EvalynOS

[hazy saddle]

But we should be all good

[devout path]

yeah yeah

[hazy saddle]

How did this not break 😭

[devout path]

i have 0 clue LMAO

do u have commands like hexdump and stuff so we can try breaking it :))

[hazy saddle]

In my kernel or my system

[devout path]

kernel

oh wait that runs on

[hazy saddle]

Sadly no

[devout path]

oh

aw

[hazy saddle]

i only have bad apple

[hazy saddle]

i should be able to make this take arguments when starting the thread right?

[devout path]

i am not entirely sure

[hazy saddle]

or i can have some external vars you fill in and then start it

and then mutex it so it can only be changed once its consuimed it

i do it with the other vars

i could have args there

and then have a var that gets written to once its safe to start another thread and change thoes args

[devout path]

yea actually

yea yea

[hazy saddle]

And if it’s not ready spin because it won’t take that long

Like fuckin maybe 500 cycles of spinning at worst

[devout path]

xD

[hazy saddle]

Can always make it a hlt spin anyways

Maybe

May not even need any locking if it’s sequential too

No no it is a thread

Ehhh whatever

[hazy saddle]

I mean what’s a better solution

I have to pass data from one thread to another

[devout path]

doin tis is better

[hazy saddle]

And a shared varrible with a spin lock is quick and dirty especially for somthing like this

[hazy saddle]

True true

But it also makes me get infra for mutexes and other things

Which is good

And it’s not the most cursed thing

[hazy saddle]

Because I have to manually craft the stack

So args might be a pita

[devout path]

pita?

[hazy saddle]

Pain in the ass

[devout path]

XD

[devout path]

yeah

[devout path]

o h

[hazy saddle]

Yeah 😭

That’s how making the thread works

and all user threads are kernel threads in a trenchcoat

[devout path]

that

is

kinda not safe

[hazy saddle]

so i need a generic this with the argument of the path to load the executable

[hazy saddle]

The kernel thread sets up userspace and then jumps to it for its context

And then it’s always running inside of user space

[devout path]

oh

[hazy saddle]

See here?

[devout path]

this confuses me a little XD

yeah

hmmmmmmmmmmmmmmmmmmm

[hazy saddle]

I can probslt strip run once with a infinite loop or having it signal to free itself

Wait is there a way to make a new instance of a static function to call?

Because I don’t think freeing would work with this if the thread it’s self don’t know which thread it is

[devout path]

hmmmmmmmmmmmmmmm

[hazy saddle]

@devout path maybe I can pass a pointer to it that has a structure of all the data it needs?

But how do I construct the stack for that

Wouldn’t that be a register argument in SYS-V too?

Which I have to put into the right thing on the stack

Because the zeros I set I think are the registers

[devout path]

hmm

im tryna find a solution myself too

[hazy saddle]

this should be the stuff popped off the stack into registers

    for (int i = 0; i < 15; i++) {
        *--stack = 0x0 + i;
    }

so i could slide a pointer into here

[devout path]

u don't need an instance btw, they share em right?

[hazy saddle]

every thread needs some seperate context right

[devout path]

yes yes

[devout path]

hmm i think you dont needa craft a stack manually...

[devout path]

yeah actually

[hazy saddle]

yes. this is what tells the CPU how to jump to userspace

it uses the crafted stack to jump to the entry point

[devout path]

like u can use kmalloc containing the args, pass the pointer there and the thread reads the data?

[devout path]

in theory should work? not as messy

[hazy saddle]

yup and then the kernel part of the thread would use that

[devout path]

yes

[hazy saddle]

this matches with my logic

the crafted stack has all of that stuff

[devout path]

oooh

[hazy saddle]

the GPRs and RFLAGS

So I should just be able to slide in that sneaky pointer to pass in the function

[sudden mirage]

is using xchg possible for context switches?

[devout path]

yeah and wat if you craft the initial stack of your new thread so it looks like it paused inside da thread switch

[hazy saddle]

Just put in in RDI I assume

[hazy saddle]

What

[devout path]

would u like me to explain it in tech jargon or in simple terms

[sudden mirage]

can i use xchg?

[hazy saddle]

The stack is useful here no?

[hazy saddle]

I mean what pausing?

[sudden mirage]

yea its pretty useful after i read your code, the only risk is bad code which causes a program to use a ton of stack which means if the memory below the stack is not mapped you get a #PF when switching threads

[devout path]

wat reference are u using?

[hazy saddle]

Wikipedia

[sudden mirage]

so ig its better to map memory below the stack, maybe

the pros of using the stack method is that i only need to store rsp in a program's PCB

[devout path]

so uuh u wanna switch thread a to thread b right? u can pause the stack by doin mov [rdi], rsp

[hazy saddle]

Guard pages are pretty important too so that can be used in combination

[devout path]

as you've mentione here

[hazy saddle]

I mean when crafting the stack

Where RDI should be in the push and pops

I put in that pointer instead of a 0

[hazy saddle]

So you would have memory that the kernel can use but the user program would PF right below the stack if it were to overflow

And then you can detect that

[devout path]

huh im confused a little ;-;

i don't have much context

wait gimme a moment to read

[hazy saddle]

I’m confused too?

