn't that bad imo the main issue is the double newlines

[mortal yoke]

thanks for testing it tho

[mortal yoke]

that's an issue with the uacpi glue, I do a println in there not realizing that uacpi adds newlines itself lol

[frank canopy]

what is the ideal cache type for what uacpi maps btw?

rn im using writethrough

[dense steppe]

WB is sufficient on x86 because of MTRR

[frank canopy]

oh cool

[dense steppe]

no... cursed because IMO caching should be explicit

[frank canopy]

lmao

agreed but also makes it easier imo

[dense steppe]

but there's no easy way to get the correct caching mode to apply to a memory range on x86, besides maybe getting attributes from the EFI memmap

[frank canopy]

RIP

[dense steppe]

(that I'm aware of; I'd be delighted to be wrong)

[frank canopy]

i mean you can probably read the MTRRs

but also

[dense steppe]

double cursed

[frank canopy]

yeah

[gentle peak]

you can get most of the way there by just determining whether the given address is in a usable or acpi reclaimable memory region

obviously prefer the attributes from EFI mmap if available but it's a decent fallback

or on x86 scan the mtrrs but at that point just map as wb

[lofty dragon]

the last thing you said is unironically brain damage sorry lol

if anyone does that they are just doing things to look smart

[lofty dragon]

usable?

i don't think it should ever map anything from a usable entry though

are you sure you didn't mean reserved/non-present-in-memmap?

where you map that as UC and you map the ACPI reclaimable as WB?

[gentle peak]

yeah exactly

[gentle peak]

I meant wb when usable or reclaimable, UC otherwise - of course, acpi should never touch usable memory, but wb is the correct memory type for it and you never know what crazy firmware dev hacks you might have to accommodate

[dense steppe]

if AML tries to access anything within usable, kill it.

[lofty dragon]

i don't think it should ever ever ever touch usable entries

that's insanity

like literally kill it

[fiery turtle]

aml can map acpi reclaim memory

and acpi nvs memory

both are normal ram

[lofty dragon]

yeah

those are WB

[fiery turtle]

ye

[gentle peak]

idk I wouldn't be surprised if there's systems out there that try to detect the os by some magic signature that it happens to put in a fixed location in physical ram

[dense steppe]

like maybe you can be lenient and allow it access into <1M usable regions that the kernel otherwise doesn't use

[fiery turtle]

lol probably

i mean linux allows aml to map itself

[gentle peak]

it's awful and please kill it but it's out there probably

[fiery turtle]

as RO

[dense steppe]

if this, get better firmware

[gentle peak]

side note but I don't understand why so many people blindly ignore all memory below 1M, if not doing so causes problems it's a bootloader bug

[fiery turtle]

too much work for little gain

[dense steppe]

I don't think I ignore low memory atm.

I currently only reserve like the zeropage

as in

page zero

[lofty dragon]

ok so map ACPI NVS/reclaim as WB, usable as WB/RO (forced), any other region as UC

[gentle peak]

I don't even do that

[lofty dragon]

i think this is reasonable

[gentle peak]

if the bootloader tells me a page is usable I will use it

[mortal yoke]

or just map everything as wb and let it be handled by the mttr's

[gentle peak]

yeah that sounds reasonable

[dense steppe]

maybe read the MTRRs and print a warning if it doesn't match what you expect

[lofty dragon]

no

that's just fancy shmancy bs

too fancy for me, waste of time (do it if you want idc lol)

[gentle peak]

it doesn't do anything except maybe indicate firmware bugs (wrong setup of mtrrs)

[lofty dragon]

yeah in which case it's not like you can do much anyways

[dense steppe]

it is better to be aware of firmware bugs than to keep yourself in the dark.

like

[gentle peak]

but even that is dubious since memory map based detection will almost certainly falsely mark some regions with a more restrictive cache mode than necessary

[dense steppe]

at least you can scream at the firmware devs

[lofty dragon]

btw RE: why some people ignore memory below 1M: i guess because GRUB and not trusting the BIOS/more primitive bootloaders

any proper sane bootloader with a spec like e.g. Limine's will guarantee things that don't take into account below or above 1M

so you shouldn't really care either

[gentle peak]

makes sense

[fiery turtle]

daily ad complete

but yeah true

[lofty dragon]

i mean i assume it's the same for Hyper and Ultra

[fiery turtle]

it is

it allocates backwards from max phys anyway so i doubt u would see anything below 1m allocated

but its marked as loader reclaim

[lofty dragon]

Limine allocates itself below 1M on BIOS + the SMP trampolines

[fiery turtle]

oh yeah hyper also allocates itself

nvm

[torpid root]

@fiery turtle

[north holly]

infy you're such a relocatable ELF file

[fiery turtle]

i have tons of stuff scheduled for the leaderboad now

btw ironclad should be added too now or?

[fiery turtle]

id hate myself if i was one

[lofty dragon]

not yet

[fiery turtle]

5M not enough?

[lofty dragon]

streaks isn't happy with it yet, and i don't think he has committed most of the changes in a final form yet

[fiery turtle]

ah sure

[hallow marten]

gib list

[kindred beacon]

@fiery turtle you haven't posted about the 1.0 release in the osdev.org thread

[fiery turtle]

True

Idk if anyone even looks there tbh

But I should anyway

[kindred beacon]

yeah there were two recent replies

[fiery turtle]

Oh yeah? Haven't looked there in months lol

[fiery turtle]

Thanks for replying btw

[kindred beacon]

obligatory mlibc shilling included

[strong heath]

I have fixed all timing innacuracies in Ironclad

and the score stabilized at 5 million

it does get very slow on real hardware though, because the allocator behaves much better in small bursts

[surreal current]

what was causing them

[strong heath]

from 5 million to 1

TSC innacuracies and bad calibration code

TSC is a mess for wallclock, I just moved it to default to HPET for those usecases

[strong heath]

ill publish it later today along with conditions

[slim panther]

wait your uacpi is slower on real hardware??

