EvalynOS

[viral bison]

and it failed the build, oooof

i hate that it's not CLion compatible XD

[hazy saddle]

Damn

[viral bison]

yeah i'm on VSCode rn

so it can't really check the errors

[hazy saddle]

Do you have bear?

You can run the make command with that

And it will fix LSP

[viral bison]

nope

btw, how do i make a patch after?

[hazy saddle]

Just give me the files it changed

My working tree is messy as FUCK

[viral bison]

well, progress

i have to go home now, so cya later

[hazy saddle]

Did this work after two hellos?

Because it beaks if you do
echo “whatever”

Then

echo “test”

If the second one is shorter

And help and then help breaks

I wonder how it patched the bashrc thing

[viral bison]

Ah

I just tried 1

After i'm home i'll try

[hazy saddle]

Kk

[cyan bison]

why r u printing \n if it is received

[hazy saddle]

for terminal echo?

[cyan bison]

yes

u dont do it unless echo is enabled in the termios

unless u of course do not have ttys

[hazy saddle]

yeah well i have no VFS or termios ot ttys

💀

im trying to fake it til i make it rn

[wanton grove]

you still can have termios because termios set and get is sysdeps

[hazy saddle]

sure. but i CBA when i can just set a sane default it wants

:p

[hazy saddle]

i dont think this is a termios thing?

[wanton grove]

why are you return EACCESS

[hazy saddle]

its a byproduct of some other crap

ah no its this

[cyan bison]

tf

[hazy saddle]

[wanton grove]

worst open impl which i saw in my life

[cyan bison]

local osdev has no tty and no vfs and expects bash to just go with the flow

[hazy saddle]

Hey hey

It kinda works

And I don’t expect bash to use FS stuff rn

That’s why I only do bultins

[cyan bison]

huh

well of course bash does fs stuff

everything does fs stuff

[hazy saddle]

I don’t expect it do work*

I expect fucking echo to work

And help

And for it to not collapse in on its self

[cyan bison]

yuh

well first get tty

[hazy saddle]

i dont have a VFS to work with TTYs

[cyan bison]

because bash sets termios to change the output and echo behavior

then get a goddamn vfs

[hazy saddle]

after bash

[cyan bison]

idk what's up with this lasting joke about u procrastinating ur vfs

cuz it was funny like 3 weeks ago now ur just not making good decisions concerning ur kernel

[hazy saddle]

its on my todo list

[cyan bison]

put it first then

[hazy saddle]

im too far in deep with bash

[cyan bison]

not really

git stash

done

[hazy saddle]

im also not eating an onion

for not porting bash

end of week

[cyan bison]

riight

[hazy saddle]

i made bets

[cyan bison]

so if it runs and compiles just call it a day

there were no specific terms were there?

[hazy saddle]

it needs to like be somewhat stable obb

[cyan bison]

LGTM

[hazy saddle]

.

[cyan bison]

im a pro osdev i approve of this bash as ported

tty problem not bash

[hazy saddle]

True

[cyan bison]

u dont have sufficient termios flags implemented (as in none at all)

so fixing that by the end of the week would force you to rush your vfs

[hazy saddle]

I just return ENOSYS when it writes to write termios

🥀

[cyan bison]

mhm

[hazy saddle]

Okay so

I need to make a fucking TTY

That accepts termios

And then it should work magically

What termios does bash want even?

[cyan bison]

bash just needs a raw shell iirc

at the least it needs to be able to disable echoing characters on input

[hazy saddle]

I’m giving it a canonical line buffered thing

[cyan bison]

I dont really remember what it needs u could check your host

[hazy saddle]

Because it didn’t echo etc

[hazy saddle]

I could also just printf whatever shit it gives

[cyan bison]

your problem

your solution

[hazy saddle]

True

[cyan bison]

I just think strace bash is a lot easier than printing the termios syscalls

because then u have to recompile mlibc and all

[hazy saddle]

I mean il just implement it so my kernel gets the call and print there

[cyan bison]

ok do that then

or dont

[hazy saddle]

Yeah

Sorry

[cyan bison]

its good

I need to go i have a train to catch

[hazy saddle]

Idk why bash restarts and shit tho?

Idk

[cyan bison]

points to a context switching bug

[hazy saddle]

Well not a restart restart

But it reprints it’s no job control complaints

[cyan bison]

gl

[hazy saddle]

Like

This looks strange

🥀

My context switching should be fine

Doom and shit runs fine

And stress tests

And it should crash way more crazily

[cyan bison]

btw the fact it recheck for bashrc probably means it is restarting

like from the entry point I mean

[hazy saddle]

But it dosnt trip mlibc sysdep madness

It would be printing a lot of other BS

If it was a full full restart

[cyan bison]

id investigate what sysdeps mlibc complains about

[hazy saddle]

Mostly VFS stuff

[cyan bison]

iirc bash might need the dup2 syscall

[hazy saddle]

That it returns ENOSYS on automatically

[cyan bison]

u really cant be doing this without a vfs

[cyan bison]

i know how mlibc works dw about me

[hazy saddle]

Ahh

[hazy saddle]

Damn

Idk how it even got this far

Should I do a proper TTY layer this time

Then VFS

And I will just adhoc it in until the VFS is ready

[cyan bison]

gl

[hazy saddle]

What should the TTY thing do?