[hazy saddle]

These represent registers

I set them to 0

So when I execute ret it fills the registers properly right

If I make the one that equals RDI to the pointer

According to the sysV abi

That’s the first argument

Which means when it starts up and that gets pushed into the stack it should be given to the kernel thread

Let me do a demo

i will make them all 1s

and it wont crash

[devout path]

im actually braindead rn i completely forgot that you made that

my bad

[hazy saddle]

but i can add some logging

LMAO

[devout path]

gimme a moment to reread this LMAO

[devout path]

yep

this is exactly what i was thinking

man i swear

[hazy saddle]

what the fuck

[devout path]

aight my badlmao

[devout path]

uuuuuuuuh

[hazy saddle]

huh

[devout path]

uuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuh

[hazy saddle]

with a zero

its consistent

[devout path]

wdym?

[hazy saddle]

its always off by 5

time for advanced investigation

because im fuckin lost

[devout path]

i

am

HUH

WAT

this aint the thing that im gonna investigate at 2:40 am

[hazy saddle]

every register should be becoming that value

[devout path]

LMAO

[hazy saddle]

leme do a panic for a register dump

[devout path]

wait what value exactly?

like the value itself when you set it?

[hazy saddle]

the value i set on the stack

[devout path]

oh

[hazy saddle]

my reg dump may not be accurate

[devout path]

heh

[hazy saddle]

later

because i aint either rn

i have a headache

[devout path]

heh

err it's working no?

don't question stuff when it's working :))

[hazy saddle]

no

its not working'

the argument isnt passing right

[devout path]

where is rdi again?

wait lemme backread

[hazy saddle]

im setting all regs

all the way

to be safe

[devout path]

xd

waht if

hear me out

what if we don't use the loop XD

[hazy saddle]

wait

no

im so stupid

oh my fucking god

[devout path]

wat

[hazy saddle]

do you see it

[devout path]

was it about hte rdi

[hazy saddle]

PLUS I

😭

[devout path]

B

R

U

H

LMAOA

[hazy saddle]

that was a reminant of debugging ages ago

[devout path]

heh xd

[hazy saddle]

i do need to fix my reg dumps tho

here we go

[hazy saddle]

I stg these kinds of mistake are so common in programming 😭

@devout path it diddnt send this here

But we are all good

[devout path]

noice

[hazy saddle]

Now I can malloc a structure and get it passed in there

[devout path]

yeaaaaaaaaaah

[hazy saddle]

And then it can operate on that to know how to do things

Like telling the scheduler to end its self

[devout path]

heh

[hazy saddle]

Or even then pass this on to userspace for a sub set of things

But the system call handler can probslt check the current thread

So that won’t need this

[devout path]

yeaah

[hazy saddle]

https://tenor.com/view/computer-break-computer-funny-rage-destroy-gif-25478291

Programming be like

Over three fucking charaters XD

[devout path]

LMAO

ngl

for some weird reason

i love programming on crt monitors more than my oled screens

[hazy saddle]

Apparently text is ass on OLEDs?

That’s what IPS can do better

[devout path]

oh damn

[hazy saddle]

@devout path

[devout path]

WAT DA HEEELL

[hazy saddle]

I’d maybe pickup a decent 75hz bussness IPS display for coding?

[devout path]

i THOUGHT I NEED GLASSES BUT THAT'S JUST BAD LMAO

[hazy saddle]

And keep the oled for gaming

[devout path]

yeaah

[hazy saddle]

It’s what like 100 200 bucks for a decent one?

[devout path]

yup

[hazy saddle]

Cheaper than glasses

[devout path]

actually glasses here are cheaper LMAO

[hazy saddle]

LMAO

[devout path]

u can get a decent built glasses for 40 dollars

in usd

[hazy saddle]

Dang

But yeah plus it can’t hurt to have an extra monitor

I have three

[devout path]

imma just add my old crt monitor on my desk atp

[hazy saddle]

I really wana get a CRT to play deltarune

The game is a pixel art RPG and it’s 4:3

[devout path]

u can get one for idk 20 dollars secondhand online XD

[hazy saddle]

And I’ve seen videos of it played on CRT and holy shit it’s good

[devout path]

even for free if ppl are decluttering

yeaah

[hazy saddle]

Yeh yeh

[hazy saddle]

Yet my friends would still clown on me for emulation and not using a regular controller

I want that steam controller

Looks so nice for pc gaming

[devout path]

the new one with the pads?

i have no idea how that's comfy but eh

[hazy saddle]

its like a steamdeck without the screen

[devout path]

ooh

fair enough

[hazy saddle]

@devout path

working on the second user app

a mandlebrot

[hazy saddle]

