EvalynOS

[wanton grove]

give

[hazy saddle]

Drop this in the folder in /src/initramfs

And unzip etc

Should work

Uhhhhh

https://en.wikipedia.org/wiki/Heisenbug

[wanton grove]

how to make this error

[hazy saddle]

Uhhhhh

Well nobody can reproduce the faults anymore

💀

But

I still have my buggy patch files

Which are other issues

the AM speech one is the latest

[wanton grove]

i need to patch both or only one

[hazy saddle]

Only one

The “FIREWORKS” test command is on those patches

And reproduces some strange ass faults

@rotund furnace May have more patches based on thoes fucks that caught more

crashes and the errors*

[rotund furnace]

I have no patches

[hazy saddle]

Are you on the dev branch?

[wanton grove]

i cloned this

[hazy saddle]

What does git status say

By default it clones master iirc

[true oak]

power of friendship

thats how you got fork

[hazy saddle]

I ain’t even forking yet

Or threeking

Or twoking

Or oneking

Or zeroking

[true oak]

i need a group like that too lmao

[mortal bane]

fork,
threek,
twok,
onek,
spoon

[hazy saddle]

99 little bugs in the code. Take one down, patch it around 117 little bugs in the code.

[cyan bison]

at least credit klange there

[hazy saddle]

wait

thats klange?

[cyan bison]

#1141057599584878645 message

[hazy saddle]

i stole it from reddit

Does this mean

Kevin Lange

Is a Redditor

[mortal bane]

ayo what 😭 reminds me of clobbering the rcx and r11 during sysret

[hazy saddle]

I think I’ve almost got them all

I think I’ve got heap corruption

(Not mllibc)

I caught my kernel trying to use fucking user memory addresses for shit

And some null pointers

And the heap I have is known to be dookie 💀

[mild glacier]

fr 😭

they also forgot "and 25 errors in the code"

[hazy saddle]

If your code dosnt compile it’s your fault

Like deeply lmfao

Getting it to compile is easy

[mild glacier]

should i compile everything with -Werror 🤔
coz undefined references to functions are not errors for no reason

[hazy saddle]

It can be good but I don’t have it enabled for NASM because it keeps yelling at me for no reason

[rancid viper]

I'm sure there is a reason

Out of curiosity what does it yell at u

[hazy saddle]

some relocation crap

[mortal bane]

you compile with PIE or PIC?

[hazy saddle]

nope

[mortal bane]

How

just how 😭

[mortal bane]

do you link with pie ?

[hazy saddle]

no

[mortal bane]

do you use the -no-pie IN your ld flags?

[mortal bane]

what about ldflags

[hazy saddle]

added it

same thing

[mortal bane]

oh wait this error is about big calls?

more then 2GB

[hazy saddle]

idfk

it shoudlnt really ever be above 2gb anyways

[mortal bane]

you are calling functions more than 2GB far away

[hazy saddle]

the kernel is less than 2gb in size and it has to fit in the higher half

and if i were things would be fucking up bad

[mortal bane]

its not about size, show me your linker script if any

[hazy saddle]

as an aside

how tf to i get the build id to work

i need my builds to have this in them

so when shit fucks up i can tell what build etc

and i can see if it properly reflashes etc

[mortal bane]

try the large one -mcmodel=large

instead of kernel

its a linker warning

[hazy saddle]

thats coming out of NASM

[hazy saddle]

okay so

my kernel is fucked

BUT

its not over

im rewriting in C++

with @wanton grove joining maybe for some things

[cyan bison]

bruh

[hazy saddle]

im gonna get c++ flanterm and nanoprintf

and then get more stuff fixed

[hazy saddle]

what

we are also gonna do proper arch abstractions

and shit

[cyan bison]

c++ evil

[hazy saddle]

C++ ❤️

[cyan bison]

C++ 🤮

[wanton grove]

c++ ❤️

[cyan bison]

c++ 🤮

[urban mulch]

C--

[hazy saddle]

i am using c++

and NOBODY can stop me

[cyan bison]

I can

im adding a new clause to my license

no taking my code and putting it in c++ code

now u cant steal my allocator like u did before

[hazy saddle]

then i wont steal that allocator

[cyan bison]

u got me there

[hazy saddle]

ya know what

if you fix my kernel

i wont use C++

[cyan bison]

no

[hazy saddle]

then C++ it is

[cyan bison]

inb4 rewrite in 3 weeks or less

[hazy saddle]

i am doing a rewrite

[cyan bison]

no i mean another rewrite

[hazy saddle]

is that what i think it is?

c++?

