Zag | Operating System Development | Page 2

fallen bobcat Mar 28, 2026, 4:14 AM

#

yes

heavy sandal Mar 28, 2026, 4:15 AM

#

the endgame plan for mine is to only attempt a trylock-and-flush if moving to dispatch is a raise

#

if its already above dispatch it just skips the flush

fallen bobcat Mar 28, 2026, 4:15 AM

#

huh

heavy sandal Mar 28, 2026, 4:15 AM

#

and then ill just make the idle thread call flush KEKW

fallen bobcat Mar 28, 2026, 4:16 AM

#

well having threads do it is what linux does iirc

heavy sandal Mar 28, 2026, 4:16 AM

#

fallen bobcat huh

so raise to dispatch in the trylock but if already above dispatch dont bother flushing

fallen bobcat Mar 28, 2026, 4:16 AM

#

I guess I can have a periodic flush or something

heavy sandal Mar 28, 2026, 4:16 AM

#

i might make it schedule a dpc if its above dispatch when it goes to flush

fallen bobcat Mar 28, 2026, 4:17 AM

#

that's not a bad idea

heavy sandal Mar 28, 2026, 4:17 AM

#

tbh i think i might make a helper soon for stuff like this

#

takes a function pointer and if its below dispatch it raises to dispatch and calls the function and if its at or above dispatch it queues a dpc that calls the function

fallen bobcat Mar 28, 2026, 4:18 AM

#

what's neat is that if its already locked it means your thing will get output(ted?) anyway

fallen bobcat Mar 28, 2026, 4:18 AM

#

heavy sandal takes a function pointer and if its below dispatch it raises to dispatch and cal...

you can just do that in dpc_enqueue

heavy sandal Mar 28, 2026, 4:18 AM

#

fallen bobcat what's neat is that if its already locked it means your thing will get output(te...

yep exactly, thats why i use trylock because if its taken then its the other threads job anyway

heavy sandal Mar 28, 2026, 4:19 AM

#

fallen bobcat you can just do that in dpc_enqueue

yes but there are actually cases where i want to queue a dpc from low irql (mostly when i do a quick dpc self-test during init)

fallen bobcat Mar 28, 2026, 4:20 AM

#

why would you do a self-test

#

if dpcs dont work youre fucked anyway lol

heavy sandal Mar 28, 2026, 4:20 AM

#

its not a true self-test so much as me just printing a test message from a dpc because im not confident enough in my code yet to trust em lmao

#

i do a lot of things like that, quickly using a thing early in init so if i break it it breaks early

fallen bobcat Mar 28, 2026, 4:21 AM

#

I use DPCs a lot so I think they're not broken

#

the fireworks test ironed out a few bugs

fallen bobcat Mar 28, 2026, 4:22 AM

#

fallen bobcat that's not a bad idea

mhm actually idk if I wanna do this

#

because enqueuing a dpc that will do essentially nothing when theres little contention is kinda icky

heavy sandal Mar 28, 2026, 4:22 AM

#

yeah

#

linux does a thing where a context can steal the flushing process

#

idk how it works

#

it locks in a way that makes the other thread break out of the loop when it starts waiting idk how

fallen bobcat Mar 28, 2026, 4:25 AM

#

fallen bobcat because enqueuing a dpc that will do essentially nothing when theres little cont...

no this isnt bad considering a DPC can only be enqueued once

heavy sandal Mar 28, 2026, 4:25 AM

#

yeah

#

do you do broadcast DPCs btw?

fallen bobcat Mar 28, 2026, 4:26 AM

#

meaning?

heavy sandal Mar 28, 2026, 4:26 AM

#

send a dpc to all processors, or all except self

fallen bobcat Mar 28, 2026, 4:26 AM

#

no why would i do that

heavy sandal Mar 28, 2026, 4:26 AM

#

instead of a single target processor

#

idk why you would lol, i have em but they make things so messy and i dont really have any uses for em so i might remove em

fallen bobcat Mar 28, 2026, 4:27 AM

#

I have cool lockfree cross core calls tho

heavy sandal Mar 28, 2026, 4:27 AM

#

damn nice, ill have to take a look at how you do that

fallen bobcat Mar 28, 2026, 4:27 AM

#

not yet in zig

#

tho

heavy sandal Mar 28, 2026, 4:27 AM

#

rip

#

i could make my DPCs lockfree if i was ok with making the queue a stack instead (aka prepend the DPCs)

fallen bobcat Mar 28, 2026, 4:29 AM

#

you can make them lockfree if you dont need to dequeue them

#

but dequeuing is needed for e.g unloading drivers

heavy sandal Mar 28, 2026, 4:29 AM

#

wdym dequeuing?

#

because im dequeuing them when i go to run them lol

fallen bobcat Mar 28, 2026, 4:29 AM

#

cancelling a dpc before it fires

#

on an arbitrary CPU

heavy sandal Mar 28, 2026, 4:30 AM

#

oh i dont really plan to support that at this point

fallen bobcat Mar 28, 2026, 4:30 AM

#

otherwise you dont need locks as they are cpu-local

heavy sandal Mar 28, 2026, 4:30 AM

#

oh lol true

#

also its still fairly easy to cancel a dpc on another cpu with a single atomic bool

#

just set a canceled flag in it that makes it get skipped

fallen bobcat Mar 28, 2026, 4:31 AM

#

well yeah but thats clumsy when you have tons you need to skip over

heavy sandal Mar 28, 2026, 4:31 AM

#

yeah

fallen bobcat Mar 28, 2026, 4:31 AM

#

also you might wanna free those dpcs

#

so

#

use after free

heavy sandal Mar 28, 2026, 4:32 AM

#

way easier than trying to make a lockfree removal from the middle of a linked list tho

fallen bobcat Mar 28, 2026, 4:32 AM

#

just do it with a lock

heavy sandal Mar 28, 2026, 4:32 AM

#

yeah ig

#

that is what i do now lol

#

my main problem rn is my wait code is a fuckin train wreck of edge cases and race conditions

#

and deadlocks actually i ran into a deadlock when trying to release a parked thread from an interrupt handler on the same cpu in the middle of queuing it to wait

fallen bobcat Mar 28, 2026, 4:37 AM

#

at which point do I have too many files in the same folder

#

cuz i kinda wanna have log.zig and console.zig separate meme

heavy sandal Mar 28, 2026, 4:37 AM

#

that is entirely up to you tbh

#

i made em separate lol

#

i tend to prefer more files over big files personally

fallen bobcat Mar 28, 2026, 4:39 AM

#

yeah nah fuck it im keeping it in the same file

heavy sandal Mar 28, 2026, 4:43 AM

#

does the fireworks thing require a framebuffer btw?

#

cause i might be interested in trying that once i get my scheduler going more but i dont have a framebuffer yet

fallen bobcat Mar 28, 2026, 4:47 AM

#

heavy sandal does the fireworks thing require a framebuffer btw?

yeah

fallen bobcat Mar 28, 2026, 5:03 AM

#

https://github.com/rdmsr/zag/commit/614d28d513a1f2fece3e9ae4e4e18b6894ca6e87 very nice

GitHub

feat(ke): logging and consoles · rdmsr/zag@614d28d

#

probably need to describe the implementation of the ring in more detail tho

#

oh also need to include kconfig with it

heavy sandal Mar 28, 2026, 5:05 AM

#

oh if you type up good comments for the log ring thatll be great

#

that shit is so hard to understand without comments if you arent an atomics knower

fallen bobcat Mar 28, 2026, 5:06 AM

#

i tried explaining the orderings at least

#

but the wrapping arithmetic is pretty weird too

heavy sandal Mar 28, 2026, 5:06 AM

#

ill read thru it when its not 1am lol

#

i dont think i could explain it or remember it offhand but the wrapping logic i found pretty understandable, but ig thats just compensation for how i look at any atomic ordering other than whatever x86 atomics are equivalent to and my brain just short circuits about it

fallen bobcat Mar 28, 2026, 5:09 AM

#

I will try writing a module comment like the scheduler one

#

well not now cuz it is, in fact, 1 am

heavy sandal Mar 28, 2026, 5:10 AM

#

top level doc comments slap tbh

#

as a feature

#

and yeah 100% fair lol

weary jetty Mar 28, 2026, 5:17 AM

#

heavy sandal way easier than trying to make a lockfree removal from the middle of a linked li...

non-blocking queues exist, it's basically just translating the paper into code.

heavy sandal Mar 28, 2026, 5:17 AM

#

weary jetty non-blocking queues exist, it's basically just translating the paper into code.

yeah ik but its a pain to actually write that code

weary jetty Mar 28, 2026, 5:18 AM

#

heavy sandal yeah ik but its a pain to actually write that code

Have an AI agent translate the paper into code for you.

fallen bobcat Mar 28, 2026, 5:20 AM

#

lockfree queue is not worth it in this case

heavy sandal Mar 28, 2026, 5:26 AM

#

weary jetty Have an AI agent translate the paper into code for you.

i have never and will never use ai for anything, period. no exceptions

#

but yeah also not worth the effort in this case anyway

weary jetty Mar 28, 2026, 5:32 AM

#

The thought is more that non-blocking queues are sort of a solved problem.

fallen bobcat Mar 28, 2026, 11:59 AM

#

mhm wondering if it's sane to put the console stuff in Ke

#

Doesn't really make sense imo

languid canyon Mar 28, 2026, 12:01 PM

#

fallen bobcat https://github.com/rdmsr/zag/commit/614d28d513a1f2fece3e9ae4e4e18b6894ca6e87 ver...

did u end up understanding why we need acquire for the head?

#

or the seq num thing

fallen bobcat Mar 28, 2026, 12:04 PM

#

languid canyon did u end up understanding why we need acquire for the head?

Yes

fallen bobcat Mar 28, 2026, 12:04 PM

#

languid canyon or the seq num thing

wdym

languid canyon Mar 28, 2026, 12:04 PM

#

.

fallen bobcat Mar 28, 2026, 12:05 PM

#

Well I made this with plain ops and it works so

#

Only the state transition needs to be ordered which make sense

languid canyon Mar 28, 2026, 12:06 PM

#

wait so the issue u had was with relaxing something else?

fallen bobcat Mar 28, 2026, 12:07 PM

#

Yeah

#

The first head load was relaxed

#

I should've made it acquire

languid canyon Mar 28, 2026, 12:07 PM

#

where

fallen bobcat Mar 28, 2026, 12:07 PM

#

In desc alloc

#

You already do it I think

#

I thought I could be smart and make the first load relaxed

languid canyon Mar 28, 2026, 12:08 PM

#

ah

fallen bobcat Mar 28, 2026, 12:08 PM

#

I'll explain why it's wrong in more detail later am on phone rn

languid canyon Mar 28, 2026, 12:08 PM

#

sure

fallen bobcat Mar 28, 2026, 12:08 PM

#

Also what do you think about flushing

#

Right now what I do is try lock the console lock and if it succeeds it flushes it

languid canyon Mar 28, 2026, 12:09 PM

#

bad idea

fallen bobcat Mar 28, 2026, 12:09 PM

#

But that might not be good for stuff like nmi or whatever

languid canyon Mar 28, 2026, 12:09 PM

#

you can just starve a random core that happened to be unlucky

#

if other cores keep feeding logs

fallen bobcat Mar 28, 2026, 12:09 PM

#

Ah so I can still lose logs

#

hmm

languid canyon Mar 28, 2026, 12:09 PM

#

no

#

not lose

#

just have a core that hangs because it keeps flushing messages that keep appearing

#

thats why a takeover mechanism is important

fallen bobcat Mar 28, 2026, 12:11 PM

#

ah

languid canyon Mar 28, 2026, 12:11 PM

#

so if u have multiple cores that keep spamming messages they all steal each others flushing work

