EvalynOS

[devout path]

i can finally sleep in fuckin peace

Lord so help me

[hazy saddle]

thank you so much

[devout path]

you're welcome!

[hazy saddle]

o n o

[devout path]

wat

[hazy saddle]

nothing

go sleep

[devout path]

imma read more of ur kernel before i do

[hazy saddle]

Is my kernel really that cool?

[devout path]

maybe document this in a md file : )

[hazy saddle]

[devout path]

yes :D

[hazy saddle]

What was the fix btw?

I’m also sleep deprived

[devout path]

@viral bison

pls explain

[viral bison]

what?

the fix you did?

[devout path]

wat was the fix

yes

i want someone else to figure it out

[viral bison]

from what i saw, you added back the saving of all the registers

[devout path]

y e s

correct

[viral bison]

i still don't know why that works

[devout path]

n e i t h e r d o i : )

that's why i sa i d

d o n 't j u d g e : )

this is me taking shots at random shit

[hazy saddle]

The holy hand grenade

https://tenor.com/view/monty-python-and-the-holy-grail-open-holy-hand-grenade-gif-13890860417931861339

[devout path]

when u have nothing, do random bullshit

until u actually do get it

[hazy saddle]

You might wana checkout my HRMR codebase if you like my kernel code

[viral bison]

@obtuse loom why would saving all the regs on ctx switch fix this?

[devout path]

im presuming that this doesn't return to garbage anymore

[hazy saddle]

That’s some of my c++ work

[devout path]

oke imma do that

[hazy saddle]

(My not horrible c++ work actually)

Wait

@devout path @viral bison

The real hardware test

[devout path]

hi

oh god

[hazy saddle]

Let me generate a patch file

[viral bison]

send me the elf directly

[hazy saddle]

[viral bison]

and the initramfs

[hazy saddle]

True

Remember when my kernel was like 5MB

Because the bad apple was bundled inside of it

[devout path]

XD

[hazy saddle]

Oh also remember when it was kernel.bin and not kernel.elf

When I had my own bootloader

[viral bison]

oh god

[obtuse loom]

fix what? 😩

[devout path]

isn't it bcuz every thread expects to get back exactly the cpu state it had?

[hazy saddle]

That never wanted to work between real machines

[viral bison]

i nuked the kernel before the cli initialized

[hazy saddle]

Because universal my ass UEFI

[viral bison]

the stack corruption

[hazy saddle]

This right?

[devout path]

im juggling between what i know and what theory would be

and both contradicts atp

i have no fuckin clue anymore

[obtuse loom]

can I have some more context

[devout path]

yeah supposedly

[viral bison]

btw @hazy saddle pressing enter or whatever key before it says press enter, will lock up the kernel

[hazy saddle]

Thread one saves data in R9
Kernel switches to thread 2
Thread 2 puts its own shit in R9
Kernel switches back to thread one
Thread one gets wrong data from R9 and fucks over the stack

[viral bison]

scroll up a bit

the framebuffer was getting nuked after a bit

[obtuse loom]

link plox you guys chat a lot

[hazy saddle]

Ehhhhh that check is gonna get removed soon actually

[viral bison]

so was the whole stack

[hazy saddle]

Once I get a user space shell

[devout path]

try mine

[devout path]

y e s

[viral bison]

huuh, sec

[devout path]

im losing hair chat

[hazy saddle]

This chat has so many messages compared to how mature others are because my kernel has been so buggy and fun to debug

[obtuse loom]

lovely

[viral bison]

@obtuse loom from here basically

[hazy saddle]

It’s held up with sticks and stone and ten pounds of Elmer’s glue

[viral bison]

btw, it's working on real hw

[hazy saddle]

LETS GOOOO

[viral bison]

tho i hear my cpu crying in pain

[hazy saddle]

[devout path]

im going to cry

[viral bison]

pls put pause on the quit thread

[hazy saddle]

It’s CPU rendering the entire bas apple

[obtuse loom]

okay cool

[hazy saddle]

[obtuse loom]

what does the context switch look like

[devout path]

mmmmmmm

[viral bison]

before it looked like this

[devout path]

before it was this

global thread_switch
global thread_switch_user
global switch_to_user

section .text
thread_switch:
    pushfq
    push rbx
    push rbp
    push r12
    push r13
    push r14
    push r15

    mov [rdi], rsp
    mov rsp, rsi

    pop r15
    pop r14
    pop r13
    pop r12
    pop rbp
    pop rbx
    popfq

    ret

USER_STACK_TOP equ 0x0000000080000000

switch_to_user:
    mov ax, 0x30 | 3
    mov ds, ax
    mov es, ax
    mov fs, ax
    mov gs, ax

    push 0x30 | 3
    mov rax, USER_STACK_TOP
    push rax
    push 0x200
    push 0x28 | 3
    push 0x4000
    iretq

now this

[viral bison]

now it looks like this

[devout path]

global thread_switch
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, 0x30 | 3
    mov ds, ax
    mov es, ax
    mov fs, ax
    mov gs, ax

    push 0x30 | 3
    mov rax, USER_STACK_TOP
    push rax
    push 0x200
    push 0x28 | 3
    push 0x4000
    iretq

[obtuse loom]

how do you declare/call thread_switch

in C land

[hazy saddle]

Switch to user will get rewritten btw so I can load them anywhere and maybe even have ASLR?