[hazy saddle]

i still have some stuff to do

for inital commit

but its looking good

okay so for this revision

i am dropping support for old add machines

because im going to be implementing lock free agloritms

including for the PMM

oh also

im going for multi arch

x86_64 aarch64 and riscv-64

Things are going to be abstracted from the start

So I don’t shoot my self in the fucking foot

Again

im also gonna support MB2

[peak cloak]

Prometheus was punished by the gods for cursing humanity with the Multiboot protocol. He was sent into the bowels of the earth and pecked by birds.

nice tho

good luck

also grub specifically doesnt report the proper amount of memory when using the basic memory info function

limine is fine with it but grub aint

like under the mb2 protocol

[hazy saddle]

Hopefully this new version of the kernel

Won’t need rewrites

Because if it does

I’m doing it in rust

You hear that @cyan bison

[urban mulch]

You should ping multiboot2 shill

[hazy saddle]

cc: @clear bison

[peak cloak]

multiboot3

[cyan bison]

nope

[clear bison]

[hazy saddle]

can you tell me abit about how it works

this is what i got so far

gonna get the riscv and and arrch64 ports goin

[clear bison]

so you got a multiboot2 prekernel I'd guess

[hazy saddle]

i need a prekernel for it?

[clear bison]

i mean if you wanna start up with paging

and other stuff

[hazy saddle]

i mean i am possibly fine doing that inside my bootloader init stub?

eg

[clear bison]

should be yeah

in mb2 you start in 32-bit pmode with no paging

[hazy saddle]

idk if i wana do this anymore

when mb3

[clear bison]

you don't have a reliable stack, gdt or idt

[hazy saddle]

nah nah i can do this

il have to make that part in assembly and c

[clear bison]

so I'd recommend you first get some preallocated area to have a stack

[hazy saddle]

and have it jump to my kmain

[clear bison]

btw are you going to go to lmode or stay in 32-bit pmode?

[hazy saddle]

im gonna be goin to long mode

[clear bison]

then I'd recommend that your "prekernel" or whatever you call it also sets up stuff like an hhdm

[hazy saddle]

yup

i will have to do a full pre kernel i think

[clear bison]

mine does in both x86-32 and x86-64 but eh, the 32-bit one is like 512mb

[clear bison]

the hhdm i mean

[hazy saddle]

i could actually support i686 with this then

[peak cloak]

i think you can support any 32 bit intel

unless you did i686 gcc

in which case you ARE lame

[hazy saddle]

im not doing i386

[peak cloak]

[peak cloak]

fairn't

[clear bison]

nah i386 is kinda hell

if anything I'll only support i586 at a minimum

[hazy saddle]

i could do i585

[peak cloak]

pentium-1

[clear bison]

pentium mmx and shi

after booting an mb2 kernel you have an MBI (which is basically an small header + an array of information tags) which is placed in EBX, and in EAX you have the MB2 signature

[hazy saddle]

/home/evalyn/Documents/Programming/C++/evalynOSprime/kernel/deps/nanoprintf/nanoprintf.h:748:(.text.npf_vpprintf+0xbc6): undefined reference to __clzdi2'
fuck

this is for a riscv compile

[peak cloak]

ahhh, free at last

[hazy saddle]

i have to disable binary format speisfies

i can prolly add an emulation for that tho

[clear bison]

you can probably add an abstraction to that

and have an arch-specifc implementation

[hazy saddle]

i just have to make a weak symbol for __clzdi2

or somthing iirc

[hazy saddle]

this emulation works

i think

i stole it

i forgor this exists

lmfao

why tf does npf use the other intrinsic tho

yup workin cross platform

[hazy saddle]

got buildIDs workin

[hazy saddle]

how cursed would it be for my pre kernel to provide the main kernel the limine API?

[rotund furnace]

congrats you made a limine compliant mb2 stub

why bother with mb2 at all tho?

[hazy saddle]

compability

or some crap

[hazy saddle]

honestly not a bad idea

[hazy saddle]

i wonder if im making a mistake

redoing all this

[rotund furnace]

prob not

you rushed the fuck out of osdev

[hazy saddle]

im so sorry

[rotund furnace]

wdym

[hazy saddle]

idk

i feel bad

[rotund furnace]

why?

[hazy saddle]

for rushing osdev?

[rotund furnace]

who cares?

[hazy saddle]

feeling like an imposteor or somthin

CPUID is working now though

@rotund furnace though genuinly is this better than the old one

Ignoring the c++ memes

I’m gonna have arch::setup()

Which will do all arch specific setups

[rotund furnace]

how much of that is c++ and how much of that is having the foresight of osdev

[hazy saddle]

And then I’m gonna add in utils::panic() which goes to an arch backend

Maybe arch::panic()

[hazy saddle]

I don’t know

[clear bison]

idk the boot stage of my kernel provides some stable API to the rest of it, so i can have a bootloader agnostic kernel

could technically make it boot with the linux protocol if i wanted to

or with any other protocol tbh

[hazy saddle]

Though a MB2 limine stub would be useful for other projects too

And be mostly portable

Drop it in get MB2 working?

[clear bison]

it'll be probably be simpler to just make your own abstraction over the bootloaders

[hazy saddle]

My abstraction plan was to have functions that would interact with each bootloader seperatly

So I could make a basic MB2 pre kernel

Have it get some memory HHDM and map the kernel properly + identity map its self

Pass some data to the kernel before jumping into it

[hazy saddle]

https://git.evalyngoemer.com/evalynOS/evalynOSprime

new repo

may replace the old one later

[soft snow]

huh why C++?

