Nyaux

[surreal path]

real

[kind root]

wait did i miss a convo

[finite summit]

no it was in our dms

[kind root]

damn

[finite summit]

currently I have gotten it down to:
his computer is haunted

[kind root]

i wanna see an obyaux kernel ngl

[finite summit]

I mean I don't feel like writing two kernels alongside each other

or rewriting obos

so he should come work with me on obos :^)

[kind root]

yeah

[finite summit]

he can uhh

not write drivers

[kind root]

maybe nyaux should just work on obos instead

[finite summit]

not mess with core kernel components

[kind root]

(if hes okay with that)

[finite summit]

ye

[finite summit]

writing syscalls/mlibc sysdeps can't go that wrong, right?

[finite summit]

not mess with the vmm

(not even I'd let myself mess with the vmm)

I mean there is one relatively simple thing in the vmm that could be done

which is updating the swap to disk interface

and it's actually like really simple

you'd just need to reimplement 5 functions

set some function pointers

shite

I never rewrote Mm_ChangeSwapProvider

anyway, I'll stop literring the nyaux thread

[cinder plinth]

come work on my kernel

[finite summit]

wasn't that like nyx/gaia

[cinder plinth]

yes

nyaux

nyx

perfect

[finite summit]

nyaux without the au

[cinder plinth]

fwiw the name Nyx existed way before he did

so

[finite summit]

nyaux but not australian

[elder shoal]

vinix/lyre/whatever copied their sysdeps had xorg very unstable because of one specific mistake on the sysdeps

[finite summit]

O_O

[cinder plinth]

what was it

[elder shoal]

they were passing the timeout to ppoll wrong

[finite summit]

actually some sysdeps in obos might translate into multiple syscalls

which require you to like know the kernel in-and-out

[elder shoal]

https://github.com/managarm/mlibc/commit/de3adba6892e742ef5d1b8c5647e296422a62ed0

[finite summit]

like fork

probably will translate to like 10 different syscalls

idk the exact number

1 to fork the current vmm context
1 to create a new process object
2 to make a new thread + one bootstrap function to get the thread in the same state as the thread doing the fork
1 to allocate a stack for said thread
1 to clone the handle table unto the new process

[kind root]

sounds slow

[finite summit]

if I did it in the kernel, the only difference is that there would be less switches from ring3->ring0

[haughty kite]

yeah which is the expensive part of a syscall

lol

youd also save a bunch of syscall validation

honestly just do batch syscalling :^)

[kind root]

io_uring in obos

[finite summit]

have any kernels actually done this

[haughty kite]

dont think so

iouring isnt quite the same thing

[kind root]

u can extend it to do arbitrary syscalls

[edgy pilot]

syscall_uring

[finite summit]

what's the man page for io_uring

[haughty kite]

man io_uring

big brain

[edgy pilot]

https://man7.org/linux/man-pages/man7/io_uring.7.html

[finite summit]

struct obos_syscall_args
{
    uint32_t number;
    uint64_t pars[5];
};
void Sys_RunSyscalls(struct obos_syscall_args* syscalls, size_t nSyscalls):```

or even better

use AML

so I can chain syscall results

[edgy pilot]

do not guarantee the order of syscalls

randomise them

[finite summit]

genius

or I can be sane and just have a Sys_Fork in-kernel

to save ring 3 -> ring 0 switches when forking

[kind root]

spawn a thread per syscall

[finite summit]

and not make a batch syscall mechanism

[kind root]

i mean a kernel worker thread

[finite summit]

I'll just design a bytecode mechanism to batch syscalls (not)

[edgy pilot]

run everything in ring 0 and you won't have to do cpl switching

[finite summit]

(pseudocode)
NewCtx = RUN_SYSCALL Sys_HandleClone, HANDLE_CURR_VMM_CONTEXT
RUN_SYSCALL Sys_ProcessStart, newCtx, HANDLE_INVALID```

anyway that's enough schizophrenia while doing osdev

(for the day)

[edgy pilot]

5343 packages to upgrade

[finite summit]

I now must go implement PCI

[finite summit]

when you forget to upgrade your packages after a week

[edgy pilot]

I'm just upgrading fedora 40 -> 41

[finite summit]

tomorrow I have school

[edgy pilot]

me too

[finite summit]

ofc when I actually want to work on OBOS

school comes

and then after school and homework and shit I'm too tired to work

on OBOS

[edgy pilot]

the only time I'm motivated enough to code is 2-4 am

but I need to sleep

[finite summit]

go to sleep

[elder shoal]

imagine having motivation for osdev

[finite summit]

it's called userspace dev in osdev

and being bored of porting shit

the fun stuff lies in your first couple ports and writing the kernel

or just come work on obos

[edgy pilot]

kerneldev is fun
osdev is not

[finite summit]

this

[plush hearth]

Is nyauxmaster Australian?

