Nyaux

[surreal path]

but with rust

can i do the same

or not

[finite summit]

you can

it's because you're taking the address of a symbol you set in the ld script

[surreal path]

right

i need both the kernel base and kernel bytes

which i will get

okay so limine gives me the base

which is a virtual address

and a physical address?

idfk

my old code is hard to read

not hard

okay i see

to do

then i map the hhdm

maybe first i should make a map function

okay i see a magic number here

im guessing this is to get the index to the next page table or page entry or whatever

right?

im guessing yea

okay i did something similar

i need to allocate a table

[surreal path]

what the hellllll

[thorn bramble]

pointer needs to be aligned

this isn't C

[surreal path]

wha

[thorn bramble]

what?

im sorry

what do you not understand?

i can try and help

[surreal path]

im confused, what needs to be aligned

[thorn bramble]

the pointer

you are dereferencing

any code you just added

audit it

you can read unaligned memory with some functions in core::ptr

[surreal path]

i do .offset

[thorn bramble]

okay, that will be aligned

assuming original pointer is aligned

and the type is correct

[surreal path]

its literarly a raw ptr to a u64

[thorn bramble]

do you know what code causes that

[surreal path]

im unsure

[thorn bramble]

then start commenting out code

or adding prints

see where it dies

literally binary search

start putting prints everywhere

once you stop seeing new prints go back and add prints inside the functions

until you narrow it down to that one pointer dereference

[surreal path]

its the map function

for some reaosn

reason*

[kind root]

least cursed rust code

[thorn bramble]

true

start printing pointers

find out which pointer is unaligned

it really shouldn't be lol

[surreal path]

okay

[thorn bramble]

i would help if i could but im playing with friends rn so

cant really build your kernel and do it myself rn

[frigid cliff]

You wrote it wrong/hj

[thorn bramble]

he does not know how to write rust (yet)

give him a break

[frigid cliff]

I know lol

[surreal path]

how am i supposed to write it

lime im so confused

like*

[frigid cliff]

the way I would write it would be something like this (assuming it's setup similarly to this)

but in your original implementation, you did not need the (**q) because you were derferencing a reference and then dereferencing the pointer, but rust would do the first dereference for you because it's a reference and references are safe to use, so it would only be necessary to write (*q)

well I'm not 100% sure how your implementation works so leaving None in q.next's place might be wrong, but I think I have some near identical code to what I wrote above in a linked list in my OS somewhere

also the way that you write rust in Allman style

[surreal path]

Isn’t box used for allocating things on the heap

Also

This still doesn’t explain why my kernel poops itself trying to map everything

[ionic jetty]

give it a diaper

[surreal path]

bro 😭

[surreal path]

Not asking you to spoonfeed me 😭😭

I just wannna understand

what the fuck is going on

[surreal path]

who’s allman

[median goblet]

Putting open braces on a new line from the function header / statement beginning

Like

void f()
{

instead of
void f() {

[surreal path]

BEST WAY TO CODE 😎😎

FIGHT ME

😡

[median goblet]

I like it too

[surreal path]

W

what

what the fuck is going on

panics getting the memory map????

i didnt switch to the page table so what

what the fuck

commenting this out causes a triple fault

lovely

im STILL

not switching

to this page table

so like wtf

okay maybe its cause im blindly copying code from old nyaux without caring

ill just

make new code then whatever tf

so rust can stop being a meanie towards me

this better be worth it writing in rust

for all the "safety" features

do these addresses look aligned chat

they look aligned

[edgy pilot]

yes they do

[surreal path]

so what the fuck

wheres my all done over here print messages

what the fuck is going on

ummmm like

erm what the sigma

https://tenor.com/view/yes-thank-you-sweet-dreams-happy-birthday-woman-lachen-gif-11074463367286076865

commenting out the map function

fixes the issue sooo

most be the map function right?

no???

[edgy pilot]

uuuuuuumm
I don't know rust but from the error message maybe try using usize instead of isize

[surreal path]

offset does not like usize

.offset

only takes in

isize for some fucking reason

help any big rust devs

what the fuck is going on

im dying

[edgy pilot]

do you have the code on github

[surreal path]

i will publish right now

[surreal path]

here https://github.com/rayanmargham/NyauxKT

[edgy pilot]

unfortunately I won't be available at least for a few hours

so I can't take a look yet

[surreal path]

okay thats perfectly fine

glad your even gonna take a look for me 😭

my eyes alone cannot figure out the root cause of this issue, everything seems to be aligned

my guess is probs the rust compiler doing something i dont know thats weird

[molten grotto]

I'd bet you are trying to create a slice thats larger than isize::MAX

[surreal path]

im not even

making a slice

😭

and how am i supposed to fix that

.offset only takes isize

[molten grotto]

there is .add but it still can't be larger than isize max

Ill take a look at the code

[surreal path]

okay

[molten grotto]

its inside limine rs when it tries to create the memory map entry slice but idk why is the count so large and I am not home rn so I can't really do any further debugging

[surreal path]

what the fuckkk

so not my fault

[molten grotto]

eh it most likely is because its your kernel that contains the request where it gets the count from

[surreal path]

what?

