Nyaux

[hollow goblet]

I tried writing a kernel in C++ before

it was also fun

it was fun for different reasons tho

it was fun to like, implement String and shit from scratch

[edgy pilot]

I implemented whole std::string from scratch

with sso and stuff

(I don't remember exactly but I think I copied some stuff from libc++ libstdc++ and msvc++)

[hollow goblet]

i copied serenity's array stuff (those stuff are pretty simple, their String is much more complicated) and String was just a glorified array of characters with helper shit

[unkempt relic]

modern gcc, you just --disabled-hosted-libstdcxx or something like htat

and it will just work

[flat nymph]

I'm just planning to use libstdc++ for that in my kernel

[unkempt relic]

i will be using exceptions in this mock kernel i am putting together to make fun of hackernews

[flat nymph]

I've just removed exceptions from mine

[edgy pilot]

here: https://github.com/ilobilo/string
it's not terrible
I believe it supports almost all C++23 stuff

I haven't encountered any bugs yet

you just need freestanding-cxx-headers and <memory> with std::allocator implemented

[surreal path]

yea we are doing c++

if that doesnt work C

[edgy pilot]

I suggest you do at least some userspace C++ first

[flat nymph]

You can just look at how I'm doing it and do it the other way around

[surreal path]

actually nah

C

[flat nymph]

boring

[cinder plinth]

how is that a problem

[finite summit]

because I said it is

[molten grotto]

eh

they are fine

[finite summit]

if you know what you're doing yeah

[molten grotto]

just don't allocate any memory in them if you are going to run them before you have an allocator

[thorn bramble]

what

the rfuck

am i reading

[thorn bramble]

???

[finite summit]

say you initialzie the allocator in there

[flat nymph]

Aren't they static?

[finite summit]

but also allocate memory in another global constructor

[molten grotto]

[finite summit]

you might assume a specific order

[edgy pilot]

just call them after the allocator is available

[thorn bramble]

that

[flat nymph]

I don't know how they work, but I assumed it's just a pointer to a static table at the beginning of the object

[molten grotto]

then you have to be careful about not using any of them before you have an allocator available tho or else you are going to get screwed

[cinder plinth]

I've never had to worry about vtables

idk why you're making such a big deal

[finite summit]

yeah I just took some random thing

[thorn bramble]

global constructors are just an array of function pointers you have to call in order to initialize global objects

[edgy pilot]

I had that issue two days ago lol

[molten grotto]

I just chose the option of not allocating memory in them as you usually don't need to

[flat nymph]

Are we talking about global constructors or vtables?

[cinder plinth]

I don't use global constructors

[thorn bramble]

i was confused at first too

[cinder plinth]

I use frg::manual_box

[thorn bramble]

vtables are fine too

like everything is fine

u just need to know

the fucking language

[molten grotto]

I use that kind of thing for objects which do need to allocate or which depend on other things

but for things like the log eh

[thorn bramble]

jsut use constexpr/consteval constructors 💯

[flat nymph]

Global constructors are absolutely fine

[thorn bramble]

consteval operator new/delete coming to a c++26 near you soon

[cinder plinth]

and are irrelevant

[flat nymph]

The wiki is just very outdated as usual

[cinder plinth]

like

they dont do anything

[thorn bramble]

yeah no they dont do anything on their own

[cinder plinth]

apart from maybe being slower

[edgy pilot]

what will they do

[cinder plinth]

yea

[thorn bramble]

it's just an abstraction over a bunch of function pointers

you can do it manually in C

[edgy pilot]

I kinda get it

[cinder plinth]

yes

[thorn bramble]

or you can ask the language to do it for you using virtual

[cinder plinth]

but virtual functions will be slower to call

[edgy pilot]

imaginary memory in the compiler

[thorn bramble]

you can allocate memory at compile time

[edgy pilot]

yes

[thorn bramble]

not really

it's necessary for stuff liek vfs anyway

[molten grotto]

they have one more indirection vs inlining the vtable inside the object

[edgy pilot]

it's like calling a function pointer

[cinder plinth]

yes but in any C++ program where you need absolute performance you shouldnt use virtual classes

[thorn bramble]

yeah sure but we are talking about a hobby kernel

[cinder plinth]

yes

[thorn bramble]

the performance hit is basically non existent

[cinder plinth]

yea ofc

like i said

they're irrelevant

when doing osdev

[edgy pilot]

my vfs was basically just virtual functions

every part of it

[median goblet]

it basically compiles to class->vtable[off_myvirtualfunc](parameters);

[cinder plinth]

yeah mine is just a bunch of virtual classes

[median goblet]

so its like calling a function pointer

[cinder plinth]

yes

[median goblet]

its Not That Bad

[cinder plinth]

I never said it was

[median goblet]

you should really start worrying about performance of trivial things like this, only when you can't do anything else to optimize your stuff

[cinder plinth]

https://stackoverflow.com/questions/449827/virtual-functions-and-performance-c

"It's not a performance problem until you can prove it."

[edgy pilot]

inline call isn't a call

[molten grotto]

can't do that with my std::function impl

[edgy pilot]

its like 2.1 times slower than a regular call

[elder shoal]

just dont do oop and have a giant switch statement ez

[molten grotto]

[edgy pilot]

genious

[cinder plinth]

it's not an issue unless you're calling those functions millions of time per second

[edgy pilot]

and you'll probably never do that in a hobby kernel

[cinder plinth]

"A virtual function call first needs to load the vtable pointer from the object. This can result in a data cache miss. Then it loads the function pointer from the vtable which can result in another data cache miss. Then it calls the function which can result in an instruction cache miss like a non-virtual function."

nowadays it's def not an issue

[median goblet]

i know your comment is sarcastic but for the record, it's basically the same

[cinder plinth]

for 1. hobby kernels and 2. fast cpus

[flat nymph]

Just use final where appropriate?

[edgy pilot]

?