#

the last core to log does the flushing

fallen bobcat Mar 28, 2026, 12:17 PM

#

languid canyon the last core to log does the flushing

How do you do this

#

The stealing

languid canyon Mar 28, 2026, 12:17 PM

#

i dont even have smp

#

so i dont do it

fallen bobcat Mar 28, 2026, 12:17 PM

#

It wasn't pointed at you but like in general

languid canyon Mar 28, 2026, 12:17 PM

#

as for how linux does it just check the code

fallen bobcat Mar 28, 2026, 12:18 PM

#

ah ok

languid canyon Mar 28, 2026, 12:18 PM

#

lemme send a link

fallen bobcat Mar 28, 2026, 12:18 PM

#

also do you think the logging stuff belongs in the core kernel or outside of it? Like the console registration and stuff doesn't really make sense to be there idk

#

I suppose I could figure out a better way to layer them nicely

languid canyon Mar 28, 2026, 12:19 PM

#

https://elixir.bootlin.com/linux/v6.19.10/source/kernel/printk/printk.c#L3061

#

basically check everything that uses handover here

languid canyon Mar 28, 2026, 12:20 PM

#

fallen bobcat also do you think the logging stuff belongs in the core kernel or outside of it?...

i dont do nt larp so everything is the core kernel for me

#

but its hard to say

fallen bobcat Mar 28, 2026, 12:20 PM

#

Gotta love their printk file being bigger than my entire kernel

languid canyon Mar 28, 2026, 12:20 PM

#

since even the core kernel logs it probably belongs to the core kernel

fallen bobcat Mar 28, 2026, 12:21 PM

#

The ringbuffer is in the kernel for sure

#

The rest I'm not sure about yet

languid canyon Mar 28, 2026, 12:21 PM

#

fallen bobcat Gotta love their printk file being bigger than my entire kernel

i think its because it contains the /dev/kmsg interface impl etc

fallen bobcat Mar 28, 2026, 1:14 PM

#

fallen bobcat I'll explain why it's wrong in more detail later am on phone rn

what can happen here if the first load is relaxed is that the head load is moved after the tail load, which means the prev_id and stuff are stale and it tries to advance the tail on something that is not actually the tail

#

I think what I can do to sensibly flush logs is:
on early boot, when no consoles are set up and threading and whatever is not initialized: flush synchronously messages in the ringbuffer up to an upper bound (to avoid starvation)
when fully booted, simply set an event when there is a new message and a worker thread waiting on that event will flush the logs

fallen bobcat Mar 28, 2026, 1:47 PM

#

fallen bobcat I think what I can do to sensibly flush logs is: on early boot, when no consoles...

actually no, I dont need to flush synchronously at all, on early boot I can do:

publish_msg();
print_msg_to_platform_debug_write();

#

and all the console stuff can be put somewhere else

fallen bobcat Mar 28, 2026, 2:40 PM

#

the temptation to add new features instead of porting to amd64

pastel citrus Mar 28, 2026, 2:42 PM

#

what new features are you thinking of

#

i would try to port it to amd64 but sadly i can't be bothered to install a separate zig version for this project

fallen bobcat Mar 28, 2026, 2:43 PM

#

I use zvm

#

I just do zvm i master --zls

#

then zvm use master

#

then zig build meme

pastel citrus Mar 28, 2026, 2:43 PM

#

i just installed zig from arch repos lol

#

ill look at zvm

#

why master btw

fallen bobcat Mar 28, 2026, 2:44 PM

#

pastel citrus what new features are you thinking of

actually those can wait

#

I think

#

I wanted to reimplement the allocator and stuff

#

but the amd64 early init doesnt need it

#

I think I can get up to early init at least for now

fallen bobcat Mar 28, 2026, 2:45 PM

#

pastel citrus why master btw

I want to stay up-to-date with latest zig and dont wanna end up having to rewrite lots of code

#

so I do it incrementally

pastel citrus Mar 28, 2026, 2:47 PM

#

ok cool it builds with master, thanks

heavy sandal Mar 28, 2026, 2:47 PM

#

fallen bobcat then zvm use master

zvm install already does use as part of it

fallen bobcat Mar 28, 2026, 2:47 PM

#

idk just to be sure

pastel citrus Mar 28, 2026, 2:47 PM

#

oh yeah i didn't have to do the use part :^)

fallen bobcat Mar 28, 2026, 2:47 PM

#

pastel citrus ok cool it builds with master, thanks

run ./zig-out/bin/kernel-x86_64 -g -t 10 to see fireworks on ten CPUs

languid canyon Mar 28, 2026, 2:47 PM

#

pastel citrus ok cool it builds with master, thanks

Damn you working on zag now??

pastel citrus Mar 28, 2026, 2:48 PM

#

fireworks

fallen bobcat Mar 28, 2026, 2:48 PM

#

yes

#

it shouldnt hang

#

if it hangs tell me and I will cry for 3 hours then try to fix it

pastel citrus Mar 28, 2026, 2:48 PM

#

it does not hang

pastel citrus Mar 28, 2026, 2:48 PM

#

languid canyon Damn you working on zag now??

idk i'll try to do amd64

languid canyon Mar 28, 2026, 2:48 PM

#

Cool

pastel citrus Mar 28, 2026, 2:48 PM

#

and then i'll see maybe i'll work on my own kernel

fallen bobcat Mar 28, 2026, 2:48 PM

#

I can do the early init work honestly if u dont want to

pastel citrus Mar 28, 2026, 2:49 PM

#

idk up to you, i can try but idk what i'm getting myself into

fallen bobcat Mar 28, 2026, 2:49 PM

#

it's annoying boilerplate

#

one thing I wanna try in the future tho is boot using hyper and maybe an efistub

#

that would be cool

languid canyon Mar 28, 2026, 2:50 PM

#

pastel citrus fireworks

Who creates this window btw

fallen bobcat Mar 28, 2026, 2:50 PM

#

sdl

#

-g sets up the bootloader parameters so it has a framebuffer created by sdl

languid canyon Mar 28, 2026, 2:50 PM

#

Interesting

#

Do u emulate limine or smth

fallen bobcat Mar 28, 2026, 2:50 PM

#

no I abstract over it

languid canyon Mar 28, 2026, 2:51 PM

#

Ah

fallen bobcat Mar 28, 2026, 2:51 PM

#

#

um also does this which is funny

languid canyon Mar 28, 2026, 2:52 PM

#

Sounds like what you would expect

fallen bobcat Mar 28, 2026, 2:52 PM

#

also @pastel citrus you might wanna pull and rerun the config thing because I changed kconfig

#

now the log buffer size is set using it

pastel citrus Mar 28, 2026, 2:53 PM

#

cool, thanks

fallen bobcat Mar 28, 2026, 2:54 PM

#

but it's really up to you, if you know other architectures id like that more than amd64 I think

#

because I already have amd64 code that I can rewrite so you dont have to think about it

#

aarch64 would be very cool if you know it

languid canyon Mar 28, 2026, 2:55 PM

#

Naaah dawg

fallen bobcat Mar 28, 2026, 2:55 PM

#

what 💀

heavy sandal Mar 28, 2026, 2:55 PM

#

smh

#

aarch64 hater infy

languid canyon Mar 28, 2026, 2:55 PM

#

Real

fallen bobcat Mar 28, 2026, 2:55 PM

#

languid canyon Naaah dawg

ring buffer assumes 64 bit
opinion rejected

#

I say aarch64 because I have a mac so running it virtualized would be cool

languid canyon Mar 28, 2026, 2:56 PM

#

no committed state in the ring buffer

heavy sandal Mar 28, 2026, 2:56 PM

#

infy just hates arm in general iirc

languid canyon Mar 28, 2026, 2:56 PM

#

KEKW

fallen bobcat Mar 28, 2026, 2:56 PM

#

languid canyon >no committed state in the ring buffer

no one needs commit

pastel citrus Mar 28, 2026, 2:56 PM

#

[iretq@raptor zag]$ zig build iso
iso
└─ install generated to myos.iso
   └─ run xorriso (myos.iso) w
xorriso 1.5.6 : RockRidge filesystem manipulator, libburnia project.

Drive current: -outdev 'stdio:/home/iretq/sources/zag/.zig-cache/o/a9bd2e42e4b51123d6fa024dbf022427/myos.iso'
Media current: stdio file, overwriteable
Media status : is blank
Media summary: 0 sessions, 0 data blocks, 0 data,  948g free
Added to ISO image: directory '/'='/home/iretq/sources/zag/.zig-cache/tmp/0340e873cb132937'
xorriso : UPDATE :      11 files added in 1 seconds
xorriso : UPDATE :      11 files added in 1 seconds
ISO image produced: 2101 sectors
Written to medium : 2101 sectors at LBA 0
Writing to 'stdio:/home/iretq/sources/zag/.zig-cache/o/a9bd2e42e4b51123d6fa024dbf022427/myos.iso' completed successfully.

failed command: xorriso -as mkisofs -R -r -J -b boot/limine/limine-bios-cd.bin -no-emul-boot -boot-load-size 4 -boot-info-table -hfsplus -apm-block-size 2048 --efi-boot boot/limine/limine-uefi-cd.bin -efi-boot-part --efi-boot-image --protective-msdos-label ./.zig-cache/tmp/0340e873cb132937 -o /home/iretq/sources/zag/.zig-cache/o/a9bd2e42e4b51123d6fa024dbf022427/myos.iso```

#

🤔

languid canyon Mar 28, 2026, 2:56 PM

#

heavy sandal infy just hates arm in general iirc

As a platform yes

#

Not as an arch

fallen bobcat Mar 28, 2026, 2:56 PM

#

pastel citrus ``` [iretq@raptor zag]$ zig build iso iso └─ install generated to myos.iso └─...

it says fail because xorriso is retarded and returns 1 on success

pastel citrus Mar 28, 2026, 2:56 PM

#

wow

#

wtf

fallen bobcat Mar 28, 2026, 2:56 PM

#

zig build run should do it all

#

I dnt think it actually builds tho

languid canyon Mar 28, 2026, 2:57 PM

#

fallen bobcat it says fail because xorriso is retarded and returns 1 on success

Hmm pretty sure it doesn't, at least my python thing never tripped on that

fallen bobcat Mar 28, 2026, 2:57 PM

#

main.zig is a stub dont use that, the entry code should be in platform/amd64/entry.zig or smth

pastel citrus Mar 28, 2026, 2:57 PM

#

makes sense, gotta adjust build.zig then

#

i saw that it picks between platforms/um/entry.zig and src/main.zig meme

fallen bobcat Mar 28, 2026, 2:58 PM

#

actually ideally theres something like platform/boot/limine.zig with a limine_entry() or limine_main() that sets up the boot info and passes it to the platform entry

#

thats probably better

#

thats what i did with my old code

heavy sandal Mar 28, 2026, 2:59 PM

#

one thing i always like is having the arch declare a const with the calling convention

#

and then the limine main or whatever can pull that to set the callconv on it

pastel citrus Mar 28, 2026, 2:59 PM

#

fallen bobcat actually ideally theres something like platform/boot/limine.zig with a limine_en...

well you can just comptime { _ = limine.kmain; }

#

in amd64/entry.zig

fallen bobcat Mar 28, 2026, 2:59 PM

#

what does that do

pastel citrus Mar 28, 2026, 2:59 PM

#

just references the symbol

heavy sandal Mar 28, 2026, 3:00 PM

#

forces it to be included in the compilation

#

and not be culled by lazy eval

pastel citrus Mar 28, 2026, 3:00 PM

#

limine being the shared module that does the setup

heavy sandal Mar 28, 2026, 3:00 PM

#

i use it to force include the arch init stuff on mine but leave my krnl.zig as always the module root

fallen bobcat Mar 28, 2026, 3:00 PM

#

I think it should be something like:

if config bootloader is limine:
  entry = platform/boot/limine.zig

then in limine.zig y ou do something like:

fn limine_main():
 convert limine to boot info
 pl.impl.entry(boot_info)

pastel citrus Mar 28, 2026, 3:01 PM

#

actually that's not too stupid

#

seems easier than my thing

fallen bobcat Mar 28, 2026, 3:01 PM

#

that's what I did in my old code

#

acutally I can send you a tarball if youd like to take a look

#

most things are structured similarly

#

it's in C

pastel citrus Mar 28, 2026, 3:01 PM

#

i'm too lazy ngl

fallen bobcat Mar 28, 2026, 3:01 PM

#

alright meme

pastel citrus Mar 28, 2026, 3:03 PM

#

looks like i need to add some more code to config.zig

#

to extract shit like bootloader from the config

fallen bobcat Mar 28, 2026, 3:05 PM

#

no

#

yes

#

you can

#

if you want

#

like getPlatform

pastel citrus Mar 28, 2026, 3:06 PM

#

hm, i added it as a field to platform

#

optional field ofc

#

i can change it later

#

i'm gonna eat now, dinner came up

#

if you want it done relatively quick then you can do it tbh, i usually work pretty slow meme

#

i'm mainly doing it for fun so don't feel bad

fallen bobcat Mar 28, 2026, 3:10 PM

#

pastel citrus if you want it done relatively quick then you can do it tbh, i usually work pret...

