EvalynOS

[hazy saddle]

I am back to the original way with the timer calling the schedular

I have got stack smashing protection implemented now and I made the panic work a bit better

It now goes through assembly for panics not from interupts

And it will build a frame and pass it to the function for interupts

So it’s a lot cleaner for things

[rotund furnace]

no error code for gpf :P

oh nvm you do print the error code

selector index 4 in your gdt is cooked

or not wtf why is bit 16 set?

[hazy saddle]

i dont fucking know

[rotund furnace]

idk something is broken

[hazy saddle]

https://codeberg.org/EvalynGoemer/evalynOS/src/commit/567dbece0eb2bc9a291334c0f0ef1bedf0eaadf6/src/kernel/utils/panic.c#L241
https://codeberg.org/EvalynGoemer/evalynOS/src/commit/567dbece0eb2bc9a291334c0f0ef1bedf0eaadf6/src/kernel/utils/panic.c#L122

this is how it prints that btw

and im 99% its corrrect

index 4 should be okay right?

[rotund furnace]

ignore it

[hazy saddle]

https://codeberg.org/EvalynGoemer/evalynOS/src/commit/567dbece0eb2bc9a291334c0f0ef1bedf0eaadf6/src/kernel/drivers/x86_64/gdt.c#L42

[rotund furnace]

for now atleast

the errorcode is trashed for some reason

oh found it

error > 16

😭

[hazy saddle]

XD

why tf is that there

how did that get there

how long as that been there

[rotund furnace]

i assume that's meant to be >>

[hazy saddle]

yup

here we go thats better 💀

[rotund furnace]

alright

so yeah selector index 4 in your gdt is borked

[hazy saddle]

Il push to git tomorrow with that fix if this isn’t fixed tonight

Sleeping in about an hour

[hazy saddle]

[rotund furnace]

lemme check

uhhh

why are your limits 0

aren't they meant to be 0xfffff

[hazy saddle]

They are ignored on 64bit though right

[rotund furnace]

maybe

yeah

also npf has binary support btw

[hazy saddle]

Oh shit really

[rotund furnace]

yeah

[hazy saddle]

@rotund furnace also still dies

[hazy saddle]

What’s its syntax for it

[rotund furnace]

works on my macine 😭

[hazy saddle]

WHAT

[rotund furnace]

unless you didn't push something

[hazy saddle]

Dev branch is up to date

[rotund furnace]

😭

I cloned dev lmao

[hazy saddle]

What the fuck 😭

@rotund furnace Does it run on real hardware?

[rotund furnace]

I'll have to check

[hazy saddle]

Is this a fucking distribution or qemu or toolchain bug?

Everyone has been banging at thier computer over this

AND IT WORKS FOR YOU 😭

https://tenor.com/view/slam-smash-computer-slow-rage-gif-15806211

[rotund furnace]

lmfao

[hazy saddle]

@devout path @viral bison look at this shit XD

[rotund furnace]

how do you boot this shit on real hardware

[hazy saddle]

@rotund furnace To be fair it did make me add a good chunk of debugging shit for the dev branch

[rotund furnace]

you don't make an iso file

[hazy saddle]

Make a USB that’s GPT formatted with a FAT32 partition with BOOT ESP set

And copy over the EFI folder the limine config the kernel.elf and initramfs

[rotund furnace]

kk

[hazy saddle]

@rotund furnace What’s the prognosis doc?

[rotund furnace]

im dealing with linux being stupid hold on

[hazy saddle]

All good

[rotund furnace]

ok well the usb drive I tried it on is dead ig

[hazy saddle]

[rotund furnace]

what the fuck why is this shit so broken

ok well it has the files on it so fuck it

[hazy saddle]

Holy shit

How long is it running

It’s actualy stable?

Can you run bad apple twice to have two of them on?

[rotund furnace]

how the fuck do I run it twice lmao

[hazy saddle]

Even if the logs fly by it should still track key presses right

[rotund furnace]

oh lmao

[hazy saddle]

Just do “badapple” again

[rotund furnace]

ah it fails on my macbook

it doesn't even boot

[hazy saddle]

MacBooks don’t have a PIT

[rotund furnace]

(intel mac)

[hazy saddle]

Iirc

[rotund furnace]

oh it failed on serial

[hazy saddle]

Lmao

[hazy saddle]

But holy shit it’s been working this entire fucking time 😭

What’s wrong with my computer

[hazy saddle]

how dows my compiled version work

I wana check if it’s my QEMU or if it’s my compiler toolchain

[rotund furnace]

let me see if it finishes

[hazy saddle]

It will loop btw

[rotund furnace]

oh

[hazy saddle]

I don’t have a way for apps to uhhh exit

[rotund furnace]

well if it does that's a success ig

[hazy saddle]

But now I’m scared on what the fix it to get it working for my PC

[rotund furnace]

they are diff binaries

so idk

mines smaller????

anyway yeah using that binary works

and it's looped like once or twice by now

[hazy saddle]

Huh

Thats the one I complied with a few changes for the GDT

And some formatting changes

So WTF

[rotund furnace]

well it works on my machine lmao

[hazy saddle]

What’s your kernel version and QEMU version and destro?

[rotund furnace]

qemu-system-x86_64 -version?

[rotund furnace]

6.17.8-2-cachyos and qemu 10.1.2

[hazy saddle]

6.18.1-2-cachyos
QEMU emulator version 10.1.2

What the fuck

[rotund furnace]

LMAO

my kernel is out of date

on a rolling release distro

[hazy saddle]

Do an update XD

[rotund furnace]

yeah you right

[hazy saddle]

oh shit

Free gigabyte of storage space thank you

@rotund furnace I mean it’s also broken on TCG

So fat chance it’s a Linux kernel bug right

[rotund furnace]

hmmm what

[hazy saddle]

(Imagine if we found a Linux kernel bug)

[hazy saddle]

If you tell QEMU to emulate instead of virtualization

[rotund furnace]

well it works on my machine 😭

with tcg

[hazy saddle]

what

the

fuck

Did you do the kernel upgrade and reboot and test?

[rotund furnace]

it gonna take a while to update my shit rq

[hazy saddle]

We are both on the same distro and it should fucking work

[rotund furnace]

not a problem on nixos btw

[hazy saddle]

Immutable or somthing

[rotund furnace]

wait but does it work on real hardware for you?