@devout path it seems to crash when running both at once

its triggering a page fault

if they run alone thry work though

[hazy saddle]

If I run mandelbrot first

And then bad apple

Bad Apple won’t execute

But mandelbrot restarts

[hazy saddle]

My kernel may be dead

I dunno how to fix it

[hazy saddle]

Well I managed to fix some other things

They both now “properly” fault when executing them because a bad variable name

And I got AVX and other things working to accelerate the code

For the userspace apps

[devout path]

oops

i feel

asleep

[hazy saddle]

It’s fine lol

[sudden mirage]

i feel like you need to add extra space between two lines because its all squished

[sudden mirage]

bro cast to (uint8_t)

[hazy saddle]

No I mean it’s not properly dumping registers

It’s inaccurate

[hazy saddle]

Why

[sudden mirage]

i was talking in general ngl its pretty hard to read if all text is white and squished together

[hazy saddle]

My registers are color coded and I can read it fine and it also dumps into the serial port on QEMU

[sudden mirage]

you are writing one byte to the stack and C treats 0x1 without a cast as a normal int unless you end it with ULL or UL or something like this

[sudden mirage]

how many pixels between row 0 and row 1?

[hazy saddle]

That’s intended

I am meant to be pushing 8 byte values

They are the registers

[sudden mirage]

cast to uint64_t or end with ULL

[hazy saddle]

Part of building the stack

[sudden mirage]

0xFULL means (uint64_t)0xF

[hazy saddle]

That + i was the invalid thing too btw

It’s meant to set them to zero

Let me read this again

The stack is cast to a uint64 already

So when I operate on it it should be 8 byte values no

the 0x1 is a debug value to set registers too

like the + i was

but if this wasnt working all threads would be fucked

not just user threads

and one user thread works

[hazy saddle]

its an 8x8 font used in flanterm

its one px it seems

[sudden mirage]

Why not use the default one?

[sudden mirage]

Too smol

[hazy saddle]

more text on screen

[hazy saddle]

not for me

[sudden mirage]

Do you take the screenshots with QEMU Full-screen?

[hazy saddle]

an all the stuff comes on seriel

[hazy saddle]

what

[sudden mirage]

Ctrl+Alt+F then take a scrnshot?

or do you take it with qemu windowed?

[hazy saddle]

i use my got ket win + s to take it while windowed

i know how to make it full screen but that dosnt matter?

[sudden mirage]

usually qemu stretches screen when you fullscreen, idk why it doesnt update the resolution but it is what it is

[hazy saddle]

because the guest os dosnt get notified

[sudden mirage]

qemu resizes?

[hazy saddle]

yes the qemu window gets a new resolution

but the guest os dosnt

it would have to request the new video mode

or have some kind of guest additions things

[sudden mirage]

ig its usually handled with a gpu driver

[hazy saddle]

even with a GPU driver it might not know etc

[sudden mirage]

so its usually polled

[hazy saddle]

uhuh. if i were to open a linux vm it probaly wouldnt change unless i change it in video settings

maybe it was changed

[sudden mirage]

makes sense you cant stretch a monitor irl

[hazy saddle]

Look are you here to talk pedantics about font sizes and qemu screenshotting and giving bad advice about stuff when you don’t have the full code context or are you here to talk about this project in a meaningful way to have it actually progress?

[sudden mirage]

saying that the font is kinda unreadable if i were to use the kernel

[hazy saddle]

Because I don’t think this convo is going anywhere

[hazy saddle]

Then? Don’t use the kernel?

The font works for me

And you can easily change it in the source code if you wanted to work on it

[sudden mirage]

it was just a suggestion

[hazy saddle]

I can even make it a compile option for the font

-DBIGGER_FONT

I think there are bigger issues afoot than the font?

The font that’s working

There is the case of the userspace failing spectacularly

[devout path]

Chill xd

what did i just come back to

[hazy saddle]

I don’t know XD

[hazy saddle]

Anyways @devout path how’s it going

I just wrapped up one of the parts of another project while you talked here

[devout path]

ooh im doin alright

finals exam week so haven't been doin anythin

[hazy saddle]

Ooo

[devout path]

noice

[hazy saddle]

Honestly my os might go on hold until I can fix this

But I am working on some cool thing for my internal services and might be good for others