Like keep track of the cursor and shit?

[cyan bison]

no

not at all really

[hazy saddle]

I mean for when stuff gets printed to ir

And it echos

Or is that for the layer behind

Eg flanterm

[cyan bison]

man termios.3 and figure it out

should be easy™

I put my full trust in the manpage's judgment

[hazy saddle]

That command didn’t eork

[cyan bison]

what do u mean bash doesnt print the sysdep shit

when it restarts

it literally does

[hazy saddle]

I was making a joke

And that’s diffrent sysdep shit

It depends on command

🥀

[cyan bison]

least cursed bash port:

[hazy saddle]

If I just do where

Enter

It prints this

@viral bison if you manage to figure this out still lmk

[true oak]

thats not bash thats rash

[hazy saddle]

I fixed it

[soft snow]

what the

[hazy saddle]

Yeah

🥀

@warped ermine

Yeah it works now lol

[soft snow]

ohh shi leaked some discord servers

rip

well there you have the discord servers im in

@hazy saddle is there a way to actually see the build output and not just get loading when building with jinx?

or should i rephrase

do you see your building process when building with jinx without meson, or is that just a global thing with jinx?

[hazy saddle]

Wait

Why are you making meson call jinx?

[soft snow]

Why not

Jinx is just for ports

not for kernel

[hazy saddle]

Jinx can build the kernel

Meson isn’t really designed for this like that

[soft snow]

true

[hazy saddle]

Also I don’t even make jinx build it

Look at my root makefile

And follow the chain of commands

[hazy saddle]

(Though it can)

[hazy saddle]

Just a bit more work

Before people can’t laugh at me for a kernel shell anymore

https://tenor.com/view/caine-i-am-god-the-amazing-digital-circus-tadc-im-a-god-gif-8296037756389483754

[soft snow]

hahaaahahaahahahahaha bro made a kernel shell 👈 😆

[hazy saddle]

I will rend you apart

Also it’s not really a shell

Just for debug commands

🥀

Like it launches bash

And launches doom

[soft snow]

"not" a shell

[hazy saddle]

You can’t even run help

Or do any shell like things

[soft snow]

bash can do that right

[hazy saddle]

Nope

No fork

And no VFS

[soft snow]

wait rly

[hazy saddle]

Yeah

Like

[soft snow]

crazyyy

[hazy saddle]

The only thing userspace can VFS kinda

[soft snow]

even my shitty os has it

[hazy saddle]

Is stuff from the tarfile

Stdin and out are hard coded to go through syscalls

[soft snow]

what is all bash builtin commands btw

[hazy saddle]

Run help

[viral bison]

fixed?

[hazy saddle]

Yes

[viral bison]

gg

what was the problem?

[hazy saddle]

Somthing with the sysdeps giving bad data to the kernel

[viral bison]

oh

lol

[soft snow]

were going to make counter gave up after 5 min

i have US keyboard on the OS so yall can use it

but me myself use swedish keyboard

so writing shit is impossible.

[hazy saddle]

Damn

[soft snow]

wait how does linux handle keyboards

and keyboard layouts

[hazy saddle]

evdev

And locale settings etc

[soft snow]

do they have a keyboard layout file which holds the keyboard layout or what?

[hazy saddle]

I guess

[wanton grove]

its done by xorg

evdev just sends ps2 scancodes for keyboard

[warped ermine]

Bash

_

[hazy saddle]

it works ;3

okay so

its VFS time

[sterile plank]

me ow

[hazy saddle]

well technicaly generic bitmap allocator time first :p

for FDs

then i need to archetcture a proper VFS

it should be FD which points to an FDR which points to a vnode

yeah id also want a dirent cache for open() etc

so i can traverse the FS etc

[sterile plank]

dont overengineer it :v

[hazy saddle]

but fd's and the file table and the vnodes can just point directly after resolotuion

[hazy saddle]

im doing the normal UNIX VFS :p

the vnode should also track what things MMAP'd it

for the page cache i think

that way i can do reclaimage easy

[true oak]

yay bitmap user

[round shoal]

wait. there is even a WAITPID

[hazy saddle]

Not for a PMM :p

There is a time and a place for a bitmap

[true oak]

still happy

[hazy saddle]

I’m also gonna make mine fast

By using CPU intrinsics

https://www.felixcloutier.com/x86/lzcnt

[kindred stirrup]

@lzcunt mentioned

[hazy saddle]

💀

Also I may end up using the C23 built in or clz

Because it isn’t supported on older CPUs

3:

[round shoal]

lyneve os

[hazy saddle]

Considering doing a rewrite

One that has a little more thinking put into it

[hazy saddle]

It’s going to properly support SMP from the start

And gonna do proper hardware abstraction

So I can port to loongarch64

I’m possibly going to start with the PMM

and making a buddy allocator

A good one

And then making my paging code not be a mess

And making proper abstractions for it

Like paging_get_next_level(table, index)

And il have propped asserts for things and error handling

And il try and make a good kernel and take it slow

And not burn myself out

https://tenor.com/view/spongebob-the-essay-hilarious-haha-gif-3565665

[hazy saddle]

The first step will be to do a GDT Init OK

As is tradition

Tomorrow

Or this afternoon possibly

And I’m not going to use that fucksss font anymore

And possibly get better logging functions and systems to start

I will possibly reuse some things from this kernel

On the new project

Like the build system

Oh and il properly make it work on gcc and clang

[hazy saddle]

Rewrite is going well so far

Getting the basic structure crap setup

With proper arch abstractions

I won’t fill them all in

But it should be enough

[hazy saddle]

It’s now up enough to boot

I am considering a name change

So I will take ideas

Preferably RuneScape related

I’m gonna clean up a bunch of this

And get loongarch working

I also need to refine the panic thing I half assed to make UBSAN work

(To make it arch specific)

Then I’m gonna get CPU local working on both

And then gonna make the PMM and paging code later

Get a decent heap going

Then scheduling

Etc etc etc

I should then be able to bring up mlibc and userspace

Then il work on a real VFS

[hazy saddle]

(This includes SMP)

[sterile plank]

meow

[normal turret]

Coolio

[hazy saddle]

Meow

[hazy saddle]

needs a few tweaks

but its lookin good

This is finalized I believe

I do hopefully hope to make sure panics won’t happen that much

Because il make things stable

Now I just need to make this support loongarch64

Good thing everything is arch abstracted

So it shouldn’t be that bad

I just need to save and restore interrupts for the spinlocks

Because flanterm has one from the start this time

Because I concern myself with thread safety

[hazy saddle]

And then implement the assembly stub to push registers to the stack

[hazy saddle]

I think I should add a proper RFLAGS parser to this

Mainly for the important bits atleast

also an i686 port or other 32bit architectures may not be off the table

But I will not be designing around that for now at least

[urban mulch]

32bit is pain

[hazy saddle]

Exactly

If it ends up not being too big of a deal with my current abstractions sure later

and there isn’t really a point in it

Especially in satans year 2026

[hazy saddle]

Loongarch64 port is done

It’s on feature parity

[true oak]

deafbeef

[hazy saddle]

That’s normal :p

[hazy saddle]

It’s called a stack smashing protector

And of course it’s gonna be in the registers

[hazy saddle]

today im gonna work on finishing up the first bit

and do the x86-64 IDT/GDT

and loongarch64 exceptions

exception handlers will be arch specific but will call hooks for things where needed to make things work

and il try and abstract IRQs as much as i can

then il work on the PMM

then paging

since im using a buddy allocator

well going to make one

to bootstrap things i will use a tiny bitmap allocator just to setup page tables and the pfndb

And I will make a generic bitmap thing so it can expand its size when it’s about to be empty

Eg pick largest memory region in the system to use for this

Take the first page and make it into a bitmap struct

Then when it needs more if ever it can allocate the past page and link it to the end

[floral kettle]

for that you could just bump allocate from the mmap

[hazy saddle]

idk why water/wux said to use a bitmap

also

GDT INIT [OK]

This code is hella copied (from old kernel)

Except I abstracted it so non BSPs can allocate from the memory management and crap

Well they should be able to with this

You pass in the array and a pointer to the TSS to the functions

being this

For the IDT it will mostly be the same and it will jump to an arch specific common handler so it can dispatch to internal exceptions or generic driver code registered to an interrupt

And I will pre fill all 256 entries of the GDT IDT which may be annoying with how I do it but eh

Just 256 function calls

[hazy saddle]

So I can catch any inerrupt

And I’m just filling them in with stubs

[neon karma]

re-read your message lmao

I was so confused

[hazy saddle]

😭

Fuck

I meant IDT

I’m too GDT INIT OK brained

[marble ridge]

i love this

[hazy saddle]

Do you like the loongarch64 port?

[marble ridge]

hell yeah

[hazy saddle]

Also thank you ^^

[marble ridge]

np!

[hazy saddle]

Il try and keep the loongarch64 port on parity where possible after getting things working on x86

and then once things look good il look into riscv

[marble ridge]

my condolences

[hazy saddle]

I have friends that will want me to do riscv

Hopefully it’s portable enough

[wanton grove]

do aarch64 and become arm developer haters

[hazy saddle]

waow

I’m gonna properly take in all exceptions

And not get a fucking “unknown exception” and get screwed

[neon karma]

do a loop

[hazy saddle]

i did a loop for for the other IRQs

and im using nasm

[neon karma]

yeah I don't think you can use ; for comments in gas

[hazy saddle]

is there a better way to do do this in C

[neon karma]

make an array in asm (with a macro loop)

[hazy saddle]

that is an array?