With how buggy it is though I don’t know if we should have ASLR XD

[devout path]

[hazy saddle]

That’s you for helping

[devout path]

xd

[obtuse loom]

this looks fine to me

[devout path]

yeah and then i decided to change it

[viral bison]

somehow saving all the regs fixed it tho?

[obtuse loom]

I think your diff just happens to make another bug not happen

[devout path]

yeah and then i saved every reg

[hazy saddle]

But then why wouldn’t a bug like this happen?

[obtuse loom]

you probably have some situation with some corruption hitting the stack of another task

that's my best guess

and changing the layout like this happens to make the important values move somewhere that's not overwritten

[hazy saddle]

[obtuse loom]

all the registers you added to the push/pop list are callee saved

so the compiler should manage it correctly

[devout path]

y e s

[obtuse loom]

unless you call it incorrectly

like from inline asm without clobbers

[devout path]

well i think it was with the create_thread funct earlier it had dupe pops

so i was like

: )

am i right

m literally just

completely guessing

[viral bison]

tho that function worked

i was using it fine

[hazy saddle]

Until it didn’t

[devout path]

y e s

[viral bison]

it didn't when you changed something inside badapple

[devout path]

it had dupe pops so the ptr after the switch wasn't pointing at a valid saved rip

[viral bison]

huuh

[obtuse loom]

the old thread_switch is correct, can't say the same for the rest of the codebase

:^)

[devout path]

hhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhh

REALLY

so it wasn't the ret jumping to whatever the fuck?

[viral bison]

no

[devout path]

im gonna stop thinking

[viral bison]

it was working

badapple was even starting correctly

[devout path]

but it did it randomly

oh

fair

wait my bad

[hazy saddle]

Question? How much of a performance penalty is there from manually putting the pushes and pops verses the compiler doing it for you

[devout path]

: )

[obtuse loom]

depends on what the calling code looks like, it increases register pressure which makes register allocation give worse results in functions with a lot of values

well I mean

you're not changing the declaration

so the calling code will still save all those registers for you

and then you're saving them again

[devout path]

so there wasn't any missing regs or anything??!

[obtuse loom]

so you're not getting the benefit of the compiler not saving them for you either

[viral bison]

no

[obtuse loom]

not that I can tell

[devout path]

im damn hallucinating

[obtuse loom]

only the ones listed as Preserved across function calls need to be listed

[devout path]

oh yeah right

[obtuse loom]

the compiler knows calling an external function will clobber all the other registers

as I said, as long as you're not misusing inline assembly to call it

[hazy saddle]

I don’t think I am

[obtuse loom]

as long as you're calling it from normal C there should be no bugs related to this

[hazy saddle]

this would be proper for setting up the inital stack though right?

and feeding in an argument

[viral bison]

don't need pagemap there

[devout path]

abi rules only apply to funct calls no? not switches?

[hazy saddle]

i may of removed that

[obtuse loom]

yes

switches are not functions

[hazy saddle]

yes i did remove that

[obtuse loom]

oh right fuck

this only applies for sys-v

what's your target

[devout path]

yeah

[hazy saddle]

should be sys-v

[devout path]

huh

[obtuse loom]

okay

double check that

[devout path]

y e a

cuz

[hazy saddle]

how would i do that?

[obtuse loom]

if you're targeting some UEFI callconv or something

then you're fucked

[hazy saddle]

i dont think i would be

[devout path]

o m h

[obtuse loom]

I assume you would know :^)

[devout path]

i mean

this is sys-v

it's compiling freestanding to x64 so

absolutely it is

[obtuse loom]

honestly should set an explicit target but yeah

[devout path]

yeaaah

ELF Header:
Magic: 7f 45 4c 46 02 01 01 00 00 00 00 00 00 00 00 00
Class: ELF64
Data: 2's complement, little endian
Version: 1 (current)
OS/ABI: UNIX - System V
ABI Version: 0
Type: EXEC (Executable file)
Machine: Advanced Micro Devices X86-64
Version: 0x1

i had to do readelf xd

[obtuse loom]

okay you're good

[hazy saddle]

if i put this back in i gett an uhh page fault right away

[devout path]

HUH

[viral bison]

well, did you also change the stack?

[obtuse loom]

cool, tried debugging it?

[hazy saddle]

no

right

[viral bison]

yeah well

[hazy saddle]

fuck

i need to shave it to 6 registers right

[viral bison]

yes

[devout path]

yeah

[viral bison]

• flags

• rip

[devout path]

it should work theoretically

[hazy saddle]

@devout path

[devout path]

hi

wat happened

[hazy saddle]

dramatic drum roll please

[devout path]

dramatic drumroll

[hazy saddle]

It works

[obtuse loom]

amazing

[hazy saddle]

Thank you for your help @obtuse loom

[obtuse loom]

🎉

[viral bison]

so it wasn't the registers

lol

[devout path]

;-;

[obtuse loom]

no worries

[hazy saddle]

It was the friends we made along the way

[obtuse loom]

what was it

[hazy saddle]

God if I know

[obtuse loom]

😂

fair enough

[devout path]

it was about saving the flags?

[viral bison]

it's time for me to finish implementing syscalls

[hazy saddle]

Who needs system calls when you have int 0x69

[obtuse loom]

shouldn't be needed either

[devout path]

lesson learned: 67

[hazy saddle]

NOOOO

[devout path]

XD

wait can u send the patch for what u just did @hazy saddle

[obtuse loom]

but if you want it to simplify initial state setup then like whatever

[devout path]

fair

[hazy saddle]

one push isnt hurting modern CPUs

[obtuse loom]

honestly yeah

it's fine

but it's not needed

up to you if you want to remove it

[hazy saddle]

how would one go about doing that btw

[obtuse loom]

removing pushf and popf?