[strong heath]

yeah

but that is because of my code

[slim panther]

hm

weird because on real hw im around twice as fast

[median crest]

I mean it depends on the aml it runs

[slim panther]

similar amount of ops

1900

i think

[median crest]

Yes but different ops probably

[strong heath]

thing is, as I said, Ironclad's slowest link is the allocator

and its because its bursty with the latest changes ive done

so smaller AMLs -> less allocations

less allocations -> shorter loads

real machine AML is much much bigger, thus it gets slower

[median crest]

Damn ironclad got 5 mil

I should ask infy to remove astral this is EMBARASSING my score is so low I gotta lock in 😭

[median crest]

Is it bursty to be real time?

[slim panther]

generic unix clone gets generic score /j

[median crest]

So true

Did infy do any big perf improvements on 1.0 I havent tested in so long

[strong heath]

it is bursty because of how the algo works, it doesnt have to do with real time

[median crest]

Oh

Have you written anything about the funny seal os anywhere Im curious about what you do

[strong heath]

I do talks about the OS regularly in big events

if you are curious about the internals its a good starting point

https://ironclad.nongnu.org/community.html

[median crest]

Epic Ill look into it at some point

Right now I am in "alarm set to 2 hours from now" territory which means I should probably sleep

[strong heath]

good night mate

[fiery turtle]

You really cant compare the runtime of two completely different aml programs lol

Like it can literally call sleep() and your score will go down

[slim panther]

lmao

[fiery turtle]

I'll probably add this comment in pci_device_open

might be a bit too verbose but

at least it explains well why it's needed

[slim panther]

i forget, is uacpi-rs a thing yet

[left orbit]

i dont believe so

i think there was an attempt by @\shinribo or smth

idr the username

[fiery turtle]

it's definitely a thing, just not completely finished yet

[slim panther]

ah

i'll use bindgen for now then

[fiery turtle]

it's linked in pins

[slim panther]

thx

[rustic compass]

im working on it

[fiery turtle]

no more 99999999999999999999 ops/s

[left orbit]

it will not print the stats if timer returns bad values?

[fiery turtle]

yup

[left orbit]

noooooooooo

how will i surpass proxima now

[fiery turtle]

static int invocation = 1;
return invocation++;

[dense steppe]

1705 ops in 0ms (avg 1705000000000/s) letsgo

[torpid root]

this post was fact checked by real osdevers:

[lofty dragon]

Vinix will be #1 just you wait :^)

[dense steppe]

I'm sorry, I cannot let you do that

[median crest]

@fiery turtle

[fiery turtle]

depends on your current rev

[median crest]

Like 2-3 months old

[fiery turtle]

yeah dunno

i was gonna update astral anyway

soon ™️

[median crest]

I will continue working on astral soon™️

If the urge to rrwrite doesnt overcome me

[fiery turtle]

what do u want to change?

[median crest]

Mostly just fix the unreadable code and actually make good designs

[fiery turtle]

well ngl sounds tempting

if u unfuck the build system as well

[median crest]

Yeah

It would take so long to get back to where I am which sucks but eh

Especially now that I dont have as much free time as a a year ago

[fiery turtle]

yeah

[median crest]

I would probably snatch some more recent stuff over like the new locking code

[fiery turtle]

yeah you could probably copy over a lot of things

just rewrite what actually matters

[median crest]

Yeah Ill think about it more and see if its something I really wannaa do

[vast kestrel]

Is there any aarch64 board with full acpi? And not just hardware reduced acpi?

Also can you configure qemu with full acpi?

[fiery turtle]

not to my knowledge

full acpi is only x86 and itanium afaik

[vast kestrel]

When running on hardware reduced do I need to choose it at compile time, or would uacpi compiled without the ifdef still work?

[north holly]

you need the compiler define iirc

[kindred beacon]

@slim panther was pumped about getting some aarch64 board with UEFI today, I don't recall whether it had ACPI tho

orion o6 has ACPI afaict?

not sure whether it's hw reduced or not

[vast kestrel]

I think it might be nice to have runtime support for both hw reduced and full acpi if it's not too much. I think the fadt has a detection bit for it no?

[gentle peak]

I'm pretty sure the define is for when you want a uacpi version that's only capable of handling reduced hw, and the full version can still handle reduced hw machines

[vast kestrel]

I see, nice

[slim panther]

it does

it came today

ill hop on vc tonight

[kindred beacon]

based

[slim panther]

70,34€ taxes

[kindred beacon]

bruh

[calm latch]

btw did you get a serial adaptor for framework laptop?

[slim panther]

yes

need to solder the pins

but otherwise yes

[fiery turtle]

It would autodetect it of course

[left orbit]

the define is there to compile out non-hw-reduced stuff i assume

so if you know you will only run on reduced acpi hardware you can use that?

[fiery turtle]

Yeah just so you can strip some code on aarch64 etc

[left orbit]

aarch64 is hw reduced?

cringe

[fiery turtle]

Yeah

[left orbit]

what does that even mean

[fiery turtle]

No events and stuff like that

[left orbit]

oh

so cringe

got it

[fiery turtle]

Yup

[left orbit]

but u still get power states and shit

[fiery turtle]

Nothing stops them from shipping a full acpi implementation I think tho

[left orbit]

u can suspend, sleep and all

[fiery turtle]

There just hasn't been one yet Iirc

[fiery turtle]

Yeah

[left orbit]

nice i see, thanks

[fiery turtle]

Np

[vast kestrel]

All I asked for is an acpi timer for calibrating the systick 😔

[flat badge]

On arm, you can usually just get the timer frequency from the dt

Doesn't acpi have an alternative mechanism for that?