[hazy saddle]

Idk

Is it looking good though?

Also I wokeup and I’m going back to sleep

[soft snow]

ohh shiii mb gn

[soft snow]

ehh personally I dont like C++ but thats just a preference

[hazy saddle]

Shit I think I found the bug

That was crashing the C verison

This is corrupting the heap

Lmfao 💀 @soft snow

Well I’m gonna go to sleep

Fix this in the morning

I’m still gonna do the C++ rewrite though tbh

[soft snow]

loool

sleep well gn

my mlibc is doing well

[hazy saddle]

Nice node

[hazy saddle]

The C version is still really rushed

And not scalable

But atleast I can fix it

And leave it in a usable space

Maybe get bash ported still

[magic saffron]

only if you dont use what you learned from the previous kernel.

interesting choice to support multiboot 2, whats the thought there?

[wanton grove]

32 bit support

[mortal bane]

EW C++ sounds like CPP

[clear bison]

booting on grub probably

grub can't boot the limine protocol

[hazy saddle]

I atleast learnt to
A. Don’t make my own bootloader
B. Make proper abstractions
C. Don’t rush shit
D. Think ahead for things

[echo cloud]

Dawg

😭

[hazy saddle]

😭

I mean if it works it works

[hazy saddle]

It’s nothing like that

[hazy saddle]

yeah no its still bugged

[rotund furnace]

this hurts to look at

num_files being at byte offset 1,100,000,100,000 is insane

that's a terabyte of memory per directory

[hazy saddle]

That’s fine imo

5 level paging

Make it sparse

Best file system design ever

Just add a pointer to next file blocks

[hazy saddle]

Okay so

EvalynOS

Is gonna be archived

And the new C++ version

Is being renamed

To “Prism”

Unless anyone else has better name ideas

I may come back to this project later

[hazy saddle]

https://tenor.com/view/breaking-bad-aaron-paul-jesse-pinkman-sad-tears-gif-15925597

[wanton grove]

https://tenor.com/view/truth-nuke-truth-true-soyjak-wojak-gif-14828627546768133163

[hazy saddle]

Once I pull up my energy to make the heap

I’m gonna make the new channel

• new repo

[marble ridge]

rip EvalynOS

[mortal bane]

EvalynOS is ded

[hazy saddle]

Nah I may revive it one day

Though it would possibly be from scratch

Planning for the new version is going well

And the code that’s being done is pretty nice

[rotund furnace]

what's planning

sounds like some nerd shit

just jump right into programming and figure out those details later

[hazy saddle]

Stuff like Linux syscall compat and our own system calls

[hazy saddle]

Yeah well that’s what I did and it caused this mess

:p

[rotund furnace]

ew binary compat

no microkernel :P

[hazy saddle]

Il do a micro kernel later :p

Once I’ve actualy learnt shit

I’m doing a monolithic

And none of that hybrid shit

[hazy saddle]

Okay so

I have almost recovered enough to make heap

For multiboot 2 support I am going to work on the pre kernel

And I will be making it limine compliant

So hopefully any kernel can just import that project

And do a small bit of linker file changes

And then have multi boot 2 with no extra changes

If you already have been using limine

https://tenor.com/view/simpsons-the-simpsons-butterfinger-crispity-crunchity-homer-simpson-gif-16660768017786514314

[hazy saddle]

@clear bison so how should I get started with multiboot 2

Also @rotund furnace @wanton grove this will including me making a heap for this shit

Which I will port to the kernel

So I’m not wasting time

[hazy saddle]

i love gnu.org taking ages to load

wonderful

[wanton grove]

im waiting

[hazy saddle]

im reading the MB2 spec rn

[wanton grove]

https://tenor.com/view/all-roads-lead-to-rome-white-rabbit-gif-13308924322367947549

[hazy saddle]

then im gonna get it going into C and then long mode

and then loading the kernel

well the limine template first

then the real kernel

as i have to make a bunch if trabslation shits

[wanton grove]

solution: copypaste from limine

[hazy saddle]

nah

im making this public domain

which also technicaly means

public domain paging code for people

[wanton grove]

will this be for pmode

[hazy saddle]

im doing the 64bit mode one first

for the mb2 -> limine stub

[clear bison]

read the mb2 spec and make sure that you use the structures in the example headers

also, the memory map includes the regions containing the kernel and other stuff that you would probably want for it not to be there

[hazy saddle]

btw do you know how to make meson properly compile for this with assembly and stuff

I can’t find the limine meson template

[clear bison]

no
idk meson

[hazy saddle]

As it’s been fucking thanos snapped

FUCK

[clear bison]

so make sure you pre process it or something to remove, mark or whatever else you want to do to those regions

[hazy saddle]

Yes yes

[hazy saddle]

I know this existed

Because it’s in my fucking browsers auto complete

[hazy saddle]

Yeah yeah. I’m converting to limine stuff too anyways

Gonna be doing somthing like this @clear bison

And is there anyone that happens to be in here that knows how the fuck meson works