I have other stuff to do anyway

pastel citrus Mar 28, 2026, 3:10 PM

#

that's good, i'll see how far i can get with this after i'm done

fallen bobcat Mar 28, 2026, 3:10 PM

#

but like if u can get up to first_init() working that would be great

#

I can do the rest

pastel citrus Mar 28, 2026, 3:11 PM

#

sure sounds good

fallen bobcat Mar 28, 2026, 3:11 PM

#

oh fuck you might need to remove the stuff in kern/init.zig for it to build

#

I just assume std.heap.page_allocator exists

#

so remove all the thread stuff (or put it behind a if (false))

pastel citrus Mar 28, 2026, 3:12 PM

#

i saw some errors earlier related to that lol

fallen bobcat Mar 28, 2026, 3:12 PM

#

moslty just comment out the function and the related code

pastel citrus Mar 28, 2026, 3:12 PM

#

should be easy, thanks

pastel citrus Mar 28, 2026, 4:20 PM

#

error: ld.lld: undefined symbol: memcpy

#

thonk

#

now Why the fuck would that ever happen

heavy sandal Mar 28, 2026, 4:22 PM

#

HUH

#

never seen that one before

fallen bobcat Mar 28, 2026, 4:22 PM

#

pastel citrus now Why the fuck would that ever happen

you need to define it for freestanding

heavy sandal Mar 28, 2026, 4:22 PM

#

no you dont

#

not in zig

fallen bobcat Mar 28, 2026, 4:22 PM

#

yes you do

pastel citrus Mar 28, 2026, 4:22 PM

#

i just copied them from main.zig to root.zig and it works

heavy sandal Mar 28, 2026, 4:22 PM

#

well

#

memcpy the symbol you do

#

but @memcpy exists and __builtin_memcpy exists and is defined in compiler_rt

#

even on freestanding

fallen bobcat Mar 28, 2026, 4:23 PM

#

pastel citrus i just copied them from main.zig to root.zig and it works

thats fine for now but eventually I'll put it like in rtl or something

#

dunno

heavy sandal Mar 28, 2026, 4:23 PM

#

why arent you using @memcpy

pastel citrus Mar 28, 2026, 4:23 PM

#

also, xorriso doesn't return 1 on success

fallen bobcat Mar 28, 2026, 4:23 PM

#

im not even using memcpy!

pastel citrus Mar 28, 2026, 4:23 PM

#

it just prints to stderr

heavy sandal Mar 28, 2026, 4:23 PM

#

and its siblings @memmove and @memset

pastel citrus Mar 28, 2026, 4:23 PM

#

and in zig terms that means it "failed"

#

for some reason

fallen bobcat Mar 28, 2026, 4:23 PM

#

I dont use memcpy anywhere

#

I dont think so

pastel citrus Mar 28, 2026, 4:23 PM

#

zig does

heavy sandal Mar 28, 2026, 4:23 PM

#

fallen bobcat im not even using memcpy!

then what are you linking that needs it??????

fallen bobcat Mar 28, 2026, 4:23 PM

#

zig

heavy sandal Mar 28, 2026, 4:23 PM

#

zig defines it!

fallen bobcat Mar 28, 2026, 4:23 PM

#

the runtime does im sure

#

bruh then idk

heavy sandal Mar 28, 2026, 4:24 PM

#

in compiler_rt!

pastel citrus Mar 28, 2026, 4:24 PM

#

clearly not 😭

fallen bobcat Mar 28, 2026, 4:24 PM

#

llvm might rely on it

pastel citrus Mar 28, 2026, 4:24 PM

#

something is wrong then

heavy sandal Mar 28, 2026, 4:24 PM

#

fallen bobcat llvm might rely on it

zig defines it for llvm in compiler_rt

fallen bobcat Mar 28, 2026, 4:24 PM

#

its probably an llvm thing, but like I stole that from the zig template and it also defined its own stuff

heavy sandal Mar 28, 2026, 4:24 PM

#

im literally linking c code and didnt need to define memcpy/memmove/memset

#

whats your target triple?

#

and your build script for that matter

pastel citrus Mar 28, 2026, 4:24 PM

#

fn targetQueryForPlatform(plat: config.Platform) std.Target.Query {
    var q: std.Target.Query = .{
        .cpu_arch = plat.arch,
        .os_tag = plat.os,
        .abi = if (plat.os == .freestanding) .none else .gnu,
        .cpu_model = .baseline,
    };```

#

.cpu_arch = .x86_64 and .os_tag = .freestanding

heavy sandal Mar 28, 2026, 4:25 PM

#

wait does baseline work for x86_64? ive had problems there

pastel citrus Mar 28, 2026, 4:25 PM

#

yeah it should

#

otherwise it'll do native

#

because you aren't cross compiling

#

that's the logic in the build machinery

#

if (plat.os == .freestanding) {
        kernel.linkage = .static;
        kernel.pie = false;
        kernel.bundle_compiler_rt = false;```

#

bruh

heavy sandal Mar 28, 2026, 4:26 PM

#

theres your issue

#

compiler_rt is what provides memcpy/memset/memmove and all that

fallen bobcat Mar 28, 2026, 4:26 PM

#

this is from the zig template

#

dont blame me

#

trl

pastel citrus Mar 28, 2026, 4:26 PM

#

idk what kind of a shit template you found that does that

heavy sandal Mar 28, 2026, 4:26 PM

#

well then i blame whoever made that template being stupid

fallen bobcat Mar 28, 2026, 4:26 PM

#

https://codeberg.org/CyCl0ne/limine-zig-template

Codeberg.org

limine-zig-template

A template to create a kernel written in Zig using the Limine bootloader.

#

i took that and removed everything apart from the target stuff

pastel citrus Mar 28, 2026, 4:27 PM

#

well there you go

#

should've been using my template

#

(outdated but oh well :^))

fallen bobcat Mar 28, 2026, 4:27 PM

#

yours uses make

heavy sandal Mar 28, 2026, 4:27 PM

#

lmao

pastel citrus Mar 28, 2026, 4:27 PM

#

at least it gets the kernel building right

#

unlike this abomination

fallen bobcat Mar 28, 2026, 4:27 PM

#

hey my build system is really nice

#

i think

pastel citrus Mar 28, 2026, 4:27 PM

#

except for the part where xorriso seemingly fails

#

but it doesnt

#

fallen bobcat Mar 28, 2026, 4:28 PM

#

xorriso issue

pastel citrus Mar 28, 2026, 4:28 PM

#

it's because it prints to stderr

heavy sandal Mar 28, 2026, 4:28 PM

#

yup

pastel citrus Mar 28, 2026, 4:28 PM

#

but it's really annoying ngl

heavy sandal Mar 28, 2026, 4:28 PM

#

thatll do it

fallen bobcat Mar 28, 2026, 4:28 PM

#

doesnt matter it still works

pastel citrus Mar 28, 2026, 4:31 PM

#

😭

heavy sandal Mar 28, 2026, 4:33 PM

#

😭

pastel citrus Mar 28, 2026, 5:00 PM

#

one _ = mkiso.captureStdErr(); later:

#

we ball

fallen bobcat Mar 28, 2026, 5:02 PM

#

very cool

heavy sandal Mar 28, 2026, 5:05 PM

#

nice

pastel citrus Mar 28, 2026, 5:37 PM

#

/// Panics!
pub fn panic(comptime fmt: []const u8, args: anytype) noreturn {
    // FIXME: this is probably broken.
    std.log.err("KERNEL PANIC: " ++ fmt, args);

    while (true) {}
}```

#

meme

#

beautiful

heavy sandal Mar 28, 2026, 5:37 PM

#

nice lol

weary jetty Mar 28, 2026, 5:39 PM

#

fallen bobcat I wanted to reimplement the allocator and stuff

For CXL support?

fallen bobcat Mar 28, 2026, 5:40 PM

#

weary jetty For CXL support?

no I wanted to reimplement my allocator i had written

#

its a fairly standard bonwick slab allocator

#

with magazines and stuff

#

I'm leaning towards making something more freebsd-esque though, their allocator is quite nice

heavy sandal Mar 28, 2026, 5:41 PM

#

one of these days im gonna have to get my head around magazines tbh

weary jetty Mar 28, 2026, 5:41 PM

#

fallen bobcat its a fairly standard bonwick slab allocator

I use that on my end. What's nice about FreeBSD's allocator?

fallen bobcat Mar 28, 2026, 5:41 PM

#

one thing that is cool is they integrate it with their safe memory reclamation mechanism

#

they also claim it is faster than linux's SLUB

#

https://www.youtube.com/watch?v=ZXUIFj4nRjk&themeRefresh=1

YouTube

Samy Bahra

Adhoc: a virtual event to explore systems engineering

Go to the 10:40 mark to skip us fumbling around.

https://youtu.be/ZXUIFj4nRjk?list=PLxYop2KBDBV3pAsLgfo6b2ZiEFse_KZSH&t=637

▶ Play video

heavy sandal Mar 28, 2026, 5:42 PM

#

slub is what ive got on my end

weary jetty Mar 28, 2026, 5:44 PM

#

fallen bobcat https://www.youtube.com/watch?v=ZXUIFj4nRjk&themeRefresh=1

Holy cow, I knew him maybe 15 years ago.

fallen bobcat Mar 28, 2026, 5:44 PM

#

he's a very cool guy

#

he's also the one who wrote their scheduler

weary jetty Mar 28, 2026, 5:45 PM

#

fallen bobcat he's also the one who wrote their scheduler

Oh wow, I had no idea he had got into kernel hacking.

#

He was always a better programmer than I by far.

fallen bobcat Mar 28, 2026, 5:47 PM

#

do you mean jeff?

#

he's also a farmer or something

weary jetty Mar 28, 2026, 5:47 PM

#

fallen bobcat do you mean jeff?

Samy Bahra, I don't know Jeff.

fallen bobcat Mar 28, 2026, 5:47 PM

#

ahh

#

I thought you meant jeff

#

the guy whos doing the talk

weary jetty Mar 28, 2026, 5:47 PM

#

I haven't heard from him in 15 years either.

fallen bobcat Mar 28, 2026, 5:48 PM

#

fallen bobcat he's also a farmer or something

dude casually wrote like 1/4 of modern freebsd as a hobby while working as a farmer 😎

weary jetty Mar 28, 2026, 5:51 PM

#

Ah, so Samy is just randomly putting up the video of Jeff's talk on his channel? Well, he always did have good taste in algorithms, too.

fallen bobcat Mar 28, 2026, 5:51 PM

#

weary jetty Ah, so Samy is just randomly putting up the video of Jeff's talk on his channel?...

it's a systems conference or something, and jeff is doing the first talk for like 15 mins

#

about freebsd's memory allocator and safe memory reclamation

weary jetty Mar 28, 2026, 5:52 PM

#

fallen bobcat about freebsd's memory allocator and safe memory reclamation

Hmm. Samy was always big on epoch reclamation. Maybe it's connected to that.

