Zinnia

[near tartan]

• finally implemented freeing page tables (yes)

• munmap (smp to be fixed later)

• virtio_gpu ghosting (fixed by uploading the fb before setting scanout)

• a bunch of stupid pty issues

[grave peak]

Nice

[hybrid island]

Nice

[drowsy tendon]

Nice

[somber solar]

still no zinnia logo?

[near tartan]

yea

the logo is missing colors iirc

[near tartan]

@drowsy tendon @somber solar what was the state of that anyways

i forgor

[drowsy tendon]

i couldn't make one (skill issue) but ilo had their own generator do one

i can dig it up and see how to add color to it

[near tartan]

:3

[somber solar]

#1344765673246822430 message

[near tartan]

[somber solar]

make sure to remove the trailing spaces at the end of the lines

adding colours shouldn't be hard

[near tartan]

good enough

i might use this as a base for more

[drowsy tendon]

@near tartan

hopefully the screenshot doesn't butcher the color quality

[near tartan]

😭

[drowsy tendon]

bruh 💀

gimme a moment

[near tartan]

bruh discord moment

it works on laptop

[drowsy tendon]

oh

crazy

[near tartan]

ts so peak

[drowsy tendon]

i thought it was Windows snipping tool being shit

anyways here are the files

[drowsy tendon]

also idk why the color palette characters are fucked up on my end

hopefully it's just a simple issue in the config file but i cba checking now

[drowsy tendon]

https://tenor.com/view/one-peak-one-piece-one-piece-cry-its-so-peak-gif-13413926760013023520

its so peak

[near tartan]

now to patch this in

[somber solar]

🔥

[near tartan]

@somber solar

[near tartan]

gcc!

[somber solar]

puts \n

laugh at this user

[near tartan]

yeah yeah

[hybrid island]

no newline at EOF?

https://tenor.com/bVyg2.gif

[near tartan]

what

[hybrid island]

when you cat hello.c

I see it does not have a newline at EOF

[near tartan]

yeah

because i didn't have one

i edited this file in zinnia

[hybrid island]

good good

[lofty copper]

Why is everyone suddenly getting GCC

[near tartan]

shrug

man every time i see your nickname i get sad that i didn't continue the microkernel

[somber solar]

how does one make nano colourful like that

[near tartan]

idk

it just works

[somber solar]

default is monochrome

is it not

[near tartan]

but this is the weston terminal, which defines TERM=xterm

[somber solar]

I have xterm-256color defined

[hybrid island]

Cuz GCC isn't actually that syscall-demanding

Especially compared to the likes of Wayland

[near tartan]

yeah gcc is much tamer

i just messed up my path

[lofty copper]

But Wayland requires a lot of Linux stuff

[near tartan]

depends

libudev i have fixed now

you do need signalfd, eventfd, timerfd and epoll

but those are relatively easy to implement

suffering from success

my weston works but now X can't find any screens anymore

im crine

[near tartan]

i think it's time to run os-test

[drowsy tendon]

it unfucked itself?

[near tartan]

no

bunch of stupid little fixes

[drowsy tendon]

ahhh i see

good job anyways <3

[near tartan]

i added a basic rtc driver

so now it also shows the correct date

gnumake still doesn't work

rup

[drowsy tendon]

sickkk

[near tartan]

rip

[drowsy tendon]

aw

🥀

[near tartan]

i have no clue what make's problem is honestly

it always says nothing to be done for 'all'

oh wait

maybe i should fix the EINVAL

💀

[near tartan]

code for wayland has been merged

[eternal wharf]

wayland init... ok

[near tartan]

but i'll take a break from this for a while

i'm going to mess with my shitty microkernel design

just to feel something

idk why but now that i have implemented all these features i'm getting really bored of rust

[hybrid island]

marvin try not to switch projects challenge (impossible)

[near tartan]

it's just always the same

idk i just don't enjoy writing it that much anymore

[lofty copper]

Just rewrite it in C++

[shy turret]

Zig time

[hybrid island]

have you, like, taken a proper break from programming, ever?

[near tartan]

considering i'm a full time SWE not really 🤔

[hybrid island]

💀

give yourself a literal break man

because like this you'll end up burning out

[shy turret]

+1 on taking a break

[near tartan]

oops

wtf happened here

[hybrid island]

forgot to add renamed file destination?