Making some proofs of concept for the major 3 bits I need to mix

My threadpool works nice

And it can generate this fractal using all my cores in 12 seconds

[devout path]

ooh noice

[hazy saddle]

i had more code after this patch i just threw out

but i found the problem

my stack pointer was being set to zero

i think i am setting up user threads poorly

i should make a better bootstrap function for them in the firstplace

that sets up a proper stack etc for userspace

@devout path i think i got this

i dont think anything is wrong with my switching logic

global thread_switch
global thread_switch_user
global switch_to_user

section .text
thread_switch:
    pushfq
    push rax
    push rbx
    push rcx
    push rdx
    push rsi
    push rdi
    push rbp
    push r8
    push r9
    push r10
    push r11
    push r12
    push r13
    push r14
    push r15

    mov [rdi], rsp
    mov rsp, rsi

    pop r15
    pop r14
    pop r13
    pop r12
    pop r11
    pop r10
    pop r9
    pop r8
    pop rbp
    pop rdi
    pop rsi
    pop rdx
    pop rcx
    pop rbx
    pop rax
    popfq

    ret

USER_STACK_TOP equ 0x0000000080000000

switch_to_user:
    mov ax, 0x33
    mov ds, ax
    mov es, ax
    mov fs, ax
    mov gs, ax

    push 0x33
    mov rax, USER_STACK_TOP
    push rax
    pushfq
    pop rax
    or rax, 0x200
    push rax
    push 0x2B
    push 0x4000
    iretq

[viral bison]

why are you pushing all those regs?

[hazy saddle]

It’s for preemption right

[viral bison]

this is mine

you only need like 6 regs

[hazy saddle]

Hmm

I may be dumb

[viral bison]

the other regs are discarded

[viral bison]

hmm, why push 0x4000?

and why 0x33?

[hazy saddle]

I have a fixed load address in userspace for programs start function and a fixed stack address top

[viral bison]

0x33 seems so strange there

[hazy saddle]

That’s the RFLAGS thing needed for ring3 right?

[viral bison]

show your gdt

[viral bison]

my gdt is:
null
kcode
kdata
ucode32
udata
ucode64

[hazy saddle]

I don’t have any kind of 32 bit at all

[viral bison]

you have to have that segment anyway

for syscalls

[hazy saddle]

But I have null kcode kdata tss ucode udata

[viral bison]

oh

tss in the middle

[hazy saddle]

Is that bad

[viral bison]

well, you have to swap ucode and udata

[viral bison]

shouldn't be

[rancid viper]

why

[hazy saddle]

Why does ordering matter there?

[rancid viper]

it doesnt

[viral bison]

syscall msrs

[rancid viper]

they dont use syscall

[viral bison]

it does

well, in the future

[rancid viper]

yes but its irrelevant here

[hazy saddle]

https://github.com/EvalynGoemer/evalynOS/blob/24cc1ed11d4ac0a5a938b0ca972094bc0783fcd5/src/kernel/drivers/x86_64/gdt.c#L58

[hazy saddle]

This patch when applied to my repo should get you to where I am at btw

I am quite vexed

https://tenor.com/view/koffa-house-md-gregory-house-house-eric-foreman-gif-3254275732664605731

[viral bison]

@hazy saddle maybe the missing [] here?

[hazy saddle]

What does the [] even do again

[viral bison]

i think it gets the value from the pointer

[rancid viper]

basically deref

[hazy saddle]

Aswell wouldn’t this brick kernel threads too

[rancid viper]

and no you shouldnt deref it

you are passing the rsp

not a pointer to it

[hazy saddle]

yes

thread_switch(&previous_thread->rsp, current_thread->rsp);

btw @viral bison you on my source tree from github or the patchset

[viral bison]

yes

yours+

pagefault seems to be caused by this

hmm

[hazy saddle]

why tf is IP even there

is that under CPL3 or 0

[viral bison]

idk

btw, all user code starts at 0x4000?

[hazy saddle]

yes

[viral bison]

hmm

that might give you problems later

[hazy saddle]

it gets loaded here
fs_read("/badapple.bin", (void *)0x4000, num_pages * page_size);

[hazy saddle]

im gonna de hard coder it later

would be easy too

[viral bison]

i use r15 to pass it

when i set the first thread stack

[hazy saddle]

i use a somewhat custon binary format

il do static elf loading later

ENTRY(main)

SECTIONS
{
  . = 0x4000;

  .text :
  {
    KEEP(*(.entry))
    *(.text)
  }

  .rodata :
  {
    *(.rodata)
  }

  .data :
  {
    *(.data)
  }
}

[viral bison]

k

why no .bss?

[hazy saddle]

it gets hoisted like this

[hazy saddle]

its implied by my loading

when fs read is told to read more bytes than there are in the file it dumps nulls

eg 0

int main() __attribute__((section(".entry")));
int main() {

this hoists my main function to the start if the binary

gcc -ffreestanding -fPIC -nostdlib -fno-asynchronous-unwind-tables -fno-stack-protector -c badapple.c -o badapple.o
ld -nostdlib -T badapple.ld -o badapple.elf badapple.o
objcopy -O binary badapple.elf badapple.bin

cp ./badapple.bin ../../initramfs
cd ../../../

[viral bison]

that RIP happens when i press enter

[hazy saddle]

and this will extract the binary

[hazy saddle]

huh

[hazy saddle]

how

[viral bison]

idk

it crashes there for me

[hazy saddle]

did my diff generate / apply right

did you change anything

[viral bison]

i did git apply <patch>

and edited a sec this

[hazy saddle]

i used git diff > whatever.patch

[viral bison]

that's to generate the patch

[hazy saddle]

yeah so it should be good

[viral bison]

ok it somewhat got fixed

how do i trigger your crash?