#

I'm making it cargo fmt and warning clean, will just make sure I can log in on all 3 architectures & then do the initial push to GitHub even without the documentation written yet.

fallen bobcat Mar 28, 2026, 5:56 PM

#

curious to see what you came up with

#

is it all in rust? I know you mentioned sel4 but I'm unsure how it relates

weary jetty Mar 28, 2026, 6:04 PM

#

fallen bobcat curious to see what you came up with

I appear to have accidentally backed out my own pagetable sharing bits, but the critical VM idea is page clustering for a software-configured allocation unit larger than the MMU/TLB mapping size to do 2 things with superpages:

guarantee superpage allocations will not fail due to external fragmentation for superpage sizes less than or equal to the allocation unit
bridge wide gaps between the MMU/TLB base page size and the first superpage size by choosing allocation units to make it easier, thereby trading internal fragmentation to reduce external fragmentation

pastel citrus Mar 28, 2026, 6:05 PM

#

kewl

#

time for some smp ig

#

actually, i'm not even sure if i can do that considering there isn't an allocator

#

fireworks also need an allocator :(

weary jetty Mar 28, 2026, 6:08 PM

#

fallen bobcat is it all in rust? I know you mentioned sel4 but I'm unsure how it relates

It's all in Rust except, of course, for the assembly. The relationship to seL4 was just broadly basing the general design of seL4's.

fallen bobcat Mar 28, 2026, 6:08 PM

#

pastel citrus kewl

yeah which is why i told you to only do early init

#

the allocator thing

pastel citrus Mar 28, 2026, 6:09 PM

#

yeah that is fair

fallen bobcat Mar 28, 2026, 6:09 PM

#

for smp I think it will be bootloader dependent idk

#

like in limine.zig I can probably boot up other cpus

#

and for other bootloaders do it manually

pastel citrus Mar 28, 2026, 6:09 PM

#

yeah, that's kind of where i put ke.ncpus = 1; and ke.cpus = ...;

#

not ideal but whatever

fallen bobcat Mar 28, 2026, 6:10 PM

#

in the future I'll do away with cpus[] entirely

#

I think I'll do the linux thing and only work with ids

weary jetty Mar 28, 2026, 6:10 PM

#

fallen bobcat I think I'll do the linux thing and only work with ids

What is this?

fallen bobcat Mar 28, 2026, 6:11 PM

#

linux uses ids to access cpu-local variables

#

what I did before having cpu-local sections is I had a global array of cpu structs and used that instead

#

but now I'm transitioning towards ids

weary jetty Mar 28, 2026, 6:13 PM

#

I've not done too much thinking about per-CPU affairs. I don't even remember how my magazines find their CPU-local data.

fallen bobcat Mar 28, 2026, 6:14 PM

#

what I do for magazines is store an array of ncpus magazines per kmem cache

#

which is also what solaris does

fallen bobcat Mar 28, 2026, 6:15 PM

#

pastel citrus kewl

also is this running the thread?

#

based

pastel citrus Mar 28, 2026, 6:15 PM

#

yeah

#

https://github.com/48cf/zag/tree/amd64-stuff

#

there are some hacks to get it to work tho

weary jetty Mar 28, 2026, 6:17 PM

#

I wonder why not to have a virtual range dedicated to per-CPU data.

fallen bobcat Mar 28, 2026, 6:18 PM

#

that also works

#

I pretty much have that actually

fallen bobcat Mar 28, 2026, 6:19 PM

#

pastel citrus <https://github.com/48cf/zag/tree/amd64-stuff>

cool but you should've put the thread stuff in ke/amd64

#

nooo

#

nvm it is

#

im stupid

#

yeah no i need to adjust a bunch of stuff (like the per-cpu stuff is just plain wrong) but this should be a nice start

#

thanks

pastel citrus Mar 28, 2026, 6:21 PM

#

it works!!! (in the one-cpu case meme )

fallen bobcat Mar 28, 2026, 6:22 PM

#

it should use the gs offset trick

#

sec

pastel citrus Mar 28, 2026, 6:22 PM

#

you mean where gs is the percpu_offset?

#

and you just end up doing gs:variable

fallen bobcat Mar 28, 2026, 6:22 PM

#

this

#

I guess for now its fine since smp stuff isnt done

weary jetty Mar 28, 2026, 6:24 PM

#

There's no way legacy features like segment registers can't have huge legacy emulation overheads.

pastel citrus Mar 28, 2026, 6:24 PM

#

why lol

#

that's literally how memory works on x86

heavy sandal Mar 28, 2026, 6:24 PM

#

yeah

pastel citrus Mar 28, 2026, 6:24 PM

#

memory addressing in general

heavy sandal Mar 28, 2026, 6:24 PM

#

even on long mode its all segments still

pastel citrus Mar 28, 2026, 6:25 PM

#

although segments dont affect virtual->linear address translation in long mode

#

fs/gsbase still does

heavy sandal Mar 28, 2026, 6:25 PM

#

iirc fs/gsbase is actuaslly the segment base register just extended and only directly accessible through the msrs

weary jetty Mar 28, 2026, 6:25 PM

#

I think there's a default segment register. I don't know that other segment registers suffer large legacy emulation overheads.

heavy sandal Mar 28, 2026, 6:25 PM

#

thats why i keep separate gdts percpu

#

because the fsgsbase is tied to it somehow

weary jetty Mar 28, 2026, 6:26 PM

#

heavy sandal thats why i keep separate gdts percpu

That also sounds relatively reasonable.

fallen bobcat Mar 28, 2026, 6:27 PM

#

also minor nitpick: thread context can be kept fully on the stack

weary jetty Mar 28, 2026, 6:28 PM

#

fallen bobcat also minor nitpick: thread context can be kept fully on the stack

I thought TLS was mostly done like an ELF section.

fallen bobcat Mar 28, 2026, 6:29 PM

#

no

#

I'm talking about context switching

weary jetty Mar 28, 2026, 6:30 PM

#

Oh, x86 context handling is insane enough that I have to do it straight out of the book every time.

#

The tech debt exists, but squashing it doesn't have to be done by hand, so in some respects the rate-limiting factor is my tech debt assessment & spotting speed.

fallen bobcat Mar 28, 2026, 6:41 PM

#

weary jetty It's all in Rust except, of course, for the assembly. The relationship to seL4 w...

ah I thought you were building on top of sel4

#

so is it a microkernel still?

weary jetty Mar 28, 2026, 6:43 PM

#

fallen bobcat so is it a microkernel still?

It's supposed to be. I should probably go over it with a fine-toothed comb & make sure all of the design goals are being honoured.

fallen bobcat Mar 28, 2026, 6:43 PM

#

@pastel citrus do you plan on continuing the port or not? if so idt and gdt setup would be cool otherwise I think this is enough for now for me to go clean up and implement the allocator and stuff before we move on to a full port

#

mhm well i guess setting up the boot info for real would be good too

heavy sandal Mar 28, 2026, 6:45 PM

#

i assume yall already know this since i think ive mentioned before but if you use export for the limine protocol stuff dont forget to call std.mem.doNotOptimizeAway so the compiler doesnt constant-fold the null default response pointer

pastel citrus Mar 28, 2026, 6:46 PM

#

fallen bobcat <@796462670995390524> do you plan on continuing the port or not? if so idt and g...

uhh yeah i could still work on it, but if you want to do stuff now then i'd just get that merged in and continue yourself because i need to leave for a bit

fallen bobcat Mar 28, 2026, 6:46 PM

#

yeah that would work

pastel citrus Mar 28, 2026, 6:46 PM

#

do you want me to make a pr?

fallen bobcat Mar 28, 2026, 6:46 PM

#

yeah

pastel citrus Mar 28, 2026, 6:46 PM

#

alright

fallen bobcat Mar 28, 2026, 6:46 PM

#

Ill merge it directly and clean it up

pastel citrus Mar 28, 2026, 6:46 PM

#

https://github.com/rdmsr/zag/pull/1

#

sounds good

fallen bobcat Mar 28, 2026, 6:46 PM

#

or I can do like you did with my limine PR meme

pastel citrus Mar 28, 2026, 6:47 PM

#

nooo

#

but yeah, i can always work on some other stuff later

weary jetty Mar 28, 2026, 6:49 PM

#

Just a quick interactive login touch test on all 3 architectures now supported needs to get passed & the initial git push is imminent.

fallen bobcat Mar 28, 2026, 6:49 PM

#

aaa you broke um

pastel citrus Mar 28, 2026, 6:49 PM

#

wait how

#

oh it's probably me commenting out fireworks

#

and the make_thread thing

fallen bobcat Mar 28, 2026, 6:50 PM

#

no

#

the build system

pastel citrus Mar 28, 2026, 6:50 PM

#

wait really?

fallen bobcat Mar 28, 2026, 6:50 PM

#

its ok I fixed it

pastel citrus Mar 28, 2026, 6:50 PM

#

is it the platform entry file thing

fallen bobcat Mar 28, 2026, 6:50 PM

#

like it shouldnt return error.noBootloader if on linux

#

and shouldnt use platform/x86_64

pastel citrus Mar 28, 2026, 6:50 PM

#

oh ☠️

#

ah i see how that works

weary jetty Mar 28, 2026, 6:53 PM

#

fallen bobcat aaa you broke um

I'm not yet but it's coming fast.

pastel citrus Mar 28, 2026, 6:53 PM

#

what

fallen bobcat Mar 28, 2026, 6:54 PM

#

idk if having it be boot/limine.zig is a good idea (that's not what you did though)

#

because like if i end up on an architecture where theres only one bootloader isnt it weird to have boot/weirdbootloader.zig

#

idk

fallen bobcat Mar 28, 2026, 6:55 PM

#

fallen bobcat because like if i end up on an architecture where theres only one bootloader isn...

ah well in that case I can just set it up in its own entry.zig

#

ok this is fine

weary jetty Mar 28, 2026, 7:04 PM

#

The pool_cache(9) manpage doesn't describe the implementation in too much detail.

fallen bobcat Mar 28, 2026, 7:10 PM

#

anyways this is really cool thanks a lot @pastel citrus

#

now I have to do homework

pastel citrus Mar 28, 2026, 7:54 PM

#

im glad i could be of some help

#

even tho it was broken KEKW

pastel citrus Mar 28, 2026, 7:56 PM

#

fallen bobcat idk if having it be boot/limine.zig is a good idea (that's not what you did thou...

oh yeah, i forgot what exactly you wanted lol, it’s a trivial change tho

#

but in that case you can just have a function that decides the entry file path based on arch+bootloader

fallen bobcat Mar 29, 2026, 8:07 PM

#

ok time for more amd64 stuff

#

also its crazy how discord hides my own thread after like 1 day of inactivity

#

I have to go search for MY OWN thread 😭

heavy sandal Mar 29, 2026, 8:07 PM

#

it does that for all threads

fallen bobcat Mar 29, 2026, 8:08 PM

#

but cant I set it so that it doesnt disappear for me

heavy sandal Mar 29, 2026, 8:08 PM

#

i have a discord server that only im in and i just made a link to my thread and put it there lmao

#

yeah itd be nice to keep your own threads there longer

pastel citrus Mar 29, 2026, 8:08 PM

#

you might be able to easily search for it if you press ctrl+t in the desktop app

#

i dont think there is a shortcut for that on the web version tho

fallen bobcat Mar 29, 2026, 8:09 PM

#

there's hide after inactivity thing

#

but I think there was something in #announcements not to do that

weary jetty Mar 29, 2026, 8:54 PM

#

fallen bobcat ok time for more amd64 stuff

I've been porting to LoongArch64 and MIPS64, generalising things so it's not merely assumed that there's only one superpage size, I dropped in some easy cases for pagetable sharing & will probably let things spin doing fixes for LTP (Linux Test Project) overnight. I wonder if there are things like that for other kernels so I can get some other personalities going. *BSD? Solaris? Windows NT? OpenVMS? NFI.
I cooked up a data structure BTW: https://github.com/NadiaYvette/hybrid-radix-btree-adaptive
I also pushed telix upstream to my GitHub just in case you didn't catch it.

