EvalynOS

[cold barn]

that's kinda dumb for a DSL though if your kernel is expected to have lots of modules

[peak cloak]

also have the instruction manual idk

[cold barn]

if not, then that's fine, just use a checkbox

[peak cloak]

true honestly

[peak cloak]

i mean i have checkboxes too

[hazy saddle]

I want to get a module loader for my kernel at one point

[peak cloak]

theyre pretty much obligatory

[cold barn]

no actual manual 😔

[hazy saddle]

Especially for something like XHCI

[cold barn]

why?

[peak cloak]

yall are fine with limine, so yall will survive borscht

[hazy saddle]

Idk just would be cooler? And if I support reloading could make testing easier depending on the environment

• it would be good if you don’t need the USB module enabled etc etc

[peak cloak]

cooler

[cold barn]

just don't execute that code then

[hazy saddle]

[cold barn]

and/or have a kernel option to disable it

[hazy saddle]

But like tiny code sizes

I mean it would also be fun to get modules working anyways

[cold barn]

make it a build time thing then

I mean yeah, that's fine

but there's probably no technical reason why you need/want modules

[hazy saddle]

Yeah

[peak cloak]

i only want modules for fonts rn

i dont want to have a built in font idk why

[hazy saddle]

It certainly is cool to have though

[cold barn]

just embed one and then load the others using a userspace interface?

[peak cloak]

no

[cold barn]

ok then wait until userspace to load one then

[hazy saddle]

Have your kernel read the initramfs and load it from there?

[peak cloak]

also no

[hazy saddle]

fonts.ko

[cold barn]

eh imo the kernel should limit how much it directly interacts with files

in terms of like, booting at least
it should basically just:
execute /init//sbin/init//bin/init or whatever

[peak cloak]

have the initramfs be loaded by a bunch of modules

[hazy saddle]

Plus parsing elf files it’s given etc?

[cold barn]

yeah but that interaction is prompted by userspace

[hazy saddle]

Though a font.psf being auto loaded isn’t too bad

Epically if you don’t want to embed a font

I do agree Userspace doing it is better though

[cold barn]

PSF files could be considered modules

[peak cloak]

psf is tolerable

i might use it

[cold barn]

so you just specify the path on the commandline, or in the bootloader config, and load it as a pseudo-module

or ttf, or bitmap, or whatever

[hazy saddle]

I also should port UACPI soon

[peak cloak]

and i dont need acpi

[hazy saddle]

I want to do my heap rewrites though

Before that*

[cold barn]

do it

wait werent you supposed to be going to bed

[peak cloak]

i keep making the joke about you needing a programming language to turn off your computer except i know thats false its just funny

[hazy saddle]

I was just finishing up my commit and got distracted here

It’s not too strict

Crappy ATT router please let NAT loopback work properly so I can commit 💀

[peak cloak]

at&t routers suck

one died in a thunderstorm and the other one just bricked itself

im typing from a hotspot

were just getting an actual expensive router instead

[cold barn]

yk you can use another pc (or vm??) for a router right

[peak cloak]

wait what

[cold barn]

unless the problem you're testing only occurs on hardware, then it becomes a little harder to do that but not impossible

[peak cloak]

connect it to The Wall(TM)

[hazy saddle]

My git server is on the same network I am on

And I CBA to setup wire guard in my laptop when a few attempts makes it go through

And it’s my grandmas house

And the router makes you use it anyways with passthrough mode

My clothes get out of the dryer in half an hours

So I should stay up til then to take a shower

[cold barn]

just edit your /etc/hosts to point git.evalyngoemer.com to 192.168.deez.nuts

[peak cloak]

base36 ipv4 addresses

ipv4-36

[hazy saddle]

True true

Then remove it when going home

I already got it commited tho lol

https://git.evalyngoemer.com/evalynOS/evalynOS/commit/c675cd186f50e3cf0e1d92cfecf1b7738ccf3c8a

[peak cloak]

oops wrong thread

[peak cloak]

virtual alloc?

[hazy saddle]

Yes

Also technically if I’m using mlibc and toybox would my OS be mlibc/toybox/evalynOS?

Since the kernel will of course support any userspace you throw at it

Just that I have a designed one in tree etc

[peak cloak]

borscht/evalynOS

[hazy saddle]

That’s part of the built system not a core component of the running system

[peak cloak]

it soon will be

borscht based package manager

https://cdn.discordapp.com/attachments/1132341895717208116/1475556674977927351/speechmemified_images.gif.gif

[hazy saddle]

I’m gonna make my own package manager

In Lua

[peak cloak]

LAU

911, can you get mao zedong on the line

also will this support 32 bit ever

[hazy saddle]

This one no

Maybe later tho

Il make a 32 bit branch

[cold barn]

why lua, specifically?

bash is a pretty common choice for a source package manager

[peak cloak]

because lua is based

[severe marlin]

Bash is meh for a package manager imo, lua is cool

[peak cloak]

id rather make one in C

like a real based person

[hazy saddle]

It would be easier for me to deal with rather than bash considering how you saw my bash scripts lol

[cold barn]

true but I think it would also make you a LOT more comfortable with bash scripts

[hazy saddle]

Trueeee

Though Lua would be easy to embed into the binary that would handle things

& could defer logic to it easier?

[cold barn]

wdym embed into the binary?

[hazy saddle]

Like having C code as the main part of the manager and having Lua embedded like you do with a game engine etc

I am about to sleep now

I am all tucked into bed

Good night :3

[cold barn]

gn!