[near tartan]

not really

i think jinx fucked up

the seatd patch is 7k lines lol

[near tartan]

that should be fixed...

okay that's now also merged

lets see what CI thinks

[near tartan]

fuck it we ball

[near tartan]

ran os-test on zinnia

my everything handling is abysmal

lol

[hybrid island]

lol

[near tartan]

my score is higher than redox tho

[hybrid island]

lmfao

show summary

[near tartan]

nvm

it's not really comparable

it didn't even get to run certain tests

as you can see basic only ran 234 tests

redox

[fossil yoke]

isnt your kernel a modular kernel

[near tartan]

yes

right now it is

but i have an older version that uses a completely different codebase

[fossil yoke]

ohh

btw does being modular on compile time counts as modular

or do you need to load .o's at runtime

[near tartan]

no

that's just being configurable

[fossil yoke]

rip

[near tartan]

i have modules as dylibs

i actually load full shared objects as drivers

for the microkernel I'll not use any AI at all to write the code itself

i have to admit i got a bit lazy with delegating debugging to the clanker

[near tartan]

i deserve this

@lofty copper indeed

c23 my beloved

[lofty copper]

C++26

[near tartan]

i dont even have a cross compiler that has support for that

[grave peak]

jai

[median forge]

rewrite zinnia in c++

[near tartan]

libipc.so

[lofty copper]

c++23 is also quite good

[shy turret]

i think gcc 16 has some more c++26 support

[near tartan]

https://tenor.com/view/javascript-saw-meme-programming-js-gif-5946925740875118368

[near tartan]

i have 15.1

I'd need to update

[lofty copper]

"I'm using llvm 19 and it's fine"

[near tartan]

i have llvm 21

[lofty copper]

I haven't bothered with rebasing because I haven't really had any issue with it

[near tartan]

iirc mesa wanted it

[lofty copper]

Also, why use gcc in the first place?

(My gcc patches are for gcc 12 )

[near tartan]

i use both

[lofty copper]

Like I just use clang + lld for everything

[near tartan]

gcc is faster to compile

i like having a second opinion

[lofty copper]

I stopped using gcc becuase of a compiler bug with C++ templates

[storm bobcat]

Rewrite in zig

[near tartan]

naw

[near tartan]

tfw jinx just nuked my kernel changes

[near tartan]

microkernel is doing microkernel things

uacpi from userspace

no idea why it's so slow but w/e

the allocator is pretty bad

[grave peak]

how is 2M slow ☠️

[near tartan]

this is userspace

it should be faster

no?

[grave peak]

if u optimize it really well i guess

[near tartan]

tru

[grave peak]

but 2m is pretty decent

[near tartan]

fair enough

i still need to reason about my IPC model

especially moving big buffers

[marsh holly]

Microbenchmark it harder

[grave peak]

i mean he said allocator is pretty bad

[near tartan]

it's eagerly mapping things and every anon alloc is a syscall basically

plus it's not single threaded

there's initd and posixd running at the same time

[near tartan]

one stupid idea i had was to make it so you can "offer" another process memory

or well

i can have all data only IPC be shared memory

then have a syscall to transfer object handles

rn my channels can pass a buffer of bytes + buffer of handles

[lofty copper]

managarm's ipc is very cool

[near tartan]

yes

[lofty copper]

But the issue with this (that I had) is that the other process/whatever needs a way to accept them (explicitly)

[near tartan]

yeah

[lofty copper]

Otherwise you can just litter its namespace

[near tartan]

right now i just listen to a channel with a byte and handle buffer

like, you can say you are expecting at most N handles

then you don't get more

[lofty copper]

That's how my IPC works

[near tartan]

dam

[lofty copper]

But it's hardcoded to 4 handles + reply right

Then if you don't want them, you can reject them

[near tartan]

why 4?

[lofty copper]

And I have async send + sync receive, and my messages take ownership of handles while being sent

[lofty copper]

Idk, just seemed like a reasonable number

[near tartan]

yeah my messages also take ownership when you send them

as in, unless you clone them, you lose them

[lofty copper]

Sometimes you need to send a reply right + 2 of them, so I made it 4

If you want more, just send it in several messages

[lofty copper]

Yeah

But then of the thread dies, it frees its messages and handles within them

[near tartan]

is that a problem?

[lofty copper]

No

But it creates a system, where you either receive a reply or a notification that the handle had been deleted, so if the server crashes or whatever, before it accepts the message, you're not left waiting for the reply forever

[eternal wharf]

this is similar to fuchsia's ipc as well iirc

[near tartan]

gee i wonder why

(totally not inspired by it)

[eternal wharf]

:^)

[near tartan]

the only other idea i had for IPC was nova's UTCB thing

[lofty copper]

I was loosely inspired by mach

But I changed my IPC 2 times now

Like not fundamentally, but just had 2 big additions

[near tartan]

i will probably change things as i go on and learn

[eternal wharf]

also is the zinnia that ran weston kil?

[near tartan]

it's still up

i'm not replacing it until i get at least to the same state

aka until i can run weston again

i wanted to get it done in a monolithic design first so i have a thing to base it off of

which will probably make implementation easier, i just need to translate it

[eternal wharf]

port the entire monolithic zinnia kernel to a single server :^)

[near tartan]

[lofty copper]

managarm?

[eternal wharf]

?

