OBOS (not vibecoded)

[flint idol]

i also might remove the dependency on the popcount instruction

for timer IRQs I will be using the lapic timer on the bsp

[flint idol]

oh and while im making changes to the kernel im also gonna make it so that my tcp sockets can shutdown on close asynchronously to avoid having to send a reset

what I currently do on tcp socket close is reset the connection

https://github.com/OBOS-dev/obos/blob/master/src%2Foboskrnl%2Fnet%2Ftcp.c#L1061-L1066

https://cdn.discordapp.com/attachments/1271481080343695486/1443350118387679403/togif.gif

[real pecan]

Obos doesn't concern itself with stability

Or maybe its stable now actually

[flint idol]

depends on the moon phase

and what u r testing

for example it might hang on real hardware with smp

or maybe it will hang on real hardware with smp disabled

it may also crash if you blow on it wrong

but generally it is better

than it was in February

there have been more safety measures added for multithreading

mostly with initialization of waitable objects

i.e., mutexes, events, etc.

oh yeah a day or two ago I added file moving to the ext2 driver

what i did is i made it delete the old entry and replace it with a new entry in the new parent

and when im done my timer interface changes

I think i will finally

add usb support to obos

starting with xhci

[real pecan]

That's good

[flint idol]

maybe I'll also add ehci and uhci support once I understand how usb works

on a general basis

after that I might port sshd

or i might work on fvwm3

[flint idol]

(I'll also add a usb keyboard driver)

[flint idol]

bruh

why is cpuid 0x15 returning all zeroes

in qemu

cpuid leaf zero

reports

it exists

cpuid 0x80000007 reports invariant tsc as existent

[flint idol]

im gonna continue this

tomorrow

gotta sleep

[flint idol]

ive mastered the art of zoning out

im not gonna be consciously aware of my actions at school until I get back home

its really convenient when I want to skip through parts of the day

currently i have a few hours left until I fade back into consciousness

[uncut narwhal]

i dont think its required to give nonzeros, and qemu doesnt implement it ever in any form ive done a ton of digging for that

you can get tsc freq on qemu from the hypervisor leaves if youre under a hypervisor instead of using tcg though

i spent like a week trying to figure this exact issue out and getting mad over it lmao

qemu just doesnt implement it for some reason shrug

it sucks

[flint idol]

i implemented the thing

for calibration

and it's pretty accurate

and i just pushed the inv. tsc code

[uncut narwhal]

god i wish qemu implemented leaf 0x15

might try to make a patch for that sometime tbh

[flint idol]

ok so i made doom use rdtsc

and the performance is exponentially better

and also the kernel feels a bit more responsive

[flint idol]

kms

why

does

it

have

to

be so hard

for the kernel to keep track of a GODDAMN NUMBER

FHSDUGHJDSLIHNFDIP

ok fixed it

[stark sigil]

lol pi

[stark sigil]

yay

[flint idol]

lol where

[stark sigil]

thats the only number i can think of that might be hard to track

[flint idol]

im writing an ifconfig command

[knotty totem]

unironically why did linux switch to iproute2 🥀

[flint idol]

(this won't be similar to the actual ifconfig command)

[knotty totem]

then name it smth different

netconf maybe

[frigid tide]

configip

[flint idol]

fun fact

obos cannot communicate with another obos

over tcp

[flint idol]

i named it ||obos-ifconfig||

LoC update

to whomever it concerns (absolutely no one)

[vale nymph]

Crazy

[dense vigil]

Just FYI even this leaf is bunk on real hardware

[flint idol]

I forgot we cant have nice things on real hardware

oh well I'll just add a kernel cmdline option to allow using cpuid 0x15 for the tsc frequency

[dense vigil]

is it even worth it when calibration + maybe 3 rounds + average is fine?

that laptop is pretty new too. almost no CPUs properly support said leaf

[trim bay]

this code will most likely not work due to an integer overflow

and due to integer division inaccuracy

instead of ecx * (ebx / eax) (which does a 32 bit multiplication with one of the operands being rounded) you should do ((uint64_t)ecx * ebx) / eax, which does a 64 bit multiplication and division and only rounds the final result

[dense vigil]

Oh wait wait was that why? let me test that actually

It was close to an okay value

Oh wait my USB is missing il test it later if i find it

But that is probably why it broke

Bonk me with a stick

[flint idol]

i didnt actually think i would get this far

i haven't done driver dev

for basically a year

god damn it

[flint idol]

these people are fucking crazy

this specification is insane

wtf is a trb and why do we need so much of them wtf

okok that makes sense now

[flint idol]

im hoping by the end of the day ill have so much to show as like. a completed transfer on xhci, or at least port enumeration

i feel like the people at intel coulda designed the TCBs in xhci to be better

but i got no better suggestions rn so i cant really be talking

[flint idol]

man is writing structs dull

ok done writing structs

[stark sigil]

man its fun

you get to think for hours about what crap you put into a uint64

[dense vigil]

uint8 flag: 1

There it’s a single bit

And you can put more for more flags and it will pad it properly to not take up too much

And you can access them without bit shift ops

[stark sigil]

FLAG FOR WHAT cool

also yeah ik bit fields exist

[dense vigil]

[flint idol]

bad practice

[dense vigil]

Really?

