Astral

[supple hatch]

what do u have to say abt that

[obsidian pivot]

astral will be left in the dust

[grand shadow]

Astralbros its over billions must use nyaux

[supple hatch]

!!!

[grand shadow]

Same libgcc issue as before, the binutil thing didnt fix it

So I will look more into that configure recipe later

[craggy tide]

what libgcc issue?

[grand shadow]

Its something about astral with gcc's configure script, it doesnt actually set ORIGINAL_AS_FOR_TARGET in gcc/as and so the libgcc build fails because of it

[craggy tide]

ah

[supple hatch]

mintsuki back to old pfp???

no way

u are no longer cirno???

[craggy tide]

i was cirno

i am cirno, no more

[lean kettle]

where cirno

[supple hatch]

no im cirno now

[rugged kraken]

Are those real ports or just a dummy/rewritten ver?

Openbox

[sacred gyro]

bruh's really necropinging from 10 months ago

and yes, those are really things he ported

[grand shadow]

Real ports yeah

These are from the old version from them to now its been like rewritten in most parts

[rugged kraken]

but yeah cool os

[grand shadow]

Ok Ive been procrastinating this for too long

I really need to get astral to compile gcc and then move on to the scheduler rewrite

[grand shadow]

screaming

yeah I have 0 motivation to work on this atm

[obsidian pivot]

Downfall of Astral

Obos will rise!

[supple hatch]

no

downfall of none, nyaux astral and obos will all rise 😎

take ur time mathew

[regal stone]

what abt mango :3

[supple hatch]

release source code and implement memory managment plus scheduling plus vfs plus usermode and now we are talking

otherwise

not rlly

dw tho we all start somewhere !

[regal stone]

source is available already

silly

[supple hatch]

post it on ur thread

then

lmao

[regal stone]

it is on my thread

ultra silly

[regal stone]

this’ll take ~52 years to do btw

(/j)

[supple hatch]

no it doenst

i did that in a month

just letting u know

it was buggy af

sure

but i reached usermode in 1 month before

i believe in u

just like parappa would say

u gotta believe!

https://tenor.com/view/parappa-discord-discord-server-parappa-discord-server-sony-gif-21769056

[obsidian pivot]

yes but your kernel was dogpoop

[supple hatch]

yea

but again

I DID

[crude agate]

true

[supple hatch]

reach usermode

just take ur time with it @regal stone

and like

[obsidian pivot]

sir your kernel could not map memory with different prot bits

[regal stone]

nah

[supple hatch]

try ur best

[regal stone]

a kernel is a kernel

[obsidian pivot]

wrong

[supple hatch]

I KNOW 😭

[regal stone]

even if it does nothing

it’s still a kernel

[crude agate]

obos is a kernel

[obsidian pivot]

a kernel is actually a kernel

[supple hatch]

so fucking stupid ik

[supple hatch]

nyauxkt was a kernel that had ub everywhere cause rust is shit

[crude agate]

nyaux is a"kernel"

[supple hatch]

[regal stone]

even if it bombs your pc

[obsidian pivot]

then you slapped the title "VMM" onto that

[regal stone]

still a kernel

[crude agate]

yeah

[supple hatch]

onto the original nyaux's "vmm"

yea pretty muchhh

[obsidian pivot]

rust is still shit no matter what

[supple hatch]

^

[crude agate]

just out of curiosity, if we removed pointers in C wouldn't we also have a very secure language?

[regal stone]

if we removed pointers

[obsidian pivot]

no you'd have an unusable one

[supple hatch]

ah yes safety

[regal stone]

nothing would be possible

[crude agate]

yeah that's the point

[regal stone]

your int main() literally makes use of pointers

[regal stone]

char **argv

[crude agate]

at one time I tried learning Rust, and I was like go fuck yourself

same for Go

[obsidian pivot]

the safest C program

[crude agate]

main;

[obsidian pivot]

jk

there is no such thing

[crude agate]

main;

this actually compiles

and is very safe

[supple hatch]

just remove the ablity to even write code

safety

[crude agate]

tbf you can omit that

[crude agate]

still better than rust

[supple hatch]

!!

[crude agate]

how do you use pointers in rust with unsafe? Like what's the syntax

[obsidian pivot]

based opinion

[fleet osprey]

only with -Wno-error=implicit-int

but once you do - yes, yes it does, and I hate that

[proper meteor]

Do not force yourself to work on it. Otherwise you will start hating it

[proper meteor]

The type is *const Type or *mut Type. You dereference with * like in C. You have & to take a reference which you can then cast to a pointer (that operation is actually safe. dereferencing is unsafe)
You can then do pointer arithmetic with the functions implemented on the pointer add, wrapping_add, sub, etc...

[errant zephyr]

and keep in mind that the semantics for it are crazy

:^)

[proper meteor]

nah it's fine

[tribal dust]

yea the semantics are not hard

it's basically stuff like "if you violate lifetimes it's UB"

"if you try to create a ptr from thin air it's UB"

"if you do pointer arithmetic that goes out of the bounds of the structure the original reference points to, it's UB"

it's sound to do lots of things with pointer arithmetic

but since what the pointer points to must be valid for the pointee type, it's easy to footgun yourself when doing pointer arithmetic

but thankfully there are helper macros like addr_of! and addr_of_mut!

[obsidian pivot]

B-but I love accessing random pointers

[analog berry]

[grand shadow]

astral has fallen billions must use obos

I've been procrastinating poking at the gcc configure process to see wtf is going on

as in

I havent done that yet

[craggy tide]

billions must use nyaux

[neon crane]

billions must use boron whenever thats somewhat done...