[near tartan]

posix subsystem :^)

[lofty copper]

I mean the microkernel doesn't become less microkernel if you do this

[near tartan]

would probably improve throughput

[lofty copper]

I guess it depends on your design, but maybe

[near tartan]

maybe vfs and posix stuff in one server

and all drivers in another

[lofty copper]

yeah

[near tartan]

that was my initial idea

[lofty copper]

Idk, I'm making it be a server per thing

[near tartan]

i probably just want to have one server for the system then

[storm bobcat]

That goes against the whole idea though

[near tartan]

what idea

[storm bobcat]

Microkernel

[near tartan]

but like

splitting vfs and posix things makes little sense for example

because both go hand in hand

[storm bobcat]

Yes

But you wouldn't put a filesystem server in there

Or a disk driver

[near tartan]

well yeah

[storm bobcat]

Maybe filesystem is ok but you're pushing it

[near tartan]

i wouldn't put drivers in there

filesystems are drivers imo

[storm bobcat]

Idk you said you wanted one server in the system

[near tartan]

no

for the system

[storm bobcat]

Which kinda goes against the point of a microkernel

[near tartan]

"system" being the core component

[somber solar]

microkernel but servers are in ring 0

[near tartan]

free overhead

[storm bobcat]

unironically a cool idea now is doing that

but with protection keys

[somber solar]

no protection

[storm bobcat]

That's not been really explored in an entire OS yet

Though

I'd guess mostly because of lacking hardware support

[near tartan]

hear me out

ok nvm i'm retarded

unless

let me experiment with this

i think i found a funny ipc method

[brisk totem]

protection keys are ass

and cant really be used to make this secure

[near tartan]

oh man i missed C so much, no longer have to wait 15 seconds to do a cached build

[storm bobcat]

Why?

[grave peak]

max 16 drivers at a time lol

[storm bobcat]

There's some paper that added them to isolate Linux drivers

It's pretty cool

[storm bobcat]

no

[brisk totem]

in c you can just yolo everything

[brisk totem]

they cant do proper isolation afaik

you can only make memory not writable

[grave peak]

ideally thats all u want to prevent bad drivers from corrupting the kernel

[brisk totem]

so you cant stop rop iirc

[grave peak]

its not for like rop or isolation

[storm bobcat]

The point is to catch bugs more than malicious code

[grave peak]

yea

[brisk totem]

apple has a feature like that but it lets you configure everything separately

so its useful for security

[grave peak]

iirc the fil-c guy said it wasnt that great either

[brisk totem]

iirc the kernel variant is pretty good

[grave peak]

maybe

[marble salmon]

Managarm does not have the equivalent of a monolithic kernel in userspace

posix-subsystem is not equivalent to Linux

[near tartan]

linuxd

[marble salmon]

posix-subsystem does precisely:

• process management
• some pseudo FSes

and nothing more

[storm bobcat]

Also I wasn't talking about a particular implementation but rather the concept

[marble salmon]

It does not do on-disk file systems or other devices

[storm bobcat]

I think Apple's is better than intel's

[somber solar]

get linuxed

[near tartan]

i think for now, doing the channel based message passing is fine

[brisk totem]

how should i deal with cycles

[lofty copper]

detect them

[brisk totem]

like if you send a handle to a channel on that channel

[brisk totem]

how lol

[lofty copper]

don't send handles to channels

[brisk totem]

well no i do need that

[lofty copper]

just see if the handle is for the channel is in the channel

[brisk totem]

how tho

[lofty copper]

like it seems like a simple check idk

[near tartan]

wdym

[lofty copper]

how do you send stuff?

[brisk totem]

this can be very indirect

[storm bobcat]

I know how to do proper synchronous IPC but idk how to do async

[brisk totem]

and you have complicated locking

[lofty copper]

do you mean handles to send or to receive?

[brisk totem]

i only have a combined "call" handle

and also a "call target" handle that im just going to make unsendable

[lofty copper]

idk, so far I've been fine with non-transferable receive channels, and it's not a problem for send rights (however that would translate to other designs)

I mean I'm thinking about making my send optionally synchronous as well, and I would have different deadlock problems there as well

To which the solution is to either trust that userspace doesn't do stupid stuff and make it its fault for deadlocking itself, with no detrement to the kernel, or to somehow be nice and detect it, like what else can you do

[marble salmon]

send weak handles

[near tartan]

[marble salmon]

that's what Managarm is doing in some cases

[brisk totem]

or i can just give the handles at transfer time so i never own them in kernel structures

[marble salmon]

yeah

[brisk totem]

ok after further analysis, im also going to have to restrict processes

to prevent resurrection; that is, after a process exits, the resources are always released and you cant bring anything back

[marble salmon]

you can implement revocation trees

that is, track from which capability another capability is derived

and then have a revoke() syscalls that revokes all derived capabilities

this is done by various L4 like microkernels

also, a system that I added in BlueRock's microkernel system is ranking of capability spaces

where you assign a rank to each capability and to each capability space and you can only put a capability C into space S when rank(C) < rank(S) (or similar, you could also reverse the direction)

this prevents cycles by design

[brisk totem]

hmm

that solution feels a lot like what you have with universes in lean and similar theorem provers