[edgy pilot]

Irish

[plush hearth]

Ah

[flat nymph]

If only someone could properly implement TLB shootdowns and working set in my kernel...
(The kernel is only 15k LoC )

[edgy pilot]

if only I was working on this kernel with someone

[surreal path]

no im working on nyaux

myself

[surreal path]

♫ 99 little bugs in the code ♫
♫ 99 little bugs ♫
♫ take one down, patch it around ♫
♫ 137 little bugs in the code ♫

♫ 99 little bugs in the code ♫
♫ 99 little bugs ♫
♫ take one down, patch it around ♫
♫ 137 little bugs in the code ♫

nyaux theme song

[next tusk]

♫ 137 little bugs in the code ♫
♫ 137 little bugs ♫
♫ take one down, patch it around ♫
♫ 93,428,471,930,587,234,682,318,095,723,456,932,108,653,294,867,394,583,495,715,902,847,523,905,

[silver yarrow]

♫ 99 little bugs in the code ♫
♫ 99 little bugs ♫
♫ take one down, patch it around ♫
♫ 137 little bugs id♫i♫
♫,g♫ goll ♫♫ behi tepl 3wo9
nt ioutaul♫
ete9h sitttcdno9kch 7ceed ee cu u19e t ond♫tbo sa e a regn tb h
nnli♫tls tdi

[edgy pilot]

somebody forgot to use locks

[next tusk]

pthtahchh 9937. ni coSegmentation fault

[crystal scarab]

mmm

[surreal path]

Hi guys got a stomach bug rn

I don’t think I can today

Maybe tmrw if it gets better

I hope anyways

Anyways love y’all !!!!

cya tmrw :D

[edgy pilot]

debug your stomach

[surreal path]

LMAOOOO

[frigid cliff]

But didn’t we just hear about what happens when you try to do that

[surreal path]

bug has been solved

uacpi now works

[kind root]

what was the bug

also welcome back

[surreal path]

someone on the osdev forums helped me spot it

was with the hhdm pages being incorrectly set cause of one stupid fucking variable

it was with the page tables

so

yea

no more vmm region corruptiuon uacpi boots

i will try turning on ksan

ksan still doesnt work

[elder shoal]

2097152

[surreal path]

what

yea thats the length of a 2mb entry

thats 2mb

[elder shoal]

girl the gods gave you macros use them smh

magic numbers are CRINGE smh

[surreal path]

no :)

anyways i want a break for today i had a really bad mental breakdown cause everyone was making fun of me in valorant earlier and i was crying and stuff

im not in the mood to do much rn

tmrw ill work on scheduler

sorry timers*

then smp then scheduling

ykyk

[surreal path]

i know its stupid to cry over some people on valorant but

they were

[tender gorge]

take time for yourself pls

[surreal path]

really fucking mean

they said a lot of things, they even found my parents facebook

and made fun of my family

and they were like

[tender gorge]

thats fucked

[surreal path]

yea

[tender gorge]

yeah they're giant assholes

[surreal path]

thanks

[tender gorge]

sorry you had to go through that

[surreal path]

people on valorant are actually horrible people

[surreal path]

@kind root

[kind root]

cool

[surreal path]

yea

[surreal path]

are u like mad at me or anything (sorry if this is a stupid question)

[kind root]

not really lol

why

[surreal path]

i just wanted to make sure

[kind root]

nah

[surreal path]

okay thats good that ur not mad

either way going forward nyaux wont be stuck anymore

and i hope to reach bash soon

[surreal path]

why am i so sure? because im not going to make the same mistakes i did last time

ive learned from my hell

[kind root]

thats good

[cinder plinth]

average valorant player

[thorn bramble]

had a similar experience once but i only witnessed it, it wasn't directed at me but my friend, there was this guy on our team and he was so bad everyone just kept flaming and insulting him but for some reason he decided to bully my friend lol

he found my friend's ig and tried finding more info but just ended up embarrassing himself, said he'd type some fucked up stuff to her mom but ended up doing nothing, later i found his discord and it turned out he was selling some shitty minecraft hacks to 12 year olds or something lol

[surreal path]

i kinda wanna steal the linux hashmap implmentation

cause i cant find any good hashmaps that work well in a freestanding c enviorment

[surreal path]

thank god i can use the bsd 2 clause

what do u guys think

[cinder plinth]

just make your own 😎

[surreal path]

nah im too lazy for that 😎

[cinder plinth]

its like

200 lines

max

but sure

[surreal path]

nah ima steal it :)

WAIT

FACEBVOOK

fucking facebook

yk what

ill make my own

:)

[finite summit]

idk about good

but I found some hashmap library online

and did some stuff to make it work

freestanding*

[surreal path]

lemme see

[finite summit]

https://github.com/OBOS-dev/obos/blob/userspace-work/src/oboskrnl/utils/hashmap.h
https://github.com/OBOS-dev/obos/blob/userspace-work/src/oboskrnl/utils/hashmap.c

[surreal path]

i think ill just make my own

[finite summit]

and it depends on this

https://github.com/OBOS-dev/obos/blob/userspace-work/src/oboskrnl/external/fixedptc.h

[surreal path]

forgot about hashmaps

ill just use a 2d array

no nvm

ill just use a slow linear array

idfk

anyone have the elf spec for x86 64

all im finding on google is i386

(32 bit)

[cinder plinth]

man elf

[surreal path]

i am not using man pages

[cinder plinth]

its the best spec

and why not

[surreal path]

cause i swear to god there was a better spec

that was for x86 64

and shit

idfk where that spec has gone

[cinder plinth]

the linux manpage is pretty complete

enough to make a loader

[surreal path]

no thanks

ima just keep looking for the damn spec

[tawdry mirage]

wdym, the sysv abi?