[hazy saddle]

Ahh the good ol waking up at 4/5 am lmfao

I wonder if I should do the VFS work or keep working on the port if PCC is wanting to do it. Or I could do the lock free heap rewrite which could be fun

[cold barn]

I think you should do vfs work, my brain is numb right now and I'm playing games until I'm tired enough to go to sleep XD

[hazy saddle]

XD and then I fell asleep

[hazy saddle]

ended up testing on some more PCs and had to patch my seriel port driver

[hazy saddle]

Why 3:

[peak cloak]

wrong gif

https://cdn.discordapp.com/attachments/1132341895717208116/1475556674600435773/61d.gif

better

[peak cloak]

nevermind mb

[hazy saddle]

What was I thinking?

The seriel port on the desktop wasn’t being init right

So I had to slap on some io waits to slow it down enough :p

[peak cloak]

yeah i see

[hazy saddle]

https://tenor.com/view/cai-winson-gif-18740869

[peak cloak]

IO abuse

i need wifi router BAD

[hazy saddle]

I mean it’s pretty standard practice

The Linux kernel uses IO waits too

[peak cloak]

flex tape: holds all the hostages in place

[hazy saddle]

The device just is too slow to keep up with my configurations of it

[peak cloak]

fair

[hazy saddle]

So I slow it down by like 10 micro seconds for every write

[peak cloak]

fair

[hazy saddle]

I even use it here if the ACPI timer reads zero to add a little delay before checking it again

https://git.evalyngoemer.com/evalynOS/evalynOS/src/commit/c675cd186f50e3cf0e1d92cfecf1b7738ccf3c8a/src/kernel/acpi/acpi_timer.c#L76

Ahh I think my code ordering there is bad lol

[peak cloak]

RIP

man why do elf binaries make sense

anyway

[hazy saddle]

It shouldn’t hurt anything because it will fall back after that

But I should fix it :p

[peak cloak]

also acpi has a timer???

[hazy saddle]

Yes

[peak cloak]

woh

why

[hazy saddle]

For timing things

[peak cloak]

uhhh kay

[wanton grove]

yes

[hazy saddle]

It’s just a device that’s part of the specs

[peak cloak]

fair

shameless open fartware ripoff

shame unto thee intel

[cold barn]

why do you reverse apply the mlibc patch?

if git apply --reverse --check ${PATCHES}/mlibc-evalynos.patch; then
    git apply --reverse ${PATCHES}/mlibc-evalynos.patch                                                                                     fi```

[hazy saddle]

So it can cleanly apply again when redoing the patch to handle if the patch files changes

[cold barn]

ah

[peak cloak]

what does reverse application even do

[cold barn]

it undoes the patch

I feel like that's pretty obvious 💀

[coarse oasis]

git unapply

[peak cloak]

no it aint

applies it in reverse

go from the last patch to the first one

[cold barn]

...yes

that's the same thing

[peak cloak]

no it starts at the end of the file instead of the start of the file when applying it

you said it undoes the patch

[cold barn]

yes that's the same thing

[peak cloak]

yes

[cold barn]

it starts at the end of the patch and the end of the file, and works backwards

I know you're trying to be funny but you're explaining it correctly

[cold barn]

(well, theoretically at least)

[peak cloak]

bruh why are my jokes actual reality now

whats next, gcc age verification??

[cold barn]

don't give them ideas!!!

[peak cloak]

i already did a joke about that its too late

[cold barn]

wtf is wrong with you 😔

[peak cloak]

i also made one about bash scripts requiring you to be over 18 😔

we're cooked

[hazy saddle]

Gonna have to make a build system in scratch now

[peak cloak]

no but why would someone need this though

[peak cloak]

its like the same thing as a regular patch application

[cold barn]

...to undo a patch? you realize not all patch are applied to a VCS tree right?

[cold barn]

no it's not

working backwards = undoing what you did

[peak cloak]

oh

no i just meant that it goes from the end of the file and goes backwards while parsing and applying the actual patch it doesnt undo it

im not sure if we're talking about the joke reverse patching or actual reverse patching

if actual reverse patching then yes i understand how someone would need that

[hazy saddle]

Top to bottom or bottom to top should apply a patch the same

[peak cloak]

i made an os on scratch, someone made a RISC-V emulator on scratch, so now we port qemu to scratch

[hazy saddle]

Btw PCC rate this fix lmfao

[peak cloak]

its a fix

it fixes

[hazy saddle]

At some point I may make outb and inb do an iowait automatically tbh

[peak cloak]

@cold barn compiler

[cold barn]

what the fuck why

that's really weird

[peak cloak]

it looks normal to me

[hazy saddle]

It fixes the issue lmfao for adding all those IO waits

[peak cloak]

because x86 sucks balls and IBM is better

[hazy saddle]

Do you propose a better solution?