[hazy saddle]

I don’t have a USB stick

It’s fine

Gone

[rotund furnace]

damn

[hazy saddle]

I’m on a 5700x tho

[rotund furnace]

im on a i7-11700

I feel like that wouldn't change anything

[hazy saddle]

Yeah

[rotund furnace]

that would be crazy if it did

[hazy saddle]

Plus syscall/stsret is an amd64 thing

That was adopted by Intel

[rotund furnace]

after they ditched sysenter/sysexit

oh well

stange

im updating 600 packages atm

[hazy saddle]

I have 1gbps internet so it’s fast as fuck unless you have one bloated ass package looks at amd rocm

[rotund furnace]

AHAHAHAHAH

[hazy saddle]

And it’s symmetrical fiber too

[rotund furnace]

gigabit

if only

[hazy saddle]

I used to be on 10mbps upload for years

[rotund furnace]

so uh quick question what do you do on boot failure

because I forgot to shutdown my laptop

and now it's hot to the touch lmao

[hazy saddle]

I think I have forgotten some halts in some places 💀

[rotund furnace]

wait it died on serial

which isn't fatal

oh it spun somewhere

im gonna assume ps2 setup because it's right after

[hazy saddle]

Ohhh it’s probaly stuck spinning checking serial port?

[rotund furnace]

no it printed that it failed to init serial

[hazy saddle]

That’s normal

If there is no seriel port it says that

It’s prolly this

Because I don’t have a timeout here

[rotund furnace]

oh prob

😭

yeah just burn my cpu while you are at it

[hazy saddle]

SORRY 😭

I think I did leave off some halts on critical panics as a way to detect if it crashed while you were away

“If the fans are loud something is prolly wrong”

(Not the best idea)

[rotund furnace]

MY MACBOOK DOESN'T HAVE FANS

😭

[hazy saddle]

I also have a way to make the kernel panic make a Wii noise when it dies

[rotund furnace]

that also wouldn't work on my pc because the fans are also at 100% always

[hazy saddle]

https://youtu.be/_FGxzfsQ3B0

If the panic handler panics you can have it make a noise like this

[rotund furnace]

oh im sure that doesn't get annoying

[hazy saddle]

That’s why I made it a toggle 💀

@rotund furnace How did the update go

[rotund furnace]

oh it just finished

sick

now I gotta update my hyprland plugins 😭

yeah it still works

I even rebuilt and it still works

[hazy saddle]

The fuck

What’s wrong with my computer

[rotund furnace]

lmao

[hazy saddle]

@rotund furnace How do I switch to the dev branch and get the proper files?

Because I cloned it and switched to dev and it dosnt have the updates

[rotund furnace]

iirc you need to pull after switching

[hazy saddle]

I did a reset and it worked

[rotund furnace]

the kernel?

[hazy saddle]

And it’s still fucked

The kernel broke

[rotund furnace]

wtf?

[hazy saddle]

I updates all my packages

[rotund furnace]

i just did git clone ssh://[email protected]/EvalynGoemer/evalynOS.git --branch dev

[rotund furnace]

cpu bug?

[hazy saddle]

Shits bad

[hazy saddle]

I guess?????

[rotund furnace]

nah

wouldn't be

something must be broken

[hazy saddle]

Who do we talk to about this

I feel like this is above my pay grade

[rotund furnace]

I would help but I can't even reproduce the bug?

try cloning the repo on the dev branch and seeing if that works

[hazy saddle]

Do you have any amd systems on zen3?

[rotund furnace]

nope sadly all intel

[hazy saddle]

Yeah maybe???

I did try dev branch clean

@rotund furnace

still fucked

[rotund furnace]

what the fuck

[hazy saddle]

I did this with your exact command

Then cd into the folder

And then make

[rotund furnace]

oh the error code is broken again 😭

[hazy saddle]

What 😭

Oh that wasn’t pushed yet

The > 16 to >>16

[rotund furnace]

oh yeah it's broken in my build as well

[hazy saddle]

So uhhhh well shit

[rotund furnace]

yeah

uh

very much

[hazy saddle]

I’m cursed?

My code is an SCP now?

SCP 0xdeadbeef

[rotund furnace]

maybe try reinstalling qemu

because it seems like a qemu problem if not even tcg works

[hazy saddle]

which qrmu package did you get

[rotund furnace]

pacman tells me local/qemu-system-x86 10.1.2-3

[rotund furnace]

what the fuck

[hazy saddle]

sudo pacman -Rs qemu-full 
sudo pacman -S qemu-full

That’s what I did

[rotund furnace]

😭

[hazy saddle]

What the fuck

@rotund furnace On a scale of one to 10 how fucked am I

[rotund furnace]

I have never seen this before so you get to name it

[hazy saddle]

Ligma

My kernel has Ligma

[devout path]

y was i mentioned here xd

[devout path]

ligma ballz

[hazy saddle]

kernel not work for me

but work for water

@rotund furnace it works in TCG for me now

just not KVM

[rotund furnace]

what the fuck

[devout path]

H u h

What the hell is happening to ur kernel

[hazy saddle]

fucked

[dense cargo]

rainbow kernel panic

[hazy saddle]

alright i am back to do more work

im gonna checkout an older commit before i redid interupts

just to see if somthings iffy with thoes

im on commit a84fd7997a integrating newer code and ive got the GDT in and i found some potental issues with it now as i had to change some stuff

[hazy saddle]

yeah idk where i was goin with that. i dont have half the things i need there because of pure c based innterupts

[hazy saddle]

thats an uhh not so happy stack for sysretq

it somehow diddnt fault on that?

ohh thats sysretq

not iretq

why is it working if i single step in GDB
and it skips over the syscalls and wont show me what happens when the IP changes

@rotund furnace you mentioned it worked on real hardware before right? leme test this myself

it only crashed when i put continue??