[hazy saddle]

oh thats it

@devout path this is it

asude from the | to an &

in the vmm

[devout path]

oh

[viral bison]

task quit should not be needed here

[hazy saddle]

What happens if the thread attempts to return to nothing though?

[viral bison]

it shouldn't

i think

[hazy saddle]

Better safe than sorry?

[devout path]

[hazy saddle]

Task quit could also start auto cleanup of the thread on return

[viral bison]

i have a reaper thread just for that

(you should too)

[hazy saddle]

but what if the thread its self was also the reaper?

[viral bison]

it might end up very badly

[hazy saddle]

[devout path]

these what if shouldn't be discussed when it hasn't happened yet : )

i love making problems for myself

[hazy saddle]

but hell yeah

[viral bison]

turns out i set the scheduler in the same interrupt as my PIT

LOL

[hazy saddle]

this causes the PF unless i remove an extra thing from the stack and then it locks up

[obtuse loom]

wack

[hazy saddle]

Wdym

[obtuse loom]

unexpected

[hazy saddle]

Ah

[devout path]

wat featuers would u like to have in ur OS @hazy saddle

[hazy saddle]

I eventually want it to be able to compile its self

[devout path]

i copuld probably dev somethin

oh

fair

[obtuse loom]

try to debug it maybe?

[hazy saddle]

And to run a window manager

[hazy saddle]

I’m too lazy for saving a few CPU cycles

[hazy saddle]

Wdym?

[devout path]

develop somethin

[hazy saddle]

And run doom

[devout path]

i mean

[viral bison]

small cheat to skip int 0x80

lol

[hazy saddle]

I use int 0x69

Because I can

[viral bison]

0x80 is the standard in linux apps

for 32 bit

[hazy saddle]

[viral bison]

btw, that code is for IOApics

so if they ever reach 0x80, i just make them go up by 1

[hazy saddle]

Can’t that u8 be a Boolean

[viral bison]

boolean is an u8

[devout path]

imma jus do a pr

is anyone workin on ur os too or nah?

[hazy saddle]

A PR to what

[devout path]

just u

[hazy saddle]

I still have some cleanup to do rn

Of debug logs n shit

[devout path]

oh for stuff that i'ma try to build for this os xd

o h

oke

[hazy saddle]

Wait what are you building

[devout path]

ext2 support : )

[hazy saddle]

Don’t I need a physical disk driver tho?

[devout path]

f a i r

[hazy saddle]

I was gonna hold off on that until PCIe and USB so I can have a driver for nvme ssd and mass storage devices

[viral bison]

you'll need uAcpi

and prob IOApic

and Apic

[hazy saddle]

[devout path]

true

[viral bison]

and some better timer

like hpet and tsc

• the apic timer

[hazy saddle]

[viral bison]

so you can get accurate timings

[hazy saddle]

But yeah ext2 would be really low priority

And honestly jumping the gun to ext4 would be based

https://tenor.com/view/whatever-go-my-scarab-scarab-assyst-gif-9837798423909404507

[devout path]

yea ;-;

[hazy saddle]

Perhaps you can work on the APIC and co drivers?

[devout path]

hhhhhhhhhhhhhhhhhhmmmmmmmmmm

yeah why not

[hazy saddle]

After I do some basic cleanup on this and push

[devout path]

ye yes

that' swhat i was plannin to do

[hazy saddle]

Can you also use sata in a legacy mode to make it more easy to talk to too?

That could work for ext2 perhaps

[devout path]

yes

[hazy saddle]

And I can get a cheap SSD and throw it in my PC for testing

[viral bison]

me who wanna support HPFS

[devout path]

uuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuuh it'll prolly take me a week though ;-; but yes yes ofc ofc

[viral bison]

or was it APFS

[hazy saddle]

And then ext4 can come with the USB + PCIe + NVMe stack

[viral bison]

it was APFS

[devout path]

yeaa

[viral bison]

https://developer.apple.com/support/downloads/Apple-File-System-Reference.pdf

if you wanna read this

lol

[devout path]

xd

[hazy saddle]

Linux DRM driver port when

Sarcasm btw to be clear

[devout path]

;-;

[obtuse loom]

if you're going to API match linux you might as well ABI match and not have to port anything

[hazy saddle]

And at that point I should just get inspired by Linux /s

https://media.discordapp.net/attachments/766703671504797696/1198114808902451220/ezgif.com-video-to-gif_2.gif

[viral bison]

i will have syscall support for linux and my os

[obtuse loom]

I mean sure lol the license allows it

[viral bison]

• i plan to have macos and windows ones too in the future

[hazy saddle]

Yes here is my kernel which is just Linux but I found and replace the entire thing

Has anyone ever done that before here

[obtuse loom]

yes of course

very unoriginal

[hazy saddle]

[obtuse loom]

there are even people who do it without releasing the source and breaking the license

[devout path]

i copy linux (the kernels) subsystem's architecture : )

[devout path]

true

[hazy saddle]

Honestly my kernel should be able to get doom once I get a libc done?