[molten grotto]

the count it uses is from the mmap response limine gives and I don't think its an issue with limine rs (and certainly not limine itself)

anyway you can take a look at the request and the response in gdb and see if they look sensible (well most likely the response doesn't because of the garbage count)

[surreal path]

but its a static variable?

im not

modifying it

at all

like look at my pmm.rs code all i do is setup the pmm freelist with the memory map

its not even mutable

and it was only used once here

[molten grotto]

limine modifies it to include the response ptr

[surreal path]

okay but i still dont get why theres gonna an issue at all

[molten grotto]

I am not sure either

[surreal path]

will i open an issue on the limine-rs repo

I think i will

[molten grotto]

I still don't think its an issue with it

I can try to take a look when I get home in like 20min

[surreal path]

okay

also why is it any kernel i write is cursed

😭

[molten grotto]

lol

[green elbow]

every kernel is cursed

[surreal path]

based

[green elbow]

just wait until you get race conditions or stack corruption

thats where the real fun starts

[surreal path]

very fun sounds like

Totally did not experience stack corruption before when handling interrupts

[green elbow]

good shit

[surreal path]

with old nyaux

[haughty kite]

but rust!

[green elbow]

character development

[surreal path]

yes

[haughty kite]

its safe!

[surreal path]

now it’s rust !! It’s safe !!!

[haughty kite]

no race conditions are possible!!!

[surreal path]

indeed !!!!!

[haughty kite]

well i mean except the ones that are

lol

[green elbow]

unsafe has entered the chat

[haughty kite]

yeah

unsafe impl<T> Send for T {}

[green elbow]

rusts memory model is not ready for page tables being mangled

[surreal path]

I’m only doing it in rust for the build system tbh

crates are pretty nice to have

[green elbow]

cringe

meson is nice to have

[haughty kite]

lmfao

meson

nice

[surreal path]

rahhhh

[green elbow]

meson my beloved

[haughty kite]

this is like why nobody likes c

and c++

[green elbow]

this is why nobody likes you

yeah

that's right

[haughty kite]

meson is bad

its just bad

[green elbow]

no u

[surreal path]

I like C and c++ kinda but honestly build system is a mess

[haughty kite]

no functions?

[green elbow]

no fuck functions

leads to complexity

meson is the perfect mix between functional and declarative

why tf should you have functions

[surreal path]

one thing I like about rust is being able to have inheritance like c++ but not like c++

[green elbow]

lmao

I mean it's not inheritance

[surreal path]

it’s new to me cause I come from pure C so

[molten grotto]

traits are nice yeah

[surreal path]

yea

[haughty kite]

to... encapsulate blocks of logic???

idk man theres a reason literally all other build systems have those

and meson is just

not good for anything except what the authors decided you want

which sucks

[green elbow]

point being that you shouldn't have so much logic in your build system you need to encapsulate it

wtf are you doing that it required that much logic

[surreal path]

anyways you guys should move this to #programming this is kinda hogging nyaux chat lol

[surreal path]

you home yet?

[haughty kite]

idk like building two binaries?

with a bunch of settings that are shared

thats like literally enough to want to have functions

or how about idk like

symbol postprocessing?

[green elbow]

that's all declarative

and can already be done easily in meson no need for functions for that

[haughty kite]

unless your logic isnt supported

[green elbow]

that's fair ig

[haughty kite]

in which case you have to do ugly hacks

also

compiling bytecode

or just partial compiling in general

e.g. compile half the codebase -> do some analysis -> compile the rest

v8 does that with torque

they first build the torque compiler, compile all the torque code, build turbofan and torque generated code, run it, embed the results into the final binary

[green elbow]

how is your logic not supported in this case it's literally just compiling something

[haughty kite]

what if you need to use a compile flag meson doesnt let you use?

(because it forces a contradictory one)

[green elbow]

mlibc literally includes compiler search paths as part of its build system how much more do you need

[haughty kite]

etcetcetc

[green elbow]

??? you can add arbitrary flags

[haughty kite]

but not remove them

and meson can and does pass some automatically with no way to disable it

other than MITM on the compiler invocation

like idk i would want a build system that like

lets me build how I want to build

not how the build system wants me to

[molten grotto]

yes Ill take a look now

[surreal path]

okay great!

[unkempt relic]

try muon, they seem less opinionated

they even provide you with functions

[haughty kite]

i'm now using rust

so

i might try zig's build system, it's apparantly pretty capable?

[surreal path]

Did u spot anything

[molten grotto]

yes you are somehow zeroing the memory where the response is

[surreal path]

wtf

????

[molten grotto]

and there was at least one bug in the pmm but that didn't fix it, you were doing for e in 0..=page_amount when it should be exclusive (or else you are going to be writing one page past the free region)

[surreal path]

Oh right

but still wtf

[molten grotto]

doc for write_bytes:

Invokes memset on the specified pointer, setting count * size_of::<T>() bytes of memory starting at self to val.
and you are doing that on an u64, I bet this is the bug that's causing it to get zeroed

[surreal path]

Right so I should treat it as a pointer

when I write bytes

[molten grotto]

you need to divide the count by 8 because you are using an u64

[surreal path]

oh okay

4096 / 8

thank you for helping me with this qwinci as usual your very smart

[molten grotto]

also I'd make the pmm return usize or a wrapper struct instead of a pointer that you can't even access without first adding the hhdm offset

and then there is likely a lot of ub because you are dereferencing random memory addresses as a pointer without first writing anything to there (there is a .write method on pointers that can be used to create a new object in there)