[marble salmon]

yeah

[brisk totem]

i like that

it also very nicely deals with pagers

[marble salmon]

i may have stolen it subconsciously from type theory

[brisk totem]

lol

[brisk totem]

its also the same as lock ranking 🙃

[brisk totem]

a pager must have a rank that is ≤ the rank of its associated process

and you can only map memory with a rank < your processes rank

and you can only do ipc if the source region has an appropriate rank

[brisk totem]

yeah so i also have one more rc cycle

an async op needs to point to the target queue

because thats how my locking hierarchy works

but then if you send a message to its own queue and then close the handle you deadlock

actually i think you can solve this by using a globally sharded lock

and then once you hold that lock, the pointer must be valid or something

[near tartan]

weston inside of weston is kinda funny

i've reworked the virtual memory allocation so no more bump vmem allocator for userspace or kernec

i really want to be able to host my own website on this

that includes

• get smp back working
• fix make
• virtio-net driver
• tcp/ip stack
• refactor the socket api so it's not superaids to write code for

i really want to run os-test to completion at least once, but for some reason it sometimes starts hanging while running sub makes

[marsh holly]

wtf unmap() finally implementeed?

[marsh holly]

I should do that on my vps, just have a nginx reverse proxy or whatever that goes into an astral vm running a web server

[steady echo]

would be funnier if the QEMU was running in weston

[near tartan]

yes

(no tlb shootdown ipis lol)

[near tartan]

this

i also really need to fix mesa, not sure what its problem is

i have to run weston with pixman gl renderer

[near tartan]

@nocturne rampart how does lwip work

[nocturne rampart]

Isn't zinnia rust

Or are you going to hotglue some c rust interop

[near tartan]

no

oh wait it's just something you integrate in the kernel?

[hoary cave]

yes

[near tartan]

L

I'm not using that then

uACPI is the only case where it's reasonable to have C interop

[hoary cave]

you just implement sysdeps and callbacks

[near tartan]

i should implement udp first

before even thinking about it

[hoary cave]

lwip is a netstack

it gives you tcp and udp

[nocturne rampart]

Lwip does that for you gang

[hoary cave]

yeah, as i said

you just give it a few callbacks that let it talk with the nic

[nocturne rampart]

Yeah

[near tartan]

ok then s/udp/virtio nic/

[nocturne rampart]

Yeah implement some nic ig

[hoary cave]

did bro think lwip is a bunch of drivers or what 😭

[nocturne rampart]

I'd say go for e1000

[hoary cave]

fuck that

virtio is so much simpler

[nocturne rampart]

But since you already have virtio gpu

Yeah net makes sense

[hoary cave]

it's not like he's got any e1k hardware

[nocturne rampart]

Mainly because it's supposed by every virtualization software in some way

[near tartan]

i already have virtqueues

[nocturne rampart]

Yeah then do net ig

[near tartan]

no

i thought it was a userspace thing

I didn't read any docs

[hoary cave]

i still dk what you thought it did in userspace but all good

[nocturne rampart]

Oh it can run in userspace

But yeah it's an entire network stack essentially

[hoary cave]

uacpi can also run in userspace fwiw

[near tartan]

idk im at work and it's doing my head in

im leaving now tho

[hoary cave]

just go home

[marsh holly]

Lwip is the cowards way out

[near tartan]

bro i already said im not doing it

im not mixing more c code

[brisk totem]

smoltcp is the slightly less cowardly way out (at least its not in c + i think its a slightly better library for kernel at the cost of being rust shit)

[near tartan]

I don't really want to make a libOS

i already have few dependencies

e.g. syn, bitflags and seq

[brisk totem]

thats fair

[near tartan]

i will look at it for reference tho

[magic charm]

read TCP/IP illustrated vol 2

if you read it too carefully you will exactly reproduce the BSD Net/3 networking stack but that's no great loss

[near tartan]

ARP

[near tartan]

icmp

[near tartan]

@hoary cave i fixed MSI-X

nvme is now using it too

instead of looping

i just had it as an infinite todo

but turns out you can't really poll a nic on a single cpu system, so i had to fix it

i also discovered a very funny scheduler bug

that's a me not knowing rust issue

and leaking thread Arcs

[hoary cave]

zamnnn

[lofty copper]

I had a funny bug in my scheduler once, where it was randomly and quietly erasing threads from the existence, because all of the smart pointers were unreferenced

[near tartan]

lol

[lofty copper]

And like the kernel was doing a great job of reclaiming resources and stuff, so like stuff was just disappearing with no trace

[hoary cave]

in rust it’s easy to mess up too, but like it’s just a few isolated unsafes so it’s also pretty easy to verify it

[near tartan]

yeah it was in an unsafe block

of which there only a few in the scheduler code

i really need to fix smp

i also found a bug in waitpid, which probably explains the odd make behavior

maybe once all those are done i can finally complete an os-test run

[somber solar]

smp when

[grave peak]

KDE first, then SMP

the fishix route

[near tartan]

@grave peak hello

[lofty copper]

wait, how does it not have smp?