[hollow elm]

you can even get it from a system register

[flat badge]

Ah i see

[hollow elm]

cntfrq_el0

[vast kestrel]

Til

Yeah but I hate dt

[vale isle]

same on risc-v

timebase-freq

does acpi have a mechanism for that in risc-v? afaik there's no other way to get that

[gentle peak]

yes

it's in rhct iirc

[fiery turtle]

Just implemented "support" for the cursed Unload opcode, which has been deprecated and removed quite a long time ago, but now at least it won't be treated as invalid, and will instead be a no-op

[median crest]

stop being a COWARD and unload the table smh

[fiery turtle]

this is what ACPICA does

and then proceeds to actually unload the objects

so if you have a reference to some node you will just UAF it

[median crest]

Unload( GDT )

[fiery turtle]

Unloading GDT... SUCCESS

[sterile egret]

I remember someone saying that the Surfaces have a lot of weird AML, do you know of other examples off the top of your head?

[fiery turtle]

yeah, it uses some esoteric operation region types

[loud ice]

LMFAO what

[fiery turtle]

at least it logged a warning right

[fiery turtle]

lol

ah

"vacation"

[vast kestrel]

[fiery turtle]

thats why my PR hasnt been reviewed in a month i guess

[sterile egret]

Lmao

[hollow elm]

lol my friend has a laptop where linux fails to shut down due to acpi shenanigans

seems to be stuck in an infinite loop with a Stall(118) in the middle since acpica complains about the stall time exceeding the maximum

[dense steppe]

Unironically I think my 2024-ish desktop (a damn good computer) has ACPI problems. Last time I tried, reboot didn't work. It just kinda gets stuck somewhere. Shutdown works fine though.

Note that this was on Linux/ACPICA and not a hobby kernel/uACPI or whatever. I'm unsure if NT reboot works.

[fiery turtle]

Bruh

[dense steppe]

Actually no, it doesn't. When doing Windows updates and it wants to reboot the computer, it turns off video output and just kinda sits there, spinning and not turning off. However with a short press of the power button, it powers off.

[fiery turtle]

Would be nice to test e.g. Managarm on it

[calm latch]

you can try pmOS

though it has a major timers/scheduling issue atm, so it might be broken on its own so potentially not representative

[hollow elm]

not gonna bother him with this now

but i just found it funny that you can apparently fuck up shutdown

it shouldn't surprise me since it's arbitrary aml and yet it did

[sterile egret]

What's the status of uAPI?

[fiery turtle]

marvin is the "maintainer" of uAPI

so u should probably ask in the uDRM thread

[urban galleon]

Yeah on my system76 system linux entered an inf loop when trying to suspend to ram

[vale isle]

an infy loop???

[median crest]

Red circle red arrow surprised face exclamation mark exclamation mark

[north holly]

https://tenor.com/view/i-saw-what-you-deleted-cat-gif-25407007

[slim panther]

https://cdn.discordapp.com/attachments/1287456798407790684/1290809675943776298/assert.gif

made asserts not be there in release mode

[median crest]

I should do that some day

[north holly]

is Astral secretely an assertos clone???

exposed!!!

[slim panther]

assertpilled

[median crest]

always has been

if you disable asserts it wont work

[north holly]

that sounds bad

[median crest]

because I do som,e things like assert(shitpiss_init() == 0)

[north holly]

now that I remember, I do stuff like that too

[fiery turtle]

I like how u even made sure the code style is correct

[north holly]

#if OBOS_DEBUG
#    define OBOS_ASSERT(expression) do { if (!(expression)) { OBOS_Panic(OBOS_PANIC_ASSERTION_FAILED, "Assertion failed in function %s. File: %s, line %d. %s\n", __func__, __FILE__, __LINE__, #expression); } } while(0)
#else
#    define OBOS_ASSERT(expression) do { (void)(expression); } while(0)
#endif```
but I worked around it like this ![meme](https://cdn.discordapp.com/emojis/575445094589661215.webp?size=128 "meme")

[median crest]

soon all of astral will be snake case

wait infy

what do you think of the style I've bee going for

https://github.com/Mathewnd/Astral/blob/rewrite/kernel-src/lib/pushlock.c

like here

[north holly]

wait what does for (;;) do again

I forgot

uh

infinite loop maybe

[median crest]

for(all;of;eternity)

[fiery turtle]

Looks generic enough

[median crest]

I supposethatsbetterthanbefore

[north holly]

I will fix my code style one day™️

bro what the fuck happened to the ™️ emoji

[fiery turtle]

yeah

[jaunty fox]

uacpi info: successfully loaded 1 AML blob, 1705 ops in 0ms (avg 4013691/s) i'll throw in this updated submission for keyronex

inlined spl routines got that

[fiery turtle]

very cool

[calm latch]

Can I submit this?

[fiery turtle]

i dont remember the context

is it legit or not

[calm latch]

It should be

I can double check

[fiery turtle]

if it is u can submit it sure why not

[jaunty fox]

it was a stroke of genius for you to add this statistic to uacpi

now as well as an allocator test uacpi is an osdev benchmark

[fiery turtle]

lol yeah

it started as a gimmick i wasnt even sure i should've went with

[dense steppe]

I like how the uACPI benchmark has become a de-facto standard

[frank canopy]

it pointed out an issue with my code too but I was too dumb to realise that it was an issue worth investigating until I accidentally fixed it later lmao

[fiery turtle]

I think it's unironically a decent performance wellness check for a kernel

[frank canopy]

yeah

I don't care about where I am on the leaderboard but I sure do like having the number as a quick general wellness check

[calm latch]

Is calibrated frequency 104978500 vs -cpu tsc-frequency=100000000 acceptable?

This gives this result

[fiery turtle]

could you fix your calibration? idk if thats too big of a skew

[calm latch]

It's consistently above 5 million though

[fiery turtle]

its like 5% faster?

[calm latch]

I can hardcode it...