[surreal path]

i agree my fault

okay

i did that for every pointer i use

memsetting it before ANYTHING

at all

anyways i need to map the HHDM

how do i do that again

i forgot

not epic smart 1000iq

okay

so i need to go over the memory map

okay easy

pub fn new_inital()
    {
        let mut q = PageMap {
            head: None,
            rootpagetable: unsafe {PMM.alloc().unwrap()} as *mut u64,
        };
        unsafe {q.rootpagetable.write_bytes(0, 4096 / 8)};
        let size_pages = unsafe {&THE_REAL as *const _ as usize} / 4096;
ADDR.get_response().unwrap().physical_base(), ADDR.get_response().unwrap().virtual_base());
        for i in 0..=size_pages
        {
            // println!("mapping virt {:#x} to phys {:#x}", ADDR.get_response().unwrap().virtual_base() + (i * 4096) as u64, ADDR.get_response().unwrap().physical_base() + (i * 4096) as u64);
            q.map(
                ADDR.get_response().unwrap().virtual_base() + (i * 4096) as u64, 
                ADDR.get_response().unwrap().physical_base() + (i * 4096) as u64, 
                VMMFlags::KTPRESENT.bits() | VMMFlags::KTWRITEALLOWED.bits()).unwrap();
            
        }
        println!("mapped the kernel i supposeeee");
        println!("um");
        let entries = MEMMAP.get_response().unwrap().entries();
        let mut hhdm_pages = 0;
        
        for i in entries.iter()
        {
            match i.entry_type
            {
                EntryType::ACPI_NVS | EntryType::ACPI_RECLAIMABLE
                | EntryType::USABLE | EntryType::BOOTLOADER_RECLAIMABLE
                | EntryType::FRAMEBUFFER | EntryType::KERNEL_AND_MODULES =>
                {
                    let page_amount = align_up(i.length as usize, 4096) / 4096;
                    for e in 0..page_amount
                    {
                        q.map(
                            HDDM_OFFSET.get_response().unwrap().offset() + (e * 4096),
                            (e * 4096),
                            VMMFlags::KTPRESENT.bits() | VMMFlags::KTWRITEALLOWED.bits()
                        ).unwrap()
                    }
                }
            }
        }
    }

okay i think thats everything

switching to page map causing fault 🚟

😔

not epic

i mapped everything in the hhdm

it should

work :c

[molten grotto]

did you map the framebuffer (starting at the base in the fb response)

[surreal path]

entry type of framebuffer should map it?

[molten grotto]

its not in the hhdm tho

the address in the fb response

[surreal path]

oh what address do i map that to

[molten grotto]

or actually it is my bad

[surreal path]

q.map(
                            HDDM_OFFSET.get_response().unwrap().offset() + (e * 4096) as u64,
                            i.base + (e * 4096) as u64,
                            VMMFlags::KTPRESENT.bits() | VMMFlags::KTWRITEALLOWED.bits()
                        ).unwrap()

also fixed the function to where it uses i.base

still

faults sadly

[molten grotto]

whats the cr2

[surreal path]

sec

cr3 is

CR2=000000007ff57000

[molten grotto]

looks like you are trying to use something with its physical address

[surreal path]

i only call this function

which puts the address into cr3

of the root page table

[molten grotto]

and what do you do after that

[surreal path]

i know cr3 expects a PHYSICAL address

allocate something

just messing with the PMM

and using flanterm

i cannot debug the rip cause rust mangles symbols

i wish to disable that but i dont think i can globally disable that

RIP=ffffffff800108ad

rip is somewhere sane tho

[molten grotto]

at least objdump can demangle them with --demangle

[surreal path]

here

its happening in memset

[molten grotto]

with gdb you could get the backtrace

[surreal path]

most be happening in flanterm i suspecttttt

okay sec

here

it happens when something writes something with memset

i cannot debug it as its one function

thats called MANY times

okay

it faults

i cannot

FIND THe damn

bug

😭

doesnt matter if i switch to the pagemap or not

[wicked loom]

is your memset C standard compliant?

[surreal path]

i dont know im using rust here

ask rust themselves

[wicked loom]

i am not sure rust provides that in a freestanding program?

[surreal path]

it does

[molten grotto]

it does

[wicked loom]

damn

that's weird

[surreal path]

it provides a lot of stuff for freestanding programs

which is nice to haves

[molten grotto]

also its not a problem with the memset, its a problem of using a physical address to access memory

[surreal path]

yep

happening somewhere in the loop

for i in entries.iter()
        {
            match i.entry_type
            {
                EntryType::ACPI_NVS | EntryType::ACPI_RECLAIMABLE
                | EntryType::USABLE | EntryType::BOOTLOADER_RECLAIMABLE
                | EntryType::FRAMEBUFFER | EntryType::KERNEL_AND_MODULES =>
                {
                    let page_amount = super::phys::align_up(i.length as usize, 4096) / 4096;
                    for e in 0..page_amount
                    {
                        q.map(
                            HDDM_OFFSET.get_response().unwrap().offset() + (e * 4096) as u64,
                            i.base + (e * 4096) as u64,
                            VMMFlags::KTPRESENT.bits() | VMMFlags::KTWRITEALLOWED.bits()
                        ).unwrap()
                    }
                }
                _ => {

                }
            }
        }

somewhere here

WAIT

[molten grotto]

just do -no-reboot -no-shutdown and then run it in gdb and you get the exact point where it faults

[surreal path]

changed it to this now

still doesnt solve it tho

😔

[molten grotto]

well that was the other bug

[surreal path]

yep

Still tho

[cinder plinth]

rust sucks!! It doesn't keep track of Page tables for you!!