[grave peak]

whatsup

[near tartan]

ssh

[grave peak]

ssh

[near tartan]

i kinda vibecoded some part of TCP

but what it threw out was utter garbage

[grave peak]

u were supposed to save us

[near tartan]

i wanted to see if it can do it

it's half decent

the other half is braindead

im keeping the constants

they seem correct

it did shit like THIS

totally unsalvageable

[somber solar]

totally unsalvageable
salvaged

[near tartan]

no i just removed this shit

there's more of this btw

[near tartan]

okay yes

technically i did salvage stuff

i kept the socket trait impl

AGI

[storm bobcat]

How do you do this

In good rust

[near tartan]

it's not good rust

i'm still rewriting it

it wrote c++ as rust

lfg

[near tartan]

wdym

[storm bobcat]

Well can you do this better in idiomatic rust

[near tartan]

yes

you don't need this function at all

this is what i do

it just maps to the uapi constants

it does it all implicitly

🔥

this is so fucking cool

(it does require yet another mlibc ABI break)

[hybrid island]

if mlibc had semver it'd be at 40 by now

[near tartan]

probably

[hybrid island]

What model/structure do you use to represent and interact with devices?

I'm looking for ideas for device subsystem rewrite 2.0

[near tartan]

wdym

"model"

[hybrid island]

oh god xd

do you just slap some BS spaghet together or smth

[near tartan]

there's no standard model, you have FileOps at most

to expose to userspace /dev

the rest uses specialized traits for whatever they're implementing

[hybrid island]

welp

worth a shot

I would like a more structured model in my kernel

[near tartan]

i don't see the need

[hybrid island]

don't get me wrong there's no need for anything in BadgerOS to be as overengineered as it is

[storm bobcat]

look at iokit

[hybrid island]

that part is for fun purposes

[storm bobcat]

its nice

[hybrid island]

yeah I have been looking at it

looks very nice, though not sure how I'll deal with anything potentially inheritance-based in Rust

[storm bobcat]

vtables

[hybrid island]

well yea

[storm bobcat]

also doesnt rust have inheritance basically

[hybrid island]

non

Rust has traits

You can simulate inheritance

But there's no class-style inheritance unless you make it all manually

[near tartan]

it's big pain in rust to do that

OOP is very nice in a kernel

[hybrid island]

yup

[near tartan]

driver model maps well

[hybrid island]

thus why I said that

[storm bobcat]

probably better than in C

[hybrid island]

I guess so

You can still do it, e.g. with an "is a device" trait that just gets you the pointer to the base device struct, and then the actual implementations go in an impl dyn Trait + '_ {} (trait-associated functions without dynamic dispatch)

[nocturne rampart]

How

[nocturne rampart]

Is this python -m http.server?

[near tartan]

yes

[nocturne rampart]

Nice
Try netsurf next

Browse the web using Zinnia

[near tartan]

send recipe

[nocturne rampart]

I didn't make one yet 💔

[near tartan]

[nocturne rampart]

Ts is a pain to cross compile

Or well skill issues from me

[near tartan]

u don't say

i need to do dhcp

[nocturne rampart]

Same

I want to do it from userspace

[marsh holly]

dhcp in kernel ☠️

[near tartan]

dns in kernel

[brisk totem]

i mean better dns in kernel than dns in every process with no global cache ngl

[near tartan]

/dev/hosts

[marsh holly]

just spinup an sqlite database in /tmp? problem, liberals?

[near tartan]

who needs databases

it's called ext2

[brisk totem]

i was more thinking dnsfd

[near tartan]

pog??

nvme working on real hw again

[hybrid island]

nice

[near tartan]

I'm assuming Weston won't render because of drm memes

because this is not a DMT resolution

ugly

[near tartan]

🔥 mesa fixed

[near tartan]

at last

i have closed the one remaining issue on the github repo

lol?

no idea why it says invalid license

probably github dying (again)

[somber solar]

time to switch to codeberg

[near tartan]

i have made some small scheduler changes

also used clanker to debug leaked tasks

[fossil yoke]

codeberg's ui is like shit

cant they just copy github's ui or something

[somber solar]

smp fixed?

[near tartan]

yesn't

work stealing and so on is working

but no tlb shootdowns

[cyan nexus]

lol, limine switched from it back a while ago XD

bc their uptime was bad

[near tartan]

though github also has a worse uptime recently

[cyan nexus]

i haven't really noticed any problems with it

and i push like 4 times a day

[near tartan]

you must not be using it professionally

[cyan nexus]

prob

[near tartan]

just a few days ago none of the pull requests were working xd

[cyan nexus]

💀

[near tartan]

and they lost some in the process

[cyan nexus]

thank god i have my self hosted gitlab as primary

lol

[near tartan]

i have a selfhosted gitea but i won't use it for this project

[near tartan]

i need to port rust

and vim

[marble salmon]

Didn't you say that you reached your goals with menix?

[brisk totem]

you should selfhost gerrit instead

[near tartan]

when has my opinion on something ever stayed the same for more than 2 weeks

i still need xhci to be more usable

also a bunch of posix correctness fixes

but that can wait until managarm work is done

[near tartan]

for example session management and signals are catastrophically bad

to the point where shit just breaks randomly if you're not careful

and i can't even complete a run of os-test yet

[brisk totem]

why? crashes?