[fiery turtle]

thats fine for the sake of the submission accuracy

[jaunty fox]

try using the kvm clock

it's cheap and easy

[median crest]

^

[dense steppe]

I think you should rework your TSC calibration code in that case.

that's like 5% off

[calm latch]

I'm planning to eventually replace my timers subsystem

So I'm not in a mood of changing it now

[dense steppe]

sure

out of curiosity, do you calibrate against PIT or against HPET (or against something else)?

[slim panther]

i'd also like to submit a new score

[median crest]

how are yall so fast 😭

[slim panther]

what likely(x) does to a mf

[calm latch]

With hardcoded 1:1 tsc to ns

[fiery turtle]

letsgo, same cpu as last submission?

[slim panther]

ye

[median crest]

I tried adding that around but saw no improvement whatsover

weird

[fiery turtle]

can u paste the score as a number

[dense steppe]

I calibrate against HPET with a fast calibration loop of like 50ms and I consistently get measurements within like 0.1% of my CPU's clock speed

[slim panther]

i got a 200k ish boost from putting that all over my allocators

[dense steppe]

feels like off-by-one on the amount of zeroes somewhere

[calm latch]

53 403 075

Yeah

[dense steppe]

or is 53M a new record?

[calm latch]

Nope

It can't be

[dense steppe]

✨ blazingly fast ✨

[fiery turtle]

ah u fucked it up lol

[calm latch]

tsc-frequency=100000000

5354445 on Ryzen 9 5900X

[calm latch]

although with wsl2 it's it's fluctuating a lot

[dense steppe]

we optimize the shit out of uacpi_kernel APIs instead of actually doing useful work

[calm latch]

idk if that's acceptable

[median crest]

@slim panther can you try this out #1061407633745125397 message

my laptop is shit and wsl doesnt help

[fiery turtle]

If you can get this consistently

[calm latch]

it's repeatable

but sometimes dips to 4.9M or so

[dense steppe]

blame it on the host

[median crest]

better than wsl who decides that sometimes it'll give me 1.7mil and sometimes 800k

[calm latch]

I'm using WSL

[fiery turtle]

Thats fine

[median crest]

wtf

[calm latch]

or 4.6

[dense steppe]

do you have a Ryzen whateverX3D CPU or similar?

[median crest]

damn its even worse now

[slim panther]

i'm on intel

the virt is worse than amd

[median crest]

i510300h

[calm latch]

I don't have x3d

[dense steppe]

is that CPU asymmetric in some way? (efficient vs perf cores, or different cache)

[median crest]

I dont think so

[dense steppe]

I was a bit surprised to learn that on some X3D CPUs, half the cores (those with more cache) run at a lower clock speed

and so what CPU QEMU gets scheduled to affects the benchmark

[calm latch]

It's not just 3D cpus

With AMD and chiplets, if you have Ryzen 9, half of the cores are just slower

[slim panther]

could anyone with an amd cpu test this image please https://discordapp.com/channels/440442961147199490/1287456798407790684/1340475807566401548

ovmf required

qemu-system-x86_64 -smp 1 -drive format=raw,file=menix.img,if=none,id=disk -device nvme,serial=deadbeef,drive=disk -cpu host -accel kvm -machine q35,smm=off -bios <ovmf>

[median crest]

ovmf

[dense steppe]

iirc I do -drive if=pflash,index=0,file=/usr/share/edk2-x64/OVMF.4m.fd,readonly (or something similar)

[calm latch]

@fiery turtle

[slim panther]

that's better ye

[calm latch]

(On Ryzen 9 5900X)

[slim panther]

lmfao

i tried running uacpi on my pc with the broken aml for the first time

it's apparently still parsing the blobs

after like 2 minutes

doesn't timeout either

[frank canopy]

of curiosity, how can you know what PCI segment(s) exist with any certainty before the blobs are parsed and namespaces initialized?

[slim panther]

i dont

[frank canopy]

ah ok, just assuming always 0 then?

[slim panther]

yes

[slim panther]

does this really matter though

[frank canopy]

in practice usually not

segment 0 bus 0 being the root is standard but its a do you trust firmware devs thing tbh

[slim panther]

never

i would like to know wtf is going on here though

[frank canopy]

it still not finished? if so dang

[slim panther]

i could make it print to console every 5 or so seconds to see if its still alive

[frank canopy]

impressively whack aml there

[slim panther]

i know the aml is borked because acpica has a stroke on boot

does uacpi print parser errors?

[strong heath]

@fiery turtle id like to report the first score obtained from an actually comitted Ironclad kernel for leaderboard consideration, booted with qemu-system-x86_64 -M q35,smm=off -smp 1 -m 4G -drive format=raw,file=gloire.iso -serial stdio -enable-kvm -cpu host

I hope to soon bump it to top 3, but being faster than managarm is good enough for me

[median crest]

the seal grows

[fiery turtle]

Nah that's not possible

Uacpi has loop timeouts

[fiery turtle]

Yeah

I think your kernel just hung

Can you try running e.g. pmos or something on it

Also post the dump I guess

[fiery turtle]

Very cool

[fiery turtle]

Btw what should I put for the description

[strong heath]

formally verified, hard real-time capable kernel written in SPARK and Ada.

[fiery turtle]

im thinking of adding a uacpi version printout as the first thing it logs

thoughts on this?

so it's easy to see which version a person is using by their logs

I'm also making registers.h public API as apparently there are cases when a kernel needs to be able to write arbitrary FADT regs

since haiku uses that for some cpu idling stuff

I'm also going to add something like uacpi_namespace_node_get_next for people that dont want to use for_each_node (like haiku)

[left orbit]

yaaaay iterator api

[fiery turtle]

which CPU was this on?