[peak cloak]

make a queue for all the outb's and execute it on every timer interrupt

[hazy saddle]

INB4 you need outb to setup a timer

[peak cloak]

then dont use a timer for the initial outb setup

https://tenor.com/view/cai-winson-gif-18740869

[hazy saddle]

That’s still slower than an iowait which costs a few useconds

[peak cloak]

true

wait @hazy saddle you can actually put an io_wait at the end of the outb's and inb's, like in the function itself

unironically

its gonna be slightly slower but no one is gonna die from this crap imo

[hazy saddle]

Yes I can

And I already said that lmfao

[peak cloak]

oh

void outb(uint16_t port, uint8_t sendWhat) {
  // ASM routine
  io_wait():
}

i mean something like that

[hazy saddle]

Just that would take more of a rework than I wana do rn

[cold barn]

yeah that's what this was #1431140025805832343 message

[hazy saddle]

Esp when your not touching that part of the code too often

[peak cloak]

oh

vision problems

[hazy saddle]

Il put that on my todo during the code cleanup

Or I could do it rn

To procrastinate

[peak cloak]

bruh just open the function and paste a slab of code in there

why yall like waiting for rewrites so much

[hazy saddle]

Yeah but I have to remove the IO waits elsewhere :p

[peak cloak]

oh

then do it

-# incoming send patches sticker

@hazy saddle run a ctrl+r to replace all the io_wait's with spaces everywhere except the IO file itself

[hazy saddle]

That’s now how this works lol

I’m the one who sends it to you

@peak cloak

[peak cloak]

fine

[hazy saddle]

I’m gonna do it anyways it’s fine lol

[peak cloak]

oh ok

[hazy saddle]

Also some IO waits are seperate lol

And still needs

[peak cloak]

cant send patches no more

my talk is quite expensive now

[hazy saddle]

https://a.co/d/0atYxhdd
https://a.co/d/0cPj6uGJ

Look at what I just ordered :3

I’m gonna be implementing a GDB stub for my kernel

And I now have the proper serial ports for it

Also I may or may not make my GDB stub very portable if I can help it (for x86-64)

[hazy saddle]

I think I’m also gonna embed a serial port driver into this & have an option to have it be integrated into yours and or use a specified port

You will need to define ways to access random any physical memory & virtual memory (so you can do fancy stuff and handling if desired)

I will also need a way to hook onto your process system so I will have commands to give an ID to enable a breakpoint for and your kernel can check for that & send commands to let it know it should enable or disable breakpoints

Maybe better solutions that that tho

Or just not support userspace debugging

[hazy saddle]

This also means you could pipe it though XHCI or UDP etc and not just serial

[hazy saddle]

Crap I need to do actual work today

Il finish that stupid outb rework in a moment

And then maybe work on my VFS

[hazy saddle]

okay so i have these functions

ones that are "standard"

and ones with intehrated delay

to make things cleaner

why two you ask?

because its safer

@cold barn and @peak cloak what do ya think

[cold barn]

it's terrible, you should throw it away immediately and switch to web dev /s

[hazy saddle]

ngl i think webdev is harder than osdev :p

https://tenor.com/view/family-guy-css-open-window-annoyed-pissed-gif-12014506

[cold barn]

it looks much cleaner

[cold barn]

nah

[hazy saddle]

i mean for me personaly atleast

my brain cannot wrap arround it well

[cold barn]

modern web dev is, but that's because it's so new nobody has figured it out yet :)

[hazy saddle]

trueee

Btw is that script going well? Btw what games were you playing

Anything new come out that’s good

[cold barn]

haste mainly

and yeah, I think i have it 99% there I'm just trying to get patch to not kill itself

[hazy saddle]

Loll

LMK if you want me to regenerate the patches in a cleaner way btw

If it’s possible

[cold barn]

I see why you did || true now, I could've sworn there was an option to silence it

OH

I remember how I solved this last time

(I also got it basically done yesterday, but didn't switch to dev so I restarted today)

[hazy saddle]

Ahh dang

What was it if I may ask

[cold barn]

you just [ -n "$NOCLEAN" ] && rm -rf <sources>

[hazy saddle]

Huh cool

Also I wonder about storing a hash of the patch file to know when to rebuild each complement?

[hazy saddle]

Fuck you mean this patch is failing on real hardware

this fixes it @cold barn 💀

ya know what just to be safe

[cold barn]

why not make them macros?

[cold barn]

alright I think I fixed it, though I do need to add the inverse condition to patching so if it doesn't clean the sources (NOCLEAN=*), it won't fail if patching doesn't apply

I also parallelized the extraction

might do the same with the download

[hazy saddle]

Isn’t static inline about the same for it being header only anyways?

[clear bison]

So many io_wait()s

[cold barn]

oh, maybe. inline isn't guaranteed to inline it though. the onle one I think it might be necessary in is inbd unless you can restructure it to not need ret

[clear bison]

always_inline attribute smh

[hazy saddle]

I could slap gnu::always_inline? plus for this it dosnt really effect performance as it’s for legacy drivers that won’t be doing much anyways if the compiler decides to do that for say code space reasons

[cold barn]

doing it for performance isn't necessary here, yeah, but I just prefer to have the actual code inserted there if that's effectively what i want

[hazy saddle]

I like the compiler to do its thing for the most part and leave macros to more advanced things liket DBG_LOG

Also I keep seeing online “you don’t need IO waits on modern hardware”

Well clearly ya do :p

This 12th gen intel system clearly is being very picky about it

fuck you mean this breaks it?????

this is a pretty standard seriel setup right?

maybe it will like this better?

Nope

Why the hell is this so picky about the io waits

@cold barn I’m just gonna call it a hardware problem and not support that devices stupid serial port ;p this thing works fine on every other device I’ve tested it on

[cold barn]

yeah that's reasonable

guess you could also look at the *BSD code, if you can test that on the hardware

[hazy saddle]

True true

[hazy saddle]

do you know where the thing is XD

[cold barn]

which BSD is that? Open? Net?