[viral bison]

there are conspiracy theories that the kernel is being ported to rust so that companies can grab the code and edit it in closed source

[viral bison]

you wish

[hazy saddle]

I just need to get system calls for file stuff

And handle file descriptors

[hazy saddle]

https://github.com/ozkl/doomgeneric

It’s not that much more than libc and 5 functions

[obtuse loom]

... but why? 💀

[viral bison]

so they don't have to share their edits

corpos are strange

[obtuse loom]

why would they rewrite it in rust is my question

[hazy saddle]

Moss kernel?

[obtuse loom]

you can already break the license

[viral bison]

MIT license

[obtuse loom]

no one is getting in trouble for it

[viral bison]

they are giving the rust ports MIT licenses

[hazy saddle]

Huh

[obtuse loom]

plenty of companies have changed linux, not released changes and gotten away with it

it happens daily

[hazy saddle]

I think I may of broken the license when I had to modify the driver on my local device to get a controller working

[obtuse loom]

it only applies if you release a binary publically

[hazy saddle]

I should probably dig it up fix it and put it on GitHub

Ahh then I’m fine

[obtuse loom]

afaik, IANAL

etc etc

[hazy saddle]

Ehhhh so sue me

@devout path alright I’m gonna clean it up

[viral bison]

this is a strange memmap, lol

[hazy saddle]

What were the parts you were gonna work on again

[viral bison]

uefi and its strange stuff

also, you should prob not use bootloader reclaimable stuff

[obtuse loom]

I mean that literally means usable after you're done using bootloader data

[viral bison]

isn't the kernel stack in there too tho?

[obtuse loom]

how should I know

what stack are you using

[viral bison]

i leave the default one alone

[obtuse loom]

from the bootloader?

[viral bison]

yes

[obtuse loom]

then check the spec

[viral bison]

k

[devout path]

hi

just simple stuff

i plan to make a block driver for now

[obtuse loom]

are you using the limine protocol?

or something else?

[devout path]

i can make the syscal isr though

if u'd like

[viral bison]

yes, limine

[obtuse loom]

okay then yes it is

"rsp is set to point to the top of a stack, in bootloader-reclaimable memory, which is at least 64KiB (65536 bytes) in size, or the size specified in the Stack Size feature. An invalid return address of 0 is pushed to this stack before jumping to the executable's entry point."

[viral bison]

i'll leave the bootloader reclaimable mem alone then

i don't lose a lot anyway

[obtuse loom]

...or just allocate a stack in the kernel :^)

[viral bison]

i can do that?

[obtuse loom]

I mean

there are a few ways of doing that

but yes

[viral bison]

i already get the stack top

[obtuse loom]

1: Just set up multitasking and exit the init task that's using it

2: Just write to rsp in the entry point, declare it in asm

[viral bison]

oh i do that already

[obtuse loom]

there are probably many more

[viral bison]

all the tasks have their own rsps

[obtuse loom]

yes and when you're done using bootloader data (including that stack) then you could use bootloader reclaimable memory

[viral bison]

k

i might do that in the future then

[hazy saddle]

wait @devout path how should i handle returning systemcall data?
would i just do frame->rax = ...

yeah that works

[devout path]

yes

my hand was filled with cheese i couldn't reply

xd

[hazy saddle]

and now this clobber list can go bye bye now i control registers properly

i wonder what n00b thinks of this

[devout path]

heh

[hazy saddle]

[devout path]

@hazy saddle

do u know zig

the programming language

[hazy saddle]

ive heard of it

but never touched it

[devout path]

oh damn

it's better than C i tell ya

[hazy saddle]

looks too high level

[devout path]

yep

[hazy saddle]

i really like C alot tbh

[devout path]

what's weird is that u declare by making it a constant xd

same

imma hit the sack

mention me here once you're done cleanin up

[hazy saddle]

wait before you do

[devout path]

mm?

[hazy saddle]

what was the things you were gonna work on?

and yeah i will

[devout path]

oh just a simple block driver

so i can have it support ahci

[hazy saddle]

from SATA and ext2 right?

[devout path]

yes

that'll take me

more or less

a week to 3

im slow

[hazy saddle]

nice ;3
what devices will it be selecting lol

also if i make any changes il ping

so the merge conflict isnt an ass

[devout path]

uuh wdym

[hazy saddle]

like what if there are two sata drivers

[devout path]

oh that'll be a problem for me in the future

i'll think of something XD

[hazy saddle]

[devout path]

HEH

[hazy saddle]

have it check GPT parition table for the right UUID?

and pass it in as a limine boot cli arg

[devout path]

hmmmmmmm yeah prolly

mmmmmmmmmmmmmmmmmmmmm

yeah i guess

[hazy saddle]

its how linux does it right

[devout path]

LMAO

i was planning to deviate from what linux does

but if i get extremely lazy

i'd probably just do that

[hazy saddle]

but i mean wouldnt it be good to be able to spesifiy the drive and partition to mount

maybe have it handled in the initramfs?

[devout path]

yeah yeah

[hazy saddle]

maybe init system program?

it has some conf files about drives

[devout path]

maybe hmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmmm

oh

[devout path]

yeah why not

[hazy saddle]

and then that can even start the shell etc

No need for a FULL init system

But having a program to read a config on how to mount the root drive and start the shell?

[devout path]

yes ofc

more practical tbh

[hazy saddle]

That way you can make the conf file run any program

Or have it be a chain of programs to run?