[molten grotto]

well branching + static calls do have a performance advantage at least when there aren't too many cases

[cinder plinth]

yeah the performance is comparable

[median goblet]

an OOP project could gain a lot of cases in the relevant switch so it'd probably just use a jump table

[molten grotto]

yeah that's why I said with a small number of cases

[cinder plinth]

"The time it takes to call a virtual member function is a few clock cycles more than it takes to
call a non-virtual member function, provided that the function call statement always calls the
same version of the virtual function. If the version changes then you may get a misprediction
penalty of 10 - 20 clock cycles. The rules for prediction and misprediction of virtual function
calls is the same as for switch statements,"

[flat nymph]

So that virtual functions can be called directly or inlined in some cases

[cinder plinth]

what is final

[flat nymph]

final class

or function

[edgy pilot]

class that can't be inherited from anymore

[cinder plinth]

what's the point

[edgy pilot]

or a function

[cinder plinth]

can't you just

not make a base class

[edgy pilot]

what's the point of override

[flat nymph]

Sometimes not?

[edgy pilot]

so the programmer doesn't make a mistake
or you need to expose that class to the user

[flat nymph]

Like in some situations I think inheritance is more convenient that having a bunch of unions or whatever

and less error-prone

But it you store a pointer to the derrived object, then it makes virtual functions free

[haughty kite]

override isn't required though afaik

right?

[flat nymph]

clang complains if you don't add it

[flat nymph]

it isn't

[haughty kite]

only if you use any override

if you dont use any of it it will not complain

[molten grotto]

not using override is bad code anyway

it should just be an error tbh

[cinder plinth]

i thought override was needed

[haughty kite]

nop

[cinder plinth]

idk it always complained for me

maybe because of -Werror

[flat nymph]

I think gcc doesn't care

[cinder plinth]

I use clang

[edgy pilot]

it's not a warning

I use clang too

[cinder plinth]

wha

then idk

i always get stuff like "overriden function not marked override"

[edgy pilot]

if you use it on at least one method, then it complains

[cinder plinth]

ah

[edgy pilot]

'func' overrides a member function but is not marked 'override' [-Winconsistent-missing-override]

[surreal path]

or yk what

ill just learn rust properly even if it takes months

guess nyaux will be suspended for 2-3 months maybe

its sad but

its the unfortunate truth

i will see you all in 2-3 months

and when im back i should be compitent enough to write an os

[cinder plinth]

why do you even want to use rust

it sucks for osdev

if you still want fancy abstractions use C++

[finite summit]

you heard it from the rust compiler itself

[green elbow]

i mean rust is fine for osdev

just

when you use rust

don't fight the borrow checker

it's like the defining feature of rust

[cinder plinth]

Write a linked list

[green elbow]

that's also fine

[thorn bramble]

doubly linked list isn't

unless you want to spam Rc/Arc

[green elbow]

with unsafe it's fine

[thorn bramble]

or that

[finite summit]

or in C

[thorn bramble]

well, no, it's not exactly fine

but yeah whhtever

[green elbow]

well, yeah, but unsafe is simply required in some cases

it is also a feature of the language

[flat nymph]

Just use Ada

[surreal path]

or actually

ill just use C++ but without most of the C++ features, only like the classes

just not rn tho

i need to sleep

in a few hours ill setup stuff

ill get there

im a little tired tho

night night

[flat nymph]

RAII is great

[surreal path]

NVM IM STAYING WITH RUST

I JUST FIGURED OUT THE ISSUE LMAO

IM SO STUPID

[edgy pilot]

https://tenor.com/view/ive-made-a-decision-and-im-sticking-to-it-peter-charles-pc-principal-south-park-south-park-pajama-day-gif-24821716

[surreal path]

pretty much

YEP IT ALL WORKS

IT PAGE FAULTED

AS EXPECTED

YAYYAYAYAYAY

[elder shoal]

now do on demand allocation

[surreal path]

mathew dont try to torture me

😡

[edgy pilot]

what's that
if page fault, map?

[elder shoal]

yeah

how I do it is on the first page fault on an anonymous range, I map a shared zero page and then do copy on write on it if nescessary

[surreal path]

NVM IT STILL ISNT PAGE FAULTING THAT WAS SOME OTHER BUG

WTF

WHY WONT IT UNMAP

RAHHH

[edgy pilot]

back to C

[surreal path]

NO

THIS SEEMS TO BE A LOGIC BUG

of some kind

switching a dif language wont help

[cinder plinth]

that's kinda smart

I just allocate it on-demand

[surreal path]

rustc

[cinder plinth]

what

[surreal path]

boo

did u get scared

[cinder plinth]

no

[surreal path]

anyways same thing it is UNMAPPING the phys addr

still no page fault tho

😔

[edgy pilot]

check the mapped pages in qemu monitor

[surreal path]

yea thats what im gonna do

[flat nymph]

Are you flushing TLB?

[surreal path]

i am invlpging yea

info tlb shows its not mapped

i will now not dealloc it and check info tlb again

if it is mapped then

this is not my problem to deal with

could be some rust quirk idfk

[finite summit]

btw info tlb has nothing to do with the tlb

[surreal path]

wym

[finite summit]

it's just printing the mappings page by page

as opposed to in a range format when you do info mem

[surreal path]

how do i view the tlb

[finite summit]

that's the neat part

you don't

[surreal path]

fun

okay this is getting weird

even not unmamping it

its not showing up

umm

lol

oh wait

flags i forgor

still not showing up?

yea this weird lmao

should i even care if unmap works or not

i mean

im doing everything correct

self.unmap(i.base + (f * 0x1000));
                    assert_eq!(
                        None,
                        Self::find_pte(self.rootpagetable as usize, i.base + (f * 0x1000))
                    );

ill do something like this to verify

if it fails

bug

if it doesnt

idk tlb hates me lol

tlb can kiss my booty

idc then

yea no problems with the assert

so it did work

just tlb does NOT care

😭

ummm

why is there no invlpg instruction

HELLO

RUST???