[flint idol]

in c yeah

compiler can do whatever the fuck it wants with that, in theory

although it does have to follow the abi

[dense vigil]

Of course it can

Of course

Would that cause issues for internal code and that’s all compiled by one compiler?

[trim bay]

if the bitfield is only accessed by code you're compiling it's fine

the issues occur when you use bitfields for hardware stuff

like page table entries, device registers, acpi tables, etc

[stark sigil]

true

but attribute packed

[flint idol]

it's

it's all making sense now

[stark sigil]

ik thats probably gcc (and also clang) only but like

[flint idol]

xhci is making sense now

[stark sigil]

USB

[flint idol]

what was that one meme

with like

idk

how to explain it

[stark sigil]

https://cdn.discordapp.com/attachments/1305395406662012928/1461085614194688050/speechmemified_image-8.gif.gif

[flint idol]

not it

[stark sigil]

idk either 😔

[flint idol]

yeah trying to remember what it looks like is giving me SIGSEGV

OHHHHHHH it all makes sense now

[dense vigil]

You want to pack most things for external things anyways right

[stark sigil]

yes

even for internal things i do it sometimes

space savings 👍

[dense vigil]

That’s bad for some things because it slows things down with misaligned memory access or some shit?

[stark sigil]

id do anything to fit some fart into 64 kib of space ngl

i mean idk fair but still

[dense vigil]

Though speed isn’t relevant if you aren’t doing it often

Esp with this

[flint idol]

wait so if im not mistaken:
command TRBs go in the command ring
while transfer TRBs go in the endpoint TRB

yeah that makes enough sense

[flint idol]

how tf does the event ring work

specifically the "Event Ring Segment Table (ERST)"

ok i get it now

[flint idol]

ok well im testing it

and im not getting any IRQs

idk if it's the xhci driver

or the msi-x implementation

i know that msi works

and when i use the PCI IRQ pins, the kernel constantly gets IRQs

but the xhci controller never reports any IRQs anywhere

[flint idol]

i hate reading c++ code

[flint idol]

so like

this makes no sense

[knotty totem]

yeah

what's up though?

like, what are you trying to do?

[flint idol]

well im trying to get IRQs from the xhci controller

to no avail

idk if it's MSI-X or the xhci driver

and i have no way to test MSI-X

fuck it im just gonna debug qemu to see what's up

well it's not the xhci driver

it's MSI

it isn't MSI, nor the xhci driver

i am a dumbass

[flint idol]

well i can receive TRBs now

[flint idol]

today I shall continue

with xhci

and i will be initializing ports today

and handling device connections and disconnections

and after that I will be adding an interface to send TRBs

[flint idol]

of course this will be after school

[flint idol]

I will try to make it as controller independent as I can

[flint idol]

I'll probably have some enums for the different packet types when an IRP is submitted

and I think it is safe to have the different packet types be the TRBs described by xhci spec section 3.3

and i will also start on the usb stack

starting with adding kernel functions like Drv_USBDeviceAttach and Drv_USBDeviceDetach

and i will also add a field to the driver header for recognized usb class codes nstuff

for pnp

[flint idol]

and the usb stack will be responsible for making sure drivers know their device was detached

I will likely add a function to the driver function table like:

void(*on_detach)(dev_desc dev);
obos_status(*on_attach)(dev_desc dev);```

I might even use that for pci as well

[flint idol]

I will also likely have a method for userspace to be signaled that a usb device was attached so that it can syscall and get the kernel to load the appropriate drivers

i don't think im gonna expose xhci controllers to devfs

at least not directly

[flint idol]

I will also have equivalent functions for usb controllers

except those probably wont be hotpluggable

[flint idol]

ok so

queuing TRBs is not working

[flint idol]

ok queuing TRBs works

im working on port initialization

topology defined
well thanks? now how do I FIND OUT THE GODDAMN TOPOLOGY, HUH XHCI?