[prime juniper]

Billions must use mana __ensure(“Not implemented!”);

[analog berry]

i wonder if theres ever going to be a hobby os project where the author wont eventually lose motivation

[craggy tide]

i find that it is hard to be motivated when working on any project with such a broad scope and little returns as osdev

[grand shadow]

yeah its just bursts of motivation

[supple hatch]

EXACTLY ^^^^^^^^^^^^^^^

😎

[supple hatch]

i find it easier to work on little things every day

sure its slower

but u wont burn out

like for me

i worked on the slab allocator today

now i will play val, if i feel extra motivated ill do part of the VMM

what motivates me is the fact that its MY own OS

and i can do whatever the hell i want to the poor cpu

make it explode?

yessir

😎

(idk if its possible but maybe it is)

idfk

thats the joys of osdev

to me at least

but i would understand how you feel

with college

just focus on college you can always work on astral some other time :)

[supple hatch]

dont be like me in the past, speedrunning nyaux caused me major burnout and a ton of bugs to be honest

[inland cliff]

Use clang

[inland cliff]

Sometimes you just get hung up on an obstacle planted by someone else due to bad design or historic reasons.

[proper meteor]

I've got nothing else to do in my life
Been working on that for almost 6 years (yeah I'm slow), still motivated for now

[grand shadow]

I dont build it nor do I want to, Im trying to build gcc in astral but due to configure memes the libgcc build fails

[prime juniper]

I try to keep working on Managarm but real life gets in the way too often

[analog berry]

we should all quit our jobs and work on managarm

[inland cliff]

Why is clang not an option though? Not trying to convince you or anything, I just want to gain insight into the reasons. I used to use it to compile my kernel (back when it was in C), of course, didn't have userspace yet, so can't really say much about that.

[obsidian pivot]

He is trying to build gcc on his os

Not build his os with gcc

[inland cliff]

I understand. The thing is many programs can be compiled with clang instead, so it could be an option for whatever gcc would be used for.

[fervent fjord]

gcc is much easier to get running

and much easier to port

[inland cliff]

What is easier on gcc?

[gritty shore]

llvm is HUGE

gcc is smaller

and easier to port

[fervent fjord]

this is one of the things you gotta get around to and do to actually realize

[analog berry]

well clang is one build for all targets

so thats nice

[fervent fjord]

if you dont mind having to port all of those targets straight up

[inland cliff]

That's one of the reason I find it so nice

But I don't have the background and experience to really know how it is like to port LLVM

Althought I could imagine if you port it once, you ported everything, so you can build anything on anything for anything, and if you add support for your OS, you'd be able to self-host and cross-compile.

And that would probably not only make it possible to compile C but also C++, Rust, and more

[fervent fjord]

gcc allows you to compile languages with its port, llvm just lets you use llvm utils, clang is a separate thing to port

so from the get go you gotta port two things, and that only gives you C/C++

while porting gcc gives you all the gcc languages because they share the part you have to port

and you have to port every arch independently, so the amount to port clang to all targets is the same as porting gcc to all targets

the thing of having all targets in a single executable is just what the user sees, every architecture has custom porting code inside

there is just no way around that

[inland cliff]

Hmmmm I see

[grand shadow]

@analog berry astral ran on my friends pretty old thinkpad and uacpi ran there too fine

Guess what was brooen

Broken

[analog berry]

Noooo lol

Not the ps2 driver

[grand shadow]

Im starting to think the ps/2 driver just isnt real

And I hallucinate things when trying to type

[proper meteor]

I also have this issue, ps/2 not working

[analog berry]

uPS2 will save u (at some point ™️ )

[worthy crater]

omg this looks awesome

with Xorg port

looking forward to more updates

[worthy crater]

@grand shadow are you thinking of porting Wine eventually?

[regal stone]

just take a go at porting PE

[tribal dust]

porting wine is a huge undertaking considering they implement wined3d and stuff as well

but it would be cool

[carmine swallow]

I mean when you already have mesa ported then it doesn't require anything special

[tribal dust]

ah that's right

but still, wine probably expects a linux uapi

[carmine swallow]

astral has linux headers

and getting wine to work on managarm wasn't that much work

can't really know about astral for sure because managarm does implement some stuff that astral doesn't have like netlink

same goes for openjdk too

[grand shadow]

I began some signal work to do wine64 but got sidetracked...... And then got sidetracked again

Repeat that like 10 times

Now Im retrying to compile gcc with a scheduler and device driver handling rewrite pending

[worthy crater]

Good luck with that

Everything you've made is still impressive, even if its a hobby OS it could be used as a serious one

At least back in the 90s it would have been considered quite usable

Maybe not in business but in home computing? Absolutely

[analog berry]

not with ps2 not working

[tribal dust]

ps2 not working is a feature not a bug

[winged basin]

Runs fine macos, freebsd, solaris, others

[neon crane]

how about try to port boron's kb driver :^)

maybe that would work

[worthy crater]

Well aside from that

[charred sinew]

https://github.com/Mathewnd/Astral/blob/e0f1fd3e031d6749472776c9a4b306aa3b8edb12/kernel-src/io/pci.c#L104
On the osdev wiki it says that the table size is bits 10-0 which has the mask 0x7ff, but you used 0x3ff. Or am i missing smth?