😭

this is dev mode

and its like

not there

yall gotta be kidding me

[finite summit]

make sure the asm thing is volatile or smth

idk rust

[surreal path]

Instead of having to explicitly use volatile, volatile is the default and you have to opt-out if you want using pure. This is done to prevent forgetting to use volatile when you need to.

does anyone here work with rust no whats going on

[finite summit]

then make an asm function

that does invlpg

[surreal path]

good idea

[finite summit]

global invplg
invplg:
    invlpg [rdi]
    ret```

taken straight from obos

[surreal path]

yoinky splunky

ummm still objdumping no exist

WHWHY RUST HATE ME

RAHH

OKAY YK WHAT I DONT CARE IMA MOVE ON

[kind root]

werent u rewriting it yesterday

[surreal path]

nah i realized it was a bug

[haughty kite]

probably because you dont know how to write correct rust code

[surreal path]

stupid me

[surreal path]

bro im using references now

😡

[kind root]

yeah thats uh

not how that works

[finite summit]

use C

[haughty kite]

yeah

[surreal path]

nah look im just gonna ignore whats going on, rust is not compiling invlpg in but it doesnt rlly matter DOES IT

nah

[finite summit]

day n of waiting for nyaux to switch to C

[haughty kite]

it does

lol

[finite summit]

it does

[surreal path]

okay look ima see if the rust server knows abt this or whatever

is fine

[finite summit]

I should join the rust server

say "C better"

and leave

[kind root]

day n of waiting for nyaux to burn out because of rust issues and speedrunning os instead of learning it

[surreal path]

im not speedrunning tho

i know what im doing

[haughty kite]

its not even speedrunning

[finite summit]

lol

[kind root]

ur trying to speedrun

[haughty kite]

really.

[kind root]

but u get stuck every minute

[surreal path]

cause of issues with dumb me

[finite summit]

(dunning kruger effect)

[surreal path]

bru

[surreal path]

yes

i understand the concepts of what i need to implement

[kind root]

High level details rather

[surreal path]

and implementation details

its just

im stupid

lol

is all

[kind root]

No

[surreal path]

wym no?

i know how to implement a gdt and a idt and a vfs and a vmm and a pmm and a scheduler and such

[kind root]

Ur not dumb ur just lacking the patience to do anything right

[surreal path]

explain further

[kind root]

Not learning rust properly, then proceeding to use it for a thing that requires insanely advanced rust knowledge, not researching the component that you're writing and instead doing it willy nilly

[surreal path]

wdym "the component"

[kind root]

Vmm, pmm, scheduler, whatever ur doing right now

[surreal path]

but i do now

know*

[finite summit]

moral of the story: use C

[haughty kite]

or learn to program first in general

[surreal path]

um

excuse me?

[haughty kite]

the rust safety rules are incredibly complex

[finite summit]

Sir Dunning-Kruger

[surreal path]

i know this concept

[surreal path]

if i didnt know to program i wouldnt be writing an os

chat can we just stop trying to roast me into a roasted chicken

this is rlly like yall trying to bring down my moral or smthin

[edgy pilot]

sorry but clearly not enough

[finite summit]

-# I came into osdev not knowing how to make a linked list, and didn't know a thing about most data structures

[surreal path]

what

i know how to make a linked list wtf

[finite summit]

I'm saying I didn't know how

lol

[edgy pilot]

he said

I came into osdev not knowing how to make a linked list, and didn't know a thing about most data structures

[finite summit]

now I do though

[surreal path]

i knwo what he said 😭

[surreal path]

are u saying abt urself

[finite summit]

yes

[haughty kite]

dont worry at least its not rust

[finite summit]

I

[haughty kite]

aka "you cant make a safe doubly linked list" lang

[finite summit]

just work on obos, it's only half cursed

at least invlpg works there lol

[surreal path]

no cause pitust thinks i "dont know how to program"

thats kinda mean to be honest

[haughty kite]

ok sorry

[surreal path]

thanks

look ima ask the rust server abt this and we'll see whats going on

[surreal path]

maybe there are gaps in my knowledge sure, but im learning them and im improving. im not trying to speedrun im not trying to not learn i am trying to learn and understand to the best i can

[hollow goblet]

@surreal path hi fellow rust osdever

[surreal path]

hi

[hollow goblet]

I just got my basic scheduler working

[surreal path]

nice

[hollow goblet]

sleep is not there yet because I'm still not happy with my timer stuff

[surreal path]

@kind root this is my response to you as well infy

[hollow goblet]

i need a non-arch specific api for that

but otherwise sleep is easyy

[surreal path]

nice !

rust server big brain

[hollow goblet]

there's something related to clobbering memory

it is useful

check the asm! docs

if memory is clobbered llvm/rustc won't reorder or drop anything

[surreal path]

okay so i need to make it clobbered?

[hollow goblet]

idk check the docs

i know it exists

i used it in clang but i don't know the syntax for rust asm!

[surreal path]

using cloober_abi solved nothing

[hollow goblet]

no not that

ok let me check the docs

worst case, use llvm_asm! maybe they don't have such an option in asm!

ok in asm! it is enabled by default

ignore everything i said

[surreal path]

bru

[hollow goblet]

unless you pass options(nomem)

[finite summit]

echo -e 'bits 64\nglobal invlpg\ninvlpg:\n\tinvlpg [rdi]\n\tret\n' > invlpg.asm

[hollow goblet]

it does what you want it to do

[surreal path]

i lit did

and rust still optimized it out lmao

[hollow goblet]

this is something unrelated

[finite summit]

rust skill issue

[hollow goblet]

i don't know the issue but i don't think it's a rust skill issue

rust's asm! is just a different syntax over llvm's

[finite summit]

why tf would rust optimize out a function call that it knows nothing about

[surreal path]

ask llvm

not me

[finite summit]

he said he also used a function call

[surreal path]

yea

i did

include_asm!