[real pecan]

Who knows

[flint idol]

(the logs that are not prefixed with [ DEBUG ] logs are from qemu)

[stark sigil]

logs 👍

[flint idol]

logs indeed

[stark sigil]

i mean i think your driver works

[flint idol]

im writing the base for the usb stack rn

[stark sigil]

oh

cool

[flint idol]

yeah the xhci driver can initialize a port now

and i think it can also initialize the control endpoint

and it can send control TRBs as well as endpoint TRBs

[stark sigil]

fire

[flint idol]

anyway

i have no idea how usb hubs are supposed to work

but who needs those anyway?

[stark sigil]

me

i need them'

i need more usb ports than i have

[dense vigil]

Don’t some computers have internal hubs sometimes or some shit?

[flint idol]

yes

im gonna do hubs later i want to get this to work first

[flint idol]

neat, i got another star

shoutout to the guy who just starred

me

and to everyone else who starred me

tis much appreciated

[knotty totem]

Meeee x3

For some reason I hadn't started it yet

[flint idol]

typedef enum usb_trb_type {
    USB_TRB_NORMAL,
    // only legal for endpoint zero (control)
    USB_TRB_SETUP,
    USB_TRB_ISOCH,
    USB_TRB_NOP,
    USB_TRB_CONFIGURE_ENDPOINT,
} usb_trb_type;

// TRB Direction is defined by irp.op

typedef struct usb_irp_payload {
    usb_trb_type trb_type;
    union {
        struct {
            struct physical_region* regions;
            size_t nRegions;
        } normal;
        struct {
            struct physical_region* regions;
            size_t nRegions;
        } isoch;
        struct {
            struct {
                uint8_t bmRequestType;
                uint8_t bRequest;
                uint16_t wValue;
                uint16_t wIndex;
                uint16_t wLength;
            } OBOS_PACK;
            struct physical_region* regions;
            size_t nRegions;
            // TODO: Return status in some way from the status stage TRB (xhci-only)? 
        } setup;
    } payload;
} usb_irp_payload;```

this is what i have in mind

for the usb stack so far

i mean like

the first layer

or whatever

idk what configure endpoint will look like yet

[knotty totem]

Oh I probably was remembering starring the GitHub repo a while ago

[flint idol]

probably will just be empty

ah

[uncut narwhal]

just realised i somehow hadnt starred obos yet either lol

[uncut narwhal]

as you can see i fixed that

[flint idol]

yes

[flint idol]

marker

[flint idol]

yay

anyway before i continue with the usb stack

i need to implement port detaching

in the xhci driver

[flint idol]

i just pushed some code for the usb stack

as well as some xhci code

now time to implement xhci IRPs

but first

i want to add a submit IRP function for the USB stack

[flint idol]

done that

[uncut narwhal]

glad im not the only one using IRPs for io

[flint idol]

IRPs my beloved

[uncut narwhal]

its such a good pattern

so much so that the zig standard library is migrating to basically IRPs for io lmao, even outside of osdev context

[flint idol]

i just pirated the usb3 spec

letsgo

[flint idol]

todo for tomorrow: get endpoint transfers to actually work

[weary hound]

you know which root port a device belongs to be keeping track of the whole tree of hubs until the root port

and on the root port you know which one it is because you got an event for it

[flint idol]

oh ok thanks

[flint idol]

endpoint transfers work now

but my control transfers are not working (qemu asserts and dies)

[flint idol]

ok well

now i am getting CC_STALL_ERROR on my transfers

usb_xhci_fetch_trb addr 0x00000000620db000, TR_SETUP, p 0x0012000001000006, s 0x00000008, c 0x00030841
usb_xhci_fetch_trb addr 0x00000000620db010, TR_DATA, p 0x00000000618b8000, s 0x00020012, c 0x00000c01
usb_xhci_fetch_trb addr 0x00000000620db020, TR_STATUS, p 0x0000000000000000, s 0x00000000, c 0x00011021

for reference im tryna send a GET_DESCRIPTOR request

i think i see why now

ok fixed it

i now receive the device descriptor

but it seems incomplete

so im gonna actually properly initialize the device first and come back

[flint idol]

(by this, i mean that the class codes are zero)

hm

[weary hound]

class code 0 means determined by interface iirc

yep

[flint idol]

thx

lets go

the usb thing configured the port

[flint idol]

i added some code so that

the kernel can attach drivers to usb devices

although, automatic driver loading for usb is not yet implemented

i need to implement aborting IRPs if the port is detached

im thinking of having an extra event pointer in the IRP for the controller to set when a port is detached

then Drv_USBIRPWait will wait on both events

(ill probably also do the same for normal IRPs)

did that

now

in theory

i have all i need to implement a keyboard driver

so ig that's what im gonna do

implement a usb keyboard driver

[dense vigil]

That basically seals the deal for being compatible with modern hardware imo right?