Or rather make the array in assembly?

Ahh yeah yeah

Lowkey I’m gonna support FRED again too

From the start

I have to lmfao XD

I will redesign some stuff around this

To make it work much nicer

[hazy saddle]

FRED il do after i get loongarch64 interrupts

All that’s left for my first commit is gonna be to port my UART driver

Because I want UART lowkey

UART will be part of early arch unit for x86 too as it will do an initial setup for debugging

And then later set it up as a device for userspace and unhook the kernel logging

Aswell I am considering having a dedicated frame buffer thread that will take control of flanterm so things can print to it and not break even on slow frame buffers

And I can serialize things better like the cursor blinking by having separate qeues

And I can serialize entire lines as filling a buffer is way cheaper

Honestly

This rewrite isn’t going so bad

Who know knowing how to kernel dev

Makes it go by so fast

And I have code I can refer to and tweak for the new systems

Yup I can pickup every single exception right

I may also tweak some things

Later tho once I have the schedular

Where only exceptions save all registers

So it can panic with a full dump

But devices should only need me to save a few registers

[pseudo wedge]

Omg it looks so cool!!

[hazy saddle]

ty :3

im working on my rewrite rn

so not alot of stuff is implemted rn

i had bash sorta ported before

just with a ton of hacks

so im doing a rewrite with stuff done proper

[pseudo wedge]

rippp i hope it will go smoothly :3

[hazy saddle]

it is

[pseudo wedge]

anyways did you make a custom bootloader for EvalynOS or did you use a mainstream one?

[hazy saddle]

i used to have one

but i replaced it with limine

[pseudo wedge]

kewl :3

[hazy saddle]

btw oh god that printf is cursed

but i must for ralsei

[wanton grove]

i want to do this with spamton

[pseudo wedge]

Lmao it’s for the greater good dw

[hazy saddle]

[pseudo wedge]

Gah I should rlly work on my os project but im still at the bootloader stage…

[hazy saddle]

thats why you dont make a bootloader to start

I actually did work on my own sorta bootloader

Though it’s a MB2 shim

So you boot it with MB2

Then you put your kernel as a module

And it does higher half mapping and shit for you

Though I need to fix the protocol a little

Because I made some mistakes

[hazy saddle]

It’s legit the same as making a kernel but even more annoying

[pseudo wedge]

Bleh

atleast I finished it last week and I can go work on the kernel :3

[hazy saddle]

I would also support Limine

As an option

Just incase some hardware has issues your loader dosnt handle

[pseudo wedge]

https://github.com/TeamSHIRO/KURO/tree/elf-loading-alpha

if y’all are interested

[hazy saddle]

il take a look in a bit

Btw

You should support UBSAN

It will crash your kernel if you do any fucky wucky C code

And same for SSP

[pseudo wedge]

hm who is ubsan..

what**

[hazy saddle]

It’s basically free to add the runtime support

Undefined behavior sanitation

And stack smashing protector

The compiler adds checks to see if you do u defined behavior

And if it sees it it will run a function

That you can set to panic

Which can save your butt

[pseudo wedge]

Nice :3

[neon karma]

it doesn't always work

[hazy saddle]

It did make me have to do this

[hazy saddle]

Good to have it than not

[pseudo wedge]

Y’all know everything lol im such a newbie

[hazy saddle]

Given its free

[pseudo wedge]

not the heart reaction ._.

[hazy saddle]

This is what my kernel does

You don’t even need a .h file for it

[pseudo wedge]

But it’s not on the bootloader side, right?

[hazy saddle]

the bootloader and kernel could use it

You just have to compile the files into the binary

Then add these flags

[pseudo wedge]

Okay :3

thanks a lot

[hazy saddle]

    -fstack-protector-all \
    -fsanitize=undefined  \

[neon karma]

you can also just do -fsanitize-trap=all

[hazy saddle]

Somtimes it’s nice to seperate them though

it works

I also statically allocate a 4kb stack for the BSP ISTs

And statically allocate the BSP IDT/GDT

Just so I can always have proper exception handling

Which will be very good when debugging other stuff

Lowkey

I could use a buddy allocator for my heap

But running ontop of the PMM buddy

To have smaller subdivisions

Which honestly

I’m fine with

Also I could maybe do NUMA awareness too

And spawn a separate buddy per NUMA node

And if one node is OOM it can goto other nodes

Just as a basic NUMA awareness implementation

[languid geyser]

you could but then you have to parse ACPI before you have a heap (or pfa per say as you want a NUMA aware one)

[hazy saddle]

i could do a bootstrap allocator like i am for the buddy anyways

[languid geyser]

and personally, I have had enough chicken and egg problems in my Mm to deal with more.