GitHub

GitHub - NadiaYvette/hybrid-radix-btree-adaptive: Hybrid adaptive r...

Hybrid adaptive radix/B+ tree: ART for dense regions, sorted pages for sparse, bitmap leaves at depth 3. Outperforms ART 1.4-12.9x across all workloads. - NadiaYvette/hybrid-radix-btree-adaptive

fallen bobcat Mar 29, 2026, 8:54 PM

#

oh didnt see you pushed it, nice

weary jetty Mar 29, 2026, 8:55 PM

#

I never finished the docs but I got a first cut of the abstract & am going over revisions I need to make.

fallen bobcat Mar 29, 2026, 8:56 PM

#

I'm gonna be interested for that more than the code since I'm not much of a rust guy

weary jetty Mar 29, 2026, 8:58 PM

#

fallen bobcat I'm gonna be interested for that more than the code since I'm not much of a rust...

I can write FORTRAN in any language. Maybe I have to squint sometimes to get through Rust's syntax being different. I can still read & write it, albeit maybe not quite as swiftly as C or Ada or FORTRAN.

#

The Adaptive Radix Tree apparently has some relatively simple techniques to address its performance breakdown in sparse datasets, provided you're not shy about plugging B+ tree code for the sparse case.

#

I might bring it over to telix b/c it basically wins against B+ trees and ARTs by being the best of both by switching between them.

fallen bobcat Mar 29, 2026, 9:06 PM

#

@heavy sandal do you use the limine header directly or did you write your own bindings

weary jetty Mar 29, 2026, 9:07 PM

#

BTW the data structure code is in C.

fallen bobcat Mar 29, 2026, 9:31 PM

#

bruh I can name fields type

#

I thought I couldnt so I named everything kind

heavy sandal Mar 29, 2026, 9:47 PM

#

fallen bobcat <@121733282676211714> do you use the limine header directly or did you write you...

i took https://github.com/48cf/limine-zig and modified slightly when i was using limine

GitHub

GitHub - 48cf/limine-zig: A Zig library for handling Limine boot pr...

A Zig library for handling Limine boot protocol structures. - 48cf/limine-zig

fallen bobcat Mar 29, 2026, 9:47 PM

#

i ended up writing my own thing

heavy sandal Mar 29, 2026, 9:47 PM

#

fallen bobcat bruh I can name fields `type`

yeah afaik only decls cant be named type, because fields can only ever be accessed with dot-member syntax

weary jetty Mar 29, 2026, 9:48 PM

#

For what architecture(s) does it do bootloading?

heavy sandal Mar 29, 2026, 9:48 PM

#

limine? a bunch, idk the list offhand

languid canyon Mar 29, 2026, 9:51 PM

#

weary jetty For what architecture(s) does it do bootloading?

All the relevant ones

pastel citrus Mar 29, 2026, 10:00 PM

#

heavy sandal i took https://github.com/48cf/limine-zig and modified slightly when i was using...

i'm probably gonna update that soon

#

i found myself needing limine bindings for my new wip project trl

weary jetty Mar 29, 2026, 10:09 PM

#

From looking at EFI docs, I can't piece together quite enough details to figure out how to get from end to end, but it looks like the environment it drops things e.g. bootloaders into is easily rich enough to not be difficult to operate in.

heavy sandal Mar 29, 2026, 10:09 PM

#

yeah efi aint bad at all

fallen bobcat Mar 29, 2026, 10:15 PM

#

pastel citrus i found myself needing limine bindings for my new wip project <:trl:829316641191...

Wait are you using zig again

fallen bobcat Mar 29, 2026, 10:16 PM

#

weary jetty From looking at EFI docs, I can't piece together quite enough details to figure ...

The problem is manufacturers are often writing broken firmware

pastel citrus Mar 29, 2026, 10:16 PM

#

trying to, yeah, but i am not gonna stick to that project for very long

fallen bobcat Mar 29, 2026, 10:16 PM

#

Bruh don't be so hard on yourself

pastel citrus Mar 29, 2026, 10:17 PM

#

nah i'm not, i just have no motivation to work on it

#

wanting to do something cool isn't enough

weary jetty Mar 29, 2026, 10:17 PM

#

fallen bobcat The problem is manufacturers are often writing broken firmware

There's never an end to that, true enough.

pastel citrus Mar 29, 2026, 10:17 PM

#

i tried to read your and infy's log rings and i just realised i have no clue what the hell i'm doing

fallen bobcat Mar 29, 2026, 10:18 PM

#

it's not like either of us could've implemented it without the Linux code lol

#

I did just write more comments tho if you wanna check that out

languid canyon Mar 29, 2026, 10:19 PM

#

fallen bobcat it's not like either of us could've implemented it without the Linux code lol

Yep

languid canyon Mar 29, 2026, 10:19 PM

#

pastel citrus trying to, yeah, but i am not gonna stick to that project for very long

Why not

weary jetty Mar 29, 2026, 10:19 PM

#

fallen bobcat it's not like either of us could've implemented it without the Linux code lol

What is „it”?

languid canyon Mar 29, 2026, 10:19 PM

#

Log ring

weary jetty Mar 29, 2026, 10:21 PM

#

I'm not sure I know what that is offhand.

pastel citrus Mar 29, 2026, 10:25 PM

#

a lockless ring buffer for kernel logs

pastel citrus Mar 29, 2026, 10:25 PM

#

languid canyon Why not

[...] i just have no motivation to work on it

languid canyon Mar 29, 2026, 10:25 PM

#

nooo

#

Work on it for 3 days so u can surpass ultra in features

pastel citrus Mar 29, 2026, 10:27 PM

#

eh i'm not really interested in having more features

#

i want quality

languid canyon Mar 29, 2026, 10:27 PM

#

Damn

pastel citrus Mar 29, 2026, 10:27 PM

#

and i just don't have the ability to put in these hours researching stuff and learning everything like you do

fallen bobcat Mar 29, 2026, 10:27 PM

#

infy

#

uLog when

languid canyon Mar 29, 2026, 10:27 PM

#

Never lol

pastel citrus Mar 29, 2026, 10:27 PM

#

:(

languid canyon Mar 29, 2026, 10:28 PM

#

U should make one

fallen bobcat Mar 29, 2026, 10:28 PM

#

lol no

languid canyon Mar 29, 2026, 10:28 PM

#

Your ring is more general purpose since its 32 bit

pastel citrus Mar 29, 2026, 10:28 PM

#

i'd be the first to put together a kernel made out of all these uXXX libraries

#

fallen bobcat Mar 29, 2026, 10:28 PM

#

It's pointless to make that an embeddable library

languid canyon Mar 29, 2026, 10:28 PM

#

pastel citrus and i just don't have the ability to put in these hours researching stuff and le...

Why?

languid canyon Mar 29, 2026, 10:28 PM

#

fallen bobcat It's pointless to make that an embeddable library

I could see some utility in that

fallen bobcat Mar 29, 2026, 10:28 PM

#

idk the thought of not NIHing stuff scares me

pastel citrus Mar 29, 2026, 10:28 PM

#

idk, adhd maybe (diagnosis pending)

languid canyon Mar 29, 2026, 10:28 PM

#

A lot of people want one

fallen bobcat Mar 29, 2026, 10:29 PM

#

To me it seems lazy

#

uACPI is where I draw the line

#

cuz I ain't writing that shit

languid canyon Mar 29, 2026, 10:29 PM

#

Maybe

fallen bobcat Mar 29, 2026, 10:29 PM

#

but like flanterm and stuff ehh

languid canyon Mar 29, 2026, 10:29 PM

#

pastel citrus idk, adhd maybe (diagnosis pending)

Fair enough

fallen bobcat Mar 29, 2026, 10:29 PM

#

languid canyon Maybe

My goal is to learn

#

If I reuse a bunch of libraries I don't learn anything

languid canyon Mar 29, 2026, 10:30 PM

#

Yeah but you already made the code

#

U can now make it a library

fallen bobcat Mar 29, 2026, 10:30 PM

#

Yes but the philosophy of reusing code others wrote in general annoys me

#

idk

languid canyon Mar 29, 2026, 10:30 PM

#

Ig

fallen bobcat Mar 29, 2026, 10:32 PM

#

pastel citrus idk, adhd maybe (diagnosis pending)

What I do if I'm reading an OS book for example is read like 10-15 mins a day

pastel citrus Mar 29, 2026, 10:43 PM

#

i hate reading

#

probably part of why im bad at this ;D

fallen bobcat Mar 29, 2026, 10:55 PM

#

paging time soon nooo

#

oh wait no i need gdt init.. ok

#

first

weary jetty Mar 29, 2026, 10:55 PM

#

MIPS is misbehaving here.

fallen bobcat Mar 29, 2026, 10:59 PM

#

ffs i hate doing architecture stuff

#

its so boring

weary jetty Mar 29, 2026, 10:59 PM

#

Usually MIPS has been the best-behaved of all architectures for me. I can't spot what's off about things that's making the first context switch to userspace fail.

weary jetty Mar 29, 2026, 10:59 PM

#

fallen bobcat its so boring

It helps if it's not the same architecture every time.

fallen bobcat Mar 29, 2026, 10:59 PM

#

yeah true

#

I've only really done x86

weary jetty Mar 29, 2026, 11:00 PM

#

fallen bobcat I've only really done x86

I used to have a lot of cool stuff. Nowadays all I've got is qemu.

languid canyon Mar 29, 2026, 11:02 PM

#

Lucky for you qemu emulates a ton of arches

weary jetty Mar 29, 2026, 11:04 PM

#

languid canyon Lucky for you qemu emulates a ton of arches

To a very limited degree. A number of the systems' values were that testing on them was desperately needed.

languid canyon Mar 29, 2026, 11:04 PM

#

@fallen bobcat in your UM target you do context switching right? Since it's just one target how do u handle different host arches?

fallen bobcat Mar 29, 2026, 11:04 PM

#

I use ucontext

pastel citrus Mar 29, 2026, 11:04 PM

#

ucontext

#

oops

languid canyon Mar 29, 2026, 11:04 PM

#

how does that work?