Yeah it crashes on real hardware for me @rotund furnace

Let me test on an Intel machine

[rotund furnace]

what the fuck

I think you are doing some ub fuckery that just works on intel

I don't have any amd hardware so I can't test that

[hazy saddle]

Yeah that’s what I’m trying to check

But my mom isn’t responding and she is the only one with an Intel laptop

[rotund furnace]

thats like the only thing I can think of

it would have to be some fucking ub that causes something

[hazy saddle]

But what about TCG

[rotund furnace]

hm

yeah

[hazy saddle]

@rotund furnace I did get an ISO file generated manually and I may automate it soon but I sent it to some friends to test

Gonna try and get a better sample size

It did triple fault on KVM on my laptop too @rotund furnace

Which is strange

Just right away

Let me see how it goes on my laptop bare metal

[rotund furnace]

maybe implement ubsan and see if that says anything?

[hazy saddle]

Damn my laptops USB ports are being funky

[hazy saddle]

How do you do that?

I did get stack smashing protection stuff added

[rotund furnace]

just add -fsanitize=undefined to your compiler flags

then just implement all the functions it tries to link in but fails because you don't have the lib for it in freestanding

you also should add __attribute__((no_sanitize("undefined"))) on the ubsan functions

[hazy saddle]

https://tenor.com/view/slam-smash-computer-slow-rage-gif-15806211

The ISO won’t boot on any other machine

[hazy saddle]

Il do this real quick

Why is the drive stick at 6MB max now

Stuck

WTF

Nvm gparted bug

@rotund furnace Yup it works on Intel

I wonder if anyone else has had this kind of issue?

I doubt I can google syscall amd vs intel because of the 32bit thing

[rotund furnace]

wait does it crash on sysexit???

[hazy saddle]

iretq

[rotund furnace]

oh

[hazy saddle]

@rotund furnace what arguments should they be taking?

[rotund furnace]

oh because in the amd manual it says

and you don't do that

[hazy saddle]

whats RPL again? thats the MSR right

[rotund furnace]

the rpl bits of the selector iirc

[hazy saddle]

that would be when doing GDT things right and changig thoes

[rotund furnace]

well it's telling you that for the IA32_STAR msr you need to set the rpl of the ring3 selectors to 11b

and you don't do that

so maybe you should

[hazy saddle]

should my panic also be unsanitised

[rotund furnace]

uh maybe

[hazy saddle]

so i need to do an OR on thoes bits to set them

[rotund furnace]

yeah

[hazy saddle]

leme finish UBASAN

its good to have anyways

UBASAN is not hit

[hazy saddle]

you know a good tool to selete bitfield things and get a hex value

[rotund furnace]

for what?

[hazy saddle]

like to get the hex value to OR with

[rotund furnace]

oh python ig

[hazy saddle]

i mean like to select checkboxes for bit numbers

[rotund furnace]

oh

[hazy saddle]

i dont wana do like 0b00000001100000000 or some thing

[rotund furnace]

wux made https://osdev.thenest.dev/

and that's pretty nice

[hazy saddle]

danke

here we go

HOLY SHIT

IT WORKS

[rotund furnace]

intel doesn't say anything about having to set the rpl to r3

fr?

what did you change

[rotund furnace]

ohhh

you could like

not do that

[hazy saddle]

Just the OR

[rotund furnace]

and just do (0x18 | 3) << 48

[hazy saddle]

[rotund furnace]

this is why you keep both the intel and amd manuals on standby

intel clearly doesn't give a shit if you load a ring0 selector

[hazy saddle]

AMD certainly gives a shit though

Too many shits

Too many god damn shits

[rotund furnace]

I mean

amd made syscall/sysret

[hazy saddle]

True

[rotund furnace]

so they are in the right here and intel is wrong

[hazy saddle]

atleast this made me get stackguards and UBASAN

And UBASAN catches it before the stack smashing protector

[rotund furnace]

so there is ub 😭

[hazy saddle]

No

That’s the command that triggers UB

On purpose

[rotund furnace]

oh

[hazy saddle]

I have to test it to make sure it works

@rotund furnace thanks for helping me find this

i would of never found this due to my team redness

@rotund furnace i wonder why it was so intermitent and heisen bug like though

but holy shit i can actualy start work on a real userspace and libc and stuff

[rotund furnace]

All good

[hazy saddle]

wait this sounds bad out of context

[rotund furnace]

This is why you test on more then just intel or amd

I mean too be fair it only works on intel because they implemented it wrong

[severe marlin]

Evalyn is a communist confirmed (I'm pretty sure you meant in terms of us politics, but communism was my first thought lol)

[rotund furnace]

Pretty sure they were just talking about amd

[severe marlin]

Yeah but the out of context bit

[rotund furnace]

Oh

[hazy saddle]

The other red

Yeah USA

But yeah I meant AMD

What’s funny is my first laptop had an AMD APU

Second laptop I get with an Intel CPU and an nvidia GPU

The screen is busted in a few days nothing I did

[severe marlin]

Yeah fair I'm also on amd

[hazy saddle]

I had to return it and I got my current AMD laptop with an nvdida GPU

And now I have an AMD desktop CPU and GPU

[severe marlin]

I have an amd + nvidia desktop and then now an amd + integrated laptop as well

But my old laptop was intel and then the power button stopped working

I was full ready to replace the button and solder on a new one but unfortunately it was an issue in the motherboard somewhere and the button itself was fine

[mild glacier]

imagine that ascii art being your "oops! kernel panic! system will now explode-"

[hazy saddle]

That is my kernel panic ascii art lol

[hazy saddle]

This is a kernel panic triggered by undefined behavior sanitization

[hazy saddle]

@rotund furnace It’s now on codeberg and my git server

Thank you git merge —squash

[hazy saddle]

its building at the very least for my kernel
-rwxr-xr-x 1 evalyn evalyn 13M Dec 23 20:41 ./doomgeneric

yeah but no

maybe i should just use mlibc

instead of whatever the fuck this jank is

[mild glacier]

yay i love it
doesnt seem like it fits in 80x25 mode D:

[mild glacier]

i heard something about an iso but just to make sure
it can run on hardware, yes?

[hazy saddle]

yes but i dont build isos by default

you manualy parition a USB and drop files on it

[mild glacier]

sounds like my solution to a distro
just throw files into a partition and boot it via GRUB lol

[hazy saddle]

it should with minor edits but take up most of your screen

i use an 8x8 font size

[mild glacier]

oh so it does fit in 80x50
coz im using 80x25 which is most definitely 16x8 px

[hazy saddle]

my screen is more than 80x50

or it can be

[mild glacier]

rly? how many chars w/h

[hazy saddle]

i have a raw frame buffer

[mild glacier]

ooo

[hazy saddle]

its given for free with 64bits

• limine

[mild glacier]

oh
brah

[hazy saddle]

and flanterm and nanoprintf give you proper tty stuff

i did have my own implementations before

but they werent as good / complete

[hazy saddle]

The cross compiler can be built properly now!