[near tartan]

well

it's hard to describe

it runs like 4-5 tests, then gets stuck doing whatever until i press ctrl+c

[brisk totem]

okay so hang

got it

have you tried gdb

[near tartan]

no use yet

[brisk totem]

like seeing what the thread is stuck in

[near tartan]

it goes to sleep it seems

[brisk totem]

yeah what kind of sleep

lol

[near tartan]

like idle

sti hlt

[brisk totem]

okay yeah but then why is the thread not schedulable

[near tartan]

haven't debugged that

[brisk totem]

i see

[near tartan]

it's likely an issue with waitpid

[marble salmon]

os-tests is not really forgiving when it comes to hangs

The tests run w/o a timeout so a single hanging test will stop the entire testsuite

[near tartan]

hm

oh another funny thing

in curl, when you don't get a response it just stalls

I don't think there's a timeout happening

[brisk totem]

i tried running your event in miri with a mock scheduler

ubbbbb

and i imagine this doesnt help either

i have no idea if its a bug in my scheduler mockup but uhh

yeah idk the scheduler wakign functions look sketch

[near tartan]

oh i already fixed these locally

but thanks

i rewrote the scheduler with the ULE paper

[brisk totem]

ah okay

[near tartan]

(and copied some parts but from the freebsd impl but shh)

[brisk totem]

lol

[brisk totem]

oh so what you are saying is your Event is no longer UB?

[near tartan]

i had the clanker check out the event handling

it's slightly less ub

[brisk totem]

send code

[near tartan]

not home

[brisk totem]

ah ok

[near tartan]

how did you test this btw

what's the setup like

i need to test my stuff more for ub

[brisk totem]

miri

and i made a very bad mockup scheduler

[near tartan]

lol

[brisk totem]

that knows how to make tasks wake up vaguely correctly

[shy turret]

Whatever forge Limine is on just kinda dies for some reason

[near tartan]

we're too chonk

[near tartan]

i have started working on refactoring my block io system

it allowed me to get rid of a lot of unsafe blocks in driver and fs code

[near tartan]

It appears that device hotplug kind of requires userspace to get notified of such events

I'll add a freebsd devd compatible API to the kernel

I can't really reuse the code directly sadly, but this will go hand in hand nicely with libudev-devd to provide libinput data

[fair lintel]

port systemd-udev to FreeBSD atp

[near tartan]

believe it or not, devd is very simple

code wise

[grave peak]

do u have any use for device hot plug in mind already?

[near tartan]

usb

[grave peak]

ah

[somber solar]

implement netlink

[near tartan]

netkink

fuh no

that doesn't solve anything

just more systemd kanker to deal with

[somber solar]

it's used for notifying udev

[near tartan]

yes but udev is hard to cross compile

[somber solar]

skrill tissue

[near tartan]

sorry for not being a 1:1 linux larp

I don't want to do sysfs

udev pretty much hard assumes linux everything from what I've heard

[marsh holly]

Ewwwwwdev

[near tartan]

exactly

[marble salmon]

udev hard assumes linux, yes

It's not meant to be portable

All OSes have their own device enumeration mechanisms, more or less

[near tartan]

exactly

[marble salmon]

Implementing netlink is not hard but you'll be forced into linux's device model or you'll need to create shims for all classes of devices (the latter is what Managarm does)

Since your /sys layout needs to be linux compatible

and "compatible" really means "quirk compatible". For example putting the same device at a different spot or with a different name will not work in all cases

[grave peak]

just embrace being a linux

join the linux squad

[somber solar]

[near tartan]

[somber solar]

ok mr smp after wayland

[marsh holly]

Larp as a unix workstation, do NOT join the linux larpers

You need to larp correctly marvin

[near tartan]

W

[lofty copper]

I agree

[grave peak]

thor binary compat or bust

[near tartan]

[storm bobcat]

Be actually unique and larp nothing like me

[near tartan]

i mean I'm just mixing and matching apis how i like it

[marsh holly]

Larp as a larper. Make zinnia look like a kernel shell os

[storm bobcat]

Well I do larp but I larp a bunch of stuff at the same time

[lofty copper]

ngl I need to look into bragi, maybe I can reuse it

[magic charm]

the right way

[near tartan]

dead os

it's safe to say that SMP is working somewhat stably

so far zero crashes

just the usual signal memes

[near tartan]

yooo

zinnia was able to detect my B580

[hybrid island]

What's that, a motherboard model?

[near tartan]

no

intel arc b580

[hybrid island]

huge

Are u gonna make drivers for it?

[near tartan]

not for now

especially not by hand

it's just way too much code lol

if i did, i would get clanker to analyze the linux xe driver and see where to start

[near tartan]

I HAVE XFCE

LETS FUCKING GO

@marsh holly

i am now using libudev-devd

[marsh holly]

holy

[near tartan]

no clue why this is so fucking slow

takes ages to run anything

also where decorations

[near tartan]

fucking hell do i need librsvg for this shit

[near tartan]

yo it can even detect the correct resolution

it's is horribly slow tho

it takes like a minute to open a window

[marble escarp]

Do you know why it's slow?