[surreal path]

lol

[molten grotto]

literally just take a look at the fucking gdb backtrace 💀

[surreal path]

OH MY GOD

IM THE BIGGEST

FUCKING IDIOT

I’m not home but I need to test the code

is there a way I can

Work on my os on my phone

[molten grotto]

I mean termux would probably work but it might suck

[surreal path]

I don’t have termux cause this is an iphone

not an android sadly

[wicked loom]

running a jinx build in termux

phone catches on fire

[surreal path]

I wish I had something like termux but I have an iphone 😔

[surreal path]

Oh yeah I should setup jinx with the nyaux rewrite

Soon enough

[molten grotto]

I mean I have once compiled llvm on my phone and it didn't get that hot

[wicked loom]

yeah tbh phones nowadays are more than capable

[molten grotto]

and it only took a few hours

[wicked loom]

fuckers are literally almost more powerful than most PCs

[molten grotto]

also have more ram than eg. chromebooks or whatever

[surreal path]

can I at least edit my code and GitHub push it

Or something

Is there an app for that

[wicked loom]

github app :^)

[surreal path]

My family forcing me to go with them to the water park I just wanna osdev

oh yeah sec

downloading GitHub now

guys I think we will have user mode within week trust

10000%

world record speed run

If I don’t fuck it up with my depression

[wicked loom]

bruh just enjoy the water park

you can always osdev when you're back home

[surreal path]

no :^)

Okay I fixed the code

Qwinci can you test it if you can I’m not home

[frigid cliff]

But it’s against the rust style guide

[surreal path]

NYAUX HAS VMM WORKING

!!!!

LESS GOOo

[finite summit]

LESGOoooooooooooooooooooooooooo

[surreal path]

LESGOOOOOOOOOOOOOOOOOOOOOOOOOOOOO

[finite summit]

WWWWWWWW

[surreal path]

WELL MORE OF PAGE TABLE MANAGER, I STILL NEED TO READD THE WHOLE REGION SHIT BUT IMA DO THAT RIGHT NOW AND THEN WE DO WHATEVER IS NEXT UHHHHHH UACPI INTERGRATION MAYBE @kind root ?

[finite summit]

scheduler

[surreal path]

to which we need the lapic timer

to which we need to configure the apic

to which we need

uacpi

so uacpi is probs after

just lemme readd all my vmm region shi

[finite summit]

you can get (and should) get the LAPIC address from the MSR

and you can still parse ACPI tables yourself

to this day obos hasn't used uACPI to get any tables

because the kernel only uses the MADT

and that code was written long before uacpi was intergrated

and it works so I couldn't be bothered with changing it

[surreal path]

wait wait wait

before anything

i dont have fucking slab allocation

shit

fuck shit pooper

okay lets do 2 things one at a time, slab allocation then the vmm region stuff

oh as well

i dont have an idt

💀

or serial

[finite summit]

who needs serial :^)

[surreal path]

serial is easy

[finite summit]

my kernel didn't have it until 3-4 months after I started

[surreal path]

im not ur kernel

[finite summit]

and it's only ever been used for my gdbstub

which is currently unused....

[surreal path]

im gonna use the same implementation of <4096 byte allocations as i have in my own kernel

as i am lazy

and we need to get to userspace FAST

[finite summit]

dam you do be speedrunning this

[surreal path]

fr fr any% speedrun

[kind root]

Yes

[ionic jetty]

best thread up to date

[surreal path]

guess whos getting there 😎

[finite summit]

nyaux will have userspace in it's first week while obos (current rewrite) is in it's 4th month and still has no userspace

[kind root]

at least its stable now ig

[surreal path]

nyaux will be far from stable most likely

[finite summit]

ig

[elder shoal]

Astral is pretty stable (until you increase the cpu count)

The cpu gets a new friend and becomes distracted and doesnt do stuff properly :(

[finite summit]

OBOS is stable with 4 cpus

any amount above is untested, but will get slower

even if the core amount is native

(I have a bunch of giant spinlocks because I only implemented mutexes 2-3 weeks ago)

[ionic jetty]

tlb shootdown issues?

[finite summit]

nope, locks

probably also tlb shootdown

[ionic jetty]

do you know where the most contentation is

[finite summit]

no idea

all I know is that there are ~82 spinlocks (based on find all) which are usually acquired at IRQL_DISPATCH

so a bunch of stuff in the kernel can't get preempted as much

causing other threads to get less execution time

[ionic jetty]

the problem isn't the AMOUNT of spinlocks but how much they're contended

[finite summit]

fair fair

idk, as I've never mesured

that

[ionic jetty]

make some cool statistics thing

when which lock is held for how long

[finite summit]

TODO:

[ionic jetty]

the ever growing todo list

[finite summit]

56 todos in the source

[ionic jetty]

amateur numbers

[surreal path]

me coding the slab allocator while yall talk

😎

[ionic jetty]

my life is 90% unfinished todos

[ionic jetty]

noice

[finite summit]

if (Kdbg_Paused)
{
    asm("sti");
    // TODO: Fix 
    Arch_LAPICSendEOI();
    Kdbg_CallDebugExceptionHandler(frame, false);
    asm("cli");
    return;
}```

FIX WHAT

[ionic jetty]

my guess would be the eoi

you once asked about it

when to send it

you probably meant this

[kind root]

fix, NOW 😡

[finite summit]

some of those todos are related to userspace

(indirectly)

// TODO: Use a function that takes in a context.
OBOSS_GetPagePhysicalAddress((void*)curr->addr, &phys);```

for example that

userspace would be running in different "contexts"

and rn that function only supports getting the one for the current context

[surreal path]

its done

the slab allocator is done

time for vmm region shit

closer

[kind root]

is that your heap allocator?