[hazy saddle]

run badapple

and then run it again

[viral bison]

how do i stop it XDD

[hazy saddle]

crash it with badapple again or run panic

[viral bison]

no i mean

how do i stop it from playing the first time

lol

[hazy saddle]

wdym?

[viral bison]

i can't run it the second time if i don't stop it from playing, no?

ah wait

i can

[hazy saddle]

the shell should still work

that running debug is a lil broken

[viral bison]

ye

it crashed at 4207

[hazy saddle]

That’s what I somewhat expect

Well I shouldn’t expect it

But yeah

This is using FPU code btw

[viral bison]

where is badapple compiled to

[hazy saddle]

But it will crash the same way if you have an infinite while loop

So it won’t do anything

[hazy saddle]

src/user/badapple

It has the make.sh

[viral bison]

found, thx

[hazy saddle]

That gets called by another shell script that’s executed by the makefile

Also I lowkey wana use some c++ in my kernel for instantiating classes for some drivers to be easier

[viral bison]

oof, i can't addr2line the .bin

[hazy saddle]

Should be able to objdump it and spesify x86-64

[viral bison]

how?

[hazy saddle]

You need to tell it it’s x86-64

Because it’s a binary blob

No headers

[viral bison]

and how do i tell it?

[hazy saddle]

Leme look it up

-m i386:x86-64

[viral bison]

still says not recognised

[hazy saddle]

Maybe drop the -S

[viral bison]

nope

[hazy saddle]

objdump -b binary -m i386:x86-64 -D ./badapple.bin this works now

207: 48 8b 45 f0 mov -0x10(%rbp),%rax

it dosnt know the offset so subtract 0x4000

[viral bison]

that's strange

why would it only happen the second time

[hazy saddle]

Why would I know 😭

[viral bison]

wait a sec

you never map the user pages

in the thread pagemap

you only map to the kernel pagemap here

@hazy saddle

[hazy saddle]

Fuuuuuuuuuuckkkkk

how should i handle passing this in there

[viral bison]

i save the currently executing thread pointer in my cpu struct

so i can access it on the local cpu

the thread then contains a ref to its process

[hazy saddle]

what if i pass the pagemap pointer as an argument to the thread?

[viral bison]

and the process contains the pagemap

[viral bison]

you already do tho

[hazy saddle]

huh

[hazy saddle]

but the actual thread dosnt have that

[viral bison]

huh?

[hazy saddle]

i need to get it passed in here?

[viral bison]

you need to get the thread struct there

[hazy saddle]

[viral bison]

or well, thread pointer

[hazy saddle]

yeah i can do that

[viral bison]

i write the cpu struct pointer to KGSBAS

so i can read it whenever i want

[hazy saddle]

which one on the stack should the pointer be again

[viral bison]

wdym

just use msrs

[hazy saddle]

when cracfting the stack

[viral bison]

it's way easier

[hazy saddle]

what

[viral bison]

write the cpu struct to the kernel GS base msr

so you can use it in other places too

[hazy saddle]

i dont have a CPU struct

[viral bison]

make one

[hazy saddle]

not rn

[viral bison]

well, then write the thread pointer there

it'll just be a temporary thing for now

[hazy saddle]

nah i got this

[viral bison]

you could use an unused register then

[hazy saddle]

ive been awake like more than 24 hours i need to just do this easily

[viral bison]

using the msr is the easiest way tbh

[hazy saddle]

nahhh fuck it we ball

[viral bison]

XD

why like that

[hazy saddle]

FUCK

[hazy saddle]

it works technicaly

[viral bison]

you write it 15 times?

lol

[hazy saddle]

to every register

[viral bison]

bruh

[hazy saddle]

so i cant possbily be wrong

[viral bison]

that is a bad idea

[hazy saddle]

[viral bison]

just write it to like r15

[hazy saddle]

i need to conform to SYS-V

[viral bison]

i use r15 and r14 to pass data to the newly created thread

tho i can use some more

[hazy saddle]

this is what it used to be

its somthing like index 6 right

[viral bison]

in theory you should just need to set 6 registers

[viral bison]

where is this code

[hazy saddle]

scherdular.c

[viral bison]

found

wait what

why pass task quit?

[hazy saddle]

if the thread exits it returns there and loops

[viral bison]

hmm

idk if that works like that

[hazy saddle]

still fucked

[viral bison]

well, are you mapping the memory?

[hazy saddle]

i need to sleep lowkey but i cant sleep knowing its broke

[viral bison]

changing 1 reg there means nothing if you don't use it

[hazy saddle]

i am

[viral bison]

show the updated mapping code

[viral bison]

bruh

[hazy saddle]

what did i do this time

[viral bison]

what reg did you pass the thread to?