[grand shadow]

oh yeah thats my skill issue

nice catch

[obsidian pivot]

@grand shadow how many syscalls do you estimate astral has

[analog berry]

at least one

[obsidian pivot]

🤯

[grand shadow]

last I checked it was almost 100

[grand shadow]

syscallcount equ 92

[obsidian pivot]

cool

[obsidian pivot]

is it okay if I took over maintaining astral then add some backdoors to it

and a crypto miner

[grand shadow]

you have to add GAMBLING to it

kernel level slot machine

[grand shadow]

Ive been a bit burnt out lately but I can feel my energy slowly coming back we might be back astralbros

[obsidian pivot]

Astral is dead

[supple hatch]

rip

[grand shadow]

Astral was never alive, wake up sheeple

[fleet grove]

astral revive when

[analog berry]

after ups2 release

[supple hatch]

release it rn

we have uhda

uacpi

ups2 now

ukernel ???

reallll

😎

[grand shadow]

Sorry I have been stuck in the depths of hell (juggling work, university and factorio space age)

[analog berry]

the astral ps/2 driver

[obsidian pivot]

Last commit 21d

It's over for astral

[analog berry]

more space for obos to take over

[obsidian pivot]

obos will have astral syscall layer

Ezpz

[neon crane]

obos should have lunix syscall layer

ez ports

[fleet grove]

the grind doesn't stop even with real-world responsibilities
only a true osdever would know that

[analog berry]

i guess nobody is a true osdever then

because every hobby osdev project is eventually abandoned

[fleet grove]

for me, i just get the itch to rewrite mine 100 times

[analog berry]

thats normal

[fleet grove]

i mean that's just the nature of them

hobby oses are very volatile

[obsidian pivot]

Linux started as hobby

It didn't get abandoned

[fleet grove]

that's true

linux has only survived though because of its large number of contributors

most oses were closed source or proprietary at the time

it was just good timing

[obsidian pivot]

(Yet)

Obos will take over

[analog berry]

because it was picked up by corporations

[fleet grove]

^

if it wern't for that, it would probably be yet another abandoned hobby project

it's the people around it that has kept it alive

most hobby projects don't have that kind of backing

not to say it can't happen, there are plenty of big projects here

but it doesn't happen often

[obsidian pivot]

obos is taking over guys

No doubt

[analog berry]

most contributors these days are @redhat or @intel people

etc etc

[fleet grove]

exactly

[analog berry]

if it werent for them it would be dead

[fleet grove]

exactly

[neon crane]

yep

[fleet grove]

that doesn't make linux a bad kernel though

[analog berry]

nah

its just not a hobby or community project by any stretch is all

[fleet grove]

yep

[grand shadow]

Astralbros wont ever recover

[obsidian pivot]

What even drove people to do that

[fleet grove]

idk man
i'm not a contributor

[obsidian pivot]

What was so special about linux

At the beginning

That made people from coorps start contribing to it

[errant zephyr]

nothing

it was at the right place at the right time to attract hobbyists

[obsidian pivot]

They just woke up and decided to?

O

[errant zephyr]

and it became the best supported free software option

[fleet grove]

so what i said basically

[obsidian pivot]

I read over that

[fleet grove]

right time right place situation

[errant zephyr]

BSDs had licensing troubles

[fleet grove]

and GNU Hurd wasn't going fucking anywhere, still really isn't tbh

[analog berry]

lol

[errant zephyr]

eh. if the timings have worked differently, it could have been a thing

[fleet grove]

meh idk, the mach kernel was never really known for being the fastest thing on the market

not to hate on microkernels, i love them, but it's always come off as bloated to me

[errant zephyr]

of course. but XNU is, and they grew out of mach

[fleet grove]

true, but XNU also has some stuff in the kernel

[errant zephyr]

yeah of course they have tons

but you can move stuff between the two

[fleet grove]

but that's fine, the problem with Herd is that everything was in userspace, so you had ipc and context switch delays

[errant zephyr]

mm, maybe

[fleet grove]

i believe NeXTStep had the same problem

i mean hell L4 literally exists because of Mach being slow

[analog berry]

common microkernel L

[fleet grove]

microkernels can be good, it's just they have more challenges to face than other kernels do

[analog berry]

yet to see one

[fleet grove]

fair enough

i think seL4 is the closest thing we have to a decent microkernel these days

[errant zephyr]

the apple secure enclave is a microkernel

and the nintendo switch os is also a microkernel

[analog berry]

the what

[fleet grove]

i believe the apple secure enclave microkernel is L4 based

[errant zephyr]

both for security + isolation

[analog berry]

switch i know yeah

[errant zephyr]

possibly

[errant zephyr]

the part of the apple silicon that does security shit

its like a TPM but actually good

[analog berry]

switch os is not a fair example its kinda limited

[fleet grove]

bro tpm...

[analog berry]

ah

[errant zephyr]

and in a single chip

and also does other things

[fleet grove]

yeahhhh

[analog berry]

but i guess microkernels can be good for niche use cases

[fleet grove]

they can

it just depends on what your needs are

[winged basin]

it outperformed 4.2BSD

[fleet grove]

oh i didn't know that

[winged basin]

Although the system has yet to be tuned, current performance appears to be in line with 4.3BSD. Some early simplistic measures of virtual memory performance are encouraging. The MicroVAX II cost of touching newly allocated memory is less than 0.7 milliseconds per 1024 bytes of data (versus approximately 1.2 milliseconds for 4.3BSD).