[hazy saddle]

openbsd

[cold barn]

hmm I think open

yeah

[hazy saddle]

the fuck?

my kernel locked up on the othrtr laptop doing nothin???

thats never happened under no load\

[cold barn]

./dev/acpi/pluart_acpi.c
./dev/fdt/amluart.c
./dev/fdt/cduart.c
./dev/fdt/exuart.c
./dev/fdt/exuartreg.h
./dev/fdt/imxuart.c
./dev/fdt/imxuartreg.h
./dev/fdt/mvuart.c
./dev/fdt/pluart_fdt.c
./dev/ic/pluart.c
./dev/ic/pluartvar.h```
try one of these?

[hazy saddle]

lookin

[cold barn]

sys/dev/ic/com.c is it, actually I believe

/*
 * COM driver, based on HP dca driver
 * uses National Semiconductor NS16450/NS16550AF UART
 */``` :)

[hazy saddle]

ooo ty ty

it looks alot more complex

what if i just goto the wiki andddd

https://tenor.com/view/yoinky-sploinky-wd-gaster-gaster-sjf-discord-gif-25028220

[cold barn]

alright we'll see if this works 😂
-# <attachment removed>

running it rn

last time it said during mlibc configuration that evalynos wasn't supported

ah that's why

wrong path to the patch

oops I forgot a .. XD

[hazy saddle]

hm it seems i am not follwong specs

that bit iirc

i dont clear DLAB right away

which may be causing the funky?

let me extract these info functions more like the BSD implementation

[cold barn]

nope nvm I was doing it right I just have no fucking clue why it's not applying properly

WHOAH

okjay

god damimt

it's git apply being dumb and slightly misleading

wellll so far it works 😂 other than that same error as last time make[4]: *** No rule to make target '../../ld/scripttempl/.sc', needed by 'eelf_x86_64_evalynos.c'. Stop.

(@hazy saddle)

[hazy saddle]

ooooo

let me test this

im reworking the seriel driver

making it alot less jank

i may have been clobbering a bunch o' shit

[cold barn]

almost half the size of the previous script :)

and way faster :P

I did end up parallelizing the download as well as the extraction

[hazy saddle]

whats MPFR and MPC?

[cold barn]

obv can't do much more than that

libraries gcc needs to build

they may not be available on the host, so that adds them

[hazy saddle]

ahhh cool

[cold barn]

it'll statically link them into gcc

[hazy saddle]

way better @cold barn \

well time to test

it works in QEMU

[cold barn]

nice

[hazy saddle]

And it fucking works!

And it broke with no changes and a reboot

Wait a darn moment

Let me test somthing

I’m going to send 5 test bytes

And check them all

Just to be safe

Two for two working

Three for three let’s fucking go

[hazy saddle]

testing in just a moment

found a bug with my DLAB and wana test one last time

was setting bit O and not 7

@cold barn huh?

[cold barn]

uhhhh

[hazy saddle]

maybe it diddnt like i was in the root folder

[cold barn]

ah

correct

[hazy saddle]

add this?

START_PATH="$(pwd)"
if [[ "$(basename "$PWD")" != "libc" ]]; then
    cd ./src/libc || { echo "./src/libc folder dosnt exist? Did you download the repo correctly? Are you in the project root directory?"; exit 1; }
fi

# at the end
cd $START_PATH

Or do you got a better way

[cold barn]

I would rather say "run the script from the dir"

[hazy saddle]

True but it makes it more idiot proof imo?

[cold barn]

and have a ./build.sh to run that one

[hazy saddle]

I could hook it into my build system and have it automatic

Let me test it with my patch and then I will throw it into the build system

[cold barn]

what's your current build system?

[hazy saddle]

A bunch of other shell scripts called from a make file

[cold barn]

ah ok

[hazy saddle]

I want to switch to meson at some point but this system isn’t too bad right now so :p

[cold barn]

so, I might be able to help convert it to an actual makefile build system if you don't plan on supporting modules

if you do, I wouldn't recommend makefiles since it gets a little messier (not inherently because of the modules but because scope creep starts to become an issue with the build system)

[hazy saddle]

I do plan to support them at some point. I do wonder how it would look like with it being a real makefile build system though

[cold barn]

you'd have to abuse the hell out of gnu make

or switch to (my beloved) bsd make, and make people install bmake on !BSD (aka linux) systems

thankfully it's packaged pretty widely

[hazy saddle]

Honestly sticking with makefiles would be nice because it does get annoying to manually add things to the list lol but I guess the reason for that in meson is the speed

@cold barn it works!

but it seems to rebuild the entire thing every run?

[cold barn]

thanks!

yes that's somewhat intentional, if you do NOCLEAN=1 (or any value other than empty/unset) it will not remove the build dirs, though it will reconfigure so it might still rebuild some stuff

[hazy saddle]

So it would say still build mlibc on patch changes?

[cold barn]

uhhh

I think you would have to un-apply the patch?

hmm, I'll think on this some

[hazy saddle]

Oh yeah that’s gonna be a bit annoying too because mlibc is a sub module and my IDE keeps showing it as changed

[rancid viper]

is there a reason why you arent using something like jinx or xbstrap

[hazy saddle]

Which is why I considered hashing the patch files and only triggering rebuild when that changes and unappplying the mlibc patches after

[hazy saddle]

It certainly still is a better script. Il just have to keep it manual for now ;3

[hazy saddle]

but i mean a bash script isnt too bad and easier for people to run no?