[strong heath]

sure

[fiery turtle]

bruh so many new osses and leaderboard changes this time

[fiery turtle]

do u want me to link the github or the original link or maybe the website

[vast kestrel]

+1

[strong heath]

ironclad.nongnu.org

[vast kestrel]

I suggest maybe even printing a shortrev

But for the love of god cut it manually and not using the actual short rev command

[fiery turtle]

that might be tough as uACPI currently requires very little from the host build system

[vast kestrel]

Hmmmm true

[fiery turtle]

otherwise id of course add -dirty and stuff

[vast kestrel]

I mean, shouldn't it be basically CFLAGS += -DUACPI_REV=$(shell git -C ...)

[fiery turtle]

yeah

[vast kestrel]

And anyone not using make would probably be able to figure out how to convert it to his favorite build system

[fiery turtle]

still this will break every existing build

[vast kestrel]

True

[mortal yoke]

you could just check if its defined

[fiery turtle]

perhaps

[vast kestrel]

If it's not defined you could either have an empty string or something

Like, at work we always print both the version and the revision

[fiery turtle]

ill probably just have this in the header at first

[vast kestrel]

Since we have had cases of stuff being compiled not from master of people actually just forgetting to increment the version lol

[fiery turtle]

lol

https://github.com/uACPI/uACPI/tree/misc-assortment?tab=readme-ov-file#notable-projects-using-uacpi--performance-leaderboards

this is the updated leaderboard (not yet merged) pls verify if you made a submission, I might've forgotten something

[mortal yoke]

I should also make a submission at some point but I want to get above davix again the average score that I have now is 6M because in the old score some cr8/gs writes were getting optimized out (it was actually a lot worse than this but I got rid of unconditional cr8 writes in lower irql and instead do a cli + check if cr8 was modified + write if it was + sti)

[fiery turtle]

lol sure

davix might score higher than proxima tbh

if its run on a 13900KS

[gentle peak]

probably not, davix's current score was measured on a 13700K, which is better than my 5800X (and on my 5800X that proxima build gets like 7.5M iirc?)

[fiery turtle]

maybe not then

[jaunty fox]

i am still in favour of someone acting as a clearinghouse for scores using a consistent testing machine and environment

[gentle peak]

yeah same

[fiery turtle]

if there's a volunteer for that with a good CPU im down

[jaunty fox]

ryzen 5800x with the turbo boost what-you-may-call-it not turned on

[mortal yoke]

though idk if its really worth it for now tbh, like its basically good enough™️ and I have other things I need to work on like APCs so I can finally get to run some proper userspace lol

[left orbit]

my 13700k also is a bit nerfed because wsl

i need to move to native linux

[gentle peak]

good point

i might try the davix iso myself later just for comparison

[left orbit]

man i hate linux desktop tho

[fiery turtle]

same

[left orbit]

i might just virtualize windows

[fiery turtle]

it just cant handle scaling man idk

[torpid root]

get a better DE

[fiery turtle]

i've tried every DE there is

[left orbit]

i might have to patch the linux kernel for anti cheats to work properly but oh well

[torpid root]

plasma 6.3 just got scaling improvements

[jaunty fox]

every time the linux desktop is about to get good they release a new major version of KDE, GNOME, GTK+, QT, etc and go back to zero features

[left orbit]

i might try wayland this time

[fiery turtle]

should probably check it out then

[mortal yoke]

does it now support different scaling between displays and dragging windows between them?

[left orbit]

lack of native discord screenshare was a big thing that prevented me from trying it out

[fiery turtle]

this is the problem that i have

[torpid root]

I haven't checked as I only have a single 1080p monitor

I had a second monitor hooked up one time and I could drag windows between monitors

[mortal yoke]

ig I can try to update to it and see, would be nice if it would

[gentle peak]

screen sharing seems to break every other discord update and get fixed the next, but a decent workaround is using the web app

[kindred beacon]

GNOME on Wayland 😋

[torpid root]

switching to vesktop fixed all my problems

[left orbit]

of course i'd try gnome

[kindred beacon]

discord has screensharing now tho

[left orbit]

well

not try

[strong heath]

except mate

[left orbit]

i am a gnome user, just not wayland

[strong heath]

mate remains the bulwark

[jaunty fox]

that's why i keep to it

[strong heath]

thats why bonkers is porting it to Gloire

[jaunty fox]

it just works and they don't badly reinvent the "uSeR eXpErIeNcE" based on nothing but vibes every 3 years

[vast kestrel]

I don't mind doing so (I have i9-13900k), as long as its just downloading a binary and running it in qemu with some basic flags

But if I do that I would prefer to do it in bulk, and ideally have it be printed to serial or debugcon

[dense steppe]

just read the backlog a bit. feel free to ping me whenever you are discussing my kernel (davix)

also I still haven't upstreamed any 'fix' for the AMD skill issues that my kernel has (mapping memory as UC- in uacpi_kernel_map). iirc this doesn't affect the score on Intel CPUs (most likely has to do with how caching modes are handled on SVM and VMX respectively)

[strong heath]

I will do it with my power station CPU with no conditions

But

I will cook results so ironclad is top 1

You will have to accept that

[gentle peak]

seems fair

all other kernels are tested while compiling gcc

[dense steppe]

your "nuclear reactor CPU", "really fkin powerful CPU" etc. has been mentioned a lot. what CPU is it?

[left orbit]

13900ks probably

[strong heath]

128 core threadripper

[dense steppe]

[left orbit]

this shit has bad single core performance tho

if you do actually have one

[strong heath]

I don't

[left orbit]

not very good for uacpi benchmarks

[strong heath]

Its the KS

[left orbit]

very specific

so i was probably right

or 14900ks

[dense steppe]

anyway I need to select a nice console font for my kernel's fbcon lol.