[finite summit]

add it to your build

then extern c it

[edgy pilot]

can rust even do that

[finite summit]

I think so

@cinder plinth might know, he's the rust compiler, after all

[cinder plinth]

Yeah it's the same as calling c functions I guess

[hollow goblet]

wait since when has that been a thing

I'VE ALWAYS USED global_asm!(include_str!()) AAAAAA

[surreal path]

srry i mean

[hollow goblet]

yes

[surreal path]

global_asm

[hollow goblet]

idk what's wrong with this server and hate on rust but yea rust has external linking lmao

[surreal path]

i have an idea

[cinder plinth]

It's not hate it's just no one knows it

[finite summit]

I don't hate rust, I despise it /j

[hollow goblet]

i guess

[edgy pilot]

how is that hating on rust

[thorn bramble]

the way everyone uses rust for osdev is shit

[surreal path]

even doing this

let mut fake: usize;
                core::arch::asm!(
                    "invlpg [{0}]",
                    "mov {1}, 5",
                    in(reg) va,
                    out(reg) fake,
                    options(nostack)
                );
                println!("{}", fake);

[edgy pilot]

It's a simple question

[surreal path]

rust still optizming it out

[edgy pilot]

did you objdump?

[surreal path]

yea

[edgy pilot]

and no mention of invlpg?

[surreal path]

yep

[edgy pilot]

can you send the binary

[surreal path]

sure

[hollow goblet]

everyone on this server (maybe except me but I don't have my stuff public

people on the rust server are not bad osdevers

there's serious stuff going on there :3

[surreal path]

@edgy pilot

[finite summit]

lemme look at that

[cinder plinth]

The only good rust os is aero

[hollow goblet]

consider: ableos

[cinder plinth]

No idea what that is

[finite summit]

damn no invlpg

[cinder plinth]

Wasn't able here

if that's their os

[hollow goblet]

he was here at some point

no idea if able is still here

[finite summit]

bruh

[thorn bramble]

osdev on the rust server is basically riscv shilling at this point

not even osdev

[hollow goblet]

heyy that's not true it's basically all people following phil-opp's blog

[thorn bramble]

czapek@raptor-wsl:~/wow$ rustc -o test.o test.rs --target x86_64-unknown-none
czapek@raptor-wsl:~/wow$ objdump -d test.o

test.o:     file format elf64-x86-64

czapek@raptor-wsl:~/wow$ cat test.rs
#![no_std]
#![no_main]

pub extern "C" fn _start() -> ! {
        unsafe {
                core::arch::asm!("invlpg [rax]");
        }
        loop {}
}

#[panic_handler]
fn panic_handler(_: &core::panic::PanicInfo) -> ! {
        loop {}
}```

what is rustc doing

[hollow goblet]

my and repnop are the only risc-v rust osdevers that I am aware of

[thorn bramble]

or what am i doing...

[edgy pilot]

what's the function name where you use invlpg

[thorn bramble]

yeah im wrong hold up

[hollow goblet]

wait

[surreal path]

unmap

[thorn bramble]

i dont think i can objdump an object file like that

[surreal path]

PageMap::unmap

[hollow goblet]

idk invlpg but is that at&t syntax or intel syntax?

[surreal path]

idk 😭

[edgy pilot]

Is that release mode

[surreal path]

no

dev mode

this may be a rustc bug actually

to be fair

[thorn bramble]

doubt

[hollow goblet]

this wouldn't be optimised out in dev mode

[surreal path]

idfk ???

[finite summit]

why doesn't unmap exist in my objdump

[hollow goblet]

is invlpg different in at&t and intel syntax?

[edgy pilot]

are optimisations disabled

[thorn bramble]

intel syntax

[hollow goblet]

maybe the assembler shits

[surreal path]

it is running devmode

[thorn bramble]

rust asm uses intel syntax

[surreal path]

so optimizations should be disabled

[hollow goblet]

is that correct intel syntax for invlpg

[thorn bramble]

yes

[edgy pilot]

yes

[finite summit]

extremely rare rust W

[thorn bramble]

invlpg takes a memory operand

[surreal path]

good question

[thorn bramble]

so [reg] should be fine

[edgy pilot]

and #[test]

[hollow goblet]

oki

[surreal path]

guess what chat

i fixed it

[hollow goblet]

@surreal path can I see your linker script

[surreal path]

now it DOES invlpg

[finite summit]

NyauxKT::PageMap::unmap doesn't exist either

[surreal path]

i fixed it

[hollow goblet]

what was it

[thorn bramble]

czapek@raptor-wsl:~/sources/zig-kernel$ objdump -d kernel/zig-out/bin/kernel | grep invlpg
ffffffff80000040:       0f 01 3c 25 00 00 00    invlpg 0x0```

yay good language

[surreal path]

zig pro

so it does invlpg

[thorn bramble]

asm volatile("invlpg 0");```

[hollow goblet]

last time I had symbols not appearing it was my linker script discarding it

[finite summit]

yay good language

[surreal path]

i see why it doesnt page fault lmao

[finite summit]

(assembly)

[cinder plinth]

What'd you change

[surreal path]

i wrote the pte to be 0

[thorn bramble]

??

[surreal path]

yea stupid ik

[thorn bramble]

?????????????

[edgy pilot]

"rustc bug" smh

[surreal path]

💀

[hollow goblet]

lmao

[surreal path]

lmao

[thorn bramble]

this guy makes me wanna quit life

[hollow goblet]

yea as normal it's just a human error

[surreal path]

why you mean :c

[hollow goblet]

this was me 3 years ago on my first osdev attempt

[thorn bramble]

ill name drop u on my goodbye note

[surreal path]

😭

☠️

[hollow goblet]

don't be mean to them

[thorn bramble]

/s

[hollow goblet]

they will improve

my first osdev attempt was just following phil-opp's blog

[thorn bramble]

same

lol

[hollow goblet]

then I wrote stuff for ableos

[thorn bramble]

its kinda shit

[finite summit]

my first osdev attempt was... questionable

[surreal path]

yea

[hollow goblet]

it is shit but

the work done around that community (the rust-osdev github org

[thorn bramble]

its basically like

[hollow goblet]

is not actually bad

[thorn bramble]

"go use this crate lololol"

[hollow goblet]

true but

they improved the ecosystem

[edgy pilot]