[rancid viper]

in the case of jinx not really

jinx is a bash script

[hazy saddle]

huh

[rancid viper]

literally the entire thing is a bash script

though it does compile a c that is here-filed into the script iirc

and some other slightly cursed things

[hazy saddle]

Huh interesting

Can I see the repo?

A Google search came up dry

[rancid viper]

https://codeberg.org/Mintsuki/jinx

[hazy saddle]

Huh looks pretty cool

Maybe for my build system rewrite

@cold barn i think its all ready for now. thew this at the top and the commit will be "improve build-libc.sh by PCC"

That good?

[cold barn]

I would do it differently, one sec

[hazy saddle]

that and these are my twaks btw

[cold barn]

yeah there is no need for START_PATH

[hazy saddle]

I can just do a push at the top and a pop at the bottom I presume?

[cold barn]

you don't even need to do that if you're not sourcing the file

it gets run in a subshell with ./build-libc.sh

[hazy saddle]

Wait huh

So I only need the check and the cd?

[cold barn]

you just need a cd "$(dirname -- "$0")" at the top to cd to the directory that the script is in

you'll get put back in your directory when the script exits

[hazy saddle]

https://tenor.com/view/now-you-know-the-more-you-know-no-kidding-no-shit-obviously-gif-1342210079425559607

bash has alot of neat things

[cold barn]

oh, and reverse applying the patch I don't think is necessary? I'm not exactly sure

you MIGHT be able to just apply the patch on top of it

although, you may just want to re-init the submodule every (NOCLEAN=1) build, or just use a tarball

[hazy saddle]

i need it for my IDE to not yell at me to commit the sub module anyways

[cold barn]

not necessarily a release tarball, but a tarball

[hazy saddle]

Yeah I’m probably gonna swap everything to tarballs away from sub modules

They are a PITA

I guess it’s time to reattempt bash & toybox?

Now that the build script isn’t total dookie

[hazy saddle]

Does sub module init wipe any modifications to it?

[cold barn]

well, if you de-init and re-init :)

[cold barn]

well I still can't build the toolchain 😔

but yeah that did improve it

[cold barn]

for IDEs and complex build systems yeah

you're trying to idiot-proof it instead of telling people they're idiots :)

[hazy saddle]

Want me to put some binaries on my website that you can use in the meantime?

[hazy saddle]

And tarballs should work quite nicely

[cold barn]

sure

and can you upload a tarball of your binutils build dir too?

including objs and logs and all

[hazy saddle]

Il zip up both the sysroot and toolchain folders

[cold barn]

I wanna compare

[hazy saddle]

Want me to do the entire libc folder?

That should have it all

may take a moment

[cold barn]

sure lol

OH

and can you make sure this is from a clean tree using my script

[hazy saddle]

let me do that

@cold barn I wonder what part is it crashing on?

[cold barn]

.

[hazy saddle]

Huh damn

Wait what make are you using?

[cold barn]

gnu

[hazy saddle]

Hm

[cold barn]

# make --version
GNU Make 4.4.1
Built for x86_64-unknown-linux-musl
Copyright (C) 1988-2023 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <https://gnu.org/licenses/gpl.html>
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.```

[hazy saddle]

Maybe the port master knows what tf is up with that?

Oh there is something you can try actualy

[cold barn]

sh is dash?

guess I'll switch that to bash

[hazy saddle]

https://distrobox.it/

You can use this to get arch Linux

Given you are on another Linux distro

[hazy saddle]

It makes a docker container for stuff

[cold barn]

I probably can't use that since I'm on WSL1, but I'll make an arch chroot

[hazy saddle]

WSL2 may work better

[cold barn]

I can't use it

[hazy saddle]

That uses a more standard VM

Ahh damn

[cold barn]

I know

[hazy saddle]

Sorry

[cold barn]

I have other software (virtualbox) that needs the virtualization capabilities

its ok lol, not your fault

[hazy saddle]

Could make a full Linux VM inside there

[cold barn]

nah a chroot should work

[hazy saddle]

🤞

@cold barn crap i forgot to switch to dev branch

rebuilding

time to reset the counter

[cold barn]

god damn arch is a pain in the ass to bootstrap

[hazy saddle]

@cold barn zipping up now

[hazy saddle]

they should give out premade chroots like ALARM does

it really helps get things started

[cold barn]

yeah I figured that out eventually

I was trying to use pacstrap/pacman and it's IMPOSSIBLE to get the gpg signing on everything to work

[hazy saddle]

crap i dont have my file server password on my laptop

💀

i can manualy copy i with a USB there

just may take a bit longer

[cold barn]

💀 no ssh keys?

[hazy saddle]

oh true i can just do that

[cold barn]

I literally just replicate an encrypted tarball of my ssh keys on basically every machine lol

need to log in from a new machine? sftp into one (with password), decrypt into /tmp, transfer, delete /tmp file

[hazy saddle]

i just manualy copy the keys over

and i have a public key on my domain name to pull easy

copying it over

copyparty may allow you to download while i upload but it may now support it with me using sftp @cold barn

[cold barn]

I'll just wait until it's done

[cold barn]

btw zstd tends to have a better ratio of compression to creation/extraction speed compared to gzip or xz

[hazy saddle]

I wonder if I can add that to the default on my KDE menu?

also I’m going to sleep now

Have a good [[timezone]]

[cunning oracle]

https://osdev.wiki/wiki/Timezone?action=edit&redlink=1

[cold barn]

goodnight!