[surreal path]

uhhh for pages bigger then 4096

yea

it maps a group of virtual addresses

to be contingious

and maps the apporite pages

and yea

once i get this done i will have working heap

[kind root]

yeah ucapi will destroy it

[surreal path]

not again 😭

nahh but i wouldnt be too worried

its the same implmentation from old nyaux

just written in rust this time

[kind root]

ah

[surreal path]

im doing that to save time cause im not fucking around with heap shangains for 12 hours again 😭 🙏

[kind root]

someone should make a uHeap (not me because i suck at algorithm stuff)

[surreal path]

true

[finite summit]

not me

[surreal path]

NOT ME that is NOT fun

[finite summit]

also isn't that just liballoc

[kind root]

idk if liballoc is necessarily good

[finite summit]

(mainly because I suck at memory safety)

[surreal path]

literarly me

[finite summit]

don't really think so

we need uAHCI

[kind root]

maybe after uGPU

wihhc is gonna take at least a few years

[finite summit]

I need to speedrun obos mesa port

[thorn bramble]

ahci is so simple i dont think you need a library for that

[finite summit]

so I can be the first one with uGPU

[kind root]

lol

[finite summit]

just like I was the first with uACPI

[thorn bramble]

managarm was

:)

[finite summit]

nay

[thorn bramble]

no?

[finite summit]

it was my third kernel

[kind root]

oberrow integrated uacpi even before it had opregions implemented

[finite summit]

before infy joined the server

[thorn bramble]

damn lol

[kind root]

like before it was supposed to be integrated

[finite summit]

#1141057599584878645 message

[surreal path]

if yall wanna see nyaux development

im streaming in vc0

[kind root]

he was also the first to even run it in kernel space

i didnt do it until way later

[thorn bramble]

that is both cool

and worrying at the same time

[kind root]

lol

[finite summit]

I'm pretty sure it was because I thought, "I need acpi, and see this random guy on some discord server is implementing an interpreter that seems complete, might as well use it"

[kind root]

i barely even had helpers for anything at that point

it was like barebones interpreter

[thorn bramble]

would you say lil was better at that time?

[finite summit]

lemme find some old code

https://github.com/OBOS-dev/obos-old/blob/old_code3/src/drivers/generic/acpi/main.cpp#L81

[elder shoal]

frigg

[finite summit]

https://github.com/OBOS-dev/obos-old/blob/d82cd2bd3e488f48e7c174040611a5dedb3ff93a/src/oboskrnl/arch/x86_64/syscall/power_management.cpp#L60

https://github.com/OBOS-dev/obos-old/blob/d82cd2bd3e488f48e7c174040611a5dedb3ff93a/src/oboskrnl/driverInterface/x86_64/scan.cpp#L187

my third kernel at one point entirely used uacpi private api

mainly because there was no public api

[kind root]

Lol

[kind root]

Yup

[median goblet]

In my 1 year of developing boron I haven't implemented a serial driver

Just wasn't needed

[finite summit]

yeah

[edgy pilot]

noice!

[elder shoal]

I just throw everything I need at 0xe9

If I end up writing a serial driver itll be to hook it up to the tty stuff

[surreal path]

i am close just needing to figure out how to properly do the inline asm cause i suck at rust inline asm

any rust experts know how to get this to work

i fucking suck at inline asm

[finite summit]

basically what you do is:
ditch inline asm
make a new asm file
throw your asm in there

[surreal path]

vmm region works now

region allocation works

[kind root]

You speedrunning this shit again lmao

[surreal path]

yes lmao

i need an idt once i have an idt i will put in lovely uacpi

we all love uacpi

dont we

[edgy pilot]

uacpi is love, uacpi is life

[surreal path]

real.

[edgy pilot]

uacpi and limine

[surreal path]

realreal.

mintsuki hear our prayers

[magic kernel]

Don’t forget mlibc

[surreal path]

of course.

[edgy pilot]

yes

[magic kernel]

[surreal path]

mlibc our beloved.

[magic kernel]

Honestly the best side project Managarm made

[edgy pilot]

no operating system is complete without those three

[kind root]

Yes

[finite summit]

I use hyper

instead of limine

[surreal path]

i also just squashed a mosquito that had my blood in it

[finite summit]

you gave birth to a mosquito then killed it?

[kind root]

Drink it back then

[magic kernel]

Snort it

[edgy pilot]

you shall be executed for treasonous acts

[kind root]

Lol

Yeah thats a bannable offence

[magic kernel]

Rule 0: one shall use Limine, uACPI and mlibc

[finite summit]

[kind root]

Anyone found with a hyper.cfg inside their image will be prosecuted

[edgy pilot]

based

[magic kernel]

Yes

[surreal path]

oh shit

my laptop is about to run out of charge

NOOOOO BUT I WANTED TO WORK ON THE IDT AS WELL TODAY

FUCK

[kind root]

u dont have electricity?

[surreal path]

barely in the shitty hotel I’m at

[surreal path]

alright so as i remember with the IDT i need isrs

problem

i dont know how im gonna implment the same array of function pointers i did in nyaux in rust

im kinda screwed

before anything tho

lets get my allocator working with rust

[ionic jetty]

afaik these don't pair well so good luck

[surreal path]

i know

[molten grotto]

the same way as in c?

[surreal path]

u can do the same way in c

in rust?