[fleet grove]

interesting

seems that i was a little too critical with it

[winged basin]

it is popular to dislike but it really did invent a lot of what's now modern and accepted practice in kernels

even the tlb shootdown

[fleet grove]

oh seriously? i had no idea man

[winged basin]

the vm object concept also

[fleet grove]

interesting

i mean i never hated mach tbh, i just never thought it was fast

[errant zephyr]

okay i dont want to bring up the obvious but

uh

was 4.3BSD the new/modern/fast one at the time

[winged basin]

basically, it was 1986

and they were so impressed by mach that they imported its vmm for 4.4bsd

[warped plinth]

mathewnd rn

/lh

[eager copper]

ksp too I'd say

[neon crane]

might have been celeste for me

[prime juniper]

Real

[obsidian pivot]

at this point it's not even a joke

astral is no more

with Nyaux replacing it, ofc

[grand shadow]

I got a brain worm when I was building my factory in gleba and now I can only think of the factory

[supple hatch]

slowly cause school

[hearty fulcrum]

Is it really being developed or is this a meme

[analog berry]

i have it planned, its real yeah

im not going to work on it at the same time as uacpi, so once uacpi is finished i might start that

[hearty fulcrum]

what do you think about a common kernel api for all u* projects?

[analog berry]

yeah this will be a thing

[craggy tide]

nothing ever happens

[grand shadow]

Nothing ever happens

I swear I gotta go back to working on this 😭 I got so close to having gcc compile

Local man stuck on gleba eating nothing but jelly for 10 years, what happens next will shock you

[proper meteor]

Astral running Minecraft when

[hearty fulcrum]

that would mean something happens

but everyone knows nothing ever happens

[charred sinew]

how hard would that be? ig the most difficult part is getting the jvm to run

[proper meteor]

tbh I have no idea

[obsidian pivot]

Since there has been confusion that Nyaux will replace Astral, it will not for another couple hundred years

#1230349543623757845 message

[fleet osprey]

> See new post on the Astral thread
> Looks inside
> No new progress
>

[analog berry]

gotta wait for nyaux to catch up and replace

[grand shadow]

Soon™️

[analog berry]

testing new uACPI on astral

350k ops average per second, not bad at all

no kvm is a bit slower

[grand shadow]

Real

[obsidian pivot]

Alas, we have lost him

[grand shadow]

I might do something this weekend no promises

[robust geyser]

we might be back..?

[analog berry]

uacpi resync with master will reignite the astral flame

it must right

[grand shadow]

I actually got as far as opening it up and preparing to open the gcc configure script to see whats up

but then factorio finished loading and everything fell apart

some of the things I have on my backlog are like pm timer, rewrite scheduler redo device driver handling and oither things

[sacred gyro]

some of my projects tend to be
"TODO: everything”

[analog berry]

https://github.com/Mathewnd/Astral/pull/13

aight

[grand shadow]

hell yeah merged ty

first commit in a month

(wasnt even by me)

[analog berry]

lol

[crude agate]

can

you

send .iso file pls

my ethernet is slow so git download is 200KB/s

and

[obsidian pivot]

Check pinned messages

[carmine swallow]

though to be fair if you have so garbage internet then downloading an iso is going to take a long time too lol

[grand shadow]

Pick your poison

(building from source is broken anyways

((will be fixed soon™️

[crude agate]

git is crashing

[fleet osprey]

soon™️ as in "possibly before my death", or soon™️ as in "probably within the next week or 2"

[grand shadow]

Soon as in its already like 20% done but I got distracted

[obsidian pivot]

Factorio finished loading?

[analog berry]

how?

[grand shadow]

My jinx is so old some keys are invalid

So Im going to have to do the tedious job of updating every jinx recipe

(only for mint to change how its done again in like 3 months

[analog berry]

classic

[supple hatch]

mathew deals with mint shanngains

electric boogoloo

[grand shadow]

Im slowly slogging through updating all the jinx recipes

Jfc this is boring

Very annoying but needed work

At least mint will be happy I wont be 2 versions behind

[fleet osprey]

🎉

actual things going on w/ Astral

we're so back

[grand shadow]

Yeahh

[craggy tide]

we're so back

[analog berry]

Inb4 this is the only astral commit for the next month

[grand shadow]

Hopefully not

[supple hatch]

mathew we back???

astral is back????

[grand shadow]

I want to get gcc to build finally and then go on to do actually important stuff like rewriting the scheduler and device handling

[supple hatch]

mathew are we back

[grand shadow]

Maybe

[supple hatch]

NOOO NOW NYAUX CANNOT KILL ASTRAL /j

dies

[analog berry]

Just add gcc support

[grand shadow]

ughhh build-all and sysroot being removed messed my makefile up

[grand shadow]

why does apt keep returning 503s on the container

so annoying

even on a browser the links dont work

@craggy tide do you ever get 503s while downloading imagedeps

[craggy tide]

yes

it's the debian mirror being ass between yesterday and today

Gloire CI failed the past 2 days for the same reason

i hope they sort it out

[grand shadow]

Damn

[craggy tide]

the alternative is that i'll have to purchase a VPS and mirror the Jinx snapshots myself

[obsidian pivot]

reminded me to update me packages

[grand shadow]

How much storage does that take

I have a vps but it does t yave a lot

[obsidian pivot]

but it does t yave a lot

[grand shadow]

donst

aaaa

[obsidian pivot]

doesn't

?