[clear bison]

interesting

[hazy saddle]

The 32 bit one will setup for the 64bit part

The 64bit part will then implement the limine protocol

That way any limine kernel can be booted via MB2

🤞

Or at the very least grub

[clear bison]

😂😭✌️

[hazy saddle]

😭

I mean

Technically

It should work

[peak cloak]

MAKE OS

DO PAGING

GGEZ

[hazy saddle]

IM ADDING MB2 SUPPORT

[peak cloak]

[hazy saddle]

[peak cloak]

MAKE OS IN MB2

32 BIT VERSION WE HAVE LIMINE FOR 64 BIT

[hazy saddle]

IM MAKING MB2 BOOT LIMINE KERNELS

[peak cloak]

OK FAIR ENOUGH

[urban mulch]

THIS IS CURSED BUT SURE

[hazy saddle]

I MEAN IT WOULD BE REALLY FUCKING COOL

AND ADD MORE COMPTABILITY

[urban mulch]

DOES MB2 EVEN HAVE ALL THE INFO NEEDED FOR LIMINE SUPPORT

[hazy saddle]

IT PROBABLY HAS ENOUGH FOR MOST KERNELS AND I CAN EMULATE SOME

[urban mulch]

FAIR ENOUGH

[peak cloak]

KERNELS WORK FINE ON IT SO THERES PROBABLY A WAY TO GET LIMINE INFO BUT ITS INCONVENIENT

OR OTHER CRAP THAT LIMINE HAS

LIKE MULTIPROCESSING

[hazy saddle]

MULTIBOOT2 WHAT DO YOU WANT FROM ME

😭

JUST WORK

IT CANT SEE IT

WHO TF MADE THIS CRAP

wait

it was seeing it as MB2

the command just gives no output

but this does nothing

is it even executeing any of my code

@clear bison do you know what im doing wrong to anger the multiboot2 gods

ngl why TS harder than making a UEFI bootloader from scratch

😭

i remember having an easier time

atleast in QEMU ovmf

[peak cloak]

do jmp start

instead of cli

check that way

[peak cloak]

because you need more sleep

[peak cloak]

also wheres the header at

[hazy saddle]

i fixed it

its all good

[peak cloak]

bug fixing is cheap, send bugs

[clear bison]

the error is self explanatory

[hazy saddle]

I made it work

[clear bison]

wrong header. you probably also forgot to pass the correct tags

[hazy saddle]

I just forgot to type boot

In grub

💀

That’s it

[clear bison]

MB2 has header tags which specify a ton of shi. you will want to mess with the "Information Request" tag and put in the list it has all the information tags you want on the MBI

Else the bootloader might just not pass them

[hazy saddle]

Yeah I’m probably gonna put as many requests as I can in?

Then get a basic PMM setup to get page tables for the 64bit stub

And then have that get a PMM and heap going

And then get limine shit booting

And pass it all over besides the memory map having stuff marked as bootloader reclaimable?

So then when the final conversion happens to limine stuff it’s easy

[hazy saddle]

The multiboot loader is working on real 32bit hardware

I’m going to work on the 32bit path first

Which will enable PAE and paging

And get the kernel mapped into the upper half

And pass the important info

And then I will add the limine conversion

It will still be code name MB2LMSTB

Because that’s gonna be its main purpose

But it will also be used to boot prism

[hazy saddle]

Btw said specs of machine

[hazy saddle]

Multiboot2 framebuffer gotten

🗣️ 🔥

I am also going to add the ear piercingly loud PC speaker 1khz screech

For if it can’t get a linear FB

Or if it cannot get one below 32bit (PAE)

Or if it wasn’t loaded by a MB2 loader

Oh also I’m porting flanterm and nanoprintf here

🗣️ 🔥

So I can like have debugging info

[hazy saddle]

GDT INIT BAD

But this is with nanoprintf too

So it’s got it all

I’m gonna work on the panic() function proper

So it can like dump registers and shit

Then make the GDT & IDT

Do paging

The basics

Get the kernel mapped and using my protocol

And we are going

[mortal bane]

EvalynOS ded

[hazy saddle]

Nah it’s being turned into C++

And renamed to prism.

And I’m not fully killing it off

[mortal bane]

dayum, i hate C++

[hazy saddle]

I plan to return at some point

But I made many footgun mistakes here

[wanton grove]

u hate me ???

[hazy saddle]

New kernel is gonna be much more portable to archs and shit too

[hazy saddle]

Which is why I’m working on the i686 loader

[mortal bane]

but not as bad as rust

[hazy saddle]

Bad news for you

[wanton grove]

🥰

[hazy saddle]

EvalynOSv3 will be rust

[mortal bane]

[hazy saddle]

V2 will still be C tho

But that’s after prism

And after prism16

[mortal bane]

dayum

[hazy saddle]

Or EvalynOS-16

Because I want to try out a DOS clone at some point

[wanton grove]

waiting for v2 in 2040 year

[hazy saddle]

And it will be peak 🔥

[mortal bane]

https://tenor.com/view/cmd-command-dos-windows-msdos-gif-12422793

[hazy saddle]

Nah I mean after prism gets a good few ports and shit

I can go back and forth and stuffs

2028

[wanton grove]

prism forever

[hazy saddle]

https://tenor.com/view/always-has-been-among-us-astronaut-space-betrayal-gif-23836476

I’m gonna rename the Chanel for this too

Later™

Once I finish the i686 bootloader

And make the heap

[opaque marten]

ouch i686

[hazy saddle]

Yuuup

I’m working on the multiboot2 pre kernel

Which will hopefully make things less crap

🙏

Atleast for the kernel its self

[urban mulch]

Is FRED worth supporting?