I just have to finish porting mlibc and making an elf loader

[hazy saddle]

iso generation is working

the new GCC is now building apps for the kernel properly with mlibc

i just need to make a basic elf loader now

[hazy saddle]

eyyy the first header is getting parsed

[mild glacier]

that's only 7 floppies in size!! (or 6 idk)

[hazy saddle]

6.40796444 floppies

being (8.8 mebibytes) / (1.44 megabytes)

[mild glacier]

ok so 7 flops

[mild glacier]

i think we should convert 8.8 MiB to MB or 1.44MB to MiB because MiB =/= MB

[hazy saddle]

i forget of floppies are MiB or MB

but my elf loader is elfing

[mild glacier]

oh wait um

[hazy saddle]

but the libc isnt libcing

[mild glacier]

IT CAN PLAY BAD APPLE?

[hazy saddle]

yes?

[mild glacier]

EvalynOS is a real OS now o7

[hazy saddle]

not until doom

which needs my libc to work

its being funky

[mild glacier]

right...
but it is a real media player now 🤣

i need video footage of EvalynOS playing Bad Apple pls

[hazy saddle]

I have a video somewhere

[hazy saddle]

The video data format is specialized for bad apple

[mild glacier]

[hazy saddle]

IVE DONE DID IT

MLIBC IS PORTED!

[hazy saddle]

@mild glacier

[hazy saddle]

Specially the format is a large array of bytes. And an array that had indices into the data array for every frame

And it reads in pairs

First byte is how many pixels to draw

Second byte is the color

It’s run length encoding

And I formatted bad apple to be purely black and white so it could compress losslessly decently with this

[mild glacier]

oh damn RLE
i should use that for my text distro video graphics
but i would do it like so:
bit 1 = color, bits 7-8 = amount then an optional byte for the type of block
e.g. 0b10101010 = print white 42 times in a row
for monochrome coloring :D

[hazy saddle]

Oo that could actually be more efficient than two bytes.
Could also do a bitstream and have start and stop sequences for better encoding

[mild glacier]

yea like 0x0000 | 00FF that prints 256 pixels in a row starting from top-left (pos 0000 to pos 00FF)
then define ur resolution

[hazy saddle]

If it’s a protocol for sending to a terminal RLE wouldn’t be too useful

This is used to compress it so it takes up less space on disk

[neon drum]

I have question: Why no WC on VM?

Keybard

[mild glacier]

mmmmm

[hazy saddle]

It hurts performance on my machine

[hazy saddle]

Fuck

[neon drum]

Weird, usually it increases performance because the writes are done in bulk

But eh, idk

[hazy saddle]

It’s a visualization difference I presume

[hazy saddle]

Others hadn’t had the same issue either

But ehhh

It’s my OS XD

[neon drum]

Real lmao

[hazy saddle]

But yeah doom is working a now

Discord

Please send video

[hazy saddle]

Alright it’s all working good and it’s on my personal git site and not codeberg rn

Gonna finish up some docs and maybe push to main or keep working on dev

[mild glacier]

evalyn os will never be doomed ||hehe dad joke||

can u make one of those animations that u can never see when u screenshot/stop

[hazy saddle]

A screensaver?

[mild glacier]

no no
that kind of animation where there's static but some of that static moves and u can see the shape by motion
but when u take a screenshot it's just random noise
https://www.youtube.com/watch?v=RNhiT-SmR1Q

[hazy saddle]

Ohhh the bad apple thing is just a side effect

The bad apple demo only writes pixels that have been changes since the last frame

[mild glacier]

ik ik but still pls make this kind of idea as shown in the video 🙏

[hazy saddle]

And soon always writes the entire frame regardless

I mean that’s not really important to do when there are bigger fish to fry lol

[mild glacier]

yea but would be cool to see :3

[hazy saddle]

• it might be trivial with the libc

[mild glacier]

i just need to find out where and how to write directly to an FB with uhh... linux 😭

[hazy saddle]

Doom generic has a Linux version that uses the raw frame buffer

That’s what I based my on partially too

[mild glacier]

but is it open source?

[hazy saddle]

Yes?

That’s why I was able to port it

[mild glacier]

oh ok 👍

link?

[hazy saddle]

https://github.com/ozkl/doomgeneric/blob/master/doomgeneric/doomgeneric_linuxvt.c

[mild glacier]

hi fellow mint user

[neon drum]

HI!

[sudden mirage]

thats a bad way to do it, it cloberrs rdi and rsi

[hazy saddle]

What????

Do you know what syscall/sysret is

[mild glacier]

some very 👌 instructions

[hazy saddle]

Yeah well but they certainly don’t clobber rdi and rsi

[mild glacier]

noted

[hazy saddle]

They clobber RCX and R11

RCX holds the RIP

R11 has the masked flags

And sysret uses those registers to return

And you manually handle RSP

Eg the stack

[rotund furnace]

Which is clearly better then pushing them

[hazy saddle]

Well you would push them on the syscall routine yourself

That you make

Or store them in a certain spot etc

[rotund furnace]

I mean like the cpu pushing them

[hazy saddle]

Ahh yeah yeah

What is @sudden mirage on about it clobbering rdi and rsi?

[rotund furnace]

No clue

[hazy saddle]

Because if it was I don’t think my kernel would be running doom right now

And calling system calls a whole lot to do so

@rotund furnace Mind if I DM you btw?

[rotund furnace]

Sure but I’m at the airport

[sudden mirage]

yes, but but when you invoke a syscall with your function it literally clobbers rdi and rsi because the syscall num and a is passed to this func, and afaik gcc wont preserve the other regs, i dont even thing inlining would do a difference if you think it will, so the best option is to pass the syscall args to the func

[hazy saddle]

are you aware that my kernels regular build has a diffrent systemcall ABI to Linux?

I pass the systemcall number to RAX
And j get the return value into RAX with other arguments in different registers.

If this function were wrong none of my code would be working.