[cold barn]

yeah building it under arch works

idfk

[cold barn]

OMG @hazy saddle I think I figured out a way to continue using submodules (my beloved ❤️) without having to fuck around with them

just fork any that you need to modify, do the modifications there, and vendor the latest change in your evalynos repo

I was seriously not thinking about it that hard since I tend to just not need submodules 😂

but yeah right now I'm just working on getting evalynos built inside arch

I might eventually take a look at your patches to figure out where the fuck the arch-dependency is happening

if you don't get to it first :)

[hazy saddle]

And I already have forks on my mom codeberg accounts lol

Il do a poll for this a bit later I think

[hazy saddle]

I can try an alpine Linux build in a moment and see maybe if it’s just an older packages thing?

I will also test WSL on my grandmas computer for this

[hazy saddle]

@cold barn compiled just fine on alpine Linux edge with a minor change because OVMF paths

Crap I am on master

I thought I did the command

I did git branch and then checkout

So it must of broke shit

But it was your script file PCC

It is gonna have to be WSL2 but I’m gonna test it on Debian there

[hazy saddle]

Oh shit KVM works under WSL2???

Let me just do a bit more testing as I still need to make the toolchain

And it’s a bit picky about audio server and XDG

But it’s booting

[hazy saddle]

It also wants sudo but that’s an easy fix

@cold barn I don’t know if this is viable for your setup but everything works perfectly under Debian WSL2 and your script on the proper dev branch. I’m even using KVM

But if everything works for me on 3 machines running diffrent environments

Somthing may be going badly on your end

I am honestly shocked at how good this works

I just had to relogin and add the group

And all issues went away

Props to the NT kernel team and whoever did the userspace bits of WSL

This is very good

Now make the rest of the userspace good

[cyan bison]

WSL2 is just linux VM

[hazy saddle]

Well yes

But I’m surprised that it can do nested VMs properly

And open GUI crap

Etc etc

[wooden maple]

WSL1 was cooler

[cold barn]

Void is what I'm using btw, but that might work too

[cold barn]

You can just do git switch

[hazy saddle]

[cold barn]

You might want to just vendor the ovmf builds from the ovmf-nightly builds. If you can't find it, I will in a sec. I think it's on the Codeberg mirror of that GitHub osdev0 account

[hazy saddle]

I should also vendor limine

[cold barn]

I'll start with a fresh void chroot and go from there then

[hazy saddle]

I could have it part of my get deps script

I don’t like binary files in there

[cold barn]

Hyper-V is known for that lol

[hazy saddle]

Ahh lol

Also my USB to serial cable is here

Bus 001 Device 016: ID 0403:6001 Future Technology Devices International, Ltd FT232 Serial (UART) IC

[hazy saddle]

it works :O

after a few correctess tweaks

tho i think i need to swap the flanterm locks to not disable IRQs

and possibly use a kthread for it

that way printf goes into a buffer

& that will flush it

maybe same for serial?

though thats much less intensive of a task so its fine

i wonder what linux does here

[soft snow]

What works.

I'm so confused

[hazy saddle]

Serial driver

On real hardware now

[soft snow]

Fire

[hazy saddle]

I’m gonna clean it up a bit more

As it’s a lot of magic numbers

And I should abstract things a bittt more

But it should be all good after than and be “perfect”

[hazy saddle]

magic number b gone

[peak cloak]

https://cdn.discordapp.com/attachments/1018234499865268365/1467486440534577214/ROBOTA_MU_WYGASA_JADYMY_NA_KURWY.gif

[hazy saddle]

But yeah the serial driver is now in a wayyyy better state

It used to be a copy of the one from the wiki almost

Which well is not exactly the best for real hardware and it being extendable

I’ve split everything out into functions etc to make it all nice and easy to work with

If I ever get termios etc to set baud rates etc etc etc

What’s good is I now have an Intel and an AMD machine with hardware serial ports

Which means any issue I have should be solvable on these if it’s a CPU feature crash / issue

And I should be able to help others etc

Only issue is they live at separate houses

[hazy saddle]

@cold barn I’m about to go to sleep

Did you ever manage to get it to build right lol

[cold barn]

yeah I need to do homework and shit though so I gotta lock in before I do anything else

I also got "distracted" with other obligations for most of today tbh

[hazy saddle]

Understandable

Good luck on the homework o7

[hazy saddle]

@cold barn https://git.evalyngoemer.com/evalynOS/evalynOS/commit/a923a02e3c002e8e9b8c44de2b503804948e1824
your commit is in :3

https://git.evalyngoemer.com/evalynOS/evalynOS/commit/70dbea79056d45783d9f9344d3a3d90c7762056c

and the new seiral driver is here

[hazy saddle]

Of course Amazon fucking delays the bracket by two days :p

After the window of me being able to reorder it for same day

So now I have to stay here at this house and wait for overnight shipping

[fiery tangle]

what bracket?