[surreal path]

yea

for x86 64

describes everything abt elf

all im seeing is this garbage 32 bit specs

there was like one like this but for x64

[tawdry mirage]

not the sysv abi pdf then

that only has info on x86_64 specifics for elfs

[surreal path]

yea

yea ik

[tawdry mirage]

i mean the differences between loading a 32-bit elf and a 64-bit elf aren't that huge

[surreal path]

id still rather have the proper spec

i found this "spec" but its not rlly a pdf and more an ugly webpage

https://refspecs.linuxfoundation.org/elf/gabi4+/contents.html

it is the proper spec thjo

ima just install dark reader cause this hurts my eyes

and read this one

[odd summit]

Just do a hash map

[surreal path]

no bro i made a shitty hashmap in userspace

[odd summit]

Oh good

[surreal path]

IT WAS SHITTY AND LIKE IT had corruption issues or something some keys went to 444????

and i cant get any library one working cause like

my free takes in size

as well

and i tried to modify the library but

didnt rlly work

[odd summit]

Valgrind time

[odd summit]

Probably worth fixing that too

[surreal path]

how

???

i need size4\

how tf will i know then

hmmm?

how will i know if i should use my slab allocator or my vmm region allocator

[odd summit]

metadata

[tawdry mirage]

it shouldn't be that difficult to make your code remember the size to pass to free

[tawdry mirage]

it's not necessarily wrong to require that imo

considering it's not a stdlib free

[cinder plinth]

wish C++ could somehow make operator delete[] take a size

[surreal path]

yea i have the slab metadata in the start of the page but idfk if i can trust just checking if theres a reasonable slab metadata entry at the start of the page

[cinder plinth]

I mean

turn delete A in operator delete(A, sizeof(A))

[tawdry mirage]

i think that's a thing?

[cinder plinth]

I think you need to pass the size explicitly

[surreal path]

im just not gonna do a hashmap

simple as that

[odd summit]

I guess that’s fair

Though articles on implementing malloc are a dime a dozen

[cinder plinth]

why do you want one

[surreal path]

exactly

i dont want one

[tawdry mirage]

nop https://godbolt.org/z/ejKP738ee

[cinder plinth]

what is your use

[cinder plinth]

ah TIL

[tawdry mirage]

in the past you needed -fsized-deallocation for clang

[cinder plinth]

maybe I'm mixing it up with delete[] though?

[tawdry mirage]

but it seems that it's not necessary nowadays (at least on trunk)

[surreal path]

symbol table and stack traces

yk

[cinder plinth]

eh you could just use a binary search tree for that

[surreal path]

function name and address of the starting function

[odd summit]

https://www.andreinc.net/2021/10/02/implementing-hash-tables-in-c-part-1

https://benhoyt.com/writings/hash-table-in-c/

[cinder plinth]

hashmap would be simpler tho

seperate chaining is easy af

[tawdry mirage]

not really? just sort the array by address and you can do a binary search on it

[surreal path]

im good man

[tawdry mirage]

can sort it when generating instead of at runtime

[odd summit]

These were the first two articles I found

[surreal path]

i know how they work

[cinder plinth]

yeah but that's not a real BST

[surreal path]

wanna see my shit impl?

[cinder plinth]

do this tho

[tawdry mirage]

yeah but you also don't need one :^)

[cinder plinth]

that's simple

[cinder plinth]

hear me out

rbtree

[tawdry mirage]

:^)

[cinder plinth]

no

b-tree

even better

[surreal path]

my shit hashmap impl

[cinder plinth]

(I have no idea what those are i just know they are complex (and used by xfs))

[surreal path]

https://pastebin.com/nGqmcvBi

[odd summit]

Time to implement a database from scratch to learn

[cinder plinth]

might try doing XFS one day

[tawdry mirage]

select name from symbols where starting_address=$0

just link sqlite into your kernel for this

:^)

[surreal path]

anyways im going back to writing elf structs like a dumb loser

[cinder plinth]

not a bad idea

would be doable

isnt there a mainstream kernel that does that tho

[odd summit]

Just don’t use it for satanism

[cinder plinth]

I vaguely remember Windows linking it or something, not sure if it was in the kernel

[surreal path]

oh come on

[cinder plinth]

bro

[surreal path]

WAIT

yep

im stupid

whoops

sorry

[odd summit]

What

[surreal path]

dw about it

[cinder plinth]

btw this is wrong in many ways

[surreal path]

I KNOW ITS FUCKING 4AM

[cinder plinth]

you should read the tutorials chicken sent honestly

[surreal path]

bro

😭

[cinder plinth]

would help

[surreal path]

im fine

i dont need no hashmap

ill do a binary search or whatever tf

there

[surreal path]

oh ok

dies\

why

cause like my hashmap was shitty

anyways its 4am

i think i go sleep

[odd summit]

Also some of those are supposed to be signed

[surreal path]

oh yeah whoops

ok fixed

[surreal path]

explain this then hmm

🤯

even tho

like bro

[cinder plinth]

you're not handling collisions first of all

[surreal path]

oh

yea ik

whats the second thing

[cinder plinth]

a lot of stuff

your remove() does nothing

[surreal path]

i check if something exists tho?

[cinder plinth]

you're just freeing it

you need to remove it as well

[surreal path]

oh yeah i need set to null

what

yea

set to null

done

[cinder plinth]

and yeah the problem is likely collisions

[surreal path]

yea

ok ill redo the hashmap from scratch with the tutorials

tmrw

NOW WISH ME GOOD NIGHT 😡

i go eep

[surreal path]

@cinder plinth 😡

[cinder plinth]

what

[surreal path]

say goodnight

or have fright

[cinder plinth]

gn

[surreal path]

gn

[frigid cliff]

what happened to rust

[silk wind]

ub

[spice yarrow]

it turns out writing unsafe rust like c is not a very good idea

[haughty kite]

you shouldnt use modulo

uint64_t hash_to_index(uint64_t hash, uint64_t size) { return (((uint64_t)(uint32_t)hash) * size) >> 32; }

[cinder plinth]

The compiler will optimize it anyway

[haughty kite]

no it wont

because this is a different algorithm

lol

it does a slightly different thing

it still uniformly distributes the hashes

but it saves a division which is good

and it lets you use prime sized hash tables

[cinder plinth]

Ok I shall do that too then

[haughty kite]

good

[molten grotto]

is using that with a 64-bit hash worth it tho if you just discard the upper bits?

or well ig the hash is still better?