[hazy saddle]

RDI

[viral bison]

hmm

can you check if it's not 0?

the function arg i mean

[hazy saddle]

it executes properly the first time tho

[viral bison]

yeah, that's very strange

it shouldn't even execute

bc that mem is not even mapped

[hazy saddle]

it is mapped tho

[viral bison]

before i meant

ok, well, i see 1 problem there

the 0x0 to 0x4000 is mapped to the kernel

it doesn't have the user flag

so it will throw a pagefault

tho i'm not even sure that's the cause

[hazy saddle]

because thoes pahes shouldnt be used rightr

[viral bison]

then why map those at all?

[hazy saddle]

and idk why tf it is mapped tbh

my entire VMM is stolem

i dont think anything SHOULD EVEN BE MAPPING IT

[viral bison]

wait, in create_thread, i see you are using malloc to create the stack

[hazy saddle]

yes

[viral bison]

but do you then map that stack to add the user flag or no?

[hazy saddle]

That’s the kernel stack right

[hazy saddle]

This allocates another stack in userspace

[viral bison]

oh ok

there is an easier way for this btw

[hazy saddle]

Probslt

But most of my code is written in seep depreciation

[hazy saddle]

Which I am having right now

And I’m about to falls sleep

Is there anything else you need

Because I dunno what to do

[viral bison]

maybe try to debug a bit

[hazy saddle]

Idk what I should even be debugging

[viral bison]

i think i'm starting to get somewhere, wait

nope

[hazy saddle]

I dunno what the heck is going on

[rancid viper]

I think text messages is too inefficient to debug this lmao

yall need to do some pair programming

[hazy saddle]

Setup one of those stupid IDEs that lets you edit code at the same time together too lmao

And honestly I’m so fucking burnt out from this one bug

[rancid viper]

zed editor

its my primary editor

: )

[hazy saddle]

Is it on Linux?

And it’s that one that shills AL

AI

[rancid viper]

fuck windows

is all Ill say

but they did recently start shipping windows binaries I believe too

[rancid viper]

meh

they have great agent features

I dont use them personally

[hazy saddle]

I get a bad taste in my mouth from anything that wants to integrate Ai

If I want AI il goto AI

[rancid viper]

so...

everything

[hazy saddle]

I don’t want it inside my editor

KATE dosnt do stuff like that

[rancid viper]

xd

[hazy saddle]

I use KATE

[rancid viper]

I mean u can turn off the agent panel with one setting if you wish

unfortunately today its basically expected to have AI stuff builtin

[hazy saddle]

If they are going to bend to the will of AI I don’t wana use it

I should be switching to libre wolf soon too because of the Firefox crap

Even fucking VIM has a sponsor for AI

JFC

[rancid viper]

it is futile to escape it

I just pick the best editor for me

an extra feature I turn off doesnt hurt me

[hazy saddle]

Been going fine for me

But yeah anyways

@viral bison wana VC and pair program later when I wakeup

[rancid viper]

I cant quit google

[viral bison]

k

[hazy saddle]

We got this

https://tenor.com/view/type-cringe-gif-19048218

Somthing somthing this gif

[viral bison]

in how much time btw?

[hazy saddle]

When I sleep im out cold

8 to 12. Hours probably

[viral bison]

XD

idk if i'll still be awake then

it's already 4pm here

[hazy saddle]

My parents try and wake me up and they think I’m dead 💀

[viral bison]

XD

[viral bison]

huuh

@hazy saddle

i might have found exactly where it crashes

the second or third execution of this loop

when calling allocate_page

[devout path]

hi

how are u

dam i was outside earlier

[hazy saddle]

I did not manage to fix it

But perhaps closer

[hazy saddle]

hm interesting

thats from my friends code

maybe its running out of pages to give?

but hm

and if i give the PMM more bitmap pages and give it from 512mb to 2gb memory it still shits its self

[hazy saddle]

I wonder what causes it inside this function

[hazy saddle]

And by shits its self I mean it wont panic at the very least

So it’s giving “valid pages”

I know the code is doing some integer division

Could it be accidentally giving physical pages that already have been allocated

[viral bison]

Try to switch to freelist

It's simpler and better than bitmap

[hazy saddle]

[viral bison]

@hazy saddle what ide do you use to code?

[hazy saddle]

KATE

[hazy saddle]

How do I do this btw

And not make things explode like they are

[viral bison]

If you were on vscode we could code at the same time

Or on CLion

[hazy saddle]

I can probably get it on cachyOS

The flatpak should be fine too anyways if I have a terminal open elsewhere

[viral bison]

K

[hazy saddle]

It is 2am rn tho

[viral bison]

Oh

Lol

[hazy saddle]

What would happen if I replace it with a bump allocator maybe

[viral bison]

8Am here

[hazy saddle]

And see if that will keep it working?

[viral bison]

bump = VERY bad

[hazy saddle]

Just to see if that’s the real issue

[hazy saddle]

Oh I know. But to see if the PMM is the issue

[viral bison]

tho i don't even think it's a pmm issue

try to add a check if it returns a null

[hazy saddle]

My friend said the VMM was fucked

But every time I made a VMM it always caused issues

[viral bison]

wtf

so it's already at the first time that it returns a null page??

but somehow it still works

@hazy saddle how do i make your printf print an address in hex?