"shout out to nyauxmaster for helping me speedrun life"

[hollow goblet]

like thanks to them x86_64-unknown-none is stabilised

they were attempting real mode rust at some point

all the ${cpu_architecture} crates are inspired by their x86_64 crate

and they are good crates

[surreal path]

wait shouldnt u set the pte to 0? i know this is wrong but i am not understanding WHY its wrong, ur setting everything to 0?

including the present bit

so shouldnt the mmu not complain

[thorn bramble]

wtf is it complaining about to u

[surreal path]

not complaining

as in

it shouldnt be mapped anymore

[thorn bramble]

tlb exists

but its not mapped anyone

but if it exists in the tlb it will still be "mapped"

[hollow goblet]

did you set cr3 if you're creating a new page table

[surreal path]

yes

[hollow goblet]

and did you do the invalidation stuff correctly

[surreal path]

yep

pub fn unmap(&self, va: usize) {
        let him = unsafe { Self::find_pte(self.rootpagetable as usize, va) };
        if let Some(h) = him {
            *h = 0;

            unsafe {
                let mut fake: usize;
                core::arch::asm!(
                    "invlpg [{0}]",
                    "mov {1}, 5",
                    in(reg) va,
                    out(reg) fake,
                    options(nostack)
                );
                println!("{}", fake);
            };
        } else {
            println!("not found");
        }
    }

[hollow goblet]

sorry I don't remember x86 enough

[thorn bramble]

; after unsafe block makes me cry

[hollow goblet]

lmao yes

that fake variable makes me cry

[surreal path]

yea so why isnt it unmapping hmmmm

[thorn bramble]

did you verify

that

[edgy pilot]

It is unmapping

[thorn bramble]

the INSTRUCTION

[edgy pilot]

probably

[thorn bramble]

is EXECUTED

[surreal path]

yes

i had objdump'd invlpg is executed

[hollow goblet]

you should really learn to use a debugger

[thorn bramble]

wow it magically fixed itself

[hollow goblet]

objdump doesn't tell you it executed

it tells you it is in the binary

[thorn bramble]

rustc goated

[surreal path]

okay we'll see

ill pull up my debugger

[hollow goblet]

gdb or lldb

[surreal path]

i use gdb

[hollow goblet]

ok

i was expecting you to pull up windbg or something

[surreal path]

i know how to use a debugger

[edgy pilot]

nah he uses ida pro

[thorn bramble]

i don't even know how to use a debugger beyond breakpoints and reading memory

[surreal path]

pro

[thorn bramble]

the totally legit copy of IDA Professional 9.0™️

[hollow goblet]

step/stepi/next/nexti/finish/br/delete breakpoints/i/p

[edgy pilot]

same but without the second part

[hollow goblet]

all you need to know tbh

[surreal path]

yep it got executed

invlpg did

[edgy pilot]

then what's the problem

[surreal path]

idfk ??

[edgy pilot]

move on

[surreal path]

wdym

i need unmap to work

[edgy pilot]

if there's no problem, why are you still debugging it

[surreal path]

if it isnt page faulting

[edgy pilot]

why

[surreal path]

infy or pitust can u tell u why

[edgy pilot]

maybe add a sleep() or a few pauses between invlpg and trying to write

[surreal path]

what does that even do

[edgy pilot]

idk maybe it needs some time to think lmao

[surreal path]

ah yes

okay im not gonna worry and move on

im doing everything correct so

if the cpu chooses to listen or not

not my business

anyways gonna add vmm_region_alloc into my global allocator and yea

anyways vmm done

[kind root]

he has been trying to for the last week

[surreal path]

lmao

i love giving ppl brainrot

[edgy pilot]

yes finally

[surreal path]

uacpi integration time

[edgy pilot]

no

[surreal path]

why

[edgy pilot]

you don't need it yet

[thorn bramble]

why do you need it

[surreal path]

uhhhh

[thorn bramble]

exactly

[edgy pilot]

you don't

[surreal path]

heap stresstest

lol

[thorn bramble]

that's a good damn point

but no

[surreal path]

then what should i do next

[edgy pilot]

interrupts

[surreal path]

i already have

[edgy pilot]

I mean timer and a scheduler

[finite summit]

irq interface

for your drivers

[edgy pilot]

no drivers yet lol

[surreal path]

lol

[finite summit]

so?

[edgy pilot]

too early for that

[surreal path]

scheduler THIS early?

[finite summit]

it's too late for a scheduler

if I'm being honest

[edgy pilot]

yes

why not

[finite summit]

well not too late

[thorn bramble]

scheduling is like

the most important part

of any kernel

[surreal path]

fair

[finite summit]

a wise man told me to have a scheduler asap when I start my kernel

[thorn bramble]

i started with a memory manager

[elder shoal]

scheduler is what I do after I do memory management

[surreal path]

start with a scheduler

[elder shoal]

you just have to not have the scheduler depend on the memory manager

[finite summit]

the problem?

[surreal path]

funny

[hollow goblet]

i don't have an allocator (I use someone else's allocator) and I have a scheduler lmao

[surreal path]

bru

[thorn bramble]

yeah taht will be fun to untangle later

[finite summit]

I didn't have an allocator until a week after I start dev on my current kernel

[hollow goblet]

what my kernel does rn

parse fdt

scheduler

pmm

has code for mapping pages but unused

[finite summit]

but I had a scheduler within the first 5 days

[thorn bramble]

pub export fn _start() callconv(.C) noreturn {
    arch.disableInterrupts();
    arch.serial.initialize(0x3f8);

    const hhdm_response = hhdm_request.response.?;
    const kernel_address_response = kernel_address_request.response.?;
    const memory_map_response = memory_map_request.response.?;
    const smp_response = smp_request.response.?;

    bsp_cpu_info = .{
        .arch_cpu = arch.cpu.ArchCpuInfo.init(smp_response.bsp_lapic_id),
        .idle_thread = &bsp_idle_thread,
    };

    bsp_cpu_info_block = .{
        .cpu_info = &bsp_cpu_info,
        .current_thread = &bsp_idle_thread,
    };

    arch.cpu.setCpuInfoBlock(&bsp_cpu_info_block);
    arch.cpu.initializeCpuEarly(&bsp_cpu_info);