[hazy saddle]

Yes

It’s very nice if you can get working

[urban mulch]

There don't seem to be many emulators that have it

[hazy saddle]

Intel SIMICS

#resources message

[urban mulch]

Is there a proper manual for FRED or did you need to do some digging

[hazy saddle]

The manual is very good

https://files.evalyngoemer.com/share/manuals/x86-64/fred9.pdf

[hazy saddle]

lets go

[peak cloak]

https://klipy.com/gifs/tense-table-smash

static gdt user 😔

ok ykw i dont blame you the gdt sucks bad

[hazy saddle]

I still generated it

Wellll

Using the wikis code

Just pre generated it

[peak cloak]

i remember assembling a 286 gdt by hand

it somehow worked

[peak cloak]

16 bit protected mode is big stupid

literally nobody uses that crap

[peak cloak]

g

[hazy saddle]

Do modern CPUs even have backwards compatibility with 16 bit protected mode?

Eg 286 mode

[peak cloak]

you need to enter 16bit protected mode to enter real mode

[hazy saddle]

Damn

[peak cloak]

if they removed that crap windows wouldve been 9/11'd

and thats the most important OS in history anyway

[peak cloak]

the other time i wrote a generator for a table like this was powerpc bats

for memory

[hazy saddle]

[peak cloak]

nah but they actually dont suck somehow

[hazy saddle]

Though TECHNICSLY

This is faster

To boot

And takes less code space

Okay uhhh @peak cloak what do I need now

I think I need the PMM?

And then to do paging

And then elf loading yeah

[peak cloak]

if you got the gdt and want to enter long mode then you need everything to load a binary and enter that crap

idk how you gonna load the pre kernel

[hazy saddle]

I’m not doing the long mode rn 🥀

@peak cloak this is the pre kernel

[peak cloak]

oh

[hazy saddle]

I have to load the 32bit version of the kernel

[peak cloak]

oh ok

[hazy saddle]

Prism kernel

Which should just need those things?

And il use PMM as heap

Or well il make it allocate things smaller then 4kb

[peak cloak]

Prism™(1)
-# 1. Prism is a registered trademark of @hazy saddle , All Rights Reserved.

[peak cloak]

😭

fair

i have no idea what the difference is

so fair enough

[hazy saddle]

I do wonder

How small of a stack can this pre kernel use

Because I give it 64kb rn

[peak cloak]

-### or ##### bytes probably

[hazy saddle]

Maybe 4kb of stack is enough

[peak cloak]

maybe

crap i need finish config file parser

[hazy saddle]

Oh also the boot protocol will uhh be like this

Kernel goes in upper part

512mb of HHDM on the lower part + the identity map for bootloader

And then modules get loaded under the kernel

And the structure that the kernel gets will have pointers to them

[peak cloak]

virtual pointers or physical ones

[hazy saddle]

Virtual ofc

But it will have physical ones too

I will give both

[peak cloak]

i thought you had brain damage for a moment

[hazy saddle]

I’m giving virtual because modules aren’t HHDMd

Also MB2 has a thing that makes me page align modules

So I should be good

[peak cloak]

why wouldnt you page align modules

[hazy saddle]

Because stupid

[peak cloak]

https://cdn.discordapp.com/attachments/829403502349647952/1471530638598476029/IMG_2856.gif

why is this optional

[hazy saddle]

Multiboot 2

[peak cloak]

GNUUUUUUUUUUU

[hazy saddle]

@peak cloak I wonder if I should make a bootloader

Like a full one

Nah

That will be annoying as fuck

[peak cloak]

i tried to make gregloader for x86 once

please dont unless you're particularly dedicated

[hazy saddle]

I’m not as dedicated as mint

[clear bison]

ehh

you can just disable protection and you'll still be in real mode

[peak cloak]

wrong

i did it it exploded on me if i dont do it

[clear bison]

just with fukced up segments

[peak cloak]

well this

im not messing with segments lololol

[clear bison]

which will indeed cause you to be in "unreal mode"

[peak cloak]

least real mode

[clear bison]

but you can fix that by just loading something in the segment registers lol

[peak cloak]

oh

anything?

[clear bison]

well, anything that is correct for the current context

[peak cloak]

ok fair

would 0 be fine?