[near tartan]

if i knew i'd have fixed it lol

[marble escarp]

I see. I was just curious.

Sometimes the fix(es) aren't easy.

[near tartan]

@spark surge btw

that is also fucked

@nocturne rampart

[nocturne rampart]

Sexy

[nocturne rampart]

Damn

[spark surge]

awesome, congrats!

[chrome tinsel]

what x server are you using?

[median forge]

is your open() is fast

[near tartan]

xorg

[near tartan]

idk probably, why

[near tartan]

this is working better now

i'm switching my weird event abstraction to actual rust async

[silver birch]

memory: 0 bytes 😭

[grave peak]

0 bytes ought to be enough for anyone

[near tartan]

idk why that happens tbh

[median forge]

zinnia is new 0 bytes kernel

[near tartan]

i mean if it does proc/self/mem then it won't work

[silver birch]

lmfao

the reddit post of the other day

[median forge]

where's 10k upvotes

[near tartan]

i should really get started on xhci do i can test real hw more

[median forge]

boot protocol or hid

[near tartan]

hid ofc

and well

MSD

i care more about that

my laptop has ps/2 kbd

[hoary cave]

boot protocol is hid, btw

what you're thinking of is report protocol

[median forge]

k

[marsh holly]

doing full hid, at least enough for most sane keyboards/mice, is not that much more work compared to boot protocol when you take in consideration everything needed to get to that point

[near tartan]

note to self

os-test hangs here

[near tartan]

i have ported LightDM

time to upstream all of it

[silver birch]

nice

[fair lintel]

that's an mlibc issue iirc

[near tartan]

really?

[fair lintel]

see the managarm recipe

[near tartan]

@grave peak wtf

uacpi is trying to log a non-utf8 string

[grave peak]

0 backtrace, 0 context, yeah let me answer that

[near tartan]

sorry but that's all i saw

uacpi_kernel_log gave me a non utf8 string

i didn't even know that was possible

[grave peak]

Try not dying on it and actually print it

[near tartan]

one sec

[grave peak]

Like how tf am I supposed to help without any info lol

[near tartan]

i didn't say that

just cursed that it can do that

[marble salmon]

presumably it's a bug

[near tartan]

i made it return a lossy str now

so let's see

[grave peak]

Yeah its possible for some firmware strings to not have utf8, but there's currently no code in uacpi to repair them tho

[near tartan]

what's weird is that this is the first line of uacpi logs

[grave peak]

First thing it should print is its own version which is a hardcoded C string

[near tartan]

ah wait no

it's trying to print the FACS string

looks like this on QEMU

something has gone terribly wrong with the latest kernel changes

on 2 of my laptops it just triple faults now

before i see any logs

[grave peak]

Yeah that cant print garbage since the FACS signature is verified earlier

[near tartan]

we'll see

[grave peak]

And then the next two are just integers

[near tartan]

uh

@grave peak

is this memory corruption or something?

[grave peak]

Absolutely

[near tartan]

this is all i do in this stage

[grave peak]

it's probably what you did at the before stage

i even see some ascii strings in the corrupted memory

did u reuse firmware memory for allocations?

[near tartan]

i hope not

i use what limine gives me

[grave peak]

or actually, those signatures are cached

so u probably dirtied the early buffer

[near tartan]

wait

omfg 💀

it is a heap allocation but it gets freed one it goes out of scope 💀

[grave peak]

[near tartan]

i have to mem::leak

[grave peak]

void *malloc(size_t size)
{
    char buf[size];
    return buf;
}

[near tartan]

basically

but it's on the heap

[grave peak]

u can release it only after uacpi_initialize()

but u know that i guess

[near tartan]

yeah

let me try this

technically speaking rust helped me here

because the buffer is given to an unsafe block

HOLY SHIT

LETS GO

even the time works kinda

damn

triple faults on the framework

[grave peak]

damn

[near tartan]

i have 0 clue why it would crash on the framework

like, i don't even see any output

[grave peak]

time for hang debugging

[marble salmon]

or improving each fault handling

i guess

[near tartan]

this happens before screen init

[median forge]

maybe race condition ?

[near tartan]

before ap spinup?

[median forge]

a

[near tartan]

i added an actual ramdisk device

[marsh holly]

Why?

[near tartan]

so i can boot from ramdisk

[marsh holly]

Wouldnt just a tmpfs be better? That way you dont deal with metadata overhead and it allows better integration with your memory subsystem etc

[near tartan]

?

that doesn't work

the init system looks for a block device by partuuid

this is so i don't need to write an usb storage driver

but i can still test on hw

[marsh holly]

Bruh

💀

[near tartan]

how is this bruh

this guy boots off of initramfs 💀

[marsh holly]

I mean yeah U already got a filesystem there so why would you need to make another?

[near tartan]

wat

i don't understand what you mean

it's an ext2 image loaded to ram

[marsh holly]

Why have two different in-memory filesystems?

[marsh holly]

Yeah but doesnt that cause more memory overhead than a normal tmpfs?