    // ...
}```

literally what my kernel does at the start

[surreal path]

iretq im in ur walls

[thorn bramble]

hi

[surreal path]

pull me out

i cant breath

[thorn bramble]

nahj u stay in there

[surreal path]

[finite summit]

what my kernel does at the start is initializes the gdt and idt, then initailizes the scheduler

[thorn bramble]

my walls are paper thin

[elder shoal]

astral does this

void kernel_entry() {
        cpu_set(&bsp_cpu);
        logging_init();
        logging_sethook(arch_e9_putc);
        arch_gdt_reload();
        arch_idt_setup();
        arch_idt_reload();
        dpc_init();
        pmm_init();
        term_init();
        logging_sethook(term_putchar);
        arch_mmu_init();
        vmm_init();
        alloc_init();
        cmdline_parse();
        acpi_early_init();
        arch_apic_init();
        timekeeper_early_init(0);
        cpu_initstate();
        arch_apic_timerinit();
        sched_init();

[thorn bramble]

my initializeCpuEarly sets up gdt and idt

[surreal path]

im going to steal ur cat then by grabbing it via the walls

or the floorboards

[thorn bramble]

after that i initialize my memory manager and start up other cores

[finite summit]

although the scheduler is cooperative until the scheduler timer is initialized

[surreal path]

mathew what is adpc

[hollow goblet]

#[no_mangle]
unsafe extern "C" fn _start() -> ! {
    HHDM_OFFSET.store(HHDM_REQUEST.get_response().unwrap().offset() as usize, Ordering::Relaxed);
    crate::memory::alloc::init(INITIAL_KERNEL_HEAP_START, INITIAL_KERNEL_HEAP_SIZE);
    crate::memory::pmm::init(MM_REQUEST.get_response().unwrap().entries());

    // Get a DTB and parse it, then discover devices
    // This also sets the serial console.
    let dtb_response = DTB_REQUEST.get_response().unwrap();
    let dtb = dtb_response.dtb_ptr();
    let fdt = fdt::Fdt::from_ptr(dtb.cast()).unwrap();
    crate::fdt::discover_devices(&fdt);

    log::info!("Hello serial console!");

    // CHECK FOR LIMINE RESPONSES

    log::info!("Initialising hart context");
    // ALLOCATE 4 PAGES FOR KERNEL STACK
    // HART CONTEXT INITIALIZE

    log::info!("Initialising trap handler");
    trap::init();

    log::info!("Initialising scheduler");
    crate::scheduler::init();

    THREAD1 = crate::scheduler::create_kernel_thread(thread1);
    crate::scheduler::create_kernel_thread(thread2);

    log::info!("Enabling interrupts");
    interrupts::enable();

    loop {
        wfi();
    }
}

static mut THREAD1: *mut crate::thread::Thread = core::ptr::null_mut();

fn thread1() {
    log::info!("hello from thread1 :S");
    log::info!("blocking thread1 :O");
    crate::scheduler::block(crate::thread::ThreadState::Paused);
    log::info!("hello again from thread1 :3");
    log::info!("terminating thread1 UwU");
    crate::scheduler::terminate();
}

fn thread2() {
    log::info!("hello from thread2 >:3");
    log::info!("unblocking thread1 from thread2 LoL");
    unsafe {
        crate::scheduler::unblock(THREAD1);
    }
    log::info!("terminating thread2 OwO");
    crate::scheduler::terminate();
}

[surreal path]

dpc*

[hollow goblet]

basically the entirety of my kernel

[elder shoal]

deferred procedure call

[surreal path]

ah yes makes total sense

[hollow goblet]

I had to cut some code and replace it with comments tho

[surreal path]

A Deferred Procedure Call (DPC) is a Microsoft Windows operating system mechanism which allows high-priority tasks (e.g. an interrupt handler) to defer required but lower-priority tasks for later execution.

[finite summit]

what it does is it defers code execution until a later point in the kernel at a lower interrupt priority

[surreal path]

i see

[finite summit]

in obos they run at IRQL_DISPATCH

[surreal path]

not very unix like mathew

[finite summit]

when irql goes < IRQL_DISPATCH

[thorn bramble]

linux definitely has an equivalent

[finite summit]

OBOS_NO_UBSAN OBOS_NO_KASAN void Core_LowerIrqlNoThread(irql to)
{
    Core_LowerIrqlNoDPCDispatch(to);
    if (to < IRQL_DISPATCH && CoreS_GetCPULocalPtr())
    {
        CoreS_SetIRQL(IRQL_DISPATCH, to);
        *Core_GetIRQLVar() = IRQL_DISPATCH;
        CoreH_DispatchDPCs();
        *Core_GetIRQLVar() = to;
        CoreS_SetIRQL(to, IRQL_DISPATCH);
    }
}```

[surreal path]

linux is not real unix then

😡

[thorn bramble]

its basically a per-cpu work queue

[surreal path]

linux is windows comfirmed

[surreal path]

i see

[hollow goblet]

this thread is eating my brain

[finite summit]

OBOS_NO_UBSAN OBOS_NO_KASAN void CoreH_DispatchDPCs()
{
    // Run pending DPCs on the current CPU.
    for (dpc* cur = LIST_GET_HEAD(dpc_queue, &CoreS_GetCPULocalPtr()->dpcs); cur; )
    {
        dpc* next = LIST_GET_NEXT(dpc_queue, &CoreS_GetCPULocalPtr()->dpcs, cur);
        LIST_REMOVE(dpc_queue, &CoreS_GetCPULocalPtr()->dpcs, cur);
        cur->cpu = nullptr;
        cur->handler(cur, cur->userdata);
        cur = next;
    }
}```

[hollow goblet]

im gonna leave

[elder shoal]

softint

[surreal path]

DONT LEAVE

we have snacks

not pro

[thorn bramble]

i knew it :^)

[hollow goblet]

don't ping me it makes me follow the thread

[surreal path]

but im a fellow rust user

[hollow goblet]

discord fix your app why can you ping random people and force them to follow the thread

[surreal path]

i will use nixstrap if u stay

[finite summit]

@discord fix

[surreal path]

😔

we lost a comrade today

[hollow goblet]

ok

[elder shoal]

tasklets to be more specific it seems

[surreal path]

so u will stay?