You replied to a message from like 2 weeks ago when things were broken from a a completely diffrent issue.

They are working now with doom executing properly and a libc ontop.

Also that systemcall function takes in the two arguments. For the systemcall number and one parameter. It’s not taking them from some random place.

[sudden mirage]

not aware of the ABI i thought you were doing the same one as Linux's

mb

[hazy saddle]

[hazy saddle]

@sudden mirage can i ask that you dont come to this thread again please?

[mild glacier]

i made this myself

good meme? 🔥

[hazy saddle]

@rotund furnace sleep works!

your laptop will stop burning

The timing on bad apple is fucked tho

That code is a complete and utter mess tho

Well it’s mostly fine

But a little desynced

To be fair I got the audio data from the MSDOS version and the frame and audio timings they do is ASS

[hazy saddle]

I’m pretty sure xor is the proper way because it tells the CPU to clear the register vs an actual move

https://stackoverflow.com/questions/33666617/what-is-the-best-way-to-set-a-register-to-zero-in-x86-assembly-xor-mov-or-and

[hazy saddle]

My priorities may be a bit off

Doom and bad apple

Before fucking sleep

[floral kettle]

average programmers be like:

[hazy saddle]

I mean it’s also both irl sleep and kernel sleep XD

[floral kettle]

ah you were talking about kernel sleep lol

[hazy saddle]

Yup

I now have proper idle threads and sleep() functions lol

[floral kettle]

nice

reminds me of the pain that implementing forcefully unblocking sleeping threads for the first time was, I had multiple deadlocks that'd randomly sometimes happen in my initial impl

[hazy saddle]

My sleep is implemented by my scheduler checking if the wakeup time is more than the current time.
If it’s higher find the next task

And il add waitfd and waitpid similar
If conditions not met skip

I had a bunch of issues too but I got it in the end XD

[floral kettle]

what I was talking about was for like waiting on an event with a timeout

[hazy saddle]

Ahhh

Wait how’s that work / where is that used?

[floral kettle]

if its only the scheduler unblocking threads when their sleep expires its not that bad but when an arbitrary random thread on a random cpu can wake up a sleeping thread at any point you have to be careful with locking

its used in eg. posix poll() or any other function that accepts a timeout

or well I guess it might depend on how exactly you implement it but at least the way I did it

[hazy saddle]

Ahhh. So waiting on a thing and if a timeout happens wake it up still too?
I could have it do IPC to tell the schedular PID X needs to wakeup or other.

And when checking if the thing is read also do a time check?

I don’t do SMP yet though

I’m probably gonna do that after @devout path finishes SATA and a filesystem as to not cause too many internal reworks as they work on that

[floral kettle]

yeah so that the thread can either get woken up when something is done or a timeout expires

[hazy saddle]

Yup yup

I’m trying to keep my intentions basic as to make my code easy to reference for others and me

So I’m not wanting to do tooooo many fancy things

[hazy saddle]

and way less for bad apple

i will want to get waitfd to wait for new data on them for the "kernel shell" to be more reponsive

but i may aswell get system() workin first

and then do waitfd

or is it poll?

yes i need poll

[devout path]

hi

[hazy saddle]

Hai hai

[mild glacier]

hi

[opaque marten]

that's still a lot ngl

what cpu do you have

[hazy saddle]

A 5700x and it’s in a KVM VM.

• I assume it’s got something to to do with the doom port

Bad Apple much lower

[opaque marten]

do you have polling

also i have a 5800x and it uses like 5-15% iirc

[hazy saddle]

It only checks input once per frame and if it needs to grab an extended key it polls. Although I can probably insert a sleep in there?

[opaque marten]

dunno

[hazy saddle]

And if it dosnt get input it just directly goes to the next

And it makes sure the buffer is empty

[hazy saddle]

Oh to be clear that’s 1/2 of one core

And it seems under KVM the core won’t down clock

The bad apple demo running at a few %

It still shows a high clock speed

So I presume it’s some other fuckery

[opaque marten]

idk then

[hazy saddle]

Okay so let me get my road map here
Get a bare minimum shell working in userspace to launch apps.
Redo the PMM VMM and Heap
make copying to and from userspace safe
Look into signals and pipes and making things more posix
Get CoW and fork() and others working nicely
Port bash
Port other things
…
Port x11

@opaque marten think i fixed the cpu usage tho

wait btop shows it differently

i thought the clis and stis were causin it

but i did get all sti/cli's goned

[hazy saddle]

Today I am going to get started on parsing the basic ACPI tables to get the IOAPIC and APIC stuff working to get away from the PIT

Aswell to move some folders around and push the dev branch to main

[rotund furnace]

please

please

just save your self the effort

use uacpi (in barebones mode if you don't needa parse aml)

lest you go down the rabbit hole of aml

[hazy saddle]

Not AML god no

Just the basic ass tables

[hazy saddle]

How hard is it to setup

The docs scared me

Like @rotund furnace How hard is it for just barebones mode rn

[onyx comet]

how do you handle the bad apple video? do you parse it yourself or use a 3rd party lib? and if you do it yourself, which format? :D

[rotund furnace]

it's like a few functions at most

it's 4 functions

one for getting the PHYSICAL address of the rsdp

one for mapping physical memory to virtual memory (you decide the address)

one for unmapping (it gives you in the virtual address you gave it)

and one for logging

that's it for barebones mode

it's like 36 loc for my impl

[magic saffron]

imo if you just want the tables adding uacpi is overkill

[hazy saddle]

Limine v3 does that thankfully. The other two are easy pz

[hazy saddle]

The tables are like really fucking easy to parse

[magic saffron]

it does have a nice header you can include to save yourself writing the definitions

[rotund furnace]

uhuhuhuh iirc v3 doesn't map acpi tables

[hazy saddle]

Code is open source

[rotund furnace]

yeahhhhh

ig

but if you want aml down the line it's easy

[hazy saddle]

https://git.evalyngoemer.com/evalynOS/evalynOS/src/branch/dev/src/user/badapple/badapple.c

[rotund furnace]

tbf I can prob replace it with my own parser because I'm gonna do aml in userspace anyway

(hopefully)

we need like a uacpi stripped down version that's JUST parsing 😭

[magic saffron]

well we have that, its called uacpi in barebones mode 😛

[rotund furnace]

yeah ig

[hazy saddle]

I mean the tables are just a couple structs

[onyx comet]

ah thanks

[hazy saddle]

Would take me two hours tops

[rotund furnace]

yeah but if you wanna do aml at some point you're gonna have to rip it all up

[magic saffron]

nah you can have both

you'll need tables earlier than aml

[hazy saddle]

Can’t believe I have to implement x2apic just because Intel got rid of the pic (and Apple I think?)