[molten grotto]

well idk what exactly do you mean by that

but you can have an array of function pointers

[surreal path]

how does that look like

[molten grotto]

struct Whatever {
    arr: [extern "C" fn(a: i32, b: i32); 10]
}

[haughty kite]

also you gotta #[repr(C)] that struct

possibly

[surreal path]

do i have to do that with the interrupt frame struct as well?

for it to work well with assmebly

[molten grotto]

yes

[haughty kite]

yess

compile with RUSTFLAGS=-Zrandomize-layout

[surreal path]

[haughty kite]

it doesnt cause any ub

so

its always correct to use it

[molten grotto]

well it depends on how you use it, if its not used in any asm then it doesn't matter (for an example how I do it in my kernel is that there is only one central irq handling function that the asm calls with the frame + number that then dispatches it further)

[surreal path]

that would be smarter

[molten grotto]

the reason why I did it like that is mostly because irq handler fns need to store an arbitrary amount of state (well at max like 16 bytes iirc) so they can be lambdas

[surreal path]

to do

fuck you rust

[finite summit]

common rust L probably

[haughty kite]

.as_ref()

also

jesus christ

have you heard of rustfmt

[surreal path]

whats rustfmt

[haughty kite]

its like the code formatter for rust

everyone uses it and its part of the toolchain etc

please for the love of god run it

also imagine static mut lol

[surreal path]

why not

[haughty kite]

and imagine using the alloc crate

[surreal path]

im not?

[haughty kite]

unsafe!!!1111111

[haughty kite]

whats GlobalAlloc from then

[surreal path]

from core?

[haughty kite]

ah

[surreal path]

implmenting it for rust?

[haughty kite]

still its L and cringe

[finite summit]

rust is cringe in general

C best language

[surreal path]

done now we have heap i think

[finite summit]

lesgo

there's no way you implement uACPI now

right?

[surreal path]

no not rn

i need an IDT

[kind root]

Meh u can skip it

Or rather

[surreal path]

nah

[kind root]

Implement later

But yeah whatever

[surreal path]

but

[haughty kite]

the rust stdlib allocation primitives are bad

for kernel stuff

[elder shoal]

The idt is one of the first things Id do

[haughty kite]

and like not regular userspace stuff in general

[surreal path]

so what am i supposed to do

[haughty kite]

write your own ones?

and also https://docs.rs/heapless is godsent

[surreal path]

i think its fine

anyways

IDT time

[finite summit]

same

[kind root]

I dont like how its common for rust that u just shove 99999 crates into your project by default

Reminds me of js

[haughty kite]

yeah true

[kind root]

But ig im just not used to that mentality

[haughty kite]

but also heapless is just very useful

and like i COULD write my own on-stack vectors

but like

why

[kind root]

Depends if you want to do that or not

[haughty kite]

true

but its really not like the point of the project

so

[kind root]

Fair

[surreal path]

is there a way

to check

if an interrupt is an exception

instead of

having a billion asm stubs

[haughty kite]

its below 32?

lmao

[finite summit]

he probably means if it has an error code

[molten grotto]

just have a macro to generate the asm stub and then an if inside the macro that checks if the number is one of the few ones that push a code

[kind root]

Bro never heard of macros

[surreal path]

i dont know how it works with rust asm

the macro system

[haughty kite]

its just gnu asm

lol

[surreal path]

i dont write gnu asm

lol

[haughty kite]

or you can use normal rust at const???

[kind root]

U dont write them in rust

[surreal path]

i know

[finite summit]

use nasm

[molten grotto]

rust inline asm is fine, you just have to use gas directives with intel syntax (and you can have it in separate files using global_asm!(include_str!("myfile.asm"))

[haughty kite]

or again

[surreal path]

i dont know gas directives

[haughty kite]

use rust macros to construct your stubs

[haughty kite]

its called

fucking

google

jesus christ

[surreal path]

okay i get it 😭

[haughty kite]

OR RUSTS FUCKING MACRO SYSTEM

OR THEIR COMPILE TIME EVALUATION SYSTEM ARGGHH

sorry

[surreal path]

chill 😭

[kind root]

Like if you're doing rust at least do shit in an idiomatic way

[haughty kite]

yeah

[ionic jetty]

yeah

[haughty kite]

or you know

build.rs

which lets you run arbitrary rust at build time

[finite summit]

proposal: ditch rust, embrace C

[kind root]

Yup

[finite summit]

you wanted memory safety, we had a tool for that

it was called not having skill issue

(and also kasan)

((kasan has saved my ass on multiple occasions))

[haughty kite]

i dont think u should use rust for memory safety

its good for other reasons

[surreal path]

yea

thats why im using it

not cause of "memory safety"

thats called skill issue

[haughty kite]

because we all know thats bullshit

[surreal path]

yea

[haughty kite]

though the rust semantics are massively opaque

[kind root]

Why are you using it?

[molten grotto]

also might be a good idea to first learn how to do that in userspace before writing a kernel :p

[haughty kite]

they are so much worse than c/c++

[kind root]

Well that's nyauxmaster

[surreal path]

generics, impl blocks, build system is nice fresh air

[kind root]

Hes known to speedrun shit

[surreal path]

thats me

ill figure out getting nasm to link anyway with the binary

[haughty kite]

theres of course a nasm crate that does that for you btw

[surreal path]

real?