#

is that posix or linux

fallen bobcat Mar 29, 2026, 11:05 PM

#

ucontext does have overhead though, so I'm considering just emitting different assembly stubs for every arch

#

its posix

languid canyon Mar 29, 2026, 11:05 PM

#

tbh who cares about overhead for debugging

fallen bobcat Mar 29, 2026, 11:05 PM

#

yeah true

#

ideally posix would introduce a new generic way though since ucontext is deprecated

weary jetty Mar 29, 2026, 11:05 PM

#

qemu doesn't have the access to the systems to verify the accuracy of their emulation or simulation either, and so often doesn't even attempt to emulate the systems. It just emulates the most common ones that people don't have trouble getting to begin with.

languid canyon Mar 29, 2026, 11:06 PM

#

i wanna make a um target as well, ill probably get inspired by your code

fallen bobcat Mar 29, 2026, 11:06 PM

#

please solve memory for me meme

#

so I dont have to think

languid canyon Mar 29, 2026, 11:06 PM

#

thats a lost cause

#

maybe iretq solves it for us

fallen bobcat Mar 29, 2026, 11:06 PM

#

lol

#

I do have an idea but not sure yet, I think I'll just ignore the vma limit lol

weary jetty Mar 29, 2026, 11:06 PM

#

fallen bobcat please solve memory for me <:meme:575445094589661215>

In what respect?

languid canyon Mar 29, 2026, 11:06 PM

#

how do u solve it atm

#

just random bullsiht go?

fallen bobcat Mar 29, 2026, 11:07 PM

#

well atm there is no memory at all, but in the old code I just mmaped it

#

i didnt handle switching address spaces tho which is what's actually tricky

languid canyon Mar 29, 2026, 11:07 PM

#

how do u do fireworks and stuff without memory management?

fallen bobcat Mar 29, 2026, 11:07 PM

#

well for now it uses mmap directly

#

to allocate memory

languid canyon Mar 29, 2026, 11:07 PM

#

its basically kernel mode threads now right

fallen bobcat Mar 29, 2026, 11:07 PM

#

fallen bobcat well atm there is no memory at all, but in the old code I just mmaped it

this worked though

fallen bobcat Mar 29, 2026, 11:08 PM

#

languid canyon its basically kernel mode threads now right

fireworks is always kernel mode

languid canyon Mar 29, 2026, 11:08 PM

#

oh ok

fallen bobcat Mar 29, 2026, 11:08 PM

#

its just a stress test

fallen bobcat Mar 29, 2026, 11:08 PM

#

fallen bobcat i didnt handle switching address spaces tho which is what's actually tricky

this is what I mean by "solving memory" btw

languid canyon Mar 29, 2026, 11:08 PM

#

do u have a slub?

fallen bobcat Mar 29, 2026, 11:08 PM

#

no I have a bonwick-style slab

languid canyon Mar 29, 2026, 11:08 PM

#

whats there to switch if its just kernel threads?

fallen bobcat Mar 29, 2026, 11:09 PM

#

I also have cool ideas for that

fallen bobcat Mar 29, 2026, 11:09 PM

#

languid canyon whats there to switch if its just kernel threads?

eventually you might want to run user apps

languid canyon Mar 29, 2026, 11:09 PM

#

i wonder how u would even do that

#

without it being incredibly slow

fallen bobcat Mar 29, 2026, 11:09 PM

#

the naive solution is mapping and unmapping everything

languid canyon Mar 29, 2026, 11:09 PM

#

yeah

fallen bobcat Mar 29, 2026, 11:09 PM

#

korona pointed out there was a VMA limit issue too

#

but apparently linux doesnt care

languid canyon Mar 29, 2026, 11:10 PM

#

yeah but like

#

unmap everything would be slow af

fallen bobcat Mar 29, 2026, 11:10 PM

#

what linux does iirc is mmap only as a TLB of some sort

languid canyon Mar 29, 2026, 11:10 PM

#

on every ctx switch

fallen bobcat Mar 29, 2026, 11:10 PM

#

yeah idk

#

which is why its tricky

languid canyon Mar 29, 2026, 11:10 PM

#

does it rely on userfaultfd?

fallen bobcat Mar 29, 2026, 11:10 PM

#

well I would rely on it anyway

#

for CoW

#

and demand paging

languid canyon Mar 29, 2026, 11:11 PM

#

um sounds like pain tbh

pastel citrus Mar 29, 2026, 11:11 PM

#

it is

fallen bobcat Mar 29, 2026, 11:11 PM

#

it's not

#

just dont do user apps if you dont feel the need to

#

I dont plan to have them rn but eventually it'd be very cool

languid canyon Mar 29, 2026, 11:11 PM

#

basically the only thing you gain is asan

fallen bobcat Mar 29, 2026, 11:11 PM

#

this is literally the only tricky part

languid canyon Mar 29, 2026, 11:11 PM

#

which kasan would be cheaper to do

fallen bobcat Mar 29, 2026, 11:12 PM

#

and full gdb

#

and tsan

#

and valgrind

#

and perf

languid canyon Mar 29, 2026, 11:12 PM

#

tsan doesnt work

#

for barriers

fallen bobcat Mar 29, 2026, 11:12 PM

#

ok true

languid canyon Mar 29, 2026, 11:12 PM

#

valgrind probably would struggle but maybe

fallen bobcat Mar 29, 2026, 11:12 PM

#

idk its like 200 lines of code so far and its really useful

languid canyon Mar 29, 2026, 11:12 PM

#

maybe if u omit the usermode part

fallen bobcat Mar 29, 2026, 11:12 PM

#

to me at least

languid canyon Mar 29, 2026, 11:12 PM

#

its good

fallen bobcat Mar 29, 2026, 11:12 PM

#

yeah you don't have to support user apps

#

you can also run unit tests/integration tests

languid canyon Mar 29, 2026, 11:13 PM

#

someone here literally implemented ioapic etc emulation

#

to verify correctness

#

thats a bit too far imo

fallen bobcat Mar 29, 2026, 11:13 PM

#

apache?

languid canyon Mar 29, 2026, 11:13 PM

#

probably

fallen bobcat Mar 29, 2026, 11:14 PM

#

personally I think it's a bit stupid to do that

languid canyon Mar 29, 2026, 11:14 PM

#

like for stuff like that u can just boot qemu and get stdout

fallen bobcat Mar 29, 2026, 11:14 PM

#

because you dont know if your emulation is correct, and if it is then you probably already know that your other code is correct

languid canyon Mar 29, 2026, 11:14 PM

#

yeah

fallen bobcat Mar 29, 2026, 11:14 PM

#

like you just end up writing tests you know will pass

#

so it's pointless

languid canyon Mar 29, 2026, 11:15 PM

#

regression testing is still important but yes

fallen bobcat Mar 29, 2026, 11:15 PM

#

well it's up to you if you implement it, but personally i have found it very useful

#

and I will keep shilling it

#

like if you consider most people here already move their code to userspace to test it first, I just go further and run the whole thing

languid canyon Mar 29, 2026, 11:16 PM

#

probably

weary jetty Mar 29, 2026, 11:18 PM

#

I've been running on qemu, so it's all debuggable anyway.

languid canyon Mar 29, 2026, 11:19 PM

#

id say its not phenomenal value to run it in usermode but still useful

weary jetty Mar 29, 2026, 11:23 PM

#

The ART-B+ tree hybrid seems like it should be deployable.

fallen bobcat Mar 30, 2026, 12:16 AM

#

I have come up with a nice cpuid abstraction

#

    const feature_info = CpuidRequest.execute(.feature_info);
    const extended_features = CpuidRequest.execute(.{ .extended_features = .first });
    const extended_processor_info = CpuidRequest.execute(.extended_info);
    const power_management_info = CpuidRequest.execute(.power_management_info);

heavy sandal Mar 30, 2026, 12:17 AM

#

nice

languid canyon Mar 30, 2026, 12:41 AM

#

Why is it needed since cpuid is arch specific features anyway?

fallen bobcat Mar 30, 2026, 12:41 AM

#

languid canyon Why is it needed since cpuid is arch specific features anyway?

it's just nicer to work with

languid canyon Mar 30, 2026, 12:42 AM

#

Hm

fallen bobcat Mar 30, 2026, 12:54 AM

#

meme

#

bruh my cpu doesnt support pcid

heavy sandal Mar 30, 2026, 1:52 AM

#

@fallen bobcat btw do you trigger DPCs using an interrupt or do you just handle them as part of returning from interrupts/lowering ipl?

fallen bobcat Mar 30, 2026, 1:58 AM

#

heavy sandal <@379432087620550658> btw do you trigger DPCs using an interrupt or do you just ...

the latter

#

I do it all in software

#

You can do it all in hardware (using cr8) and use self-IPIs though

heavy sandal Mar 30, 2026, 1:59 AM

#

self-ipis is what i do but im running into race condition hell about it atm

fallen bobcat Mar 30, 2026, 1:59 AM

#

On windows the irql functions are basically inlined to a mov cr8

heavy sandal Mar 30, 2026, 1:59 AM

#

(though i dont think its actually because of the route i took i think its just bad code i wrote)

fallen bobcat Mar 30, 2026, 1:59 AM

#

heavy sandal self-ipis is what i do but im running into race condition hell about it atm

What is it

heavy sandal Mar 30, 2026, 1:59 AM

#

fallen bobcat On windows the irql functions are basically inlined to a mov cr8

yeah ik, was just curious how you did it

heavy sandal Mar 30, 2026, 2:01 AM

#

fallen bobcat What is it

ive got a function that queues DPCs if youre at or above dispatch and just runs the passed function if you arent. my event sync primitive uses that to defer signalling to dispatch irql to prevent race conditions in the yield function, but im running into use-after free on that event signalling dpc if the event never gets waited on

#

which is weird because the place where it isnt being waited on is running at a level where it should use the immediate callback and be fine

fallen bobcat Mar 30, 2026, 2:05 AM

#

Why are you dynamically allocating your DPCs in the first place

heavy sandal Mar 30, 2026, 2:06 AM

#

the use-after-free isnt the dpc, its the event object being passed

#

its just that i dont understand why the dpc is being delayed so much

fallen bobcat Mar 30, 2026, 1:44 PM

#

bruh I got mails from the Linux mailing list but the mail client on my phone is ass

#

And they're all European so I'm fucked until tomorrow

languid canyon Mar 30, 2026, 1:46 PM

#

fallen bobcat bruh I got mails from the Linux mailing list but the mail client on my phone is ...

what did they say

#

oh cool u got 2 reviewed-bys

fallen bobcat Mar 30, 2026, 1:53 PM

#

Wanna tell them maybe we should merge this and do another patch for other fixups

weary jetty Mar 30, 2026, 4:21 PM

#

fallen bobcat And they're all European so I'm fucked until tomorrow

I think there's enough instability ongoing in the USA that a lot of ppl likely made arrangements to go somewhere else like I did, which might explain why they're all here in Europe now (I myself am in Berlin).

#

What's a DPC?

heavy sandal Mar 30, 2026, 4:22 PM

#

weary jetty What's a DPC?

deferred procedure call

#

basically just something you can queue up from inside an interrupt handler that then runs at a lower interrupt priority level

weary jetty Mar 30, 2026, 4:23 PM

#

Ah, C can be painful to do continuation-passing style within.

heavy sandal Mar 30, 2026, 4:23 PM

#

i think DPCs are a windows-ism originally?

#

idk maybe someone else did it before NT

#

and it isnt really continuation passing

#

its more like starting a thread or queuing something in a thread pool

fallen bobcat Mar 30, 2026, 4:45 PM

#

weary jetty Ah, C can be painful to do continuation-passing style within.

It's like linux softirqs

#

DPCs

heavy sandal Mar 30, 2026, 4:45 PM

#

yeah

#

except someone realised you can use the softirqs for other things that arent irqs too

weary jetty Mar 30, 2026, 4:46 PM

#

fallen bobcat It's like linux softirqs

Linux' softirqs got phased out 20+ years ago, no?

fallen bobcat Mar 30, 2026, 4:48 PM

#

No

heavy sandal Mar 30, 2026, 4:48 PM

#

not afaik? and in any case the nt kernel's dpcs are still being used to this day

#

(well the executive's dpcs. nt things calling it separately KEKW )

weary jetty Mar 30, 2026, 4:54 PM

#

fallen bobcat No

I seriously remember everything getting moved off softirqs and onto kernel work queues or similar.

#

I think it was even part of the Big Kernel Lock removal.

heavy sandal Mar 30, 2026, 4:55 PM

#

windows actually has something similar to that difference too

#

normal DPCs (similar to softirq) vs Threaded DPCs (similar to work queues)

weary jetty Mar 30, 2026, 4:56 PM

#

The sweep was total IIRC. There should be no softirqs left in Linux.

fallen bobcat Mar 30, 2026, 5:28 PM

#

weary jetty I seriously remember everything getting moved off softirqs and onto kernel work ...