[hazy saddle]

and then move it over and reclaim

[languid geyser]

but you can.

[languid geyser]

you have a temp bootstrap allocator?

[hazy saddle]

il do NUMA later

[languid geyser]

like one that allocates pages?

[hazy saddle]

im going to

yes

[languid geyser]

wow.

[hazy saddle]

just bump allocate off the MM for the page tables and pfndb

[languid geyser]

I just ended up only doing permanent allocations before the PfnDb and slub were online.

[languid geyser]

yeah fair, so I guess the allocator becomes strictly illegal once you have the pfndb built?

how do you handle mapping of the database?

[rancid viper]

lowk in my next design ill just thread a freelist intrusively and then after pfndb is setup ill rethread them into there

[languid geyser]

do you do zones or sparse mappings?

[hazy saddle]

Il do a virtually sparse for all pages

And il have it being zeroed be the non initialized state

[languid geyser]

yeah I do the same.

[languid geyser]

wait what?

[hazy saddle]

Then when the buddy comes arround to using them it can set them up as needed

[languid geyser]

like you zero init all the pages when booting?

[hazy saddle]

So I don’t have to fill in every pfndb entry?

[languid geyser]

or what do you mean.

[hazy saddle]

Oh right nvm

😭

Wait no

When I allocate a physical page

I zero it

So when I’m backing the PFNdb

That state can be considered init

[languid geyser]

like your pfndb entry (pfn lets say) will be valid in a fully zero state? I might be super confused, I think I am not understanding what you are doing.

I am a little stupid.

[hazy saddle]

Yes

[rancid viper]

where are you storing the linked list nodes for the buddy

then

[hazy saddle]

When I setup the buddy

Then it will do thoes

I’m going to lazy init the pages

[rancid viper]

uh sure

[hazy saddle]

So if you have a large chunk o’ ram

[rancid viper]

but wouldnt you want to merge them all

asap

and balance out the buddy

[hazy saddle]

I mean as in. Il only register pages into it lazily into the 2mb section or whatever it fits

Same as I do with my old kernels free list

[languid geyser]

I think that's what I do as well? I just put all of them onto the free list and once I need them I zero the pages on demand... and if any zero'd pages are free I push them onto the zero'd list.

[languid geyser]

ok nvm i completely don't understand what you mean.

[hazy saddle]

Maybe?

[rancid viper]

xd

[hazy saddle]

On my old freelist

I don’t add every node right away

[languid geyser]

like the bootstrap one?

[hazy saddle]

The full PMM

[languid geyser]

how does your pmm work?

[hazy saddle]

https://git.evalyngoemer.com/evalynOS/evalynOS/src/commit/e469e9581f5c0583e3618935ff4e936a88cdacd8/kernel/src/memory/pmm.c#L54
https://git.evalyngoemer.com/evalynOS/evalynOS/src/commit/e469e9581f5c0583e3618935ff4e936a88cdacd8/kernel/src/memory/pmm.c#L141

[rancid viper]

but so when you go to lookup a buddy in the pfn are u checking if it is 100% zero? and what happens when it falls outside of the sparse array

[languid geyser]

is it a buddy freelist?

[hazy saddle]

My old kernel

Is a normal free list

[hazy saddle]

Il use a safe dereference function to return an error if it’s outside

[languid geyser]

like one big freelist of all of the regions from the memmap?

[rancid viper]

how do you tell

[hazy saddle]

Yes

[languid geyser]

okie.

[hazy saddle]

If it page faults

The page fault will redirect to an error handler

[rancid viper]

oof

alright

[hazy saddle]

If it was from that safe function

So PDNdb misses can be detected

[rancid viper]

I suppose its rare enough where that isnt a perf issue at all

[hazy saddle]

Yeah it only happens if you try and free MMIO

Which is rare

[languid geyser]

how does a pfndb miss even happen?

[rancid viper]

well no it would happen if you look for a buddy that doesnt exist, I solved this by populating every page with a "max order" so they never attempt to

[languid geyser]

shouldn't all valid pages have their entry ?

oh wait is that the thing you are doing?

[hazy saddle]

That never happens

[languid geyser]

demand creating pfn entries?

[rancid viper]

how

[hazy saddle]

MMIO isn’t in pfndb

[hazy saddle]

Wdym

The page won’t get passed out

If it’s not in the buddy ever

[languid geyser]

ok i am completely lost gg.

[rancid viper]

in a buddy you lookup buddies blindly

unless u have a regions list that u check against to see whether the buddy falls outside of the region

[hazy saddle]

Sure. But that physical page will exist in the pfndb

[rancid viper]

if its sparse it might not

[hazy saddle]

But then there is no backing physical memory

[rancid viper]

yes

exactly

[hazy saddle]

Eg it won’t be in the buddy

[rancid viper]

but how do u tell what is in the buddy

that is the issue

[hazy saddle]

When I parse the memory map

And add things into it?