[haughty kite]

first, a 64 bit hash isnt that much more expensive

also, you use it to accelerate the equality check

[cinder plinth]

that's just a mixer function btw

not a complete hash function

you should totally embed xxhash

come on bro just one more dependency

[haughty kite]

yeah of course

[haughty kite]

or wyhash or murmurhash3 or something else

or siphash

[cinder plinth]

I use murmurhash2a

/*
    MurmurHash2a, by Austin Appleby
*/
uint64_t murmur_hash(const void *key, size_t len, uint64_t seed) {
  const uint64_t m = 0xc6a4a7935bd1e995LLU;
  const int r = 47;
  uint64_t h = seed ^ (len * m);
  auto data = (const uint64_t *)key;
  const uint64_t *end = (len >> 3) + data;
  while (data != end) {
    uint64_t k = *data++;
    k *= m;
    k ^= k >> r;
    k *= m;
    h ^= k;
    h *= m;
  }
  auto data2 = (const unsigned char *)data;
  switch (len & 7) {
  case 7:
    h ^= (uint64_t)(data2[6]) << 48;
  case 6:
    h ^= (uint64_t)(data2[5]) << 40;
  case 5:
    h ^= (uint64_t)(data2[4]) << 32;
  case 4:
    h ^= (uint64_t)(data2[3]) << 24;
  case 3:
    h ^= (uint64_t)(data2[2]) << 16;
  case 2:
    h ^= (uint64_t)(data2[1]) << 8;
  case 1:
    h ^= (uint64_t)(data2[0]);
    h *= m;
  default:
    break;
  };
  h ^= h >> r;
  h *= m;
  h ^= h >> r;
  return h;
}

[frigid cliff]

just one more dependency
400 dependencies later

[haughty kite]

400 dependencies
did you mean: a rust hello world?

[frigid cliff]

Lmao

If you add another zero it might just be enough for a nodejs hello world

[surreal path]

hey sorry ive been busy all friday

parents took me on a trip

we went to the cinema

!!!

well they dropped me off at the cinema

and then we had kfc

and then we went to the park

was fun

ill work on nyaux later today!

[tender gorge]

glad you had fun!

[median goblet]

but how does that work

[cinder plinth]

https://lemire.me/blog/2016/06/27/a-fast-alternative-to-the-modulo-reduction/

[surreal path]

I AM SO SORRY GUYS BUT SCHOOL

I STUDIED

ALL DAY

CAUSE I MISSED SO MUCH SCHOOL

😭

oki i work on hashmap now

[surreal path]

some of the math was complicated but im guessing u have these things called "buckets" that u index into with the hash and if u treverse the linked list to find ur value pair right?

i dont understand the growing part however

like do u just add another entry

if u have a collission

how many buckets should u have?

are u able to create MORE buckets?

so many questions

[cinder plinth]

yes

a bucket is a key value pair

[kind root]

Well ideally one per value

[surreal path]

when do u make more

[kind root]

When too many collisions

[surreal path]

okay

i can have a collission counter

if it reaches 10

ill create a new bucket

good design?

[kind root]

U can always adjust the algorithm later

[cinder plinth]

no

every time

[surreal path]

whats the better algorithm

[cinder plinth]

bro this isnt hard just look at wikipedia

do open addressing + seperate chaining

[kind root]

Never look at wikipedia

Always spam this server with questions instead

[cinder plinth]

true

[surreal path]

not funny

[cinder plinth]

do a linked list for now

you can do a fancy tree later

might use a rbtree for mine

simple bst is fine tho

[surreal path]

is it my fault my brain is stupid

seriously

[kind root]

Huh, there are rb trees in hash tables?

[surreal path]

and i dont spam

[kind root]

For what purpose

[surreal path]

who said i spammed infy

[cinder plinth]

finding a key in a bunch of buckets

[kind root]

No one, I was making a general joke

[surreal path]

i didnt find it funny

[kind root]

Is hard to make a joke everyone would find funny

[surreal path]

fair

[kind root]

U only search one bucket because u know the hash tho

[surreal path]

can u apologise though cause i kinda felt like it was making fun of me

[kind root]

Eh sure sorry if that sounded offensive to you

[surreal path]

oki forgiven

[cinder plinth]

https://en.wikipedia.org/wiki/Hash_table#Other_data_structures_for_separate_chaining

instead of linking buckets in a list you use a tree

[kind root]

Ah, I think you're using the word bucket incorrectly

[cinder plinth]

maybe

[kind root]

Bucket would be one singular tree in this case

So efficiently finding the value if there are collisions so a lot of them end up in the same bucket

[cinder plinth]

yeah that

idk the terminology

i just

did it

[kind root]

Makes sense

A bucket is just like a data structure to store values that matched a specific hash, a linked list usually for a simple implementation

[cinder plinth]

yeah a linked list is good for most use cases

but O(n) worst case

[kind root]

Ye

[cinder plinth]

meanwhile rbtree O(log n) guaranteed

😋

[kind root]

All depends on what sort of values and how many of them u want to have

[cinder plinth]

ok avl tree might be better tho

for that

[kind root]

Sometimes u literally make a dynamic array + linear search and that ends up being faster