[hazy saddle]

The thing I need to get a serial port

https://www.amazon.com/StarTech-com-1U-Rackmount-LCD-Console/dp/B001Y1F0HW

[fiery tangle]

ah ok

[hazy saddle]

This thing

[hazy saddle]

i got the thing in there

[peak cloak]

[hazy saddle]

@peak cloak okay i was jokin

but i had to remove my gpu

💀

shit was scary because i couldnt find the release

and i popped it out half way

the serial is working

but the driver failed once

[peak cloak]

gg

[hazy saddle]

maybe i will do several self test methods if one fails

ther is a scratch register i can use to test

yeah there is a register on the serial controller that gives you a byte of memory

[peak cloak]

make sure you dont write to the CPU_SELFDESTRUCT register

[hazy saddle]

no ofc not

just the serial self test failed

[peak cloak]

or try to set big endian mode on x86

(you bend 15 cpu pins or more)

[hazy saddle]

GDB stub time

I will have two modes for SMP

One for latching onto a core

And one for latching onto a thread

You will of course need a lot of kernel support

For sending IPIs etc

And if a core has IF=0 and the serial port is on there well

Not much I can do about it

If your having issues you can always dedicate a core :p

I will include serial drivers for this

But I will support generic TX/RX layers

So you can use UDP with this or whatever else etc

[hazy saddle]

Oh sick QEMU lets me use the physical serial port as the backend

So I can work on it really nicely

i think i may have to use polling on the serial driver?

if i am in a #DB context i dont want other IRQs

[hazy saddle]

Okay so

I am connected

I think it’s going okay?

But

I think it’s tweaking out

It locks GDB into an infinite loop

I am only sending an empty packet to everything but

The fuck is wait exit enter?

i also stop getting any packets or data?

yeah i am getting the same packets

i get nothing after

[sterile plank]

:hmmCat:

[hazy saddle]

i mean i should prolly like

implement some

[sterile plank]

i wanna make a joke

have you tried debugging it?

[hazy saddle]

you know whats cursed

about how im doing this btw

[sterile plank]

tell me

[hazy saddle]

i have the VM connected to my physical serial port

thats connected to my usb adapeter

and im pointer gdb to the usb adapter

[sterile plank]

Oh great heavens

[hazy saddle]

all on the same machine

[hazy saddle]

@sterile plank progress!

[sterile plank]

deltarune

[hazy saddle]

Yes I’m listening to deltarune theory while osdeving

[sterile plank]

now listen to osdev theories while deltaruning

[hazy saddle]

I now have to implement the register packets

So I’m going to have to add the right callbacks for debug and breakpoints

So when it gets one of those

It will capture it and then spin loop waiting for more GDB stuff

So it will have to nest interupts

Which is supported I think

Or maybe I don’t need that

No I will but

Okay yeah this will work

In the exception

It will store a pointer to the stack with all of the register state

So then when GDB packets come in

It can modify and read those

[sterile plank]

I wish you best of lucks :pp

[hazy saddle]

Then when that exists and the interrupt is popped

[sterile plank]

i havent been osdeving for a long time

[hazy saddle]

Then the #DB will return and have the new register information

[hazy saddle]

Ty :3

[hazy saddle]

In the spinning loop il possibly spin similar to a lock so it knows when its ready to continute

Using atomic etc because this crosses context boundary’s

[hazy saddle]

Tbf you won’t need full nesting if you just make it enable interupts on that context so then the other ones can follow in and eventually return?

[sterile plank]

i have no idea

[hazy saddle]

its oke :3

im just talking into the void of my thought process

[hazy saddle]

@sterile plank

It’s coming along well

I wonder if I could support reloading the kernel over serial

Like it would allocate the memory needed to put a new image there

And load a trompolihe

Get the new image over serial

And then reload the kernel using LBP

Reusing all the info I got earlier

[hazy saddle]

To be clear this is only one snapshot

I still need to add single step crap

Because rn GDP always is returned the one thing

[hazy saddle]

chat my schedular may be fucked

Okay well good news bad news

Good news

The debugger works

Kind of

Bad news

It gets trapped inside that loop

And won’t debug anything else

Somthing seems to be saving and restoring that context there

And switching to it

Which

May be the first thread?

It could be that

Because then it gets first launched to

Yeah yeah

Not sure the proper way to solve that right now

and well somthing switches to it

but it dosnt sleep nice

or it does idk i cant read cpuusage well

but it acts like the idle thread

then the debugger gets trapped there

okay yeah

the first thread spawned

gets clobbered

which isnt an issue

because i can just spawn idle thread first

and then it gets clobbered

with what basicaly is an idle thread

this is what it looks like when its stuck

i think i just need the schedular to unset & set the TF on any returns from the preemption

eg here

Then this way when going into any context after a preemption

It can properly enable and disable TF

If task switching to needs debugging
Enable TF
Else
Disable TF

Also @peak cloak I’m gonna add borscht

Because I’m gonna make this configurable

As a lot of this may be dead code

So having it configurable will be good for performance

With all the extra branches

[peak cloak]

YOU'RE WHAT

good luck finding instruction manual

tell me also if you cant understand it i want people to actually be able to read it

also use version 2.0c if you dont want to kill yourself

[hazy saddle]

What did you change

[hazy saddle]

Also why notttt

[peak cloak]

there's 1.3fa backport and then there 2.0c, 2.0+ has customization options and also the gui alignment doesnt suck balls anymore

[hazy saddle]

Make one

[peak cloak]

[hazy saddle]

Why do you even feel the need to do backports lmfao

[peak cloak]

blind

[hazy saddle]

One day blinding stew

[peak cloak]

maybe some esoteric idiot wants to try 1.0

its literally all bugfixes to 1.0 anyway

so 2.0 is the only actual update

the backport was kinda important anyway, if it was a bug id leave it alone but it literally is so bad that make won't read your makefile properly

[hazy saddle]

[peak cloak]

https://git.evalyngoemer.com/kvass/borscht/releases/tag/2.0c

documentation.md

https://git.evalyngoemer.com/kvass/kvass-docs/src/branch/main/specs/borscht.md

its also here

still no GREGloader commit

[hazy saddle]