To say mount drive and then run program to setup network card layer

And THEN run the shell

Maybe use TOML for the config that’s easy to parse right

[devout path]

yeah

[hazy saddle]

Or “TOON”

[devout path]

hmmm maybe only

[devout path]

XD

[devout path]

wat

was that sarcastic or

xD

[hazy saddle]

No

Not sarcasm

[devout path]

o h

[hazy saddle]

What’s the issue with TOML

[devout path]

nothing nothing

im just kinda dead rn

4 am

3 am *

[hazy saddle]

gonna need one more test on irl hardware before push

but im almost done reviewing and getting rid of stargglers

[devout path]

nice

[hazy saddle]

next commit will be a simple generic constructor for threads

what do these refer to again

ehhh il figure it out

should work with sysv

worst case inline asm grab the reg

[devout path]

For redundancy xd

[devout path]

Heh

[devout path]

Basically making sure it doesn't go to garbage

[hazy saddle]

can i also get a video recording of the many bad apples for friends?

[devout path]

oh

[hazy saddle]

yes but if i need to pass an argument

[devout path]

Lemme boot my pc again

[hazy saddle]

on real hardware

so leme finish this up

almost there

[devout path]

oke

I lost my flashdrive i can't run this on hardware ;-;

[hazy saddle]

WTF

[devout path]

why

[hazy saddle]

nuuuuuu

@viral bison you still here

[viral bison]

just finished eating

[hazy saddle]

code review looks good enough and its LGTM

[viral bison]

send it if you want me to try

[hazy saddle]

Can you also record a little video of it running many bad apples at once

[viral bison]

i'll try to see if i then can upload it here

[devout path]

Peak

[viral bison]

@hazy saddle

[hazy saddle]

Fuck yeah

[hazy saddle]

@tawdry ingot get in here and look at this

I did it buddy. It’s gonna be pushed to main soon and then you can get inspired

Your kernel is saved

@viral bison this about sums it up right

[hazy saddle]

@devout path its pushed

[tawdry ingot]

Used to pray for times like this

[devout path]

oke

thank u!

[hazy saddle]

No thank you!

[hazy saddle]

Tar.gz as initram FS might be cool at some point ngl

Il add that to the list for far things in the future

[devout path]

yea

[devout path]

you're welcome!

[hazy saddle]

If the initramFS is starting to near ~16MB il start work on compression probs

[devout path]

Aleighty

[hazy saddle]

My next steps are gonna be getting doom working

I think I can compile it as a freestanding app and point it to my headers for a libc and generate a .a file for static linking

As il need to make a really custom libc because I’m missing some semantics it would need

And il have to emulate them

[devout path]

oooh nice nice

[hazy saddle]

Because I don’t have semantics for opening and closing files or seeking or file descriptors

I’m gonna have to have the libc emulate them

And that will handle FDs and map them to the right file operations for the system call side

[hazy saddle]

@viral bison how do the system call instructions work on x86-64 btw

Like how would I be meant to set them up

I know it’s an MSR but isn’t a bit more than that?

[viral bison]

sec

[viral bison]

this is how i do it

Frame and Syscall regs contain basically the same values

[hazy saddle]

What are the values of these magic numbers that I can’t see?

Looks really easy tho

[viral bison]

which ones

the Segments?

[hazy saddle]

And then in userspace it’s just load things into register and call syseenter?

[viral bison]

syscall*

[hazy saddle]

The selectors and the MSR addresses

[viral bison]

warning, the GDT has to be like this

[hazy saddle]

The selectors are offsets to the GDT entires right

[viral bison]

yes

but it HAS to be like this

the layout i mean

[hazy saddle]

Unless I change the offsets to work for me?

Or just straight up

[viral bison]

the msrs are fucked up

they want UCODE32, UDATA and UCODE64 in that exact order

same for KCODE and KDATA

[hazy saddle]

[viral bison]

you can skip UCODE 32 tho

and just set that value to the entry before UDATA

if you don't want to use the syscall instruction in a 32 bit env

[hazy saddle]

My kernel is gonna be 64 bit only

For now atleast

[viral bison]

i do support compat mode too

[hazy saddle]

So why is it user code 32 and kernel code 64 here?

[viral bison]

(or well, it should)

[viral bison]

it sets the base entry

then as i said, the msrs are fucked up

so it gets the other entries by hardcoded offsets

[hazy saddle]

This looks like a mess but yeah

[viral bison]

it's not really

[hazy saddle]

I don’t know the magic numbers I have to fill in here

[hazy saddle]

I mean the order

[viral bison]

just change the Long Mode bit on ucode

for the compat mode

every entry is +0x8

[hazy saddle]

Ok the wiki has the right numbers

Il change to #defines and then use them

These are the 32bit ones

[hazy saddle]

And then I just need this order

[viral bison]

yes

[hazy saddle]

Plug in the MSRs real quick and have a basic assembly handler and we are all good

https://tenor.com/view/dog-accepting-fate-tiktok-at-peace-gif-567488341419321713

Should be easy

[viral bison]

we hope

[hazy saddle]

20 minute quick adventure in and out

@devout path i may wana make a dev server for github commit notis and shit so its easier for you if i am pushing stuff when you work

or just in general

i dunno if the mods here could get a web hook in a thread?

wait

wtf do flags match to

@viral bison anything wrong with this

the wiki values dont wana work

even just for 64 bit ones

wait why did this skip 4??

[viral bison]

i think a flag is wrong here?

?

XD

[hazy saddle]

the flags should be fine

[viral bison]

then wdym it doesn't work

[viral bison]

well, the last 2 are flipped if you wanna use them for syscalls

[hazy saddle]

triple fault

[viral bison]

when?