[surreal path]

im just gonna do seperate chaining for now

[cinder plinth]

yeah because

cache

[kind root]

Basically yeah

[surreal path]

ill keep a variable of how many key value pairs are in the hashmap, if it exceeds a number

[kind root]

Especially for small objects

[surreal path]

ill add another bucket

[cinder plinth]

it's like hashmaps

they are O(1) best case but that 1 can be slow

so trees might be better in theory

but not in practice

[surreal path]

i think this is better

[kind root]

Yeah I mean 1 just means its the same amount of time for any N, but the baseline can be very slow

[kind root]

Yes

[cinder plinth]

yea

[cinder plinth]

like a dynamic array

[surreal path]

yea

basically

[cinder plinth]

if size >= capacity { capacity *= 2}

[tender gorge]

please dont ruin your schooling with osdev lol

[surreal path]

i tried

im not doing a hashmap

this is pointless

thanks but no

:)

[thorn bramble]

what did you even need a hashmap for

[surreal path]

exactly

[thorn bramble]

no i am asking seriously

i wanted to know lol

[surreal path]

function name and address pairs

[thorn bramble]

for stack traces?

or something?

[surreal path]

yea

[thorn bramble]

you don't need a hashmap for that tbh

[surreal path]

exactly

[thorn bramble]

if you sort them you can just binary search them if you want it to be slightly faster

but like no need for a hashmap

[surreal path]

yea

[surreal path]

broski

ah yes

section header with type 1235125125125

elf signature is correct

like sir

???

lol

[surreal path]

okay

my suspesion is that

the calculation for the seciton in the for loop is wrong

which i dont understand

???

[molten grotto]

(Elf64_Shdr *)hdr + hdr->e_shoff

you are doing that

[surreal path]

but i wrapped it under parenthises?\

i CAN SPELL

CANT

😭

[molten grotto]

wait yeah my bad I were looking at the git code, didn't realize the image had different code nvm, it wasn't even the same line

[surreal path]

ill push it to git

[molten grotto]

also that's not what I meant, I meant the first line

[surreal path]

oki

yea that solved it

thanks for spotting that

cool

[surreal path]

lets goo

[surreal path]

i cannot even sort the fucking array

weird ah shit going on

im losing it

[edgy pilot]

https://en.wikipedia.org/wiki/Quicksort

[surreal path]

okay i got it working through a bubblesort

[surreal path]

stack traces done :)

[surreal path]

😎

coolness

[surreal path]

@kind root look!!

[surreal path]

^^^

[kind root]

congrats

[surreal path]

timers then scheduler

then vfs and file systems

then userspace

yippie!

https://tenor.com/view/adhd-creature-autism-creature-yippee-yippee-creature-the-creature-gif-10446234257473462038

[kind root]

[worldly condor]

Heapsort

[thorn bramble]

damn nyaux being worked on

good stuff

meanwhile i got a bit sidetracked and i'm building a language with a friend and i'll probably use it to write yet another kernel

[haughty kite]

what does the language look like?

[thorn bramble]

the syntax looks kind of like pascal with begin/end for blocks and an interface/implementation thing, though interface is implied and right now we're lacking any sort of class concept/object oriented programing but i think we might just add UFCS and call it a day

it's not really well thought out and we are improvising :^)

[frigid cliff]

arr.sort((a, b) => a < b)

[edgy pilot]

sorted binary tree

[surreal path]

bubblesort my beloved

[spice yarrow]

Do insertion sort, it's a lot faster and still simple

[surreal path]

nah bubblesort my beloved

https://tenor.com/view/ok-and-gif-20529255

wait thats selection sort

[spice yarrow]

You can't even recognize what bubblesort looks like

[surreal path]

umm no

https://tenor.com/view/sort-graph-bubble-sort-chart-linear-representation-gif-15796779

this is bubblesort

😡

i understand bubblesort very well

because it is easy

quicksort is easy too but

i made it poop my array

and im too lazy to debug that

bubblesort is lit 2 for loops

lol

😭

[molten grotto]

uhda on nyaux when

[surreal path]

uhda is out?

ill unironically use it if it was ready for use

not at this stage tho

[molten grotto]

well it works as far as I have tested but its only like 3 machines along with qemu, Ill have to try fixing some kind of bug on my astral oss impl so I can make a proper test image

[surreal path]

not at this stage but whenever i reach the point where i am able to run a good amount of userspace programs (bash, binutils, etc) ill port in uhda

it wont be long now

few more weeks

timers -> scheduler -> timers -> vfs -> tmpfs -> dev devices -> elf loading -> -> driver loading -> xsave/fpu shi -> syscalls -> mlibc -> bash

-> more syscalls -> more and more syscalls -> unix signals? -> uhhhh

doom port ig?

wait I forgot something

yep

[frigid cliff]

Bogo sort

[finite summit]

Miracle sort

[surreal path]

ill play val ig then

work on hpet

blah blah yk

[cinder plinth]

I shall port it

it will be my sidequest today

ok maybe not today

one day my OS will play pink floyd tho 😎

[molten grotto]

I mean you can always download + convert it to raw pcm using ffmpeg and then use it as a kernel module + play it inside the kernel

that's basically what I do for testing it

[cinder plinth]

yeah I know

but I cba to port uhda rn

I should probs port uacpi first

[cinder plinth]

what I plan on doing is just a /dev/audio

then cat a file to it

and the uhda_kernel_pci_allocate_irq i'm not sure how to do

[molten grotto]

well it's basically just supposed to setup msi/msix/legacy irq for the device and wire it up to some cpu vector (that then calls the provided handler fn with the arg when it occurred)

maybe I should document them better

also not all of them are really necessary if you don't plan on disconnecting the device like the irq deallocate function