And the allocator functions its self ?

The only time it would happen

[rancid viper]

lets say u have page 0x49000 and it wants to know whether to merge with its buddy at 0x48000

how do u tell

[hazy saddle]

Is if I valid values get passed

[rancid viper]

if that buddy is in the allocator

[hazy saddle]

It’s valid physical memory so I check the pfndb

[rancid viper]

no lets say its not

[hazy saddle]

Otherwise they wouldn’t get added

[rancid viper]

lets say the region starts at 0x49000

[hazy saddle]

Oh fuck

[rancid viper]

[hazy saddle]

Then I can just take a safe deref

[rancid viper]

I suppose yeah

[hazy saddle]

Because it’s not that often of a case

[rancid viper]

what I do is I poipulate a "max order" for all pages

[hazy saddle]

Call an assembly function to probe it

[rancid viper]

along with the order

and so that page would have a "max order" of zero

and so it would never attempt the merge

[hazy saddle]

Eg the max order is the order it was originally added as?

[rancid viper]

its just calculated at init to figure out what is the max order it can be before it attempts to merge with non-usable memory

another thing u can do is have regions

and bounds check

[hazy saddle]

True

[rancid viper]

the max order system consumes quite a bit of memory cuz for me at least its 3 bits per pages

[hazy saddle]

Il come to that when it’s time to start it up

[rancid viper]

and order is also 3 bits

for me

[hazy saddle]

I’m doing orders between 2mb and 4kb

[languid geyser]

yeah, but neither it is in the buddy, why would you try accessing the pfndb range in mmio regions...

i don't understand :(

[hazy saddle]

If you try and free memory

When cleaning up page tables etc

[rancid viper]

so 4 bits for order

[languid geyser]

oh wait your pfa allocates multiple pages?

[hazy saddle]

If you happened to map some in there

Like say something MMAPd your frame buffer

[languid geyser]

yeah, so it's in userspace region.

[hazy saddle]

You may accidentally get some MMIO into your buddy if you don’t check

[languid geyser]

or you mean if the kernel maps it?

[languid geyser]

like you try to free it?

[hazy saddle]

The kernel maps the frame buffer into userspace offen

[hazy saddle]

Yes

[languid geyser]

the virtual address?

how would that happen.

[hazy saddle]

How would what happen?

[languid geyser]

you allocate virtual address space and the pointer somehow turns into a physical one, then into a pfn index and make its way into the free function?

[hazy saddle]

When you reclaim memory from a user process. You get back all physical memory you gave it

You walk its page tables to resolve the physical addresses

[languid geyser]

you have Vads.

[hazy saddle]

And then you free the backing physical memory

[languid geyser]

to know what type of memory it is.

[languid geyser]

what...

[hazy saddle]

Sure you can do that too

[languid geyser]

ok i see.

[hazy saddle]

You can convert a virtual address into a physical address

[rancid viper]

thats not ideal

[languid geyser]

so you only track regions via page tables?

[hazy saddle]

What do you propose

[languid geyser]

Vads.

[hazy saddle]

You still have to walk the tables at some point

[languid geyser]

Maple tree.

[rancid viper]

for anon regions, store an rbtree or something with the pages in there

[languid geyser]

💲

[rancid viper]

or maple sure

[languid geyser]

maple.

[rancid viper]

same shit

[hazy saddle]

Yeah I can do that. Then walk page tables for known regions

[languid geyser]

why walk page tables?

you just walk the page tables to free the higher levels when destroying the address space.

[hazy saddle]

Less memory usage than storing all the physical regions describing the pages in an annon region or shared etc

[languid geyser]

you don't want to blindly free the pte's as they might be used by other processes.

[hazy saddle]

Which is why you check the pfndb

[languid geyser]

are you sure that it's worth it and not a micro optimization?

[hazy saddle]

Walking a page table isn’t horribly expensive

And I destroy processes asynchronously anyways

[rancid viper]

well at least in my impl its free

because once its in use its frees up the link node of the buddy

so I can use it for threading the pages into the anon one

holy misstypes

[hazy saddle]

The annon region maybe be large or oddly sized based on how userspace asks for memory

[rancid viper]

doesnt matter?

[hazy saddle]

Then it becomes “not free”?

[rancid viper]

why?

the pfndb has static size

fixed to the page count * sizeof the struct

and if u have to have a link node in the page struct for the buddy or freelist anyway

and u dont need it when the page is in use

it is free to union it for the rbtree or mapletree node as long as they are the same size

(any page in an anon region is in-use)

[hazy saddle]

What?

So your storing the anon rbtree node in the pfndb

???

[rancid viper]

yes

[languid geyser]

i don't.

[rancid viper]

the rbtree that contains all of the phys pages

[hazy saddle]

I store that in the address space object

[languid geyser]

and i don't get why you would want to do that.