[hollow goblet]

yes

[surreal path]

pro gamer

😎

blackmail my favourite past time activity

anyways

time for scheduler

or whatever tf

ill do that after i eat me hungy

[hollow goblet]

port to nixstrap fr

should I release nixstrap

[surreal path]

yes

[hollow goblet]

ok gonna clean shit up

[thorn bramble]

please don't; use nixstrap

u dont ev en know fucking nix

this guy will never finish a project

istg

[surreal path]

i will

chill

also how do u know i dont use nix

i lit used nixos a bunch of times lmao

tf

[thorn bramble]

i still doubt you know nix

[surreal path]

you doubt everything lmao

[hollow goblet]

i do too but

[hollow goblet]

a user for nixstrap = i will learn what other people want it to do and it's not just a gearbox (my kernel) thing

[thorn bramble]

cant wait for the 10 daily pings @ sanana

[surreal path]

what 10 daily pings

oh lmao ur talking abt me

[thorn bramble]

oh did i underestimate...

my bad

[surreal path]

pinging sanana

[thorn bramble]

[surreal path]

😡

[hollow goblet]

tho i would not suggest using nixstrap with rust as your first attempt at using nixstrap

rust + nix + custom targets and build-std don't really add up well

it does work don't get me wrong

it takes a bunch of effort

[surreal path]

ok fair

hi fadanoid

[hollow goblet]

there's a rust template available already tho

let's update my disko fork

that seemed to work

[flat nymph]

It can't, deal with it

[flat nymph]

(my kernel still doesn't do tlb invalidation properly)

[surreal path]

😡

oh yeah sorry im taking a break for like until weekend cause school is getting more stressful

sorry chat

but this is to avoid burn out

[kind root]

cya in two months

[surreal path]

bro

💀

infy stop being

like actually ur being so fucking mean

stop

[flat nymph]

It's the worst type of bugs if you don't get it right...

[surreal path]

its not funny anymore at this point

[kind root]

lol

im sorry

[surreal path]

thank you

[kind root]

cya on saturday

[surreal path]

nah friday

friday i have no school

😎

[magic kernel]

You got this

[plush hearth]

Lucky

[wicked loom]

honestly i back infy on this

i bet you're going to disappear for months now

it has already happened like 3 times

idk why this one would be different

prove us wrong, i guess :p

[surreal path]

nope im here

[surreal path]

ill prove yall wrong

lmao

im currently outside rn ill be back in a few hours eta 5pm irish time to work on nyaux

[surreal path]

alright

ready to start some work

im gonna do the uacpi first because i believe its just better to do so, i know yall disagree but its fine

good heap stress test anyway

[surreal path]

okay very fun

so after integrated uacpi

we got some issues

with the heap

okay more bugs

fun

this just happened

wtf is going on

now

lol

memory corruption!!

even tho im being "safe"

[surreal path]

CR2=ffffffff80001320

hmmm

fn slab_allocsearch(&mut self) -> Option<usize> {
        let mut h = &mut self.slabs;
        'outer: loop {
            println!("{:?}", h);
            if let Some(fre) = h.as_mut().unwrap().freelist.take() {
                h.as_mut().unwrap().freelist = fre.next.take();
                return Some(fre as *mut KTNode as usize);
            } else {
                if h.as_ref().unwrap().next_slab.is_none() {
                    println!("breaking out");
                    break 'outer;
                } else {
                    println!("{:#?}", h.as_ref().unwrap().next_slab);
                }
                h = &mut h.as_mut().unwrap().next_slab;
            }
        }
        println!("HELLLO???");
        let new = slab_header::init(self.size);

        let o = h.as_mut().unwrap().next_slab.take();
        println!("suc");
        if let Some(g) = o {
            new.next_slab = Some(g);
            h.as_mut().unwrap().next_slab = Some(new);
        } else {
            h.as_mut().unwrap().next_slab = Some(new);
        }
        let gethim = h.as_mut().unwrap().next_slab.take().unwrap();
        println!("{:#?}", gethim.freelist);
        gethim.freelist = gethim.freelist.as_mut().unwrap().next.take();
        let gg = gethim.freelist.as_mut().take().unwrap();
        return Some(*gg as *mut KTNode as usize);
    }

[edgy pilot]

make a scheduler first, then uacpi

[surreal path]

no this is good to stresstest my heap

im gonna redo my slab allocator

with MaybeUnIt<T>

struct slab_header {
...
freelist: MaybeUninit<Option<&'static KTNode>>
...
}

because otherwise its ub

and the compiler can do whatever the hell it wants

it optized prev out

causing dumb issue

lol

thanks to @glossy walrus for the suggestion

tho i dont think theres anything wrong with the 'static lifetime

the ktnode is going to live for the entirety of the kernel when its in the freelist

[surreal path]

okok my brain is dying

trying to figure out how to right this slab allocator properly

hmm

the mad lads at rust community server told me to use raw pointers

in this context it is SAFE

Option<NonNull<T>>

they told me to do

[hollow goblet]

raw pointers are always safe they obey the same rules as references

just you're supposed to uphold those rules instead of the compiler

the unsafe part is reading from and writing to the raw pointers

[surreal path]

i get you

[hollow goblet]

sigmas know i'm skibidi now

[surreal path]

lmao

i should talk in the rust server more they are so helpful

[surreal path]

slab allocator nearly done

just working on the last few functions then

we will have slab allocator that isnt UB

thanks to the RUST server for guiding me on how to create no UB

[surreal path]

fn slab_allocsearch(&mut self) -> Option<*mut u8> {
        let mut h = &mut self.slabs;
        for i in h.iter_mut() {
            
            unsafe {
                if let Some(friend) = (*i.as_ptr()).freelist {
                    (*i.as_ptr()).freelist = (*friend.as_ptr()).next;
                    return Some(friend.as_ptr().cast::<u8>());
                }
            }
        }
        None

this is good rust code according to them lol

[surreal path]

lol

i have tracked down the issue to my slab_header init

solved it

enabling strict provence is the best thing ive ever done

no more UB

cause the compiler errors out if theres some ptr to int to ptr cast

😎

@finite summit what is ur opinion on this information

[finite summit]

https://tenor.com/view/jfk-clone-high-i-like-your-funny-words-magic-man-jack-black-gif-18659433