Now there's tasklets

#

too

#

which are built on softirqs

weary jetty Mar 30, 2026, 6:01 PM

#

fallen bobcat which are built on softirqs

If they're really still in there, I'm surprised I'm misremembering what went on.

crimson sand Mar 30, 2026, 6:05 PM

#

fallen bobcat Now there's tasklets

https://lwn.net/Articles/960041/

LWN.net

The end of tasklets

A common problem in kernel development is controlling when a specific task should be done. Ker [...]

weary jetty Mar 30, 2026, 6:05 PM

#

My (micro)kernel never had a Big Kernel Lock & whatever was wrong with softirqs to begin with. The message-passing -based IO/VFS layers also make asynchronous IO work & M:N threading via scheduler activations makes threading facilities for various programming languages actually work, too (tested with Haskell/ghc, though I should add more languages' runtimes).

fallen bobcat Mar 30, 2026, 6:10 PM

#

crimson sand https://lwn.net/Articles/960041/

ok so it's like WQ_BH then

heavy sandal Mar 30, 2026, 6:11 PM

#

i should probably implement threaded DPCs tbh (which are more like workqueues)

fallen bobcat Mar 30, 2026, 6:14 PM

#

I will just have a generic workpool thing

heavy sandal Mar 30, 2026, 6:15 PM

#

thats also an option

#

ive been waffling back and forth between whether to implement a kernel thread pool or threaded DPCs

#

(threaded DPCs effectively act as a percpu thread pool with one worker iirc)

weary jetty Mar 30, 2026, 6:26 PM

#

I'm not sure if it's exactly applicable to telix. Things run in microkernels' version of „userspace” so a lot of things could be written in Haskell or Mercury or Idris2 or F# or F* or Scala3 or whatever.

heavy sandal Mar 30, 2026, 6:27 PM

#

yeah dpcs are only really applicable to stuff running in-kernel imo

#

so if youre doing microkernel stuff then it probably isnt as applicable

weary jetty Mar 30, 2026, 6:32 PM

#

I'm kicking off a feature completeness kick with a bunch of filesystems and fancy network things and additional architectures.

#

I can probably push Xwayland + GNOME + Firefox up a fair amount. It would probably be good to have that to demo fairly early.

fallen bobcat Mar 30, 2026, 6:51 PM

#

fallen bobcat bruh I got mails from the Linux mailing list but the mail client on my phone is ...

seems like the linux guys handled it

#

I literally just sent a 1 character patch and didnt say anything else chad

fallen bobcat Mar 30, 2026, 7:56 PM

#

my project structure is so weird zls instantly crashes now

#

nooo

weary jetty Mar 30, 2026, 7:57 PM

#

fallen bobcat my project structure is so weird zls instantly crashes now

zls?

fallen bobcat Mar 30, 2026, 7:57 PM

#

zig language server

weary jetty Mar 30, 2026, 7:59 PM

#

Oh yeah, LSP servers have trouble with projects that do funny things when they build.

heavy sandal Mar 30, 2026, 8:00 PM

#

fallen bobcat my project structure is so weird zls instantly crashes now

impressive

fallen bobcat Mar 30, 2026, 8:01 PM

#

i switched to zed like a week ago because it seemed to handle it better than emacs but now both dont like it

fallen bobcat Mar 30, 2026, 9:41 PM

#

omfg I shouldve designed everything with cpu ids from the start 😭

heavy sandal Mar 30, 2026, 9:41 PM

#

your own ids or the hardware ones?

fallen bobcat Mar 30, 2026, 9:42 PM

#

well instead of passing around a *ke.Cpu is pass around an id now

#

so the big struct has been shattered

heavy sandal Mar 30, 2026, 9:42 PM

#

oh yeah no passing around an id is way better

fallen bobcat Mar 30, 2026, 9:46 PM

#

well at least it builds now, so I can get back to amd64 stuff

fallen bobcat Mar 30, 2026, 11:49 PM

#

mhm moving the directory structure around a bit:
arch/ -> ISA primitives (like cpuid, table formats etc)
platform/ -> platform-specific stuff (board etc)

#

then each subsystem has its own architecture subdirectory

fallen bobcat Mar 31, 2026, 1:23 AM

#

bruh gdt init not ok

heavy sandal Mar 31, 2026, 1:28 AM

#

fallen bobcat bruh gdt init not ok

if youre using an "m" constraint to set the gdtr then thats the problem, if not then id need to look at it to see whats up

fallen bobcat Mar 31, 2026, 1:28 AM

#

no I fixed it

heavy sandal Mar 31, 2026, 1:28 AM

#

oh pog

fallen bobcat Mar 31, 2026, 1:28 AM

#

i forgot to pack the struct

#

I dont get the difference between packed and extern tho

heavy sandal Mar 31, 2026, 1:28 AM

#

packed is bitfields

#

extern is c abi

#

theres an open proposal to rename packed to bitpacked because that would actually make sense

fallen bobcat Mar 31, 2026, 1:29 AM

#

what is attribute((packed))

#

but in zig

heavy sandal Mar 31, 2026, 1:29 AM

#

align(1) after every field

fallen bobcat Mar 31, 2026, 1:29 AM

#

bruh what

heavy sandal Mar 31, 2026, 1:29 AM

#

theres no keyword or simple option for it

fallen bobcat Mar 31, 2026, 1:30 AM

#

I think packed works tho

heavy sandal Mar 31, 2026, 1:30 AM

#

packed requires it to be backed by an int

fallen bobcat Mar 31, 2026, 1:30 AM

#

bruhburhbruhbruhbrub

heavy sandal Mar 31, 2026, 1:31 AM

#

so if its too big or the abi cant handle the int it dies

#

what you want here is extern + manually set field alignment

fallen bobcat Mar 31, 2026, 1:32 AM

#

that's ass

heavy sandal Mar 31, 2026, 1:32 AM

#

for the gdt pointer anyway

#

bro its two fields

#

its fine

fallen bobcat Mar 31, 2026, 1:33 AM

#

ugh

#

ok

#

gdt init... ok works now

heavy sandal Mar 31, 2026, 1:33 AM

#

nice

#

byte-packed structs are def on my wishlist for zig

#

esp once they rename the current packed to bitpacked (which was accepted at some point) so the term wont be as overloaded

fallen bobcat Mar 31, 2026, 1:35 AM

#

mhm do I do zig comptime magic to generate the interrupt handlers

heavy sandal Mar 31, 2026, 1:35 AM

#

i used to do that

#

swapped to doing it with asm macros in an asm file just because it was a bit more readable

fallen bobcat Mar 31, 2026, 1:35 AM

#

idk GAS macros

heavy sandal Mar 31, 2026, 1:35 AM

#

i do still generate my isr_common with comptime tho

fallen bobcat Mar 31, 2026, 1:36 AM

#

like in nasm i can just do:

%assign i 32
%rep 224
    INTERRUPT_NOERR i
%assign i i+1
%endrep

#

idk if gas lets me do that

heavy sandal Mar 31, 2026, 1:36 AM

#

it does but its a bit more fiddly

#

so before i learned how to do that i used comptimePrint etc to generate asm for the handlers at comptime