[hazy saddle]

Like for the IOAPIC and other things

[rotund furnace]

and apic

because some machines don't have apic

but also some of the machines that have x2apic don't have apic

🙃

but apic/x2apic shit is generic

[hazy saddle]

It’s the same driver except one is MSRs and the other is MMIO right?

[magic saffron]

more or less

[rotund furnace]

yeah x2 is msr

[magic saffron]

there's one special case where ICR (register for sending IPIs) is a single register now instead of two

[rotund furnace]

yeah

that ig

otherwise you can just do something like

uint32_t lapic_read(uint32_t reg) {
    if(x2apic_mode) {
        return (uint32_t) __rdmsr(IA32_X2APIC_BASE_MSR + (reg >> 4));
    } else {
        return mmio_read_u32(apic_base_address + reg);
    }
}

void lapic_write(uint32_t reg, uint32_t value) {
    if(x2apic_mode) {
        __wrmsr(IA32_X2APIC_BASE_MSR + (reg >> 4), value);
    } else {
        mmio_write_u32(apic_base_address + reg, value);
    }
}

[hazy saddle]

I can have a generic IPI function that can handle that

[hazy saddle]

Yup exactly

I’m not gonna be doing SMP yet

[rotund furnace]

how do you do ipis?

[hazy saddle]

But a timer that’s more precise than the PIT would be nice for nanosecond sleep stuff

[rotund furnace]

see I only use the apic timer for scheduling

I calibrate it with the pit 😭

[magic saffron]

if you're planning smp support at all, I'd recommend getting that setup asap. It'll change a lot of assumptions you can make and you'll encounter different bugs.

[rotund furnace]

^^

[hazy saddle]

I’m atleast gonna wait for @devout path to finish the sata and ext2/4 driver as it will take a ton of reworks

[hazy saddle]

It’s mostly slapping locks on things that go cross core right?

And implementing thread local storage for threads

[magic saffron]

pretty much, locking is the big thing. You'll also have to worry about caches (just the tlb on x86).

[hazy saddle]

For locks I should make arch specific ones and then have a macros that calls the proper arch one I think? With some pre processor statements right

[magic saffron]

you can do it generically with atomics

gcc/clang have builtin functions for them

https://gcc.gnu.org/onlinedocs/gcc/_005f_005fatomic-Builtins.html

[hazy saddle]

I can also probably use plain atomics for some things too?

But like anything two cores may wish to touch should have locks

[rancid viper]

depends

generally though

throwing locks on things is the easy way out

[magic saffron]

its a good way to start: wrap any shared state in a lock

[hazy saddle]

Aswell for my scheduler I was thinking for cross core that threads would compare how many threads they have weighted by sleep and then pass threads between cores to make sure they are balanced.

And one core only talks to two other cores at most as to not be talking globally and or having more lock contention

[rotund furnace]

wdym pass a thread to a core

all threads should beable to run on any core

(although keeping them on one core might help)

[hazy saddle]

Like have it scheduled to another?

[magic saffron]

I wouldnt consider sleeping threads when balancing run queues, as you can re-evalate which core the thread runs on when it wakes (last is preferred, but conditions may have changed). You keep each core mostly independent, with some work stealing (if a core runs out of work -> it takes a thread from someone else's queue) and a thread that runs occasionally that scans and balances run queues if needed. If you're worried about too much cross-core traffic you can introduce a hierarchy where cores belong to a family of 4 cores, the neighbouring 3 cores are checked first, then any others (but if we're talking performance, measure first!).

[hazy saddle]

Wouldn’t the scheduler be running on every core

[rotund furnace]

yes

but if you pause a core one cpu

it should beable to resume on any cpu with next to no overhead

- cache shit

[hazy saddle]

So if the scheduler runs out of tasks it should knock on others and be like hello may I have some work and if the local cores don’t have any work go further out?

And I assume the handoff is passing over the struct

[rotund furnace]

once a task/interrupt context pushes any gprs that are caller saved, yield will push any callee saved gprs switch stacks restore callee saved and ret

that's it

[magic saffron]

the handoff is the same as taking a thread from the local queue, except its a different queue

[rotund furnace]

everything else is choosing which threads execute when

[magic saffron]

and yes

[hazy saddle]

So the exact same as I have now?

[rotund furnace]

yeah

any core that yields will just get "a new thread"

it doesn't matter which one

obv this is for the most basic of scheduling but oh well

[hazy saddle]

So before the idle thread it should go and ask others and then idle if there is nothing to do

[hazy saddle]

I’m thinking once I get my PMM VMM and heap reworked and get fork() going il work on SMP

[rotund furnace]

scheduler kinda has two co-equal parts tbh. scheduling a thread to run, and switching to it

[rotund furnace]

I mean

sure why not

[hazy saddle]

That way it will make sense to have it and il make a multi core Mandelbrot demo or something

[magic saffron]

demo oriented development lol

[hazy saddle]

Exactly XD

What’s why I got bad apple running before VMM

💀

[rotund furnace]

😭

[rotund furnace]

????????

that must of SUCKED

[hazy saddle]

My brain needs to see something working

[hazy saddle]

It was fine I just baked it into the kernel image

[rotund furnace]

the message "syscall! - 0xcafebabe" was enough from me

[hazy saddle]

And I didn’t use dynamic allocation

RLE encoding

Maybe after SMP and all the other cool stuff I should work on an Intel HD audio driver

maybe I can make it play sound effects on kernel panic

Btw how are you meant to calibrate the LAPIC timer or other if you don’t have a working PIT on a system

[rotund furnace]

iirc you can use the invariant tsc

or some other timer like hpet

although iirc hpet also needs calibration

idk

[magic saffron]

invariant tsc doesnt give you the frequency

so

[hazy saddle]

Not on every CPU

[hazy saddle]

There are MSRs for that one tho

[rotund furnace]

ah

damn

[magic saffron]

which ones?