[hazy saddle]

Which is why I don’t want that to describe fragmented physical pages

Which is why I would do the cheap cost of walking the page tables and a PFNdb resolve

[languid geyser]

what do you mean by this, describe in what way?

what is the goal of all of this?

like what is the thing this brings you?

[hazy saddle]

In order to reclaim the memory?

[languid geyser]

ok.

i think I understand now.

and your system makes complete sense.

idk why I was so confused.

[hazy saddle]

So when reclaiming the memory I walk the page tables and check the pfndb and change ref count or free if needed etc

[languid geyser]

if I understand correctly you do the most normal thing ever.

[hazy saddle]

lol

[languid geyser]

yeah.

my bad.

[hazy saddle]

It’s all good

If you ever feel dumb

Look in #osdev-misc-0

[round shoal]

why bro is rewriting

[hazy saddle]

Because my old code is kinda bad?

And I’m wanting to get a fresh clean state

And not make certain mistakes I had

[viral bison]

@hazy saddle look at this

it should make mem operations way faster

or well, it did for me

[hazy saddle]

I already do rep movsb and rep stosb

[viral bison]

oh

lol

[hazy saddle]

I don’t even check for EMRS

[viral bison]

you should

it can be used to optimize

[hazy saddle]

It’s still been faster than all naive C copies

And it’s beaten every other non SIMD method on my machines

😭

Me and mint and a few others did some benchmarks

My machines are just straight up built diffrent

💀

[viral bison]

ooof

[hazy saddle]

Like zen5 its worse

Zen3 it’s better

Zen1 it’s better

[viral bison]

HOW

[hazy saddle]

Etc etc

[hazy saddle]

Does it look like I know

😭

[viral bison]

XD

[hazy saddle]

Ofc relative

But like

Yeah XD

My kernel will just use it regardless

And it will use it regardless on the assembly failsafe version for userspace copies

Because it’s easier to implement

[viral bison]

also, for some reason the uacpi performance is atrocious

idk why it's so bad

[hazy saddle]

Did you use TCG by accident?

[viral bison]

no

kvm

[hazy saddle]

What’s your allocator looking like

[viral bison]

tcg has a lot of problems with my map and allocation functions for some reason

[hazy saddle]

Oh also

[viral bison]

https://github.com/Diamantino-Op/HorizonOS/tree/main/src/kernel/kernel-common/src/memory

[hazy saddle]

Mobsb is generally always better

Even on unaligned iirc

[viral bison]

it depends tho

i think with non ERMS

mvosq beats it

[hazy saddle]

Maybe?

[hazy saddle]

I wonder if it’s just slow memory allocation and C++’s hidden calls?

[viral bison]

idk

a second pair of eyes will help, lol

[hazy saddle]

i dont have ghidra installed rn to check a dissasembly

[viral bison]

you can disassembly with linux commands

using specific ranges

[hazy saddle]

yeah but its alot easier esp for mangled C++

[viral bison]

i might ask in general then

[hazy saddle]

maybe use a CPU local here instead of two+? function calles?

[viral bison]

wdym, put the kernel ctx in a local cpu thing?

[hazy saddle]

also generaly you can just use one context for the kernel necause its the same mappings

[hazy saddle]

CPU local varribles? you know what that is right

[viral bison]

i am

[viral bison]

ye

[hazy saddle]

then why two function calls when you should be able to use the same instance and hard code the reference

[viral bison]

the calls just return the var in the kernel class

[hazy saddle]

im just saying calls & maybe virtual method calls can kill performance

in my old C++ kernel with a few friends UACPI had similar issues

[viral bison]

so maybe cache the ref?

[hazy saddle]

also maybe use a slab allocator

verses going to the VMA constantly

[gentle quest]

are you sure your timer code is correct btw

because those numbers dont look valid

even for slowest impl on earth

under kvm it should be faster

[viral bison]

yeah idk, a check may help XD

[hazy saddle]

Maybe use the kvmclock?

[hazy saddle]

Is this a challenge /silly

[viral bison]

i am

lol

anyway, caching the ctx ref did not help

i think the optimizer already took care of that

[hazy saddle]

I just need to finish up loongarch64

And make it have the early arch init callback

And then it’s all good for a good first commit

[neon karma]

bruh you haven't commited until now?