[cinder plinth]

ok so it should be easily portable

what about uhda_kernel_pci_enable_irq?

is that just a sti

[molten grotto]

it's for the pci device's irq (msi/msix/legacy irq enable bit)

its also not really necessary if you enable the irq already in the pci irq allocation function

[cinder plinth]

ah alright

has uhda been ported to something yet

[molten grotto]

I have it on astral and managarm locally

and my own kernel ofc

[cinder plinth]

uhda = subproject('uhda')

uhda_sources = uhda.get_variable('sources')
my_kernel_sources += uhda_sources

uhda_includes = uhda.get_variable('includes')
my_kernel_includes += uhda_includes

btw

why didnt you just declare a dependency

[molten grotto]

it could probably be made a dependency too, idk how would that work with the compilation flags tho (and this is also what uacpi does)

[cinder plinth]

should work fine

uhda = declare_dependency(sources: blabla, include_directories: blabla)

then you can just do something like dependency('uhda')

[surreal path]

nyaux uses makefiles so

yea

[molten grotto]

maybe I should try that, as long as the compilation flags are inherited from the parent project then that would be easier to use

[cinder plinth]

they are

I have my own patch for e.g flanterm

that uses declare_dependency

[molten grotto]

then you just find all the files and add the include dir manually the same way you do for uacpi

[surreal path]

i have a question

for allocating an irq what does that mean?

im so confused

and uacpi right wants to install an interrupt handler?

i have to allocate an irq?

but what does that mean

and how does that work

also unrelated but nyaux device file icon concept

idfk

[surreal path]

but still what does this mean

[desert haven]

Interrupt vectors are a resource, like memory

[edgy pilot]

do you have device files yet :p

[surreal path]

soon ™️

[desert haven]

Some are available for use, some are in-use by things. So if you need a vector for a device to send you an interrupt, you allocate it. Installing an IRQ handler just means "hey call this function when the interrupt is fired"

[surreal path]

okay i see

[surreal path]

also btw if yall forgot

nyaux logo

never forget

😡

anyways im gonna make install interrupt handler and uninstall

allocating them

and give them to uacpi to use

for whatever tf it wants

AFTER I SLEEP

CAUSE I DIDNT SLEEP

YET

AND ITS 7AM

💀

ive been awake all night

i was working on the hpet anyway

which works btw!!!

lapic timer AFTER i get irq allocation shit

then scheduler then vfs then tmpfs then device files and yada

[desert haven]

Yeah you should sleep, make sure to look after yourself

[surreal path]

@kind root i have a question, uacpi_interrupt_handler doesnt take stackframe directly so it would not be able to be installed. what do i do

also what is ctx

??

is passed to the provided handler for every invocation???

huh

i want to impl this

and what am i supposed to do with handle a firmware request

[molten grotto]

you don't need to do anything there

[surreal path]

just return uacpi_status_ok?

[molten grotto]

you make a wrapper function

[surreal path]

wdym

im confused

[molten grotto]

for the interrupt, if your interrupt handler functions use a different format

[surreal path]

mhm like taking a stackframe

[molten grotto]

yes

[surreal path]

what next?

[molten grotto]

?

[surreal path]

can u like step by step explain this

cause i didnt understand

😭

[molten grotto]

well eg. if your interrupt fn is of the form c bool irq(void* arg, irq_frame* frame); then you have some kind of wrapper eg. ```c
typedef struct {
uacpi_interrupt_handler fn;
uacpi_handle ctx;
} uacpi_irq_wrap_info;

bool uacpi_wrap_irq_fn(void* arg, irq_frame* frame) {
uacpi_irq_info* info = arg;
return info->fn(info->ctx) == UACPI_INTERRUPT_HANDLED;
}

uacpi_status uacpi_kernel_install_interrupt_handler(uacpi_u32 irq, uacpi_interrupt_handler handler, uacpi_handle ctx, uacpi_handle* out_irq_handle) {
uint8_t vec = idt_allocate();
uacpi_irq_wrap_info* info = kmalloc(sizeof(uacpi_irq_wrap_info));
info->fn = handler;
info->ctx = ctx;
*out_irq_handle = (uacpi_handle) info;
register_irq_handler(vec, uacpi_wrap_irq_fn, info);
io_apic_register_isa_irq_or_gsi(irq, vec);
}

[surreal path]

so i gotta make my interrupt handlers take in a void *arg as well

okay

also whats with io_apic_register_isa_irq_or_gsi for? why are we doing that

and whats with register_irq_handker taking in info?? huh

[molten grotto]

its the void* argument

[surreal path]

ummm

[molten grotto]

you don't have to do it exactly like this, it was just an example

[surreal path]

i have an array of function pointers like this

how is that gonna work

to put something in void * how will the assembly stub code know

[molten grotto]

the irq that uacpi gives is usually an isa irq for which there can be an override if you use the io-apic, if using pic then you just use that pic irq

[surreal path]

okay i get u there

[tawdry mirage]

void *isr_ctxt[256];?

[surreal path]

oh okay

[tawdry mirage]

then mov isr_ctxt(%rip + %rax * $8), %rsi assuming it's the index is in rax?