[hazy saddle]

I could of sworn I heard about them

To get the frequency it used

[rotund furnace]

so like

idk

[magic saffron]

on newer intel cpus there is cpuid leaf 0x16 (iirc, maybe its 0x15) to get the lapic timer and tsc frequencies. Other than that you have to calibrate against another timer. On virtual machines you have pv_clock which is excellent. Then you have the ACPI PM timer, HPET and PIT you can use as reference clocks.

[rotund furnace]

this shit is too hard

yk what we need

a whole new timer that's better then the rest

[hazy saddle]

Let’s call it the x2PIT timer

I was making a joke btw

[hazy saddle]

I mean if a system has no PIT and no invariant TSC or known calibration for the timer I can just say fuck it and say no

[magic saffron]

I mean yeah you could do that, but its pretty unlikely.

[hazy saddle]

Newer Intel CPUs lacking a PIT would have the MSRs so it’s like edge case devices like MacBooks or some shit I think?

[magic saffron]

they're not MSRs, its cpuid

[hazy saddle]

Ahh sorry sorry

[magic saffron]

haha you're alright

MSRs are generally writable

which would be cool, you tell the timer its frequency.

[hazy saddle]

Would be cool if the x2APIC timer let you do just that since it uses MSRs

Read a register for the range it can handle

Then write to another for how many times per second to interupt

[hazy saddle]

That’s a working x2apic and timer

[mild glacier]

let's just ignore that ascii art in the bottom right corner 😔

[hazy saddle]

no thats on purpouse lmao

[mild glacier]

oh so that's not a fatal error?

[hazy saddle]

no i made it cause a fatal error

to make sure it was working

[mild glacier]

let's just ignore that ascii art in the bottom right corner 😔
i said this coz it's supposed to be a "let's just ignore the errors" comment

how did u get 100+ quadrillion bytes from spawn 💀

[hazy saddle]

@mild glacier my structures were wrong so I read bad memory

So it got the wrong address

And when it tried to use it

Well we all know what happens when you touch non cononaical memory

[mild glacier]

also that "Advancement Get: How did we get here?" makes me wanna use that as the error message for segfault in purpur linux
can i 🙏

[hazy saddle]

It’s one of my many random error messages

[mild glacier]

along with what other funni error messages?

[hazy saddle]

Yes

All my error messes are random

[mild glacier]

show me what others u have in ur arsenal >:c

[hazy saddle]

https://youtu.be/j4Ph02gzqmY

It’s open source

[mild glacier]

💀

[hazy saddle]

@mild glacier src/kernel/utils/panic.c

[mild glacier]

it's open-source so im just gonna steal borrow it rq

[hazy saddle]

Better credit my ass

I spend a good time slacking off making thoes

[mild glacier]

better license it bruv

[hazy saddle]

it is

it is licenced

under the MIT licence

[mild glacier]

oh ok

[hazy saddle]

https://git.evalyngoemer.com/evalynOS/evalynOS/src/branch/master/LICENCE.TXT

https://www.youtube.com/watch?v=MeXQBHLIPcw

btw @rotund furnace i cant find my usb got a moment to test my kernel on a machine with x2apic?

[rotund furnace]

I'm on vacation so no real hardware other then my macbook

oh and no usb drives

so sadly no

[hazy saddle]

damn

all good

[rotund furnace]

tbh I'm not even sure the machine I use to test has 2xapic lmao

[hazy saddle]

XD

i do get why intel depracated the old one tho

the MSR intrface is much nicer

[rotund furnace]

yep

[mild glacier]

thx for letting me use src/kernel/utils/panic.c

[hazy saddle]

gonna take alot to port the thing

[mild glacier]

but those repetitive printf("(string)\n"); could be replaced with puts("(string)"); plssss

it'll save u source code bytes 🙏

[hazy saddle]

have you considered i dont care

lol

it dosnt matter

also i dont have puts implemented

[mild glacier]

lmaooo "I like elephants and god likes elephants" as an error message is an insane reference

[hazy saddle]

wtf?????

the PIT is setup the same as before

then interupts fail

but if i stub the sleep it works

nvm my bad

i fucked somthin else up

okay well closer but not there

PS/2 fails when using the PIT

but not HPET

for some reason

the timer works though???

[hazy saddle]

@hollow crater its 8am i hsvent sleep ive been coding but here is the patch file for whatever the fucks been going on
https://git.evalyngoemer.com/evalynOS/evalynOS/src/branch/dev
thats the proper branch to apply to

[hollow crater]

aight u can (should?) go to sleep now lol
i'll see if I can figure out whats going on but I make no promise I will be able to xD

[hazy saddle]

It’s all good

I should sleep

Yes

[hollow crater]

g'nightttt!!!

[hazy saddle]

@hollow crater btw did the patch file manage to work and libc build properly?

I uhhh

Forgot to sleep

[hollow crater]

welp its still compiling gcc here

[hazy saddle]

Good old GCC

BTW

On some machines

[hollow crater]

i wanted to test it on a clean chroot so here we go

[hazy saddle]

I’ve had it fail

And you have to redo it once more for GCC

But then it worked

[hollow crater]

oh

[hazy saddle]

Only on a fresh install on my laptop though

I’ve ran the script from clean git downloads on my desktop fine one go

YMMV

[hollow crater]

well i will see how it goes
no reason to fear for now kekw

[hazy saddle]

Just a reminder my code is connected together with 14 pounds of duct tape

And you got the dev branch right

[hollow crater]

well in theory this should be the same which's bad
but i have nothing to do anyway so ill do it again if it did

[hazy saddle]

To be fair I also riced the shit out of the laptop too

[hollow crater]

relatable

[hollow crater]

looks cool tho

good old win 7 style always wins

[hazy saddle]

It wasn’t too hard either

Also I don’t plan to ever update GCC of binutils

So you won’t ever have to recompile these

[hollow crater]

well ig i wouldnt have had to do that if i just tested locally 😭
since i already have a cross compiler
but i just thought why not

[hazy saddle]

Would my cross compiler even work for yours tho

Or vice verca

[hollow crater]

well yes it doesnt seem u had it configured that different from mine
worst case it'd just need extra flags to the makefile here and there