[hazy saddle]

Nope

[neon karma]

that's not a recommended use of git

[hazy saddle]

Yeahhh

welp loongarch64 is backkk

Why is it not working

Wait maybe I need to update Limine

On this laptops install

Because base revision 6

Let’s go

after code review it’s gonna be time for release

And then I’m gonna work on my PMM/Paging

[hazy saddle]

I think for my VFS vnodes will contain pointers to the backend function implementations so for /dev/fb0 it will get routed to the /dev vnode cache and find /fb0 inside of it

And then that can contain its own MMAP implementation for that file

And for this kernel as soon as I have a basic userspace running and crapping out hello world and exiting and cleaning up etc

And on SMP

Il do my VFS

[fiery tangle]

tf did u do to ralsei 😔

[hazy saddle]

Ralsei powers my kernel

He is disappointed it crashes

[fiery tangle]

https://tenor.com/view/ralsei-gun-deltarune-plush-stuffed-toy-gif-8733171219990997497

[hazy saddle]

[true oak]

stalin reborn as a ralsei plushie

[hazy saddle]

I am going to rename the old repo to EvalynOS-old and make a new repo under the old name

And then get my code pushed

Then work on the bootstrap PMM & paging code when I stop feeling like dookie

[hazy saddle]

My old repo is a mess

looking at you 20mb C header file of bad apple video data

Which is why I don’t think il make it a branch

Any other future rewrites will be on a branch

Because my git history won’t be a mess

[hazy saddle]

i think the main branch il call /dev and then later il add the /main branch to squash dev and put onto main when things look good for releases

[hazy saddle]

Crap I need a readme

README.md
TODO

[soft snow]

crap you reminded me i accidentaly rm -rfed mine outta existence

[round shoal]

bro uses a vmm for malloc 💀

use a slab allocator, a dalloc or smthng

[hazy saddle]

Why are you necroing this just to clown a person

:p

[round shoal]

;-;

bro use dmalloc

or use a buddy alloc

[clear bison]

That's how you usually do it...? It's just that you don't call it directly and instead use some intermediary allocator, like a slab one

[round shoal]

i dont reccommend it btw

[hazy saddle]

you kinda have to at some point

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

we are so back

[soft snow]

complete rewrite?

[hazy saddle]

Mostly

A bit is reused here and there with a bit of changes and fixes

[soft snow]

I see your adding aarch support

[hazy saddle]

Nope

Loongarch64

Hopefully that port isn’t left behind horrible

It will atleast make me have good abstractions

For later etc etc

[full fern]

💀 we doing apples to oranges here

[hazy saddle]

https://git.evalyngoemer.com/evalynOS/evalynOS/commit/31cb6a1d44748149a456f738b4a7ec586261d1b0

finished the PMM

gonna do paging tomorrow or in a day or two

[hazy saddle]

FRED is so back

Thanks to tomato and Claude there is now a set of patches that allow it to support FRED

These patches can be applied onto this commit

https://github.com/qemu/qemu/commit/becd22fdc2a071783d9e04421526633772b3b98c

They are vibe coded as fuck

But they work

And it runs faster than SIMICS

And it works on both without any changes so I somehow fixed it

But

This is pretty cool

New evalynOS will be getting FRED support soon ™

But I had to go back to the old project to test this and get this working

after a make install and a symlink it can be invoked like this
qemu-fred-system-x86_64 -cdrom ./evalynOS.iso -cpu max,fred=on,lkgs=on -M q35,accel=tcg -m 2G

[hazy saddle]

Hopefully these patches can be cleaned up later and kept up to date with QEMU releases

But it will have to stay out of tree

[gentle quest]

Why

[hazy saddle]

It’s AI generated

And QEMU dosnt accept AI code

[gentle quest]

Where did it say that?

[hazy saddle]

And the patches are fully vibe coded because nobody I know can work on the QEMU codebase good enough

https://www.qemu.org/docs/master/devel/code-provenance.html

[gentle quest]

Can you send a link

[hazy saddle]

Scroll down here

[gentle quest]

Thanks

[hazy saddle]

Np

[soft snow]

freed

[soft snow]

ehh i vibecode half my patches anyways

its fine

i steal half from astral and then vibe code the rest

[hazy saddle]

well

i think cleaning this up for submission is not too feesible

so i am going to work on paging

i wonder though

Can I make paging so generic

That it’s arch agnostic

And just needs an architecture defined file to say the parameters

ehhh

i dont think its that worth it

[pseudo wedge]

hey fred

[hazy saddle]

yup

and now there is a decent emulator for it

atleast until QEMU makes a full implementation

these patches can atleast be a stopgap

[pseudo wedge]

oh i thought fred was somebody lol

[hazy saddle]

even if vibe coded

[hazy saddle]

nope its very very new CPU features my kernel supports

:3

[pseudo wedge]

kewl

[normal turret]

ah really? Coolio

[pseudo wedge]

god why did i go to osdev as a web developer lolllll
so complicated lmao

[hazy saddle]

yes

sombody vibe coded up some qemu patches

[hazy saddle]

.

apply to qemu and compile

and it should work

[normal turret]

maybe I'll try adding support for it to my OS then

[pseudo wedge]

em vibe coders

[hazy saddle]

its random patches to get emulation support for a feature

its hardly anything vital/critical

and SIMICS is the real deal emulator

its just thats much slower and annoying to work with for devekupment

so having a stopgap set of patches is useful

plus the QEMU codebase is not very easy to traverse by hand FWIW

[slate meadow]

The radix page tables part yes, definitely. The PTE I generate over two arch functions make_leaf/make_dir for those platforms