[hazy saddle]

when do this

then would this jsut work™

[viral bison]

if you set the right offsets yes

aren't the flags wrong tho

[hazy saddle]

This is my original code

That did work

But with it swapped

[viral bison]

well, did you edit all the selector values too?

like, in every other asm file

you should have to edit: GDT asm file and the ctx switch

[hazy saddle]

I don’t have a GDT asm file

[viral bison]

well, the code you use to load it

[hazy saddle]

What code do I need to change? The segments were at the same offset for atleast trying to replicate the wikis version

[viral bison]

well, you did swap ucode and udata

did you also swap their values in the loading / ctx switch?

[hazy saddle]

Wait what

[viral bison]

if previously udata was 0x30 and ucode was 0x28

now it's the opposite

[hazy saddle]

But its faulting even before my context switch code

But yeah I’d have to change that I think

[viral bison]

well, i think you did not swap those values in the gdt loading

[hazy saddle]

I’m
Too sleepy for this rn

[viral bison]

oh wait

i don't think you have to swap them there

hmm

i think the flags are wrong

[hazy saddle]

Then why do these flags work when 5 and 6 are swapped

[hazy saddle]

And these don’t

[viral bison]

try to copy mine

[devout path]

oke

[devout path]

ohhhhhhhhhhhh

[hazy saddle]

@devout path do you know if this is correct

I’m trying to get system calls to use the right instruction

[hazy saddle]

What is : 4 mean

[viral bison]

it's just 4 bits

[hazy saddle]

So like a compiler hint?

[viral bison]

no, it sets the size to 4 bits only

[hazy saddle]

Oh you can’t do that in C

Or can you?

You can

Huh neat

wait @viral bison how do these constructors work to mappings things out

[viral bison]

wdym

[hazy saddle]

i mean how does c++ map the consttor arguments to parts of the struct

this shoiuld be okay ho

void gdt_set_tss (int num) {
    gdt[num].limit_low = sizeof(tss);
    gdt[num].base_low = (uint64_t)&tss & 0xffff;
    gdt[num].base_mid = ((uint64_t)&tss >> 16) & 0xffff;
    gdt[num].access = 0x89;
    gdt[num].flags = 0;
    gdt[num].base_high = gdt[num].base_mid = ((uint64_t)&tss >> 24) & 0xff;
    gdt[num].base_high = gdt[num].base_mid = ((uint64_t)&tss >> 32);
}

[viral bison]

srry here

after the :

[hazy saddle]

https://tenor.com/view/wii-wii-sports-wii-bowling-bowling-bowling-strike-gif-24891879

TRIPLE FAULT!

[viral bison]

try to set the flags like i do

also

ltr is just wrong

ltr should be loading 0x18

oh wait no

you changed it

then yes it's ok

[hazy saddle]

then why triple fault

[viral bison]

idk

[viral bison]

we need the full log

[hazy saddle]

what log its faulting i cant get anyb data out

[viral bison]

.!c qemu

.!t qemu

[nova edge]

qemu

Run QEMU with following flags: -M accel=tcg,smm=off -d int -no-reboot -no-shutdown

[hazy saddle]

i did fix this line gdt[num].base_higher = gdt[num].base_mid = ((uint64_t)&tss >> 32);

[viral bison]

here

[hazy saddle]

[viral bison]

also -D log.txt

[devout path]

wat wat happened

[devout path]

h u h

[hazy saddle]

i did find one minor issue

the base_high vs base_higher

@devout path @viral bison

thats blocked by the server?

[viral bison]

no

[floral kettle]

yeah it got automodded

[hazy saddle]

damn

[viral bison]

i downloaded it

well, here are the 2 exceptions

[hazy saddle]

thats a tripple fault no?

[viral bison]

doesn't seem like so

[hazy saddle]

i mean without the flags it auto reboots

[floral kettle]

then its a triple fault

[viral bison]

but i only see 2 exceptions there

[hazy saddle]

last line

[devout path]

i'll take a look at this tmr if it's not fixed yet

[viral bison]

check RIP ffffffff80001e37

[devout path]

im still tryna figure out why my fix worked yesterday

[devout path]

• im ded rn fr

absolute dead

[hazy saddle]

OBJ dump my .elf?

[viral bison]

addr2line -fai -e <exe> <addr>

[hazy saddle]

/home/evalyn/Documents/Programming/C/evalynOS/src/kernel/drivers/x86_64/gdt.c:73

[devout path]

i on't think doin any static analysis would come up to somethin useful

ok im gonna regret saying that but

nvm

ok nice

[viral bison]

well, what's there?

[viral bison]

try to do it in pure asm

and stop using inline

[hazy saddle]

[devout path]

wat

oh nvm

[viral bison]

• the asm above

[hazy saddle]

https://tenor.com/view/wii-wii-sports-wii-bowling-bowling-bowling-strike-gif-24891879

TRIPPLE FAULT

[floral kettle]

this is correct

[viral bison]

wtf

[hazy saddle]

Isn’t my kernel lovely

[viral bison]

did you set the ist of every other interrupt to a non 0 or 1 value?