[surreal path]

ill have it uacpi_irq_wrap_info *blahblah[256]

[surreal path]

i can do it in c

cause it can be a global var

so no need

[molten grotto]

why? its not uacpi specific

[surreal path]

what should it be then

[molten grotto]

void*

[surreal path]

okay

[molten grotto]

then the drivers and whatever can put anything they want in there

[surreal path]

kk i see

okay this is my function

yep

[molten grotto]

you should start at 32 as 31 is reserved, idk if it's actually used by anything tho

[surreal path]

okay

[surreal path]

looks good?

[surreal path]

also thanks for catching that

[molten grotto]

except for the sharing of interrupts yes but you likely don't need that to start with

[surreal path]

sharing of interrupts?

[molten grotto]

there can be multiple devices sharing the same interrupt pin

[surreal path]

that bad or good?

[molten grotto]

well its bad because then you have to allow installing multiple irq handlers for the same idt entry + then when you actually get that irq you loop through the handlers until you find one that acknowledges the irq (and the handlers have to check if the device/other thing they are for actually generated it or not)

[surreal path]

but i dont do that

[molten grotto]

its a thing you run into especially with eg. legacy pci interrupts, there are only so many pci interrupt pins

[surreal path]

lol

[kind root]

neither

its just a thing that exists

u only have e.g. 220 idt entries per cpu

there can be e.g. a thousand devices generating interrupts

[surreal path]

okay then

how do i know when i should allocate a cpu idt entry for a handler

or

use some same idt entry

??

[kind root]

well generally never

[surreal path]

what

[kind root]

the way it works in linux for example

it has a per-cpu structure of available idt slots

when something wants a slot, it loops through cpus, find the one with least slots occupied, and allocates one out of that

[surreal path]

and what if u have thousand of devices

generating interrupts

and u dont have enough

idt entries

[kind root]

if u have thousand of devices

u probably have a lot of cpus

so u get 220 * num_cpus slots

some devices still share the same physical origin, so u have to have multiple listeners on a slot

basically u have to think of this as detached things

idt entry is not the same as the irq origin

as irq can always be redirected anywhere

but the same irq can only be redirected to one place

[molten grotto]

you can keep a list of handlers for each idt entry or whatever

nvm that reply lol, it wasn't related to that

[kind root]

well linux does return -ENOSPC

so request_irq just fails

u can easily reach this limit if u have a lot of virtio devices

because it does a per-cpu queue

so per-cpu irq

so after about 200 virtio devices you're done

[surreal path]

so basically no idt entry can share multiple irqs

if u run out of idt entries

[kind root]

not multiple irqs, no

but one irq can originate from multiple devices

[surreal path]

yea i get u on that which wont be for me to worry about

all i need to do is

have a cpu struct of

how much entries i have allocated

and such

[kind root]

u need to care about this if u want to support legacy pci interrupt routing

because that only has 4 irqs for all devices on the system

[surreal path]

should i care about this??

[kind root]

probably not

[surreal path]

okay

[kind root]

i wouldnt bother

well if u want to support older hardware

without MSI

[surreal path]

no i dont wanna support pci with no msi

that comes from

late 90s rihgt?

right*

if so im not doing it

[kind root]

kinda

[surreal path]

yea okay

[kind root]

u can do lspci -vvv on your pc

and see what has MSI and what doesnt

it would look like:

        Capabilities: [40] MSI-X: Enable+ Count=64 Masked-
                Vector table: BAR=2 offset=00000000
                PBA: BAR=2 offset=00000400

[surreal path]

yea all have msi x

well its in wsl

so some have null

for microsoft pci shit

[kind root]

lmao

wsl doesnt count

[surreal path]

which i dont care

[molten grotto]

thinking about this I realized that in my kernel I do it incorrectly I allocate a new cpu irq vector for each pci device using a legacy irq and then install it for the io apic gsi, ig Ill have to first check if the io-apic already has a configured entry for that and if yes then get the vector out of that + reuse it before allocating a new one

[kind root]

bruh

[surreal path]

okay what im going to do for now because i dont have smp yet

[kind root]

do the same thing but pretend u only have one cpu

[surreal path]

is just allocate vec if no vec allocate no handler

yea

[kind root]

https://github.com/torvalds/linux/blob/master/kernel/irq/matrix.c

this is the thing i was talking about

[surreal path]

yea i see

dw ill figure this out and change this around as i get smp working

for now

im just gonna

get this working

oh and do i put anything into out irq handle

[kind root]

well its a thing u use to identify this specific irq if its ever uninstalled, so up to u

[surreal path]

can i just put a number in it

[kind root]

sure

u can ignore it as well

[surreal path]

ill put the vec that was allocated

when i uninstall

ill allocate a int on heap

set it to vec

make the ptr in the handle

[kind root]

waiit

why not just *out_irq_handle = (uacpi_handle)vec

[surreal path]

oh fair

lol

what should i do with "irq"

ignore?

[molten grotto]

no its the actual io-apic/pic irq you have to configure

[surreal path]

uh huh

[kind root]

lol

[surreal path]

i dont have that yet

so im just gonna

not do anything

[kind root]

??