[craggy tide]

probably a few 10s of gbs

[grand shadow]

yeah I wouldnt have enough rip

[craggy tide]

actually more than that

probably in the 100s because we need it for multiple architectures

like 10 of them

[grand shadow]

finally done with all the packages

now doing another full clean build to make sure

[grand shadow]

Clean build worked

https://tenor.com/view/walter-white-walter-heisenberg-breaking-bad-walter-white-death-gif-26283180

[warped plinth]

LET'S FUCKING GOOOOOOO

[grand shadow]

alright I can finally get back to debugging why the gcc/gcc configure script fails to find as in astral

inb4 missing which

[grand shadow]

its been a whil

e

I wonder if this is it

[fleet osprey]

wha... have you gone this entire time without /bin/bash?

I'm shocked

[obsidian pivot]

Smh astral doesn't have bash /j

[analog berry]

astral does have bash

and it runs it by default

just not at this path i guess

[fleet osprey]

yeah, just not a /bin symlink I guess

[grand shadow]

I guess some configure scripts detect that its not working and use grep -e instead

[gritty shore]

still no mail

:(

[analog berry]

mathewnd dont u have friends that mail u stuff

[grand shadow]

you could have someone ssh into it with another user and send you mail

[analog berry]

can u send mail from your pc into the astral vm?

[grand shadow]

I have no idea how the unix mail stuff works

gwahh nothing feels better than diggin ten of thousands of lines deep in the gcc/gcc configure script

okok

I think

it might be something with test -x not working

DEFAULT_ASSEMBLER is empty and test -x says its good

no thats working

test -x with an empty argument gives 0

hmm

[grand shadow]

(on the linux host

[grand shadow]

nvm I'm dumb

test -x -> 0
test -x "" -> 1

test -x "" -> 0 in astral

which is wrong and bad and stupid

[analog berry]

Mlibc skill issue?

[grand shadow]

probably astral skill issue

I need to see wtf test -x "" is doing

probably just

calling access() or something

access X_OK

or whatever the flag is

[analog berry]

That is the flag

[grand shadow]

ok yeah astral skill issue

vfs_lookup with an empty path returns the reference node

I forgor why I do that

or if its even something I intended

[analog berry]

What's a reference node

[obsidian pivot]

just make it not do that

if all breaks then bad

[grand shadow]

like vfs_lookup(&result, reference, path, ...) goes through path starting on reference

[obsidian pivot]

if the bug gets fixed, keep the change

[analog berry]

So a workdir?

[grand shadow]

kinda yeah, like the path is relative to

[obsidian pivot]

if nothing happens, revert the change

[analog berry]

Well he doesnt have tests soooo

But yeah sounds wrong

[grand shadow]

I mean if you want to look up literally nothing then you should also get nothing

it makes sense lol

I should fix this

[analog berry]

if path == null or '\0' einval

[grand shadow]

ENOENT probably

[gritty shore]

is there a posix test suite even

would be cool

[analog berry]

But thats an invalid path

Or I guess depends on what POSIX says

[analog berry]

Definitely

[obsidian pivot]

just return success and go on with your life

[analog berry]

There's a paid Unix conformance test suite too

[gritty shore]

yeah but like

free

and open source

[analog berry]

Good question

Probably

[grand shadow]

[ENOENT]
O_CREAT is not set and a component of path does not name an existing file, or O_CREAT is set and a component of the path prefix of path does not name an existing file, or path points to an empty string.

so it IS the correct behaviour

UNLESS

theres like

[analog berry]

Oh well

[grand shadow]

AT_EMPTY_PATH for some functions

but like

that seems to be a linux extension

which bleh

astral is my kernel and thus I decide that I dont want to implement this linux extension, too bad

[analog berry]

Me when ill have to implement that extension for my kernel

[grand shadow]

doing the build again

time to see if that fixed it

ASTRALBROS, WE EATING GOOD TONIGHT

ITS FIXED

[fleet osprey]

Amazing

[grand shadow]

alright hopefully now it should pass the libgcc build and die somewhere else

[fleet osprey]

maybe you'll be lucky it'll not die at all

[grand shadow]

unlikely

[grand shadow]

linking cc1

[warped plinth]

:0!

any update?

[dim arrow]

how long does each build take?

[prime juniper]

Nice work

[robust geyser]

too long

[grand shadow]

UINT64_MAX

[analog berry]

the uacpi benchmark doesnt lie

[grand shadow]

What even affects the speed there

[analog berry]

allocator speed and stuff like memcpy

[grand shadow]

So just buy a better cpu

[grand shadow]

Did you run all of them on your compiter

[analog berry]

all of who

[grand shadow]

All of the benchmarks

[analog berry]

nah qookie did

but its mostly consistent between different pcs

no huge difference

[grand shadow]

How did everyone go

[analog berry]

managarm does 1.5M, keyronex with all optimizations is close to that, pmos (the microkernel) gets about 500K, obos does 20k, nyaux does 150/s

[grand shadow]

Damn my allocator sucks then

[proper meteor]

that's the number of AML instructions per second?

[analog berry]

yup

[grand shadow]

@analog berry how do you enable the benchmark (or is it enabled bt default)

[analog berry]

its not a benchmark really lol, its just the log output

X ops in Xms

it prints out during init

[grand shadow]

Ohh

been working through the night and its linking libstdc++ now

so I assume it has passed libgcc

this probabbly wouldve been quite a bit faster if I did make with more jobs but my scheduler still sucks

[gritty shore]

What is the biggest factor?

I'd say allocations maybe?