[hazy saddle]

I don’t think so

[viral bison]

that might cause problem

[viral bison]

move the df stack to like 2

and leave 0 unallocated

[hazy saddle]

like this

[viral bison]

yes

[hazy saddle]

https://tenor.com/view/wii-wii-sports-wii-bowling-bowling-bowling-strike-gif-24891879

[viral bison]

but remember to update it too

[viral bison]

where you set the ints

[hazy saddle]

TRIPLE FAULT

[viral bison]

ehm

[hazy saddle]

What

[viral bison]

you set DF to 2 here

[hazy saddle]

shit

[viral bison]

while on the ist screenshot you set it to 1

(this is not the cause btw)

[hazy saddle]

still TFs

[viral bison]

yeah well

we have to find out why it TFs

[viral bison]

how is the stack represented btw

[hazy saddle]

would this be proper qwinci?

struct __attribute__((packed)) GDTEntry {
    uint16_t limit_low;
    uint16_t base_low;
    uint8_t  base_mid;
    uint8_t  access;
    uint8_t  limit_high : 4;
    uint8_t  flags : 4;
    uint8_t  base_high;
    uint32_t base_higher;
    uint32_t _reserved;
};

void gdt_fill_entry (int num, uint8_t access, uint8_t flags) {
    gdt[num].limit_low = 0;
    gdt[num].base_low = 0;
    gdt[num].base_mid = 0;
    gdt[num].access = access;
    gdt[num].limit_high = 0;
    gdt[num].flags = flags;
    gdt[num].base_high = 0;
    gdt[num].base_higher = 0;
}

void gdt_set_tss (int num) {
    gdt[num].limit_low = sizeof(tss) - 1;
    gdt[num].base_low = (uint64_t)&tss & 0xffff;
    gdt[num].base_mid = ((uint64_t)&tss >> 16) & 0xffff;
    gdt[num].access = 0x89;
    gdt[num].limit_high = 0;
    gdt[num].flags = 0;
    gdt[num].base_high = gdt[num].base_mid = ((uint64_t)&tss >> 24) & 0xff;
    gdt[num].base_higher = gdt[num].base_mid = ((uint64_t)&tss >> 32);
}

[hazy saddle]

__attribute__ ((aligned (16))) uint8_t kernel_stack[16384];

__attribute__ ((aligned (16))) uint8_t user_stack[65536];

__attribute__ ((aligned (16))) uint8_t df_stack[65536];

__attribute__ ((aligned (16))) uint8_t nmi_stack[65536];

[viral bison]

ah

65k is too much i think btw

i only gave it 4k

anyway, let's go back to that TF

[floral kettle]

the gdt entry is too big

it should only be 64 bits

and the 64-bit int following the actual tss gdt entry is what you should set to tss_addr >> 32 like you did in the old code you had in git

[viral bison]

like this

[hazy saddle]

i copied this for that

[viral bison]

XD

that is the TSS one

lol

[floral kettle]

you don't need a separate one for the tss one anyways, if you really want to keep the gdt entry split to separate fields like that instead of just one u64 you could make it an union of a struct that contains the normal fields and then just a plain u64

and then do smth like ```c
gdt[index].entry = make_gdt_entry(...);
gdt[index + 1].tss_addr = tss_addr >> 32;

[hazy saddle]

whats a union

[floral kettle]

you haven't ever used an union in c?

[hazy saddle]

nope my first time

but i think i got it?

https://tenor.com/view/bowling-worst-luck-ever-good-shot-just-my-luck-wtf-gif-21072949

it works

unions are cool

wait nvm

https://images-ext-1.discordapp.net/external/YMr2RokKWhrwt2ag5A5wizR0l_XiGQGf8MMnKvSgEDs/https/media.tenor.com/KZDJK3uiY54AAAPo/wii-wii-sports.mp4

i think i got this one though

@viral bison

[viral bison]

New stscalls?

[hazy saddle]

Not yet

But working GDT

[viral bison]

Gg

[hazy saddle]

Could you re give the screenshots for the MSRs and stuff for the new syscall thing

[viral bison]

Sec, i'm finishing my pizza

[hazy saddle]

Oooo what kind

[viral bison]

with ham

and homemade

[hazy saddle]

Oooooo

[viral bison]

here

[hazy saddle]

thank you

[hazy saddle]

arent msrs a u32?

Note that the Kernel does not automatically have a kernel stack loaded. This is the handler's responsibility.
@viral bison should i just grab this from the threads->stack_top?

[viral bison]

they should be

i think

or i might be doing it wrong

[hazy saddle]

yours says u64

[viral bison]

sec

u64 yeah

[viral bison]

yes

[hazy saddle]

[viral bison]

you also need to enable again interrupts After you loaded rsp

[viral bison]

yeah mine just combines the value into 1 u64

and split it inside

[hazy saddle]

i mean for the first arg

[hazy saddle]

ima do that frfr rn

[viral bison]

it should be u64

[hazy saddle]

the osdev wiki says otherwise

[viral bison]

it's a 64 bit reg

[viral bison]

don't mind it

i think it's still in 32 bit

lol

[hazy saddle]

XD

[viral bison]

or well, maybe MSRs are all 32 bit for IDs

i just use 64 bit for convenience

[rancid viper]

there are no 32bit MSRs

where on the osdev wiki

[hazy saddle]

for the ID

[rancid viper]

oh

that is 32bit

[hazy saddle]

here we go

[viral bison]

huuh

rdmsr is wrong

it's the same as wrmsr XD

it's reading

so it should be returning stuff

[hazy saddle]

this is the wikis version

[viral bison]

that is using pointers

[hazy saddle]

ah

[viral bison]

you are not

[rancid viper]

here is a C one ```c
static inline uint64_t x86_64_msr_read(uint64_t msr) {
uint32_t low;
uint32_t high;
asm volatile("rdmsr" : "=a"(low), "=d"(high) : "c"(msr));
return low + ((uint64_t) high << 32);
}

[viral bison]

yeah i sent it above

[hazy saddle]

bonk me

[rancid viper]

c != c++

[viral bison]

it's basically the same

nothing changes

just the cast

[hazy saddle]

wait i also need to set bit zero here

[viral bison]

huuh

i might have to do it too

[floral kettle]

me when __readmsr(SOME_MSR_MACRO) intrinsic

[rancid viper]

xd, is that a clang one

[viral bison]

oh it exists???