[fiery turtle]

will you be available fo reruns every now and again?

maybe we should have a github issues page where people would submit their ISOs

idk

[dense steppe]

I should really setup CI or something for building davix ISOs

but I won't right now (because I'm lazy)

[rustic compass]

how to get the best uacpi score:

1. have friends that hold the world record for a overclocked x86_64 cpu
2. ask them to repeat the record with your os running in qemu
3. success

[dense steppe]

also it's literally just as simple as ./run -M q35 to build and run the kernel so it's not rocket science generating an ISO yourself

[fiery turtle]

@slim panther would this comment be sufficient to understand what is meant by this helper?

[left orbit]

very nice

if you can't understand that then you are brian damage (half joke)

[dense steppe]

"On x86, the SystemIO address space corresponds to port IO." is what I would've written, but this definitely works too

[fiery turtle]

port io seems a bit overloaded no?

[dense steppe]

maybe

[fiery turtle]

so many things have ports

but yeah idk

[dense steppe]

yeah actually this is probably better

zero probability of a misunderstanding (unless brian damage)

[rustic compass]

"On x86, the SystemIO address space corresponds to the port IO added with the i386."

[vast kestrel]

Yeah of course, especially at weekends

I'll probably automate it just so it happens automatically lol

[fiery turtle]

could be nice

[vast kestrel]

Given it should just run qemu | grep avg

[gentle peak]

i8086, actually

[fiery turtle]

like a website where u submit an image and it automatically posts on github issues lol

[vast kestrel]

:^)

[gentle peak]

that sounds like asking for things that exploit vm escapes

[vast kestrel]

Tbf I don't run my PC 24/7, too much power lol

[fiery turtle]

it could schedule images to run ig

[vast kestrel]

But yeah I would rather it's images given directly by people we know on the server

[gentle peak]

also, maybe running in docker or similar would be a good idea, to keep qemu version constant

and/or keep the latest tested image for a given kernel around to retest if qemu is updated

[vast kestrel]

Tbf even kernel version can change it

But I can just say what the kernel and qemu version I am using are

[gentle peak]

yeah so linux or qemu update triggers global retest

[vast kestrel]

I just have a standard Ubuntu 24 lel

[fiery turtle]

yeah maybe u could store the latest iso for each leaderboard entry

[vast kestrel]

I might switch to Fedora eventually but yeah

[rustic compass]

if security is a concern one could run it on a independant machine with a configured linux with custom secure boot and full chain of trust

or just auto reinstall after each run

[vast kestrel]

I do have secure boot, but I ain't trusting Linux usermode enough for that

But having a dedicated computer just for that is a bit expensive

[left orbit]

oh i am so tempted right now

[vast kestrel]

Like we could rent a server but at this point...

[fiery turtle]

that would be nested virt

which would hinder perf

[left orbit]

u can rent a dedicated server

[slim panther]

yes

[fiery turtle]

thats probably a bit too expensive lol

[slim panther]

i have a powerful vps

[rustic compass]

we could for example use a n100 board as the reference server, cheap and serial interface

[left orbit]

eh $50-60 a month isnt too bad

but yeah

[gentle peak]

imo just "restrict it to manually submitted images from people known from this server and well-established oses built locally" provides good enough security

[slim panther]

4 ops/s

[left orbit]

still i am so tempted to build a thing where u can submit isos and test then

something like managarm xbbs

[fiery turtle]

that would be cool

if it also automatically made PRs for leaderboard updates lol

[left orbit]

that would be easy enough

[rustic compass]

good enough

[vast kestrel]

I can have the script just look at issues and create pull requests on demand

[vast kestrel]

But I agree

[gentle peak]

also i don't think nested virt would be that bad for this, sure it'd be quite a bit slower than regular virt but the most important thing is environmental consistency

[left orbit]

boring

[vast kestrel]

Also given it's notable kernels, aka not every random on the server but the bigger projects on the server

[slim panther]

technically, wouldn't an actual benchmark using uacpi kernel api be awesome? like seperated into mutex, alloactions, etc

[fiery turtle]

a virtualized env is affected by neighboring VMs

[gentle peak]

ah true

[fiery turtle]

even for the memory bus bandwidth etc

[visual bridge]

t

[rustic compass]

but if it is a virtual server then the hoster could do all sort of stuff with the underlying hardware

[visual bridge]

so much activity here

[vast kestrel]

I would run it one at a time obviously

[gentle peak]

nah, test all the kernels at once

[fiery turtle]

in a virtualized env u would basically get wsl2 skewing all the time

[vast kestrel]

Also one thing to think is how much memory and how many cores we give it

Like not all kernels support smp, and some kernels like smp more than others

[gentle peak]

-smp 1 -m 4G is the standard for the benchmark i believe?

[rustic compass]

is a n100 board actually that bad, i wanted to buy one to tinker with it

[fiery turtle]

yeah

[vast kestrel]

Then I might want to pin it to a p-core as well

Some poor soul would randomly get an e-core for their benchmark :^)

[gentle peak]

also worth thinking about is how to make sure host scheduling doesn't affect it

i have a homogenous cpu and even i have multiple very different performance ranges

[rustic compass]

also try to pin the frequency

[fiery turtle]

ideally u want to isolcpus=

[vast kestrel]

True

[left orbit]

taskset

[gentle peak]

taskset isn't enough

[left orbit]

good enough

[vast kestrel]

Cgroups has a way to isolate a core

[gentle peak]

it restricts it to a specific cpu but that still leaves things like cpufreq

[fiery turtle]

yeah u just want to isolate a cpu, then allocate a cgroup, pin that cpu to it, and then run qemu there

[dense steppe]

we are really, really overengineering the benchmark right now lol

[rustic compass]

maybe disable variable frequency in uefi and set a fixed multiplier?