[grand shadow]

if I had to guess which part of the kernel makes it slow its probably the vmm since its just a linked list

[gritty shore]

I shall beat you all

Eventually

When I port uacpi

[hearty fulcrum]

allocator and memcpy speed

so just use rep movsb

[gritty shore]

I think everyone on that list uses rep movsb

[hearty fulcrum]

I get 500k but I believe my physical allocator is slow

[gritty shore]

I think mine should be decently fast

Probably close to keyronex

Or managarm

since I use managarm's slab allocator

[hearty fulcrum]

me too

[grand shadow]

booooring

[gritty shore]

I'm in the process of making a better one

[grand shadow]

I think the biggest issue with mine is that the vmm sucks

and all my kernel allocations go through it

[grand shadow]

wth is a libstdc++convenience

[wild marsh]

typing that into ddg leads to https://gcc.gnu.org/onlinedocs/libstdc++/manual/appendix_porting.html

[proper meteor]

I would definitely make sure the scheduler works well before attempting to build gcc. It's already long enough when running on Linux, I don't want to be waiting an entire day between each test

[grand shadow]

my scheduler works well its just not good

[grand shadow]

funky

[proper meteor]

Well, replace "well" for "good" in my sentence

[grand shadow]

the biggest issue with it is that it sucks for smp which means most of the -j2 benefits would be i/o

[grand shadow]

alright its going

it did libssp and its doing some math library now

did it

did it work

DID GCC FUICKIGNG BUILD

[supple hatch]

OMG

[grand shadow]

I'm syncing and creating a copy of the hard disk[

[supple hatch]

YOOOO

MATHEW LETS GOOO

[grand shadow]

make installing

[winged basin]

congrats, let's see what happens

we pray for a successful hello world

[analog berry]

Daaamn

[grand shadow]

my i/o is still super damn slow

[grand shadow]

damn

[analog berry]

Make install failed?

[grand shadow]

yup

[analog berry]

Any ideas why

[fleet osprey]

[analog berry]

Install is probably easier to debug than compilation

[fleet osprey]

no way I was right that it built, but now it failed to install

so close yet so far

[wild marsh]

just turn off the docs :^)

[analog berry]

Wdym far lol

Some bullshit step with docs fails

[fleet osprey]

I predicted that it would fully work

[fleet osprey]

^^^

[analog berry]

Ah

[wild marsh]

inb4 genhooks is a perl program and it's gonna be a massive pain

[grand shadow]

its c I think

but I dont have texinfo

so maybe its that?

[gritty shore]

nothing else matters starts playing

[prime juniper]

Damn that ain’t bad. Do we have a benchmark Linux run to compare against?

[prime juniper]

Nice

[prime juniper]

Better allocator for Managarm too?

[gritty shore]

it's probs gonna be pretty kernel specific

I'm just planning on a multithreaded slab allocator basically

with per-cpu object caches

[analog berry]

No one integrated uacpi into Linux so no lol

But we might see reactos soonish

[gritty shore]

what about in userspace?

you have tests

[analog berry]

Asan + ubsan gets me to 40k/s

Haven't tried without

[prime juniper]

Ah yeah I thought this was just a testcase one can run

Unlucky

[analog berry]

But my test runner is built for consistency verification and not speed

So it does extra stuff thats unneeded

[analog berry]

The test case is the qemu aml blob lol

[prime juniper]

Lol

[analog berry]

But I think its similar on real hw

As in the rate doesnt change all that much

[grand shadow]

guh I have no idea

it dies in some c program that the build system creates

https://github.com/gcc-mirror/gcc/blob/6866547e2421cc40606c13c2c3621a3d9121e893/gcc/genhooks.cc#L149

[grand shadow]

ahhh fuck it I'll just comment that out

just wanna see if this will work

next time I'll disable docs

[analog berry]

What does it even do

[grand shadow]

I have no idea

[analog berry]

Looks like shitty scripts in c I would write when I barely knew c to practice

With garbage code style too

[gritty shore]

average gnu build system

[proper meteor]

‼️

[grand shadow]

aaaasldosafkdaskfsdagjiosdgokjdsaigjsdakpvskodpsvksaovijs

[fleet osprey]

YOOOOO 🔥

[grand shadow]

one more step towards full self hosting done

[robust geyser]

holy fuck

gj!!

astral is so back

[analog berry]

daaamn

[grand shadow]

yeahhh

I think astral is the first hobby os to actually pull this off

[fleet osprey]

but still no /proc/meminfo or /proc/cpuinfo for neofetch

[warped plinth]

LETS FUCKING GOOOOOOOOOOOOOOOOOOOOOOOOOO

[analog berry]

indeed

very big achievement

[craggy tide]

damn very nice

next build llvm :^)

actually, that may be easier to do

since the build system is just cmake, not nearly as scuffed as autohell+crust unlike gcc

[tawny oak]

Congrats, that's super cool 🎉

[fleet grove]

Congrats ^w^

[grand shadow]

now I have two paths

scheduler or devices

I will have to do both in the end so

hm

[obsidian pivot]

scheduler

[proper meteor]

scheduler

[grand shadow]

yeah true

last thing I was doing was trying to calculate run and interruptible sleep time

[analog berry]

is astral fully self hosted now?

or do u have to port something else for that

[flat copper]

jinx ig

[grand shadow]

Define fully self hosted

[grand shadow]

Jinx is linux only sadly

[analog berry]

can produce an iso of itself from scratch

[grand shadow]

Never tried compiling mlibc or some other packages so cant really be sure

Also missing xorriso for that

But the kernel compiles at least lol

[analog berry]

xorriso is probably easy to port

i mean if u compiled gcc just fine i doubt mlibc would be a problem

but i guess jinx is the real blocker

[grand shadow]

Jinx isnt exactly needed if I unshit the kernel build so

[grand shadow]

starting next week I will be free from the shackles of oppression (university) and will have a few more hours of free time every day so if I dont spend that time playing cogmind or something I will spend it on astral

banger game btw got my first win yesterday after 90 hours

[analog berry]

mathewnd try not to get addicted to random games impossible challenge

[obsidian pivot]

Next addiction is block blast

[neon crane]

wee

nice

me too

[prime juniper]

Better port meson and ninja

Shouldn’t be too bad

[fleet osprey]

nah watch it be genshin

[grand shadow]

so I wanna add serial support

for more tty diversity and something I wanted to do

is it always guaranteed to exist something at 0x3f8 or is there any way I can check for it? acpi maybe?