[hazy saddle]

Should be the same as standard normal printf

[viral bison]

FUUUCK

wait what

oh i'm stupid

[hazy saddle]

I am using nanoprintf

[viral bison]

k, i was using > instead of <

lol

ok, alloc_page is out

it returns a valid page

[hazy saddle]

Hmmmm

I mean I should redo that later anyways to maybe a buddy allocator

[viral bison]

buddy might be better for malloc

i think freelist is the best one for pmm

[hazy saddle]

Ahhh

How does a free list work

[viral bison]

virtually no mem needed, and it's instant

basically you add blocks (that represent free pages) in a linked listr

you store those blocks inside the free pages

then, when a block is used, if a block is multiple pages wide, you just shrink that

else, you nuke that block from the list and overwrite the data on that page

so it basically uses no mem

[hazy saddle]

So a link list of free page ranges?

[viral bison]

yes

[hazy saddle]

When I take pages I break off from it

[viral bison]

yes

[hazy saddle]

And when I free pages I add to it

[viral bison]

at the start

[hazy saddle]

And it kind of just works?

Do I have to go through the entire linked list?

[viral bison]

not if you don't require more than you have on the first block

when you find a block of the size you need, you stop

that's why you add freed pages at the start

[hazy saddle]

And I can take a block and a part of another?

[viral bison]

well, it depends

if you need them contiguous no

else yes, you can

[hazy saddle]

Why would I need it to be contiguous?

[viral bison]

DMA sometimes need it like that

[hazy saddle]

Ahh

I need to make myself a generic linked list helper tbh 🥀

[viral bison]

want mine?

it's pretty good

[hazy saddle]

What’s it licensed under?

[viral bison]

huuh

my os is under GPL

[hazy saddle]

I’m on MIT

[viral bison]

idc really

i might change the license too

[hazy saddle]

You could cc0 the linked list stuff

I’m planning on having a repo with a bunch of small shit I made to be public domain

[viral bison]

i tought gpl was good, will prob change it

anyway, here is the code: https://github.com/Diamantino-Op/HorizonOS/blob/main/src/libs/hos-base/src/LinkedList.hpp

it's made with c++

so you might have to adapt it

[hazy saddle]

GPL means I would also have to be GPL

[viral bison]

even with permission?

[hazy saddle]

Technically but it can get messy I think?

Also templates

[viral bison]

sec, lemme go and change that

[viral bison]

it was the best way for c++

[hazy saddle]

Yeah

MIT still needs credit for when the source is open

But it’s a lot more flexible

[viral bison]

oh cool

[hazy saddle]

Also dosnt a linked list need a malloc

[viral bison]

well, yes

tho it can be made without

the one i sent was made to work with malloc

but if you go in my PMM file

you see the implementation without

[viral bison]

found my PMM?

[hazy saddle]

Not yet rn

[viral bison]

k

tho rn we need to find what is nuking badapple

[hazy saddle]

Or rather any extra threads

[viral bison]

wdym

[hazy saddle]

User*

It will be nuked even if two different user apps were to be loaded

[viral bison]

also, how does the ctx switch work

like, the thread queue

[hazy saddle]

It should push all GPRs switch stacks and then restore GPRs

Works for kernel threads

[hazy saddle]

Linked list

It cycles through all of them

[viral bison]

yeah i saw, but i don't see it looping i think?

when it reaches the end of the queue it will take a nullptr, no?

[hazy saddle]

I’m pretty sure the list is setup so the last element links to the top

[viral bison]

hmm

prob

ok yeah, it should

[hazy saddle]

@viral bison I wonder if it’s possible to gdb the user apps too?

[viral bison]

it is

tho idk if it will load the symbols

if they are not elfs

[hazy saddle]

I assume you can have external symbols generated

It will compile the elf too

But it dosnt load the elf

It’s elf -> binary blob

[viral bison]

hmm

let's hope the addresses are the same

[hazy saddle]

The linker says the load address

So it should be fine

Linker script says 0x4000 I put it there

[viral bison]

k yes, it recognised it

[hazy saddle]

Also for not using malloc on the PMM

What if I used a bootstrap bump allocator? Once I have malloc working I can then use that properly

[viral bison]

no need for that

you can make a linked list without malloc

look at mine

[hazy saddle]

Or I just reserve the first page for the PMM?

[viral bison]

yep, you reserve the first page

and point the first pointer to that

or well

first page of every mem region

limine gives you a memmap

use that

oh wait

debugging the app is useless rn

it never even reaches that point

lol

[hazy saddle]

Isn’t the page fault under CPL3 mode?