[surreal path]

yea i havent written code for the ioapic yet lol

[kind root]

u cant just ignore the irq number lol

[surreal path]

okay

um

yea no interrupt will fire but

:)

[molten grotto]

well you don't need that function at all unless you want events

[kind root]

if u have rebased pic its just 32 + irq

[surreal path]

no i dont have pic

im not doing pic

ill write ioapic soon

for now

ima just install the handler

whichll do ABSOLUTELY nothing

but eh

lol

[kind root]

could just do return ok

and be done with it

[surreal path]

eh nah

wha

[kind root]

the SCI irq number is basically alyways 9 btw

its a universally accepted thing

[surreal path]

oop my nbad

yep

[molten grotto]

maybe mark your printf with __attribute__((format(printf, 1, 2)))

[surreal path]

nah its okay

[kind root]

or gnu_printf if clang

[surreal path]

trust

[kind root]

not nah

do it

[surreal path]

why i use kprintf to not just print numbers

i also print my lovely text

[kind root]

and

[surreal path]

im not making it so im required to give at least 2 parameters

[kind root]

ur not??

it just makes it so the format string is verified against the args

[surreal path]

oh

[kind root]

at compile time

[surreal path]

fair

whoops

gives warning

so what tf am i supposed to do

[kind root]

gnu_printf

[molten grotto]

printf works too

[kind root]

not on clang

[surreal path]

yea that works

[kind root]

in the header

not the definition

[surreal path]

oh

done

[molten grotto]

works just fine

[kind root]

idk i remember it not working for some reason, i have these ifdefs everywhere

[molten grotto]

;compile cclang_trunk ```c
attribute((format(printf, 1, 2))) void d(const char* fmt, ...) {}
int main() {
d("%d");
}

[viscid pecan]

Compiler Output

<source>:3:9: warning: more '%' conversions than data arguments [-Wformat-insufficient-args]
    3 |     d("%d");
      |        ~^
1 warning generated.

qwinci | 30ms | c | x86-64 clang (trunk) | godbolt.org

[kind root]

maybe someone else can enlighten me

[kind root]

actually no

    #ifdef __clang__
        #define UACPI_PRINTF_DECL(fmt_idx, args_idx) \
            __attribute__((format(printf, fmt_idx, args_idx)))
    #else
        #define UACPI_PRINTF_DECL(fmt_idx, args_idx) \
            __attribute__((format(gnu_printf, fmt_idx, args_idx)))
    #endif

the other way around

[surreal path]

works on gcc

[kind root]

gnu_printf is the gcc way

[molten grotto]

yeah warning: format ‘%d’ expects a matching ‘int’ argument [-Wformat=] on gcc with format(printf

[surreal path]

yea so it works

[kind root]

so do both work or

;compile

__attribute__((format(printf, 1, 2))) void d(const char* fmt, ...) {}
int main() {
    d("%d");
}

[viscid pecan]

Compilation successful

No output.

8infy | 56ms | c | x86-64 gcc 14.2 | godbolt.org

[surreal path]

nvm

[kind root]

didnt seem to do anything

[molten grotto]

;compile -Wall -Wextra

[viscid pecan]

Compiler Output

<source>: In function 'd':
<source>:1:58: warning: unused parameter 'fmt' [-Wunused-parameter]
    1 | __attribute__((format(printf, 1, 2))) void d(const char* fmt, ...) {}
      |                                              ~~~~~~~~~~~~^~~
<source>: In function 'main':
<source>:3:9: warning: format '%d' expects a matching 'int' argument [-Wformat=]
    3 |     d("%d");
      |        ~^
      |         |
      |         int

qwinci | 34ms | c | x86-64 gcc 14.2 | godbolt.org

[surreal path]

;compile cclang_trunk

__attribute__((format(gnu_printf, 1, 2))) void d(const char* fmt, ...) {}
int main() {
    d("%d");
}

[viscid pecan]

Compiler Output

<source>:1:16: warning: 'format' attribute argument not supported: gnu_printf [-Wignored-attributes]
    1 | __attribute__((format(gnu_printf, 1, 2))) void d(const char* fmt, ...) {}
      |                ^
1 warning generated.

linuxmaster2.0 | 62ms | c | x86-64 clang (trunk) | godbolt.org

[kind root]

maybe because new gcc

[frigid cliff]

Text is just fancy numbers

[surreal path]

it is

[kind root]

text is unfancy numbers

[surreal path]

105 110 102 121 32 97 110 100 32 111 98 101 114 114 111 119 32 97 114 101 32 109 97 100 101 32 111 102 32 112 105 122 122 97

[frigid cliff]

Cringe decimal ascii

[surreal path]

😡

@hexed mango 69 20 6C 6F 76 65 20 69 74 61 6E 69 74 75 6D

@frigid cliff zoe i hope u become soggy pizza thats left in the microwave

[frigid cliff]

Ok well it’s in all caps

[surreal path]

wait

i forgot

to translate

it

fuck

[frigid cliff]

Wha

What the heck

[surreal path]

🍕

[kind root]

sometiems pizza tastes better the next day ngl

[frigid cliff]

Pizza isn’t even that good ngl

[kind root]

L take

[surreal path]

L take

[frigid cliff]

Maybe you need to burn me at the stake but it’s true

[surreal path]

chat, force her to use nyaux as her daily driver

[kind root]

ok what is your fav food

[surreal path]

https://tenor.com/view/crane-brazil-grab-poor-car-gif-17372636

[frigid cliff]

Idk

[kind root]

capital punishment

[frigid cliff]

As long as it can run gcc and write and save code

[surreal path]

worse then death penalty

[surreal path]

soon

[kind root]

if u stop playing val

[frigid cliff]

What rank are you even

[surreal path]

sliver 2

😭

[kind root]

play cs instead

[surreal path]

i didnt even play today i focused on osdev today thank you very much

😡

[kind root]

its like valorant but without the cringe abilities

[molten grotto]

I like valorant more

[surreal path]

skill issue

[frigid cliff]

Cs ranking sucks tho

[kind root]

u have faceit

[surreal path]

whats cringe is ur mega skill issue

[kind root]

which is also kernel level anti cheat

[surreal path]

with using ablities