[fleet osprey]

I know that at least on legacy BIOS it's dumped in the BDA somewhere, I used it
there's probably some way to check via ACPI but I never got far enouh into my OS to figure that out

[obsidian pivot]

acpi tells you

https://github.com/OBOS-dev/obos/tree/userspace-work/src/drivers/x86/uart

[grand shadow]

bet

also dang uacpi work wait on astral just unconditionally sleeps 100 ms for every work_await

@analog berry on startup how many times is uacpi_kernel_wait_for_event or uacpi_kernel_acquire_mutex with a timeout called

or work_await

[analog berry]

never

[grand shadow]

ah

[analog berry]

depends on aml ofc

[grand shadow]

I'm trying to figure out why its so damn slow

its probably the vmm

[supple hatch]

how fast

does astral load uacpi

[grand shadow]

200k/s on my computer

which tbf isnt that good

but still

[supple hatch]

oh freak u

you with your

1137 ops

200k ops*

[grand shadow]

infy gets like 350k/s on his computer

@analog berry whats your cpu

I wann cope with my bad design decisions

[supple hatch]

oh on kvm

we get 100k ops

not bad

SO CLOSE TO BEATING ASTRAL

😡

[analog berry]

i9 9900

[grand shadow]

someone with the latest i9 run astral

[analog berry]

11k

[supple hatch]

11k???

oh its

11k

[craggy tide]

i don't have the latest i9 but i can run it on my Core™️ Ultra 7 265K

just send an iso

[supple hatch]

run mine

[craggy tide]

i can run both

[supple hatch]

nyaux iso

[craggy tide]

qemu command?

[supple hatch]

sudo qemu-system-x86_64 -M q35 -drive if=pflash,unit=0,format=raw,file=ovmf/ovmf-code-x86_64.fd,readonly=on -cdrom NyauxKC-x86_64.iso -d int -serial stdio -no-shutdown -no-reboot -smp 20 --enable-kvm

it doenst really matter

js enable kvm

and make sure machine is q35

[craggy tide]

why tf do you run it with sudo lol

[supple hatch]

wsl

[analog berry]

nyaux doesnt know he can add himself to the kvm group

and he wont need sudo

[supple hatch]

shhhh dont tell me that

[carmine swallow]

I also ran it on my i5 13600k, this is what I get on it

[supple hatch]

damnnn

22k ops

niceee

[grand shadow]

wont fully boot because its missing like everything to make the iso less heavy
qemu cmdline
qemu-system-x86_64 -cdrom astral.iso -M q35 -enable-kvm -cpu host,migratable=off

[grand shadow]

yeah I really need to find out wtf is making it super slow

thats like double what I get in my computer tho which is crazy

[gritty shore]

@supple hatch

[analog berry]

are u running a nasa computer

[gritty shore]

for some reason -M q35 makes it slower (I get 12k)

[gritty shore]

yes

I am nasa

[analog berry]

no no use q35 lol

[gritty shore]

why

what does it change

[supple hatch]

115k ops

no use q35

[analog berry]

like entirely different aml

[supple hatch]

^^

[gritty shore]

ahh

yeah I get 12k with q35

[supple hatch]

expected

im waiting for mints results

which will DOMINATE ALL

nyaux will dominate all

[grand shadow]

I get 500k/sec with i440fx lmao

[supple hatch]

real

bru

[gritty shore]

@analog berry you should make a real leaderboard

[obsidian pivot]

I get 140k ops/s with pc

lmao

-M pc

[analog berry]

i will add it to my readme, yes

[grand shadow]

you should set up a server with a i9-69696966969k to run it

[obsidian pivot]

add it to the CI tests

on each uacpi commit

to update the leaderboard score for each kernel

with the new uacpi commit

[carmine swallow]

then you'd have to build the kernels too which might be painful

[obsidian pivot]

exactly

@fierce forge can you test obos' ec driver on your laptop

[supple hatch]

the hell is an ec

[obsidian pivot]

embedded controller

some dumb acpi thing

[supple hatch]

oh oki

[fierce forge]

ok

[obsidian pivot]

thanks

[gritty shore]

hear me out

make uacpi a shared library

relink on CI

[fierce forge]

obos doesn't boot

[obsidian pivot]

bruh

can you send a pic

so I can see how it crashed

[fierce forge]

No picture

[obsidian pivot]

nothing logged means that hyper broke

or ventoy is being stupid

[analog berry]

try uefi?

[obsidian pivot]

can you try dding the iso directly

[analog berry]

i know ventoy is quirky

[fierce forge]

And my EC driver shows no signs of life

[errant zephyr]

its defo ventoy being stupid

ventoy and stupidity is like a constant

[obsidian pivot]

ventoy=stupidity

$$ventoy=stupid$$

[frigid saddle]

oberrow

[fierce forge]

It works fine with limine...