[haughty kite]

yes lmfao

[surreal path]

wtf

[haughty kite]

its rust

[surreal path]

yall never told me this

[haughty kite]

theres always a crate for everything

yeah because youre using rust so i assumed you knew you could copypaste a crate for it lol

[kind root]

Is there a kernel crate

[haughty kite]

its called x86_64 not kernel but yes

[kind root]

Lmao

[surreal path]

bru 😭

[kind root]

Just import that

[haughty kite]

oh god

[surreal path]

ah yes

[haughty kite]

that crate is so skill issued

that group of rust osdev people are actually just

no

[kind root]

In a good way?

Oh

[haughty kite]

no in the worst way possible

[kind root]

Well their acpi crate fucking sucks so I expected as much

[haughty kite]

lol yes

x86_64 is very bad

[finite summit]

import kernel

[kind root]

Recursive and half of the opcodes are incomplete

[haughty kite]

its like maximally handholding

[kind root]

Almost cosmos framework

Or whatever its called

[haughty kite]

oh yeah its like cosmos

[surreal path]

lmao

[haughty kite]

basically

[kind root]

Lol

[haughty kite]

well its not that high level

but its quite close

they have come up with the stupidest abstraction ever

PhysAddr and VirtAddr

god

that shit is like cancer

its just bad

[kind root]

Lol

[molten grotto]

eh I don't think having wrappers for different address types is unreasonable

[haughty kite]

counterpoint

[molten grotto]

though ofc you could just use plain usize or whatever but I actually kinda like having a wrapper struct over it

[haughty kite]

you have a pointer type for pointers???

and a phys one is fine

but also like

userspace virt addrs are a thing too

[haughty kite]

which has other advantages

[molten grotto]

well yeah for kernel ones it doesn't have that much sense, except maybe in like paging code or whatever

for userspace ones it would be useful because then you can have all your safe load/store methods on it

[haughty kite]

yeah but then

its (a) contextually obvious what the userspace thingies are

(b) you cant add things to their types

(c) this is a thing that can easily leak to your whole kernel BUT its in an arch-specific crate!

[molten grotto]

yeah just make your own if you want that

[haughty kite]

you kinda always want that tho?

you never want THEIR one

oh and also

they obviously have a handholdy pagetable wrapper

[molten grotto]

I meant that if you want to use a wrapper type in general

[haughty kite]

so THEIR wrapper type is never useful

that kind of wrapper type COULD be useful tbf

[molten grotto]

yeah

[surreal path]

why no workie :c

i did exactly as it said

run the function in build.rs

[molten grotto]

you didn't link it

unless it automatically links the library

[haughty kite]

probably need to tell it to do the asm thing

[surreal path]

it doesnt help

[flat nymph]

What's with all the hate? (sorry for hijacking this thread) but I'm thinking of adding it to my OS (but may be it's a mistake so may be not)

[flat nymph]

You shouldn't
There are machines (I have one) where that is fake news

[kind root]

You're mixing up cpuid and msr no?

[flat nymph]

Hmm

Perhaps

I'm talking about that limine bug

(also, how did I miss linuxmaster coming back a week ago?)

[kind root]

That was lapic id

Not address

[flat nymph]

The address is fine

Ok I misread the message ignore me everyone

[finite summit]

because it's very rusty

[flat nymph]

@kind root would you not mind another OS using it?
The 32 bit support looks tempting

[finite summit]

it's quite amazing

[wicked loom]

oh for christ's sake

[flat nymph]

I'm thinking I could support two protocols

[wicked loom]

what's even up with everyone wanting 32-bit support all of a sudden like holy shit

sure i will make 32-bit LBP okay?

[finite summit]

I might port obos to x86 (protected mode) eventually

[wicked loom]

sounds painful but i can also get it to work

[flat nymph]

My friend has 2 Intel atom laptops which are 32 bits only

[finite summit]

Find all and Replace: uint64_t -> uintptr_t

[wicked loom]

idk why you couldn't just use multiboot2 on 32-bit but god knows

[flat nymph]

And he expressed interest in participating in osdev if we can make my os work there

[flat nymph]

Tbh I could

But limine loads SMP

[kind root]

Ultra does higher half 32 bit so a lot more convenient

[finite summit]

fyi the ultra protocol (used by hyper) doesn't do SMP

[kind root]

Sure

[flat nymph]

And with multiboot you have to jump through hoops

[wicked loom]

does ultra just not map any usable memory on 32-bit or how does the memory map look?

[flat nymph]

It only maps 1gb

[kind root]

It maps the lower gig

[wicked loom]

what about mmio

[kind root]

What about it

[flat nymph]

Why do you need a memory map for it?

[wicked loom]

do you not map the framebuffer?

by default

[kind root]

Nope

[finite summit]

hyper doesn't do any sort of special mapping

[kind root]

Its provided as physical_address

[finite summit]

or even any hardware init

[kind root]

Well A20

[finite summit]

that's not hardware is it

[kind root]

Sorta

[wicked loom]

it's hardware

[flat nymph]

In my kernel I don't use direct mappings of any sorts

[finite summit]

my kernel didn't map the framebuffer as WT until very recently

maybe a bit after I was done the VMM

[wicked loom]

you must mean WC

[finite summit]

yeah WC

mb

mix them up in my head often

when the kernel is mapped in hyper, you don't get any sort of prot bits set, just RWE

so if you don't remap the kernel yourself, you'd need to

[flat nymph]

So I just get physical addresses from limine responses and map stuff to wherever my vmm decides to, and with hyper I don't think I'll have to change anything much

[wicked loom]

how do you allocate the page tables your vmm uses

because there's this funny thing called recursive mapping

that is often (read always) used on IA-32 due to the constraints you're working with

[kind root]

Even if you did nothing would have to be changed

They're pretty similar protocols

At least on the base level

[flat nymph]

Temporary mappings

(after I switch to my own page tables)

[wicked loom]

do you just have those preallocated in the kernel binary?