[clear bison]

no

unless you placed your code at like 0x00000-0x0FFFF

when loading the CS

just put something that translates well into the physical address you're now

"unreal mode" isn't a separate cpu mode, is just real mode with the descriptor caches fucked up

loading the segment registers with something valid for real mode will get you out of it

you can't pull this shit off in pre-286 things because if i recall correctly there aren't descriptor caches to begin with lol

protected mode segmentation was added with the 286

[hazy saddle]

Wait would that mean it would check the segments every memory access?

[clear bison]

nope

the x86 cpus only really check the GDT descriptors when they get loaded

even implied loads, like when getting an interrupt

[hazy saddle]

So what is the “cache”

[clear bison]

they use the descriptor caches to check and do calculations

[hazy saddle]

Or did IA16 have no GDT

[clear bison]

i don't think IA16 is a thing :P

pre-286 stuff didn't have a GDT or IDT

no protected mode segmentation there

[hazy saddle]

Well the 8086 :p

[hazy saddle]

But it did have some segments I think?

[clear bison]

yeah, just (segment x 16 + offset)

but these aren't really similar to the protected mode ones

protected mode uses the values in the descriptor cache to first check the privilege shit, then check if we're doing a code or data access, then add the base to the offset and then check the result of that against the limit

iirc

if any of that fails you get a GPF usually

or something similar to that

[hazy saddle]

I wonder could you make a 286 kernel

Like a proper one and crap

Or would it be super annoying

[peak cloak]

probably this

286 protected mode is slow as frigg

windows 2.1's 286 version used real mode

you also have to actively use the GDT

[hazy saddle]

MB2LMSTB will require PAE for now

[hazy saddle]

i am doing a limit as i am copying the data structs to a diffrent place

that way i can for every memory region check if it has a module and or the executable in it

and ignore such pages when setting up PMM

& same for when mapping modules to special place

which wont be HHDM

aswell the kernel to be loaded here will be considered a "module"

[mortal bane]

i'm still making my modules work in Modularus

[hazy saddle]

These are just MB2 modules

So like loaded files from bootloader

[mortal bane]

that's what i'm doing too for early bootstrap

[hazy saddle]

Your bootloader just gives them to you

And it’s not too bad to parse

Mostly just boilerplate

[hazy saddle]

@clear bison does the MB2 memory map have MB2 strucutes inside "free memory" regions?

or is it good enough to just check for modules and my own image?

eg what does MBI mean here?

multiboot information?

[clear bison]

the information structure you get passed in EBX

[hazy saddle]

[clear bison]

the thing you get the tags from

[clear bison]

no it doesn't exclude that

[hazy saddle]

so how should i make sure im not fucking that up

[clear bison]

the mbi has a size field in the header iirc

so just exclude from EBX to EBX + that size

and it should be fine

[hazy saddle]

uhhh

thats just header length

[clear bison]

not this one

the MBI header

[hazy saddle]

ahh

let me check that

and print it

[clear bison]

it calls it total_size in the spec

[hazy saddle]

also is limiting the pre kernel to 8 modules a bad idea

okay yeah thats about right

il add and then round up and round down

@clear bison also since i load at 1MB il assume memory from 1mb to 2mb is reserved by me

[clear bison]

for UEFI loading shi at 2mb is safer

[hazy saddle]

damn

so 2mb to 4mb then

[clear bison]

Also this ain't really the size or the multiboot2 header but rather the MBI

just being pedantic with the naming :P

[hazy saddle]

;p

MB2 mbi should be inreserved region

my loader interface

will do this properly btw

[clear bison]

wha

[hazy saddle]

[hazy saddle]

it should be marked as bootloader reclainable or some such

[clear bison]

I don't even bother reserving it because i initialize the PMM after taking all the shit out of there

[hazy saddle]

i think il need it valid after some stuff?

[clear bison]

and my PMM is the only thing that might touch that area during the early init

[clear bison]

hm?

[hazy saddle]

to make the new memory map?

[clear bison]

the memory map i provide to my PMM already has all the stuff i need excluded

i can then trash everything that isn't the kernel and modules

[hazy saddle]

but the one im passing to the real kernel

[clear bison]

my PMM isn't for the prekernel

it's the real kernel

[hazy saddle]

dont i need PMM

for like paging

I mean I could use linker magic

To reserve some memory

And know it’s there

And then bump allocate from that?

[clear bison]

just grab a fixed area that can fit like 512 memmap entries and you should be good

whoever the fuck passes more than that probably has a broken UEFI impl

[hazy saddle]

Tell that to @rotund furnace S laptop

That had all those fucked up regions

[hazy saddle]

Yeah but like paging tables

[clear bison]

that was just limine fucking shit up actually

[hazy saddle]

To like map the kernel to the higher half

[clear bison]

because they left space between the ACPI tables

was it?

[hazy saddle]

Somthing like that?