[viral bison]

i think the bootloader does this tho

[hazy saddle]

you can never be sure

[rancid viper]

limine doesnt at least

[viral bison]

ah

what does limine enable in that register?

[rancid viper]

LME

[viral bison]

only?

[rancid viper]

afaik

yeah

[viral bison]

oh

i'll enable all the features then

[rancid viper]

could double check the proto, I dont use limine after all

[floral kettle]

msvc lol (though clang's intrin.h implements that too, idk if you can use that with other targets than windows tho might not be able to because of sizeof(long) being different)

[rancid viper]

ah 😄

[viral bison]

does fast FXSAVE enable any other instruction?

[viral bison]

and LMA?

[rancid viper]

that gets enabled by the cpu

[floral kettle]

lma isn't something you can set yeah

[rancid viper]

seems like I was almost correct

NX is also set

[viral bison]

k

[rancid viper]

I forgot that was in efer

lol

[viral bison]

so what does fast fxsave do?

and can i enable it always if i have fxsave?

[rancid viper]

im not sure, I dont enable it at least

prolly gotta read the sdm for that answer

[hazy saddle]

    uint64_t efer = rdmsr(EFER);
    efer |= (1 <<  0);
    efer |= (1 << 11);
    wrmsr(EFER, efer);

[viral bison]

hmm

[viral bison]

don't do 11

[hazy saddle]

that enable NX right

[viral bison]

limine enables that for you if the cpu supports it

[rancid viper]

yep

[hazy saddle]

[rancid viper]

I just panic out if NX isnt supported iirc

lmao

[viral bison]

btw @rancid viper

[rancid viper]

🤷‍♂️

[floral kettle]

I just unconditionally use it without even panicking would have to be a very old cpu for it to not support nx

[rancid viper]

yep

I check a bunch of cpuid bullshit in init somewhere

so happens to be there iirc

[hazy saddle]

if your CPU dosnt have NX get a cheap used office pc from ebay for cheap

[rancid viper]

will prolly come across that code soon, im revamping my init system

[floral kettle]

for the fxsave bit as far as I can tell the intel sdm doesn't say its even there, it falls within a reserved range so it might be amd only

[viral bison]

done

[floral kettle]

its probably nothing you should worry about (and you should prefer xsave/xrstor to fxsave/fxrstor anyway if they are present)

[viral bison]

i sent the explanation from the osdev forum above

[hazy saddle]

this is what i do for fxsave and it works on intel. i enabled it to fix a crash when i had kernel sse

[rancid viper]

yeah seems to be an amd meme

[hazy saddle]

i dont use kernel sse anymore tho

[viral bison]

hmm

[rancid viper]

anymore?????

xD

[hazy saddle]

[viral bison]

how i do it

[rancid viper]

I have this routine that I for the most part borrowed from somewhere

[viral bison]

also, this is cursed

but works

lol

[hazy saddle]

saving this rq

[rancid viper]

#define true (__rdtsc() % 2 == 0)

[viral bison]

?

rng boolean?

[rancid viper]

boolean

[hazy saddle]

wrmsr(SFMASK, ~0x2); why is this a NOT 0x2?

[viral bison]

so you disable every rflag except res1

[hazy saddle]

ah

[hazy saddle]

@viral bison ?

void init_syscall() {
    // enable syscall instruction
    uint64_t efer = rdmsr(EFER);
    efer |= (1 <<  0);
    wrmsr(EFER, efer);

    uint64_t star = ((uint64_t)0x18 < 48) | ((uint64_t)0x08 << 32);
    wrmsr(STAR, star);

    wrmsr(LSTAR, (uint64_t)syscall_handler);
    wrmsr(SFMASK, ~0x2);
}

void execute_syscall(struct syscall_frame* frame) {
    asm volatile (
        "movq %0, %%rsp"
        :
        : "r"(get_current_thread()->stack_top)
        : "rsp"
    );

[viral bison]

do you have a user code 32?

[hazy saddle]

yes

[viral bison]

k, it should be ok then

[hazy saddle]

then i swap this out

[viral bison]

huh?

it's already ok

[hazy saddle]

nono

from int 0x69 to syscall

[viral bison]

i don't see an int 0x69 there

[hazy saddle]

it was

[viral bison]

wait, i think the mov rsp here is a bit bad

idk if it'll work

[hazy saddle]

yeah shit

il just not inline it

its easier

change_stack:
    mov rsp, rdi
    ret

wait what else did i have to do other than chanign stacks? @viral bison

it still GPFs

at in usercode

[viral bison]

i think you are still changing the rsp wrong

[hazy saddle]

huh

[viral bison]

tho it's strange that it happens in usercode

what do the error bits say?

[hazy saddle]

external IDT

wait leme check somthin

[viral bison]

strange that is failing in the IDT

[hazy saddle]

wait

lmfao