[analog berry]

it takes so long before it reboots highly doubt its hyper

[obsidian pivot]

well it's not my kernel's fault, as the first thing it does on boot is print some info about the current commit

one of the first things

[analog berry]

idk how ventoy boots stuff so no idea

[fierce forge]

obos skill issue

(I've used balenaetcher)

Which should be dd

[analog berry]

try uefi boot to be sure

[fierce forge]

It is UEFI

[analog berry]

ah then yeah

[fierce forge]

I'm not sure this thing even supports csm

[supple hatch]

face reveal

[grand shadow]

yeah I cannot for the life of me figure out why astral takes so long to run uacpi init

skill issue very liekly

[hearty fulcrum]

allocator, memcpy, mutex/spinlock

these affect performance the most

[analog berry]

its still not terrible

[grand shadow]

@analog berry does uacpi by default use kernel or its own memcpy

[analog berry]

its own

[grand shadow]

how do change

[analog berry]

platform/libc.h

u want to add uacpi_libc.h header

which i think u have already

[grand shadow]

oh I do

blegh

[wild marsh]

how do you cope with the compiler generating it's own calls to memcpy/memset/memmove

[analog berry]

implementer's problem

if you're integrating uacpi but dont have those idk what you're doing

[wild marsh]

yeah but then why provide your own?

[analog berry]

never seen those calls autogenerated

so its a theoretical problem

[hearty fulcrum]

wait so I haven't been using proper memcpy this whole time?

[grand shadow]

finally pinpointed it down to somethin

[hearty fulcrum]

smh

[analog berry]

imagine not reading the readme

[grand shadow]

if I just return false in the semaphore acquires I get > 1mil

[analog berry]

lol

[grand shadow]

now I just have to figure out which function is responsible

[grand shadow]

400k per second

nice

[craggy tide]

and this is nyaux

[analog berry]

Damn pretty fast cpu

Even nyaux is approaching obos speeds

[craggy tide]

most of the time it actually says 36ms

it seems pretty inconsistent

[grand shadow]

seems like one of the biggest culprits is the acquiring of the semaphore lock

lockless time

probably disabling the interrupt is being very expensive for some reason

wonder if it might be something with trying to run pending interrupts or some shit

yeah I did something very stupid and just removed the interrupt_set(false) and interrupt_set(intstatus) from before the spinlock acquire

this seems to be super slow

now to figure out why

[fleet osprey]

just™️ disable all locks

[fierce forge]

I need to fix my timers 😦

When does this leaderboard thing end?

[grand shadow]

yeah disabling DOPENDING_SAVE() on interrupt_set(true) like quadruples uacpi performance

jfc

wonder if this is one of the reasons why it was so slow to build gcc

[gritty shore]

whats DOPENDING_SAVE

[analog berry]

Its perpetual

[grand shadow]

it saves current context and handles pending interrupts

[analog berry]

Hasn't started yet

[hearty fulcrum]

@analog berry I'd like to update my score, please

[analog berry]

Wait how

[hearty fulcrum]

a giant skill issue that I do not want to say

[analog berry]

Lmao

[hearty fulcrum]

this can be faster with a proper physical mem allocator

[fierce forge]

So 5kk ops/s could be credible?

[grand shadow]

hm I think its the context save that seems to be messing everything up?

[obsidian pivot]

Blame it on infy

Not me

[fierce forge]

That's what I've gotten, but I think my timers are broken

[analog berry]

No lol

[obsidian pivot]

Not me

[analog berry]

It wouldnt take this long to boot

So its obos dying

[obsidian pivot]

It literally can't be

[fierce forge]

But then Linux said that the frequency that I've measured is the same

[obsidian pivot]

The first thing it does

Is print shit

Before anything else

[analog berry]

Could be dying too fast for us to see it

[obsidian pivot]

Doubt

[analog berry]

Use hyper with debug prints

[fierce forge]

I can try porting pmOS to hyper

[analog berry]

Yeah

[fierce forge]

And I have quirky hardware

[obsidian pivot]

Albeit not the first thing

https://github.com/OBOS-dev/obos/blob/userspace-work/src%2Foboskrnl%2Farch%2Fx86_64%2Fentry.c#L213-L263

[analog berry]

This would be nice for testing and skill issue fixes if any need to be made

Make it print and halt after @obsidian pivot

First thing

[obsidian pivot]

Busy

[grand shadow]

im rertarted

I could literally just check if theres pending interrupts before calling this

oh my fucking god

[hearty fulcrum]

could anyone try mine?

qemu-system-x86_64 -serial stdio -m 512M -cpu max,+hypervisor,+invtsc,+tsc-deadline -M q35 -accel kvm -cdrom image-x86_64.iso

[hearty fulcrum]

new record 1294442/s

[obsidian pivot]

Well unless hyper gives me bad boot info

It can't really crash before that log statement, as I would have observed it

[hearty fulcrum]

yes :D

[obsidian pivot]

Unless by some miracle the framebuffer from the firmware is fucked

[hearty fulcrum]

ops/s?

[hearty fulcrum]

nice

also try running again, something might change

[fierce forge]

I'm gonna plug my laptop in

[gritty shore]

now I wanna port uacpi just to benchmark

lol

[obsidian pivot]

Now it could also be fucked firmware

[hearty fulcrum]

based

[obsidian pivot]

That has a bugged boot menu

[fierce forge]

Works in pmOS

Are you setting some resolution?

[obsidian pivot]

No

[fierce forge]

The screen is 2944x1840@90

[hearty fulcrum]

@analog berry shouldn't the scores be determined by testing on a one specific machine?