[vast kestrel]

Might as well do it right

[fiery turtle]

oherwise u kinda get unfair results lol

[dense steppe]

yea lol

[vast kestrel]

Like, as long as it ends up as a single script I come home and run I don't mind doing it even every day lol

That's the main bottleneck, ensuring everyone is using either serial or debugcon to output the info, so I won't need to manually copy things out

I don't mind running a command over serial automatically if need be, but nothing manual

[gentle peak]

i feel like thats not too much to ask from kernel devs

[fiery turtle]

like every kernel here has serial or debugcon

[vast kestrel]

[left orbit]

not pmos i think

[fiery turtle]

hm

[gentle peak]

also kinda off topic but does anyone know if there's a way to make qemu riscv virt use non-hardware-reduced acpi?

[fiery turtle]

dont think so, it's a very bare bones impl i think

[gentle peak]

that sucks

[fiery turtle]

they're actively working on it tho

[slim panther]

not a thing afaicr

[fiery turtle]

so far only itanium and x86 have seen full acpi

[gentle peak]

i was looking forward to seeing uacpi riscv performance :/

on tcg but still

[fiery turtle]

i mean there's still a namespace

[gentle peak]

nope, there's not even a dsdt

[fiery turtle]

i think your qemu is just old

[gentle peak]

i'm on 9.2.0

[fiery turtle]

because keyronex definitely did run a dsdt on riscv

[fiery turtle]

did u make this printout yourself?

[jaunty fox]

i'm not sure if you need a special qemu branch

[gentle peak]

yeah

[jaunty fox]

possibly you do

[dense steppe]

I actually plan on removing (or putting in #if 0) debugcon from my kernel once I have fbcon lol

[jaunty fox]

i thought it was merged

[fiery turtle]

ohh it's for the microkernel?

[gentle peak]

yes

[fiery turtle]

btw hardware reduced doesnt mean there's no DSDT

[hallow marten]

it's good to output to e.g serial because you can debug it on hardware if you have a serial port

[fiery turtle]

no DSDT is not compliant impl

[gentle peak]

oh really?

[fiery turtle]

yeah

hw reduced just means no global lock and general purpose events

[dense steppe]

yea. but first you gotta have a serial port lol

[gentle peak]

if your hardware has a serial port that is

[hallow marten]

yes

[gentle peak]

on all the hardware i have the most low-level debugging is framebuffer

so i just use triple fault debugging

or, if i'm trying to get rid of a triple fault, halt loop debugging

[dense steppe]

.macro hcf
42:     cli
        hlt
        jmp 42b
.endm

so fkin useful

the reason that I'm doing fbcon right now is because I kinda want to test on a very old laptop

and for that, I need output somehow

[fiery turtle]

btw how are u checking for dsdt

u do look at x_dsdt as well right

[gentle peak]

ngl i completely forgot that dsdt isn't in rsdt/xsdt

[fiery turtle]

lol

yeah it's pointed to by fadt

[calm latch]

I have ns16550 driver, but it's broken on x86 atm

[gentle peak]

yep false alarm

[fiery turtle]

ye

[calm latch]

The issue is that there's SPCR table on RISC-V, but not on x86

So like do I just log to COM0 unconditionally?

[gentle peak]

the 'proper' way to do it would be to find it in the ACPI namespace

but then it obviously can't be used for uacpi benchmark

[calm latch]

or that

but after I find the port, how do I know that it's not an UPS or a mouse, idk

[gentle peak]

no real way to tell

[fiery turtle]

state_reset(), then do it again lol

[gentle peak]

that'd work yeah but then you'd have to have a separate uacpi benchmark build mode

[calm latch]

I can actually

[gentle peak]

also the results would be invalid because the 2nd run has better cache stuff

[calm latch]

My log server buffers certain amount of lines

[fiery turtle]

true

[calm latch]

So it would just spit the result out once the serial port was found, after the benchmark

[gentle peak]

that'd work yeah

[calm latch]

Like this is the only issue

[gentle peak]

that is fundamentally unknowable

the only difference is the protocol being used

[calm latch]

But stuff like SPCR explicitly tells you that a serial output should be used for logs

[gentle peak]

you just have to trust that, when the user tells you to (f.ex) use COM1 for logging, it's not connected to a mouse or ups

[calm latch]

(or for the console)

I guess I could make it a boot option

[gentle peak]

kind of like linux's console= option

[calm latch]

Yeah

[vast kestrel]

Isn't there a debug table that the bios can provide? I think in qemu it points to debugcon

Debug Port Table 2

[north holly]

@fiery turtle is there anything new I could test in uacpi?

[fiery turtle]

Soon

New api but nothing big

Its feature complete

For the most part

[north holly]

will be waiting

or well not waiting

but more of an asynchronous read

[dense steppe]

ask him to ping you. when he does, that is an interrupt

[north holly]

yes

[torpid root]

you installed an interrupt handler

[north holly]

then if I am in bed or taking a shit, I will register a DPC so I can do it when the IPL lowers (I wake up or finish taking a shit)

[fiery turtle]

Who said you couldn't do programming while taking a shit

[north holly]

true

but I would still prefer deferring it

thank god none of my real life friends that are in this server

well there are two

but none of htem look here

[vale isle]

@ oberrow irl friends

[north holly]

[median crest]

at everyone are you oberrow friend

[torpid root]

@everyone says: yes

[dense steppe]

@left orbit was it you who tested my kernel on an i7-13700K? because I'm curious about how this image would perform:

[lofty dragon]

curious about a 13900KS?