But like

Where should I get memory to be able to map shit

Because I need to map the kernel to the higher half

And i need to map the modules too

And the HHDM

And the identity map

[clear bison]

just have some PMM

and initialize it with the memory map

[hazy saddle]

That’s what I’m trying to do

[clear bison]

then when you pass shi to the actual kernel you can just map all your crap as bootloader reclaimable

[hazy saddle]

exactly

[rotund furnace]

lenovo decided to make all acpi tables PAGE ALIGNED

not just that

each acpi table it self was in 1 page of memory

• 1 padding page

for some fucking reason

so each acpi table took up 2 pages of memory

[clear bison]

even then I'm guessing there weren't more than 512 memmap entries

[hazy saddle]

The thing fucking flooded the screen atleast :p

[hazy saddle]

Wait fuck

With the freelist

How should I handle reserving that memory

:p

I could go by the order I put things in the freelist

And count pages?

And skip those etc

Wait fuck that would be reversed

@rotund furnace Your the freelist person any ideas

[clear bison]

huh?

just, don't put it in the freeslist or use some flag?

my freelist doesn't cares about the used segments

like any good freelist would

:P

[hazy saddle]

I mean when I use memory I need to know I used it. But I can just bump allocate from a found region big enough and call it a day

[echo cloud]

setgdt and setup_gdt being separate functions is funny

[echo cloud]

If u do non static gdt ur lame

[peak cloak]

true

[hazy saddle]

One is to set the GDT registers and one is the function to call from the main file

[echo cloud]

Do you hate namespaces?

😭

[hazy saddle]

It dosnt matter?

setup_ is the namespace anyways

[echo cloud]

That makes sense if you have that namespace nvm

It’s just a lot of people do shit like “init_gdt” without actually having an init namespacing

[hazy saddle]

Everything goes under setup_

[echo cloud]

Shit pisses me off 😭

[hazy saddle]

So like setup_idt

[echo cloud]

Yeah fair

[hazy saddle]

Or setup_paging

Etc etc

[echo cloud]

I have an init_ namespace but that’s only within init.c, each name space outside of init has a namespace_init and that gets called within my init_config()

[hazy saddle]

I think this loader protocol won’t do HHDM but do an identity map for the first 3.5gb and then the upper 512mb is for your kernel

And it will map the pages of the kernel binary up there from the loaded module

• the BSS etc

Then it will give a memory map inside the identity map and the kernel will deal with how it wants to do things

And il ensure everything is inside the identity map or the loader will panic

• there needs to be a region of memory that has no reserved stuff in it for me to use

I could also instead of a PMM

Make the binary have like 8mb of BSS

I can bump allocate from for paging tables

And call it a day

And then just tell the kernel that entire region is reclaim

Which honestly may be easier tbh

[hollow crater]

u can also do like mb2 and provide nothingwhatsoever ✨

r u making a boot loader+custom protocol now damn :o

been away for a while so guess I missed a lot..

[hazy saddle]

Nah a MB2 pre kernel

To boot new kernel i686 port

So I have to provide shit

Or atleast fucking mappings

I think il do the PMM in chunks

And then when generating the final memory map to pass

I will simulate all the allocations

Checking for reserved etc

[urban mulch]

I think your best bet for the memory map response is to record the various unusable and ACPI reclaimable, etc. regions ahead of time. Then all you're left with is accounted for in the bootloader allocator, which you later read to find the usable and bootloader reclaimable memory.

This way you don't have to worry about 10 different types of unusable memory accounted for in the bootloader allocator.

[hazy saddle]

the issue is tracking the reclaim memory properly. im gonna pass a memory map thats the MB2 one but with the reclaim stuff set proper and the rest is whatever grub said™

im not sure a good method for tracking such is

i mean i have ways but they seem bad™

[urban mulch]

Basically what I'm saying is track bootloader reclaimable and usable memory in one structure (the bootloader allocator) and literally everything else in another (e.g. some heap-allocated collection).

[magic saffron]

The lbp memory map makes more promises than the mb2 memory map like handling overlapping regions, alignment and the kernel + modules not being marked usable.

Same goes for the rest of the protocol: mb2 is very barebones, limine defines a bit more.

[hazy saddle]

Yes but I’m working on the 32bit path right now. I already am handling the modules properly + using the tag to make sure they are aligned to page at the very least so I am able to see if a page is in some garbage region.

It’s just when I allocate from tree memory I don’t know how to track what I have used so I can mark it as reclaimable when I pass it to the real full kernel after mapping & such

[magic saffron]

Ah I assumed you were still doing a limine shim.

[hazy saddle]

That comes after I get 32bit working

Which still needs me to solve this issue to pass it to the 64 bit pre kernel which needs this information

[magic saffron]

Its still nice to follow those things even if its 32bit (e.g. make it a nice entry state for the kernel)

[hazy saddle]

Oh yes I am gonna be making a much nicer entry

For 32b I’m thinking lower 3.5gb is identity mapped and the 512mb upped is the minimum load address

[magic saffron]

Bootstrap your heap with some statically allocated regions?

[hazy saddle]

And il be giving a full structure with all the information needed to boot

[hazy saddle]

Issue is I’m unsure how many pages I’m gonna need to use because of the kernels BSS & other things

And MB2 puts me at a fixed physical address

So I’m limited on that afaik?

[magic saffron]

I'm not saying have your whole heap be statically allocated lol, just enough to get your pmm and p/vma up and running.