[viral bison]

how do i see that

[hazy saddle]

I dump the bits of the error code

[viral bison]

error is 0x111

[hazy saddle]

For a page fault when bit3 is set it means CPL3 right

[viral bison]

idk

[hazy saddle]

That’s not hex that’s binary

Eg 0b111

[viral bison]

oh yeah

srry

[hazy saddle]

That’s pretty useful for debugging though innit?

[viral bison]

how tf is it failing on CPL3 if it never even reaches main

[hazy saddle]

[viral bison]

seems like it's skipping main entirely

[hazy saddle]

So what the fuck is going on

[viral bison]

that's what we need to understand

well, now we know why it is pagefaulting

WAIT

[hazy saddle]

What

[viral bison]

oh no

tought you were loading the new pagemaps too erarly

bc what if it's not the new thread thta's pagefaulting

but the old one

this would make more sense

[hazy saddle]

But I only run on one core tho

[hazy saddle]

Maybe you can add a hook to the PF to log the current thread ID?

[viral bison]

try

[hazy saddle]

Sorry

[viral bison]

?

k

pinpointed the fail location

2nd time it runs allocate_page()

[hazy saddle]

Let me add some logging to that

[viral bison]

i'm stepping into it with gdb

ok

it fails before doing free_pages--

[hazy saddle]

the exact part it fails on

[viral bison]

hmm, then my breakpoint isn't triggering

what is the RIP?

[hazy saddle]

and if i make it use less pages it does the same

Interrupt Frame:
IP=0x0000000000004203 SP=0x000000007fffffa0
SS=0x0000000000000033 CS=0x000000000000002b
FLAGS : 00000000 00000001 00000010 10000111 

[viral bison]

ok wtf

where exactly is that log

[hazy saddle]

this is my panic log

to the serial

[viral bison]

no, i mean the alloc page one

show the line

[viral bison]

how the fuck is it failing in 0x4000

[hazy saddle]

userspace code ig 😭

wtf is this mapping here

this is before bad apple

HUH

[viral bison]

wtf

[hazy saddle]

and then running bad apple once

HUH

and then after once more

[hazy saddle]

flanterm bug?

this dosnt apear in serial

[viral bison]

race condition prob

[hazy saddle]

FT and sterial output is the same tho

[viral bison]

that's strange

something might be nuking the FB too?

[hazy saddle]

somthing may be up with the VMM?

thats what she said and its her VMM code

[viral bison]

vmm seems ok

it's not even failing there

[hazy saddle]

but what if its settings things wrong

[hazy saddle]

like these logs

with qemu debugs info mem

the memory map is bad

@viral bison we may need an expert to help us?

this is so fuckin strange

[viral bison]

ye

we need someone else

[hazy saddle]

fresh set of eyeballs

https://tenor.com/view/patrick-spongebob-looking-around-nosy-map-gif-15455970

@viral bison who should we get

[viral bison]

idk

someone that is available rn

something is trying to access a phys address

i think i'm getting close

it happens only when i run badapple

[hazy saddle]

interwsting

i cannot thank you enough

i dunno if i could of even debugged this far without you

[viral bison]

to get that error, i fixed this function:

this should also remove the garbage from the new pagemap

[hazy saddle]

it now page faults after running bad apple once

but i can repro the error

[viral bison]

yes

[hazy saddle]

but we are clean

i wonder what was causing the non clean bits of it

[viral bison]

they are in the kernel pagemap

[hazy saddle]

ig its some kind of just random non zero memory?

[viral bison]

but you zero it

[hazy saddle]

but i never should be putting anything in tjere

but how is this put there then

[viral bison]

also, that a address

[hazy saddle]

huh

[viral bison]

this here

something was mapping it

and it's where it is pagefaulting now

[hazy saddle]

thats the frame buffer

uint64_t frameBufferBase = 0x00000000A0000000;

[viral bison]

oh, where are you allocating it

[hazy saddle]

FUCK

[viral bison]

yep

fuck

that's why it is pagefaulting now XD

and that's what i think is breaking the second badapple

[hazy saddle]

should i have that just grab the current thread from the schedular

[viral bison]

yes

IT WORKS

[hazy saddle]

ITS DOING IT

[viral bison]

i'll pass you the code i have

[hazy saddle]

the XMM registers arent preserved

[viral bison]

it has some more fixes

[hazy saddle]

so they kinda sync up

[viral bison]

yeah well

[hazy saddle]

but it works

[viral bison]

you aren't doing xsave

nor fxsave

so that's kinda normal

[hazy saddle]

i forgor how to do that ngl

[viral bison]

it's a bit of a pain

well, simd is not even enabled in your os

[hazy saddle]

SIMD is

bad apple does use SIMD

i can convert to integer divison rq

[viral bison]

where do you enable it

[hazy saddle]

wait i may of nuked that code

i may be only doing x87?

[viral bison]

prob

i don't see any simd init

[hazy saddle]

wait im not using any FPU

still works with this

i think they all run tho

leme add a systemcall to DENGEROUSLY print a string

umm

the threads need to know thier ID maybe

[viral bison]

btw