[hazy saddle]

True true

[hollow crater]

uhh

[hazy saddle]

I use clang for kernel

GCC for userspace

[hollow crater]

ohhh

why that tho

[hazy saddle]

Clang because I use host compiler and clang is better at some things

https://cdn.discordapp.com/attachments/1323419288668995657/1439390497687867562/me_on_ur_mom.gif

I’ve heard it’s code gen is nicer

And I CBA to make an LLVM toolchain for userspace

[hollow crater]

😭
clang supremacy

[hollow crater]

fair

[hazy saddle]

If I ever get rust in userspace I will do LLVM tho

https://cdn.discordapp.com/attachments/1434603991957897258/1456665252514234419/505e9f95328cb0b7.gif

[hollow crater]

maybe add xorriso / clang as deps in the readme then

[hollow crater]

😭

[hollow crater]

i know no rust anyway
im rust free
only c/c++

[hazy saddle]

I should

I do also say it’s incomplete

https://cdn.discordapp.com/attachments/1323419288668995657/1439390497687867562/me_on_ur_mom.gif

[hazy saddle]

Yeah but it’s blazingly fast and memory safe

And I need it to spam emojis all over my readme

But like other people might want things in rust

Not bad to support it

Even though I won’t provide many things natively for it

Also @hollow crater my kernel triple faults under GCC

[floral kettle]

smells like UB

[hazy saddle]

Or clang spesific extensions

[hollow crater]

eh

[hazy saddle]

With the fucking UBSAN

Or somthing

[hazy saddle]

What

[hollow crater]

fairs

[hollow crater]

sobs

[hazy saddle]

You applied the patch to dev and not master

Right

[hazy saddle]

That sounds like a patch applied to master

[hollow crater]

ohhhh

[hazy saddle]

Because dev branch is one ahead where I added that

[hollow crater]

welp lemme fix ts

[hazy saddle]

Because I updated to limine v4

[floral kettle]

also why not clone limine in the makefile

[hazy saddle]

(Protoco(

[hollow crater]

fairfair

[hazy saddle]

It’s a sub module

I regret that somtimes tho

[floral kettle]

its not tho?

[hazy saddle]

I also get linine binaries from package manager

[soft snow]

Why not just have a limine file with everything prebuilt

[floral kettle]

I mean like the actual limine binary branch not the protocol

[hazy saddle]

Ahhh

[soft snow]

Lol

[hazy saddle]

Troll

True

[floral kettle]

it just tries to use it from the host system /usr/share

[hazy saddle]

I should do that

Same for OVMF

[rotund furnace]

clang for kernel & rustc for userspace 🔥

[hazy saddle]

[soft snow]

I just have a limine folder so I build everything once and then just update it periodically each half year.

[hazy saddle]

I’m waiting for Lily CC

[rotund furnace]

lily cc?

[soft snow]

You can copy AuroraOS folder.

[hazy saddle]

https://discord.com/channels/440442961147199490/1306765166007816363

@rotund furnace

[soft snow]

https://codeberg.org/aurora-org/AuroraOS/src/branch/dev/limine

[floral kettle]

I just do this (assuming its an early-stage os project where I don't have a meta package manager I can use to install it and I have a cmake target for running the thing) ```cmake
if(NOT EXISTS "${CMAKE_BINARY_DIR}/limine")
execute_process(
COMMAND git clone https://codeberg.org/Limine/Limine
-b v10.x-binary --depth=1 "${CMAKE_BINARY_DIR}/limine"
)
execute_process(COMMAND ${CMAKE_COMMAND} -E make_directory ${PROJECT_BINARY_DIR}/iso_root)
execute_process(COMMAND ${CMAKE_COMMAND} -E copy_if_different
${PROJECT_BINARY_DIR}/limine/limine-bios.sys
${PROJECT_BINARY_DIR}/limine/limine-bios-cd.bin
${PROJECT_BINARY_DIR}/limine/limine-uefi-cd.bin
${PROJECT_BINARY_DIR}/iso_root
)

execute_process(
    COMMAND git clone https://codeberg.org/Limine/limine-protocol
    --depth=1 "${CMAKE_BINARY_DIR}/limine-protocol"
)

endif()

[hazy saddle]

I think there is another repo for limine and OVMF

[rotund furnace]

oh shit that's neat

[hazy saddle]

Once it’s good enough to compile the kernel i will be supporting it as an option

https://media.discordapp.net/attachments/829483226770178070/1090160912587051088/cat_bounce_1.gif

https://media.discordapp.net/attachments/829483226770178070/1090160935001399326/tenor.gif

[soft snow]

I just have everything prebuilt.

Its easier.

[floral kettle]

that's also prebuilt tho

in the binary branch

[hazy saddle]

I also love when my kernel corrupts OVMF ❤️

[soft snow]

Ahh. I dont use cmake though

[hazy saddle]

Make files and bash scripts for everything

Who cares if it’s a complete mess

[rotund furnace]

fuck yeah 🔥

that's what I'm gonna do while I wait for chariot to get remote compilation

after fabricate is done lmao

[hazy saddle]

I do wonder if I should make my own build system for my kernels dependencies where it downloads an alpine CHROOT image downloads everything and builds it

@hollow crater is the GCC there get?

Yet

[soft snow]

What are y'all doin

[hazy saddle]

Compiling my kernel to help find a bug

[soft snow]

Ahh.

I feel like my kernel should be able to compile on every machine just not my machine without big efforts.

[hazy saddle]

GCC tho

@hollow crater

Oh wait I forgot it already finished didn’t it

Sorry

[hollow crater]

yea i actually got it all to work
sorry was afk 😭

[hollow crater]

😭 😭 😭

[hazy saddle]

It’s meant to like

[hollow crater]

now imma take a look at whats exactly going on

[hazy saddle]

Open a kernel shell

Oh that kind of working

XD

I narrowed it down btw

[hollow crater]

it didnt :P

its just stuck there

[hazy saddle]

The interrupt thing isn’t calling the function pointer

At all

It should be running the generic timer ISR function pointer

That different timers plug into

But it’s straight up not running

[hollow crater]

mmmmmmm
so neither the pic interrupts nor the pit ones get fired
u sure interrupt handlers are working at all? 💀

[hazy saddle]

Well if that function never calls