Thank you

This will be kool

[peak cloak]

kallegedly

KDE allegation tool

[peak cloak]

kool - KDE smart fridge

anyways tell me if it explodes somehow

the documentation is quite indeed utterly confusing

[hazy saddle]

Nah it’s not that bad

[peak cloak]

ohok

[hazy saddle]

Il have a make target that will download and compile it

And then let you use it

Otherwise it will use default options

[peak cloak]

download it and compile it

genuine insanity

and then i break backwards compat /j

"Oh no!" you scream, as you literally can't compile your project

It's just like the XAPIC

[peak cloak]

also you might get brutally segfaulted if you try to use an invalid config

not always but theres a risk

it works fine with configs that dont suck balls

[hazy saddle]

It’s called

Pinning to commits

[peak cloak]

insane

but anyway im not breaking backwards compat frfr

[hazy saddle]

https://tenor.com/view/touch-canttouchthis-ucanttouchthis-mchammer-musicvideo-gif-4723913

[peak cloak]

if i add more features its not gonna conflict im certain

[hazy saddle]

That’s what major versions are for

[peak cloak]

until i dislike a certain feature and do an x86 and remove it

intel sucks balls, i thought they at least stood by backward compat

but no

[hazy saddle]

loveley

this breaks things

i have zero idea how this should work

[hazy saddle]

Ngl I may have to move this to a feature branch

And just do my VFS soon

But I’m really close to finishing this

Just lost

[peak cloak]

RIP

[hazy saddle]

Maybe when obberow is free he can help like he said

[hazy saddle]

Wait

What if it already has been working

[hazy saddle]

And this is fucking things up

[hazy saddle]

And I’m just catching it in ifle threads

Because I don’t have breakpoints yet

[hazy saddle]

Okay yeah it seems to just be breakpoints and them being broken

I have swapped serial to be an NMI

It does not like having a “cli” inside of the #BP handler though

It got very corrupted

Though maybe I should setup ISTs

Well I should :p

[hazy saddle]

Well more so locked up badly

[hazy saddle]

I tried to add a breakpoint manually and it well didn’t go well

So I’m gonna code in hbreak support for GDB

Not sure how to tackle software ones for this :p

Hopefully it won’t break badly?

Because with the manual break it kept trapping on the same one after running s

Maybe that’s just some other issue

But it is working almost

[hazy saddle]

This will be good because it will then be on a whole seperate context

Well I hope that still means I can fuck with the frame that gets restored on IRETQ

Also instead of a CLI il swap up some handling to not set sti earlier etc

[peak cloak]

fair

crap aint i supposed to sleep

[hazy saddle]

No that’s me :p

[peak cloak]

yes

[hazy saddle]

Amazing

I woke up

It’s now randomly not working

Even after undoing changes

It’s just fucking packet overflowing constantly

I am going to have to scrap this all

Aren’t I

and restart

[hazy saddle]

I have an idea!

IRQLs

This way everything can be masked except serial

Which means I don’t have to mess with cli/sti

And messing things up

Which means things can be dealt with cleanly

[hazy saddle]

IRQLs suck

They make my kernel run doom at 1 frame per 7 seconds

Atleast during the fireworks test

Before it could atleast somewhat be playable

[fiery tangle]

[hazy saddle]

CC: @urban vale I would like to ask if you have any idea what’s causing this

wait

guess who was using cli/sti still

:p

i fixed it

now IRQLs work

[fiery tangle]

Paru!

[hazy saddle]

I am doing it the lazy way

Without CPU local

So I won’t be able to emulate it on other architectures

Sadly

But that’s a problem for future me

[hazy saddle]

Sorry about the ping lmfao

But now I can properly make serial have super high priority

And then raise to the level below it

So only serial can interupt me

I also could just not use IRQLs on CPUs lacking support

And fallback to other methods

Because for most things I’m not using them fully

Sooooo I can use a few ifdefs here and there

Or arch spesific abstractions

oh btw all my IRQLs are at IRQL2

so im not sure if anything is breaking

wait i can test this

this seems to work

wait a second

i can use a lookup table for IRQ to IRQL translations

oh and thats bad because i shouldnt change IRQL on exceptions

nvm i already handle that

[hazy saddle]

Also

That’s not how IRQLs work

😭

[clear bison]

After looking at IRQLs I can't believe i just reinvented them in my previous kernel

[hazy saddle]

Lmfaoo

They are pretty cool

And it’s gonna let me implement my debugger

Without having a ton of issues

Since it will be able to preempt everything

Even when spinlocked etc

[hazy saddle]

I set an IRQ to IRQL15 and it works

With like properly setting the vector

[kindred stirrup]

My fireworks test tends to slow everything else down with its hundreds of threads

[hazy saddle]

Nah normally it runs doom finely

My fireworks spawns a shit ton of hello worlds

In a batched way

It was because I still had spinlocks setting cli and sti

Through the schedules

Because it uses a different unlock function

[kindred stirrup]

They should instead raise IRQL

[hazy saddle]

Well yes

My inline functions for spinlocks that everyone use did that

But the schedular has a special unlock function

Where assembly can call it

Which I forgot to update

Fixing that makes it run smoothly

[fiery tangle]

me?

[hazy saddle]

No vector as in interupt

I am bad with names so I call them vectors

Instead of interupts

Also because I can’t spell

[kindred stirrup]

"Interrupt vector" is an actual technical term

The x86 real mode interrupt dispatch table is called IVT