[surreal path]

😎

[surreal path]

fun

uacpi page faults on kvm

and non kvm runs out of memory

why does the curse always follow me

why tf is parent 0x0

why is out_obj 0x0

why is args 0x0

wtf

[kind root]

the cause is wrong also

[surreal path]

wdym

[kind root]

LOAD_CAUSE_INIT is whats passed

[surreal path]

UACPI_TABLE_LOAD_CAUSE_LOAD_OP

yea

wtf is happening

UACPI_TABLE_LOAD_CAUSE_INIT

[kind root]

memory corruption

probably

who knows

[surreal path]

fucking hell

[kind root]

like literally

look at uacpi.c:438 there

u get bogus args somehow

[surreal path]

yea

all the addresses i give to uacpi

their sizes too

could it be my vmm region

i mean

it looks fine

WHY AM I ALWAYS CURSED

RAHHHH

changing between dev and release mode doesnt change anything, in tcg it works fine (only runs out of memory cause i didnt make free function yet) but kvm it page faults

cr2 is 0

as u'd except

[surreal path]

not pro gamer

[surreal path]

still debugging this

not fun

im losing my mind

[kind root]

I'm used to a lot more verbose debugging sessions

[finite summit]

nyaux thread almost at 23000

[surreal path]

yea i know

anyways i think i found the function at fault

slab_allocsearch is doing something rlly weird

i just need to find where

and we gucci

[surreal path]

crazy !!!! ik

anyways i have school tmrw

i must sleep now

goodnight chat

[steady flume]

good night nyaux :3

[kind root]

good morning nyaux

back to work NOW 😡

[surreal path]

IM AT SCHOOL 😭

[finite summit]

Too bad

Get a laptop and setup rustc

[surreal path]

they dont allow devices here

also

school nearly finished but i have tennis practice after this

ill be back around 8pm

[surreal path]

im back

[surreal path]

@kind root im back to work

im gonna write some unit tests

maybe it'll show me whats going on

[kind root]

nice

[surreal path]

its fucking hell trying to setup them up 😭

[cinder plinth]

which language are you using now

[surreal path]

still rust

[cinder plinth]

[surreal path]

one thing i found out

its not an alignment issue

i cant get unit tests working

so

im just gonna ball and not care

actually wait

i can do something brb

wait

i will read the source code of linux about this

[surreal path]

@kind root I GOT UNIT TESTS WORKING

[surreal path]

@kind root ```rs
running 1 test
thread 'mem::pmm::h' panicked at src/mem/pmm.rs:348:9:
misaligned pointer dereference: address must be a multiple of 0x8 but is 0xffffffffffffffff
note: run with RUST_BACKTRACE=1 environment variable to display a backtrace
thread caused non-unwinding panic. aborting.
error: test failed, to rerun pass --lib

Caused by:
process didn't exit successfully: /home/rayan/NyauxKT/kernel/target/x86_64-unknown-linux-gnu/debug/deps/NyauxKT-207c79480958aa03 (signal: 6, SIGABRT: process abort signal)

look what i found

extern crate std;
    extern crate libc;
    assert_eq!(1, 1);
    let data: *mut u8 = unsafe {
        libc::mmap(0 as *mut c_void, 4096, libc::PROT_READ
        | libc::PROT_WRITE, libc::MAP_ANONYMOUS, 0, 0) as *mut u8
    };
    let it = NonNull::new(unsafe {
        data
        as *mut slab_header
    }).unwrap();
    unsafe {
        (*it.as_ptr()).obj_size = 16;
    }
    let obj_amount = (4096 - size_of::<slab_header>()) / 16;
    let start = unsafe {it.add(1).cast::<um>()};
    let mut prev = start;
    for i in 1..obj_amount {
        unsafe {
            let new = start.byte_add(i * 16);
            (*new.as_ptr()).next = None;
            (*prev.as_ptr()).next = Some(new);
            prev = new;
        }
        
    }
    unsafe {
        (*it.as_ptr()).freelist = Some(start);
    }
    unsafe {
        if let Some(friend) = (*it.as_ptr()).freelist {
        (*it.as_ptr()).freelist = (*friend.as_ptr()).next;
        
        let bro = friend.as_ptr().cast::<u8>();
        
    }
    }

my test

[kind root]

allocator bug?

[surreal path]

i think

or maybe i have the wrong thigns to mmap

hey: 0xffffffffffffffff

oh

let data: *mut u8 = unsafe {
libc::mmap(0 as *mut c_void, 4096, libc::PROT_READ
| libc::PROT_WRITE, libc::MAP_ANONYMOUS, 0, 0) as *mut u8
};

mmap is giving me

0xfff

[cinder plinth]

that's just -1

so it failed

[surreal path]

why???

😔

WHY TF DID IT FAIL

😭

nvm i solved it

it works fine

@kind root

it failed !

[kind root]

congrats

[surreal path]

fun right

bru

im trying to attach a debugger

[surreal path]

well another step closer @kind root

we can debug this further tomrw

i have to sleep i have school

[kind root]

Good luck

[frigid cliff]

Why is this functions name just the letter h 😭

[thorn bramble]

h

[surreal path]

I ONLY PLAYED A LITTLE BIT I SWEAR

😭

i had to go sleep after

[surreal path]

chat i dont think i can code today i have an exam to study for