[near tartan]

sure

[marsh holly]

Since you have to store more shit like superblocks inode/block bitmaps etc

[near tartan]

but it's much easier to deploy

because i don't need to change my distro infrastructure to accommodate ramdisks

i already parse initramfses if that's what you mean

but the initramfs init will always try to mount the rootfs from a block device and chroot into it

i'd rather not hack in a special case to do this

[marsh holly]

I guess

Tung ting ting

[near tartan]

https://giphy.com/gifs/before-sahur-yogif-ul81UmmtsS0QDpidDR

💀

i removed so much clanker shit over the week

and smp works on real HW too now

@grave peak i added support for memory types in Zinnia, what type do i need for ACPI tables?

can i treat it as regular memory?

[marsh holly]

In the original video it was 9. Larpers DNI

[near tartan]

https://tenor.com/view/warframe-titania-sorry-good-argument-cry-about-it-gif-25597522

[grave peak]

For tables specifically or just generic uacpi map

[near tartan]

tables

and i guess uacpi map would be good to know

[grave peak]

Tables are wb cached

[grave peak]

You have to look at the memory map and see what its trying to map

If ram = wb, otherwise uc

[near tartan]

👍

perfect

[near tartan]

this is a terrible bug man

i triple fault on real hw before the allocator is done

:D now i gotta hack in a pre-kmalloc framebuffer terminal

[hybrid island]

damn

[near tartan]

wait what the fuck

do i need to manually enable EFER.NXE?

please don't tell me bro

[grave peak]

I mean yeah

How was it not enabled by limine tho

[near tartan]

i havent had to do that ever

[grave peak]

Because limine

[rugged shard]

i was gonna say check the bios but what year is that machine from

[grave peak]

Its a framework

[near tartan]

2024

it's an amd 7040U

[rugged shard]

ah

[silver birch]

was that the very bad bug?

[near tartan]

no

it did not fix it

[silver birch]

hmm

[near tartan]

i keep triple faulting on the framework once i switch to my own cr3

this literally worked some time ago

idfk what could've changed

[silver birch]

are you sure the cr3 is in ram, is aligned, the stack is good, the stack is big enough

[near tartan]

yes

stack is inside the kernel binary so it should be loaded fine

there's no way this cpu supports LA57 right

[silver birch]

huh

[near tartan]

i asked clanker about it and it says that maybe the paging level doesn't match but that sounds like bullshit

[grave peak]

Unless you asked limine for it you aren't getting it

[near tartan]

yea i did not

[grave peak]

But nah its not supported im p sure

[near tartan]

la57 is likely a server thing

[grave peak]

Its only server cpus

Ye

[near tartan]

like you can't even get enough ram in this to saturate 48 bits

[grave peak]

But yeah you probably have a bug in direct map building code

Or kernel remap code

[near tartan]

likely

[marble salmon]

are you booting from limine?

[near tartan]

ofc

[marble salmon]

Managarm doesn't triple fault on that machine and it doesn't support LA57

[near tartan]

yeah obviously it doesn't have la57

but i'm almost out of ideas

like, the boot has literally worked before 💀

[grave peak]

What does clanker think about your mapping code

[grave peak]

Bisect?

[near tartan]

is fine

it suspects hhdm issues

[grave peak]

Bisect should take maybe 5 boots max tbh

If you didn't do gigantic commits

[near tartan]

:^)

oh wait a fucking minute

uh

@grave peak

do i need a CR4 bit for the Global bit

💀

lol i did add the Global bit a week ago

mfw

but the question is

what do i need to enable to make this work

[hybrid island]

SDM checkment time?

[near tartan]

ughh

[hybrid island]

Alternative: stop using more and more extensions like global bit until you don't get 3fault

disable nx

[near tartan]

nx is not the problem

[hybrid island]

hell disable pat if you've got it on

CR4.PGE
Page Global Enable (bit 7 of CR4)

Looks like the G bit is indeed gated behind a CR4 bit

[near tartan]

yeah but it should be ignored even if its set

[grave peak]

yeah thats probably not it

[near tartan]

no nevermind

@grave peak it actually was the problem

global + no PGE is allowed on intel

[grave peak]

shouldnt that also have died under kvm?

[near tartan]

but not on amd

[near tartan]

i am on an intel pc

[grave peak]

oh ok

so it would've died on kvm on that pc

or laptop rather

[near tartan]

ig

[lofty copper]

Dw I've been triple faulted by it as well

[near tartan]

(it was CR4.PGE btw)

[hybrid island]

Seems like

Time to download a copy of whatever AMD calls their SDM

[near tartan]

amd64 architecture manual

[marsh holly]

thats on you for setting a reserved bit

[near tartan]

it works on kvm

[hybrid island]

kvm != baremetal, especially wrt paging

[grave peak]

on intel kvm right

[near tartan]

yes

[grave peak]

can u try on amd kvm

[near tartan]

no

[grave peak]

[near tartan]

i don't have any amd laptop set up with it

[hybrid island]

My desktop is AMD

I can test for you

Just lob image and qemu command line at me