[flat nymph]

No

[flat nymph]

I use HHDM before that

[wicked loom]

ok so that's kind of what i was getting at

it's a sort of chicken and egg problem, that's what the hhdm is mainly designed to solve

at least at the beginning

along with providing an easy way to output to the framebuffer from the beginning instead of having to bootstrap at the very least a primitive vmm to map the framebuffer

doing all that on massively constrained 4gib of address space is... well

not sure how it'd work

[flat nymph]

Oh, for vmm I have 16 list entries in the kernel binary, and I allocate more as needed

[wicked loom]

yes

that is one way of solving the chicken and egg problem

it would still mean you get no framebuffer until you can map it yourself, on 32-bit architectures at least

[flat nymph]

My framebuffer driver is in user space

[wicked loom]

also, the smp spin wait loops are in the hhdm too, so one would need to also figure out a way to handle those

[flat nymph]

So I don't get it until my kernel and few servers are initialized

[flat nymph]

Temporary mappings again

[wicked loom]

anyways read: it's kind of complicated to provide a fully featured 32-bit version of the Limine protocol

without sacrificing any feature

[kind root]

Yup, thats why I actually removed a few features when designing mine

[flat nymph]

Those are annoying with PAE

[kind root]

To make it more uniform

[wicked loom]

imho it's just not worth it

to trim features out of the Limine protocol to make a handicapped 32-bit version of it

[flat nymph]

HHDM is the issue

[wicked loom]

which is why my idea is that people would just use multiboot2 if they cared that much about 32-bit support

[kind root]

Yeah, maybe a separate limine version would be better

[kind root]

Not really because then u need an assembly stub and stuff thats annoying

Unless ur not planning to run in the top half

[wicked loom]

a tweaked multiboot2 with higher half support? :^)

[kind root]

Lol

[wicked loom]

i can do that with way less lines of code

tweaked mb2 with smp and higher half

[flat nymph]

I was solving it by booting into a (future Ring 3) server, which loaded kernel

[kind root]

Mb3 u mean

[wicked loom]

¯_(ツ)_/¯

i mean isn't that basically what ultra on 32-bit is?

what else does it provide that mb2 (with smp and higher half) would not?

[flat nymph]

What's everyone's issue with GAS?

[wicked loom]

i have no issue with gas

so it's not "everyone"

[flat nymph]

*a lot of people

[wicked loom]

my only issue with gas is when people don't call it through the compiler driver

[kind root]

Im not that good at mb2 to tell, but the way protocol works is still different , aka multiboot headers and stuff

Plus ultra has tons of extra features

To configure handoff state

[wicked loom]

like what

[kind root]

And like the mappings, whether lower half is mapped, or depth

It supports 32 bit with pae

Idk off the top of my head what mb2 does or doesn't have

But isnt limine multiboot with smp support as well then if u speak in those terms

[molten grotto]

I don't have an issue with gas itself but I do hate att syntax because its full of unnecessary annoying prefixes, non-obvious things like implicit memory accesses, cringe offset syntax and weird operand order than whats used literally everywhere else

[wicked loom]

what if i took Hyper and embeeded it into Limine

and then provided the Ultra protocol by just using Hyper

[elder shoal]

Limine subsystem for hyper

[kind root]

Lmaoo

Based ngl

[wicked loom]

i'll do it if you make Hyper build with limine-efi instead of whatever you have now and using makefiles

so the build system can be more easily integrated

it'll basically just result in embedded Hyper EFI executables

to chainload

[kind root]

I have nothing tho, it just uses the clang target or the mingw gcc

[wicked loom]

the config can be generated on the fly and shoved in some EFI variable

[kind root]

But if u tell me how I could look into it

[wicked loom]

exactly that's the main issue, with limine-efi it will also build with gcc

limine-efi is literally just headers + a small relocation magic that allows an elf to turn into a PE

you can see in https://github.com/limine-bootloader/limine-efi-c-template that it's relatively easy to get working

[kind root]

Why are custom headers needed?

[wicked loom]

they are not

you can use your own

and just use the relocation code

[kind root]

I’ll take a look, sure. I dont see why it would be a problem to add another target like that

[wicked loom]

as long as you use __attribute__((ms_abi)) for calls it's fine

for calls into UEFI i mean

[kind root]

Yeah I can just define that

[wicked loom]

if you don't have that in your EFI headers you should add it

[kind root]

I have the empty macros

So yes

Like EFI_API

Iirc

[wicked loom]

it's basically that, then build however you want, as long as you compile in the .c and .S files from limine-efi (it doesn't matter where, they just need to be in), then if you do that, and link with the linker script provided by limine-efi, you can use objcopy binary to turn the elf into a pe

[finite summit]

that would be based

and I would be able to dual boot obos and linux

[kind root]

Yeah that should be trivial to add in

[wicked loom]

i promise i will do it if you do that (or add the option to do that)

i mean i could also do it downstream but it'd be easier if it was upstreamed

[kind root]

How will you embed it?

Like whats the mechanics for that

[wicked loom]

i'll just add it as another "submodule" in ./bootstrap and then build it as part of the limine build system and embed the resulting .EFI file into the limine binary

and then chainload it

if someone uses protocol: ultra

[kind root]

Ah I see

[wicked loom]

then i guess one would need to adapt the various config options you provide into limine config options for ultra

[kind root]

Ok nice ig