Then you can use those as normal.

[hazy saddle]

Oo I could that to bump allocate a heap to hold data structures for things I allocate?

Eg the “used” list

Then I can combine that with other reserved regions to make the final memory map

[magic saffron]

You could

[hazy saddle]

Any other better ways you’d think?

[magic saffron]

I'd either do it that way or try bootstrap things using only the stack. The latter is more annoying imo but perfectly doable.

[hazy saddle]

I think I have ~13MB to work with so I can just give myself 8MB of heap and call it a day

Given a load address of 2mb

[magic saffron]

8M is massive lol

[hazy saddle]

Il tune values after

I give a 64kb stack lmfao

[magic saffron]

Numbers from the days when ram was cheap 😔

[hazy saddle]

How does 640kb sound?

That should be enough for anybody

[magic saffron]

I hope you dont need more than 10 stacks 😆

[hazy saddle]

I bet I could get away with a 4kb stack tbh?

Maybe 8kb

[urban mulch]

It's a bootloader but I'd say use an 8K stack at least

Fortunately you're not writing it in rust, otherwise you'd need 16k just for hello world

[hazy saddle]

[hazy saddle]

It’s just the pre kernel so I just need to make the stupid freelist PMM and some basic ass paging + the bump allocator for the used list then giving some info and redoing the memory map nicer

Yeahh I may want to do more than 4k

[floral kettle]

also put a guard page at the start (the lowest address) of it

its not fun to debug when you run out of stack space and start getting random corruption (granted a 4kb guard page isn't fully reliable either if you manage to create a large enough stack frame that it goes over it, though even then its likely that at some point you'd get an access to the page when accessing some variable or smth)

[hazy saddle]

I’m not sure how I should do that without an IDT/Paging? I will provide a guard page or 4 for the kernel though

[floral kettle]

ah yeah I somehow ignored the part about pre kernel lol

[clear bison]

setup a stack and push shi?

[hazy saddle]

The information being the used list so I can make a proper memory map

Which then I realized

I can just do the same thing I do for the stack

And allocate it in BSS

And then I can get PMM and track used shit etc

[clear bison]

https://tenor.com/view/ninja-ninjaishyper-low-taper-fade-massive-crazy-gif-13735679314385451792

my mb2 prekernel parses shi before enabling paging and then just pases the info through the stack while at the same time marking itself as unused in the memory map, that does mean i have 2 freelist pmm things but eh

the actual kernel one is more complex

[hazy saddle]

I mean I’m gonna parse it too before that

But after paging is enabled I have to map the kernel up there and make the final memory map

Wait I could just

Make the tables first

Then enable it?

But that leads to the same issue so

Won’t make any difference

[clear bison]

hmm?

the map is just for physical memory

after you map shi in the higher half you don't really have to care were your kernel is physically apart from not marking it as usable

[clear bison]

well, if you don't make the tables first how are you gonna enable paging in the first place :^)

you would get that sweet

[hazy saddle]

Trueeee

[mild glacier]

https://tenor.com/view/easy-steal-gif-6770495535008702461

[hazy saddle]

i may change this to "bmb2" or better multiboot 2

and i think the fact its gonna be public domain is gonna piss off the FSF

[hazy saddle]

yeah i am confused about the flanterm issue

it fully fails in anything but O0

and these vales look right

[peak cloak]

nooo big corpos can use it

[hazy saddle]

What if I want big corpos to use it huh

What if I want them to win

https://cdn.discordapp.com/attachments/1420566657541214320/1468828977706832179/ezgif-306d93eac6eb2f4e.gif

[peak cloak]

https://tenor.com/view/hollow-knight-gif-1317513878955975206

[peak cloak]

fork ritual

[hazy saddle]

Memory map is properly being parsed

And reserved regions removed

So I can make the PMM

And my memmap code is flexible enough to allow me to remove the regions I used up from that

[peak cloak]

we as humans also fork, its called mitosis

[peak cloak]

nicE

[hazy saddle]

@wanton grove time for you to make i686 paging btw

[peak cloak]

its the best one

[hazy saddle]

Jkjk

[peak cloak]

MUAHAHAAHAHAHAHAH

[hazy saddle]

😭

I’m actually just marking no PAE as unsupported

And it will say talk is cheap send patches

[peak cloak]

oh wait its pae paging

[hazy saddle]

Same as x86-64

Just missing a level

I think

[wanton grove]

fuck u

does limine c template works ?

[hazy saddle]

im not even on that yet

I’m working on the custom mb2 protocol

[clear bison]

you know PAE is basically useless in most i686 systems cuz most didn't get to 4G+

the only useful shit it could add is the NX bit

but that's not always implemented

[hazy saddle]

Yeah but the paging scheme is more similar to x86-64

And most i686 systems support it

So it’s what I’m targeting first

[clear bison]

it's also a bit slower on i686s

[hazy saddle]

Oh

Should I just not support PAE at all then

[clear bison]

i think you should just have a PAE and Non-PAE build

[hazy saddle]

Maybe?

But like