[dense steppe]

that would also be interesting

[lofty dragon]

well @strong heath has one

[strong heath]

https://tenor.com/eIITFdC6np7.gif

[dense steppe]

as long as it isn't tested concurrently with a GCC compilation

[strong heath]

whats a command you would like the davix software distribution to be run with

[dense steppe]

qemu-system-x86_64 -enable-kvm -cpu host -M q35,smm=off -m 256M -cdrom davix.iso this should work

[left orbit]

im off pc for the night

but 13900ks should do a bit better than 13700k

[fiery turtle]

Virtualize in the brain

[left orbit]

higher clocks and slightly better single core performance will do

[strong heath]

varies a lot, the average seems to be 6.4-6.5, it has lows of 5, it has highs of 7

bind PCI bro

[dense steppe]

yeah I really should implement those routines

... but there's like a million other things I also should implement

[strong heath]

such is life

[dense steppe]

indeed

like it's interesting because I consistently get results approximately in the range of 4.4-4.8M on an AMD CPU

[strong heath]

yeah well its AMD

slow ass cpu

next purchase make it intel

[lofty dragon]

if it still exists

[strong heath]

https://tenor.com/oikihwscnvH.gif

[median crest]

soon it will all be loongarch

[strong heath]

bad joke bad joke

[mortal yoke]

new score is 6.7M as I realized that cli + sti was unnecessary, basically its enough to first write the software irql to the cpu local so then all interrupts that happen after that point can only set the hw irql to the new irql (which is fine, then no cr8 change is needed inside the lower irql function) and after that load the hw irql from a cpu local + do cr8 change if its greater than new irql (as the hw irql could have been changed to the previous irql before the first write of new irql to the software irql cpu local)

[fiery turtle]

damn

i havent merged the updates yet, you can still add yours

[left orbit]

i get like 6.2-6.3M pretty consistently

[lofty dragon]

btw

to @gentle peak but to everyone else really, Limine 9.0.0 is out and it features SMP wait loops using monitor/mwait

idk why i am saying it in the uACPI thread

shit

i should've used the #1282508189698625619 thread probably

[calm latch]

uACPI thread needs to be renamed into lounge2

[kindred beacon]

updated managarm score: 5618646/s

still on my 14700K btw

[fiery turtle]

[slim panther]

fuck

i need to get my game up again

[calm latch]

beating pmOS again

[mortal yoke]

ig you can add it I think the average on my cpu is now 6.8M

[calm latch]

although not by much and on a faster CPU

[slim panther]

@fiery turtle my bad btw, the new score is on an Intel Core Ultra 7 265KF

[median crest]

@fiery turtle if I make a shitty allocator just for uacpi that only cares about speed and not about memory usage would that be accepted

[dense steppe]

if the kernel is still usable with these changes (i.e. doesn't leak memory or break concurrency or anything else), my opinion would be yes. but I'm not infy

just use the shitty allocator in the entire kernel

[fiery turtle]

Link and description

[fiery turtle]

Sure

[fiery turtle]

I guess

[flat badge]

imo making a shitty allocator that is used exclusively for uacpi is gaming the benchmark

if you use it in the entire kernel it's fine

[fiery turtle]

Yeah

[vocal geyser]

i fixed my shit allocator and got this

i have amd ryzen 5 3600 (i bought it in 2020-2021)

[hollow elm]

(3) still has 1 early reference(s)! seems a bit worrying

are you on an older version of uacpi?

[vocal geyser]

no

[dense steppe]

why do you initialize uACPI before setting up the APIC etc.?

[vocal geyser]

i initialize early tables

first

after init hpet

and finally init uacpi

here with non fixed allocator

[vocal geyser]

its fixed

[fiery turtle]

It means they leak an early table reference

Aka not calling table_unref

[north holly]

wiat what

I should probably fix that

[fiery turtle]

Basically a safeguard warning because a lot of kernels handle early mappings in a different way from late mappings

[vocal geyser]

can i get into this table 👉 👈

[fiery turtle]

I can add your kernel if you want, the pr is still open

[median crest]

dear lord its embarassing to see astral that low

[fiery turtle]

Yeah

[median crest]

I need to buy a good computer

[vocal geyser]

ok i can send link

or you already have

[fiery turtle]

Link, cpu used for measuring, short description

[north holly]

meanwhile obos being at the 2nd bottom:

I need to optimize my kernel one day

[median crest]

I got 4.5mil on streak's but he didnt send the full score grrr

[fiery turtle]

Beg until he does

[north holly]

I'd probably be able to optimize it greatly if I did one thing...

[mortal yoke]

an NT driver compatible kernel and userspace I am bad at coming up with short descriptions (and the link, though you an probably find it https://github.com/Qwinci/crescent2), score 6,818,418/s (for something that's close to the average on i5 13600K)

[fiery turtle]

Letsgo

[median crest]

wait qwinci your computer is good right

[fiery turtle]

Cpu + exact score pls

[vocal geyser]

https://github.com/cppLover0/Orange amd ryzen 5 3600 x86_64 unix like os

i trying to add a unix-like features in soon

[fiery turtle]

Sure

[north holly]

lets go

obos will be 3rd last instead of 2nd last

[mortal yoke]

decent yes, not as good as the i7's/i9's or whatever else

[fiery turtle]

Wdym, he got 3.5M

[north holly]

this guy didn't

I was talking about his kernel

[fiery turtle]

That's the non fixed allocator

[vocal geyser]

its before

[north holly]

oh

obos will forever stay at 2nd last

[vocal geyser]

now my kheap allocator is good

[mortal yoke]

do the cr8 optimizations that I did

[north holly]

will do

but I want to optimize my entire allocator interface first

[median crest]

can you see if what I did makes things better
qemu-system-x86_64 -cdrom astral.iso -M q35 -enable-kvm -cpu host,migratable=off

[north holly]

it currently kinda uses function pointers

which sucks

[median crest]

I guess it did letsgooo

[mortal yoke]

this makes sense considering streak's cpu is better than mine

[median crest]

its crazy how much cpu speed matters

on my laptop I get like 1.8-1.9 mil

[vocal geyser]

why it throws error

[median crest]

thats it trying to get the root device but failing because theres no nvme in the vm

the default options picks nvme0n1 as the root

[vocal geyser]

a ok

[fiery turtle]

maybe get qwinci to run your os as well so u get a better score

[vocal geyser]

🥹

[mortal yoke]

you are compiling uacpi test runner into your kernel lol

can you send an iso

[fiery turtle]

wait what