fallen bobcat Mar 31, 2026, 1:48 AM

#

.rept 224
INTERRUPT_NOERR %32+\+
.endr

.global __interrupt_vectors
__interrupt_vectors:
.rept 256
.quad isr\()\+
.endr

``` lmao

#

amazing syntax as always

heavy sandal Mar 31, 2026, 1:48 AM

#

yeah

fallen bobcat Mar 31, 2026, 1:55 AM

#

lesgo we have interrupts now

heavy sandal Mar 31, 2026, 1:55 AM

#

pog

fallen bobcat Mar 31, 2026, 1:55 AM

#

tfw i have fucking 600 lines log ring but no interrupts nooo

heavy sandal Mar 31, 2026, 1:55 AM

#

lmao

#

oh btw idk if you plan to do aarch64 but if you are, what interrupt controller(s) are you planning on supporting?

fallen bobcat Mar 31, 2026, 1:56 AM

#

I have no idea

#

I know nothing about arm

#

I know there's a thing called the GIC tho

heavy sandal Mar 31, 2026, 1:56 AM

#

yeah

#

but theres five different versions and for the most part they all behave very different

fallen bobcat Mar 31, 2026, 1:57 AM

#

i do have a raspi, could be cool to target it even tho it's not well documented

#

or I could get one of those open source boards

heavy sandal Mar 31, 2026, 1:57 AM

#

apart from 4 which is back-compatible with 3

#

and i think 3 is sorta back-compatible with 2?

fallen bobcat Mar 31, 2026, 1:57 AM

#

heavy sandal but theres five different versions and for the most part they all behave very di...

well I'm gonna do it per-board

heavy sandal Mar 31, 2026, 1:58 AM

#

but afaik 5 doesnt have any back compat and noone talks about 1 lmao

heavy sandal Mar 31, 2026, 1:58 AM

#

fallen bobcat well I'm gonna do it per-board

well if whatever board you end up with has GIC version 3 or 4 i got you covered for helping to understand that shit

#

i think mine is one of if not the only hobby os to have support for MSIs on the gic v3/4

#

so i got you covered there for the knowledge lmao

fallen bobcat Mar 31, 2026, 2:00 AM

#

most driver stuff goes completely over my head tbh

#

im not a big hardware guy

heavy sandal Mar 31, 2026, 2:00 AM

#

mhm

#

normally i like driver stuff tbh

#

but that interrupt controller is ass

fallen bobcat Mar 31, 2026, 2:12 AM

#

#

no ascii art

#

worst kernel ever

heavy sandal Mar 31, 2026, 2:12 AM

#

i dont have ascii art either

fallen bobcat Mar 31, 2026, 2:12 AM

#

i think this is enough info for now but eventually it'll call into the generic panic thing

#

which shows current thread, ipl, etc

heavy sandal Mar 31, 2026, 2:13 AM

#

mhm

heavy sandal Mar 31, 2026, 2:22 AM

#

fallen bobcat i think this is enough info for now but eventually it'll call into the generic p...

trick i do for that btw is to define a format function on the interrupt frame type and then i can pass a pointer to it through to the main arch-independent panic handler and because the type is still comptime known (depending on target ofc) itll call the format function

fallen bobcat Mar 31, 2026, 1:13 PM

#

I just realized I cant do the addrspace(.gs) trick

#

very sad

#

since for zig *T != *addrspace(.gs) T so things like self functions dont work

fallen bobcat Mar 31, 2026, 1:34 PM

#

ugh I think I need a self pointer or rdgsbase

slate vessel Mar 31, 2026, 1:59 PM

#

fallen bobcat ugh I think I need a self pointer or rdgsbase

I kinda stole your log ring, if that's okay

#

printk_ringbuffer documentation was an interesting read

fallen bobcat Mar 31, 2026, 2:00 PM

#

it's not up to date tho

#

which is why i sent a patch

fallen bobcat Mar 31, 2026, 2:01 PM

#

slate vessel I kinda stole your log ring, if that's okay

did you finish implementing it

slate vessel Mar 31, 2026, 2:01 PM

#

yes

fallen bobcat Mar 31, 2026, 2:01 PM

#

where

slate vessel Mar 31, 2026, 2:01 PM

#

I don't have it pushed yet

#

it's basically your code but c++ified

fallen bobcat Mar 31, 2026, 2:07 PM

#

fallen bobcat ugh I think I need a self pointer or rdgsbase

bruh ok I had to do this:

pub inline fn percpu_ptr(variable: anytype) @TypeOf(variable) {
    const offset = asm volatile ("mov %%gs:(%%rax), %[out]"
        : [out] "=r" (-> u64),
        : [self_off] "{rax}" (&cpu_self_offset),
    );

    return @ptrFromInt(offset +% @intFromPtr(variable));
}

#

no performance gain at all for using percpu stuff like that since I cant do gs loads/stores directly

#

nooo

#

I guess I could use rdgsbase if the machine supports it but it might just be faster to do that ^

languid canyon Mar 31, 2026, 3:21 PM

#

slate vessel it's basically your code but c++ified

Wouldn't it be easier to convert the c version

#

Or did u want his one specifically

slate vessel Mar 31, 2026, 3:21 PM

#

tbh I forgot you had it as well meme

languid canyon Mar 31, 2026, 3:21 PM

#

💀

#

Mine supports pr_cont but doesn't support 32 bit, his is the opposite

heavy sandal Mar 31, 2026, 3:22 PM

#

and mine supports neither KEKW

languid canyon Mar 31, 2026, 3:22 PM

#

Didn't you copy mine

#

So it should support cont if you did

heavy sandal Mar 31, 2026, 3:23 PM

#

mine internally would support cont but i dont have the exposed api for it

languid canyon Mar 31, 2026, 3:23 PM

#

The api is the easy part

heavy sandal Mar 31, 2026, 3:23 PM

#

yeah i just never wrote it lmao

slate vessel Mar 31, 2026, 3:23 PM

#

languid canyon Mine supports pr_cont but doesn't support 32 bit, his is the opposite

what's that?

languid canyon Mar 31, 2026, 3:23 PM

#

pr_cont?

slate vessel Mar 31, 2026, 3:23 PM

#

yeah

heavy sandal Mar 31, 2026, 3:23 PM

#

lets you extend a message

slate vessel Mar 31, 2026, 3:23 PM

#

ah

heavy sandal Mar 31, 2026, 3:24 PM

#

as long as its the latest written message

languid canyon Mar 31, 2026, 3:24 PM

#

Print continued

slate vessel Mar 31, 2026, 3:24 PM

#

yeah

#

reopen a committed reserve

#

I don't think I need that

languid canyon Mar 31, 2026, 3:24 PM

#

Yeah

slate vessel Mar 31, 2026, 4:16 PM

#

fallen bobcat where

https://github.com/ilobilix/kernel/blob/master/kernel/source/lib/log.cpp

GitHub

kernel/kernel/source/lib/log.cpp at master · ilobilix/kernel

Hobby kernel in modern C++. Contribute to ilobilix/kernel development by creating an account on GitHub.

fallen bobcat Mar 31, 2026, 4:25 PM

#

fallen bobcat bruh ok I had to do this: ```rust pub inline fn percpu_ptr(variable: anytype) @T...

@languid canyon do you think there's a better way to do this

languid canyon Mar 31, 2026, 4:33 PM

#

Not in zig

heavy sandal Mar 31, 2026, 4:33 PM

#

fallen bobcat <@353593948523986945> do you think there's a better way to do this

so there is a stupid workaround possible

#

which is to make the self-parameter to methods on percpu stuff anytype

#

and then itll use generics to instantiate for both gs and normal pointers

fallen bobcat Mar 31, 2026, 4:34 PM

#

I literally just thought of that

#

I can make every self anytype

heavy sandal Mar 31, 2026, 4:34 PM

#

the problem with that is you lose literally all lsp type hints which makes the development side a pita

#

but it does work

fallen bobcat Mar 31, 2026, 4:35 PM

#

The only structure I need is spinlock

#

I think

#

Yeah no ok it's not feasible

heavy sandal Mar 31, 2026, 4:44 PM

#

F

fallen bobcat Mar 31, 2026, 9:00 PM

#

Linux patch merged

#

letsgo

heavy sandal Mar 31, 2026, 9:00 PM

#

🎉

#

oh btw @fallen bobcat zig merged llvm backend incremental support

#

they dont have state saving yet so it only applies to if you use the --watch flag on the build process to have it watch for changes on the filesystem but it is there

languid canyon Mar 31, 2026, 9:05 PM

#

fallen bobcat Linux patch merged

congrats

#

is this your first linux contribution?

fallen bobcat Mar 31, 2026, 9:06 PM

#

languid canyon is this your first linux contribution?

Yeah

#

First real major project contribution

languid canyon Mar 31, 2026, 9:07 PM

#

mention in your CV meme

fallen bobcat Mar 31, 2026, 9:07 PM

#

lol

#

At least the patch looks bigger because another guy contributed to it

languid canyon Mar 31, 2026, 9:07 PM

#

did they co-author it?

fallen bobcat Mar 31, 2026, 9:07 PM

#

fallen bobcat First real major project contribution

I've been wanting to work on a BSD for a while but not sure which or what to do

#

I asked on the netbsd IRC and they gave me stuff that was already done

weary jetty Mar 31, 2026, 9:10 PM

#

fallen bobcat Linux patch merged

Congrats!

fallen bobcat Mar 31, 2026, 9:11 PM

#

languid canyon did they co-author it?

Yeah I think so

fallen bobcat Mar 31, 2026, 9:12 PM

#

heavy sandal they dont have state saving yet so it only applies to if you use the `--watch` f...

I don't use that, also does that imply incremental linking or not

#

Like does it make linking faster

heavy sandal Mar 31, 2026, 9:13 PM

#

im not sure about linking, this was mostly the llvm emit side, but i think it will might do incremental linking too

fallen bobcat Mar 31, 2026, 9:13 PM

#

Idk I think what's slow for me is linking

heavy sandal Mar 31, 2026, 9:13 PM

#

mhm

#

for me its def the emit so itll help there (albeit itll help more once they get state saving)

fallen bobcat Mar 31, 2026, 9:17 PM

#

Idk I'm not sure

#

The freestanding build is fast the UM one is not

heavy sandal Mar 31, 2026, 9:17 PM

#

mhm

fallen bobcat Mar 31, 2026, 9:59 PM

#

#

🥀 @slate vessel

#

1:1 code rewrite for that nooo

slate vessel Mar 31, 2026, 10:02 PM

#

fallen bobcat 1:1 code rewrite for that <:nooo:750466870993223750>

?

#

should I add a license header thingy?

fallen bobcat Mar 31, 2026, 10:03 PM

#

idk you do you but I just think it's kinda crazy you 1:1 rewrite a piece of code and add a small "based on XXX" lol

slate vessel Mar 31, 2026, 10:03 PM

#

oh sorry

fallen bobcat Mar 31, 2026, 10:03 PM

#

you do what you want, I know you and youre not using for commercial purposes or something

#

I dont think this counts as licensed code since its rewritten tho

slate vessel Mar 31, 2026, 10:03 PM

#

I don't really know licenses

fallen bobcat Mar 31, 2026, 10:04 PM

#

I just find it funny meme

slate vessel Mar 31, 2026, 10:04 PM

#

I'll include a permalink

fallen bobcat Mar 31, 2026, 10:04 PM

#

yeah just a link is fine

slate vessel Mar 31, 2026, 10:05 PM

#

done

fallen bobcat Mar 31, 2026, 11:24 PM

#

ok pmm time I think nooo

#

well not nooo , at least this gets me away from architecture stuff

slate vessel Mar 31, 2026, 11:24 PM

#

what kind are you planning to make?

fallen bobcat Mar 31, 2026, 11:25 PM

#

for now I will do a simple freelist but I had a plan in mind to add per-cpu lists on top of that

#

it involved having one global "bucket" freelist which is a list of lists of 64 pages each

slate vessel Mar 31, 2026, 11:26 PM

#

what about a buddy :P

fallen bobcat Mar 31, 2026, 11:26 PM

#

fallen bobcat it involved having one global "bucket" freelist which is a list of lists of 64 p...

and every cpu has like 4 local buckets

#

you allocate and free from buckets[0]

#

when you need to drain you shift all buckets down

#

it's basically magazines

fallen bobcat Mar 31, 2026, 11:27 PM

#

slate vessel what about a buddy :P

I have no need for that

#

and a buddy allocator sucks anyway

slate vessel Mar 31, 2026, 11:28 PM

#

I like it quite a bit

heavy sandal Mar 31, 2026, 11:28 PM

#

buddy is great imo

#

i use it as a vmm tho

#

not for the pmm

fallen bobcat Mar 31, 2026, 11:28 PM

#

the buddy allocator in linux infamously sucks

#

if I need higher than one page i can do an llfree-like thing

slate vessel Mar 31, 2026, 11:30 PM

#

don't you need to allocate more than one page frequently if the mmu doesn't support 4k pages

fallen bobcat Mar 31, 2026, 11:30 PM

#

what kinda architecture doesnt support 4k pages

heavy sandal Mar 31, 2026, 11:30 PM

#

yeah really

slate vessel Mar 31, 2026, 11:30 PM

#

I'm not sure

#

I guessed some arm boards

heavy sandal Mar 31, 2026, 11:31 PM

#

then dont worry about til you find one imo

#

i think arm requires 4k page support

fallen bobcat Mar 31, 2026, 11:31 PM

#

slate vessel I guessed some arm boards

it's an architectural thing

heavy sandal Mar 31, 2026, 11:31 PM

#

id have to double check

fallen bobcat Mar 31, 2026, 11:31 PM

#

not a platform thing

heavy sandal Mar 31, 2026, 11:31 PM

#

arm supports a couple different granule sizes tbf

#

as the base level

#

but i think 4k is required to be supported

slate vessel Mar 31, 2026, 11:32 PM

#

good to know

heavy sandal Mar 31, 2026, 11:32 PM

#

id have to double check

#

havent looked at that section of the manual in a bit

fallen bobcat Mar 31, 2026, 11:32 PM

#

4kb is pretty much the standard for all archs nowadays

heavy sandal Mar 31, 2026, 11:33 PM

#

yeah

fallen bobcat Mar 31, 2026, 11:33 PM

#

you could argue 16k would make sense nowadays

heavy sandal Mar 31, 2026, 11:33 PM

#

i would be unsurprised if windows used the 64k granule size if supported since like everything in windows's vmm uses 64k units if possible but idk

slate vessel Mar 31, 2026, 11:34 PM

#

I think I've written my vmm in a way that it doesn't care if the page size is 4k or not

#

of course I haven't tested it in such a way :P

fallen bobcat Mar 31, 2026, 11:36 PM

#

mhm well I need to implement my architecture abstraction first

#

I think I'm not gonna do anything fancy for now and just copy what I already ahve

slate vessel Mar 31, 2026, 11:36 PM

#

fallen bobcat mhm well I need to implement my architecture abstraction first

mine is a mess

fallen bobcat Mar 31, 2026, 11:36 PM

#

but eventually I will move to a cursor-like API over ptes

heavy sandal Mar 31, 2026, 11:36 PM

#

mines kinda meh abstraction wise but it works

#

requires the arch pte type to have an init function with a certain signature for setting up a valid pte and requires the page number to be accessible as the page field of a packed union in the addr field of the pte

fallen bobcat Mar 31, 2026, 11:39 PM

#

oh fuck im gonna have to rewrite kmem

#

fun

#

I'm kinda tempted to move to a lock-free thing like mimalloc tho

#

lmao I found my old OS and it straight up gdt init... fail

pastel citrus Mar 31, 2026, 11:52 PM

#

limine v2.x was beautiful

wet kernel Mar 31, 2026, 11:52 PM

#

[ FATAL] GDT init... FAIL

fallen bobcat Mar 31, 2026, 11:54 PM

#

when limine was actually good 😔

#

ugh I feel like the "bump alloc from largest entry" thing is so hacky

#

what if it's a bunch of small memory entries

languid canyon Apr 1, 2026, 12:11 AM

#

fallen bobcat lmao I found my old OS and it straight up gdt init... fail

Lmao

fallen bobcat Apr 1, 2026, 12:11 AM

#

idk why actually

#

it should work

#Zag