Because you don’t need a PS/2 keyboard

That’s gonna be so cool

[flint idol]

assuming the kernel boots on modern hw

then yeah

[dense vigil]

Btw do you handle a keyboard that exposes there virtual keyboards?

[flint idol]

(the kernel currently only boots on some haswell-era devices lmfao)

[dense vigil]

Because a lot of gaming ones

[dense vigil]

Mine boots on 5700xs and 12th gen intel and really old ones? What kind of stuff are you doing lmao

[flint idol]

no clue

[dense vigil]

[flint idol]

i tell myself "ill get modern hardware to boot eventually"

[dense vigil]

What does it do on modern hardware

What kind of crashes

[flint idol]

hangs sometime in really early boot

like before the fb or gdt is initialized

(fb is inited before the gdt lol)

[dense vigil]

Do you use flanterm? You can use its bump allocator to have it before any kernel shit starts to print things

[flint idol]

i think flanterm is the problem here btw

i think i dont give it enough memory for the bump allocator

[dense vigil]

Do you use a smaller font perchance?

[flint idol]

(i made the bump allocator region configurable for flanterm)

[dense vigil]

I had this exact same issue

[flint idol]

i use the default font currently

[dense vigil]

If the screen was bigger than 720p it would break

[flint idol]

and flanterm gets 16M of memory

[dense vigil]

Because i had an 8x8 font

[flint idol]

i might try allocating more memory

because if i dont see logs then it very well could be that

[dense vigil]

https://git.evalyngoemer.com/evalynOS/evalynOS/src/branch/master/src/patches/flanterm-fb-balloc-size-increase.patch
this is all i needed

[flint idol]

well

[flint idol]

.

[dense vigil]

Are you giving it too much memory??

I give 1.5mb

Aswell for the default font the normal amount should be fine because it scales on more then 1080p

[flint idol]

no such thing as too much memory

[dense vigil]

On the BSS maybe?

[flint idol]

especially when my fb is memory

[flint idol]

it uses a bootloader module to allocate

the memory

[dense vigil]

Wait huh

Why do i feel like this is cursed

[flint idol]

for limine i just make it load an empty file

maybe it is

[dense vigil]

i think this is why it’s breaking

[flint idol]

static struct {
    uintptr_t address;
    uintptr_t size;
} Arch_FlantermBumpAllocator = {};
static void *flanterm_malloc(size_t sz)
{
    static size_t bump_off = 0;
    if ((bump_off + sz) > Arch_FlantermBumpAllocator.size)
        return nullptr;
    void* ret = (void*)(Arch_FlantermBumpAllocator.address + bump_off);
    bump_off += sz;
    return ret;
}
static void flanterm_free(void* blk, size_t sz)
{
    OBOS_UNUSED(blk && sz);
    return;
}```

[dense vigil]

Flanterm has an internal bump allocator it uses

[flint idol]

x to doubt

[flint idol]

not if you override it

[dense vigil]

May be worth trying that?

Well yeah but

[flint idol]

i override it.

[dense vigil]

I’m gonna download OBOS and test it

And then make a PR of it works

[flint idol]

gl

[dense vigil]

What’s the process for building it and shit

[flint idol]

obos is known for having a shit build system

[dense vigil]

Just download and run make right

[flint idol]

cmake -Bbuild -DCMAKE_BUILD_TYPE=Debug -DOBOS_USE_CLANG=1 -DOBOS_USE_LIMINE=ON --toolchain=src/build/x86_64/toolchain.cmake .
cmake --build build

should work

[dense vigil]

repo link?

[flint idol]

https://github.com/OBOS-dev/obos

[dense vigil]

should i use the xhci branch?

[flint idol]

use master

[dense vigil]

hope this isnt a virus

shit i dont have cmake insralled XD

[flint idol]

lol

[flint idol]

u can review the 65k loc of code if u want

[dense vigil]

im too lazy

[flint idol]

tell me if u find bugs

[dense vigil]

il just sue ya

if its a vrius

[flint idol]

I cast: MIT License

[dense vigil]

this is fineee

[dense vigil]

i cast america

fuk

[flint idol]

oh shit

it booted a bit

with the bigger flanterm buffer

now it is hanging trying to initialize the scheduler timer

[dense vigil]

IL be doing the fixing

whats the build issue

[flint idol]

its mainly with userspace

[dense vigil]

how should i ignore it to get an ISO?

[flint idol]

wdym

[dense vigil]

make[2]: *** No rule to make target '/home/evalyn/Documents/Programming/C/obos/config/initrd.tar', needed by '/home/evalyn/Documents/Programming/C/obos/out/obos.iso'. Stop.

[flint idol]

create that file

[dense vigil]

XD

[flint idol]

then cd into scripts, run generate-initrd-x86_64.sh

then build again

then you should have a bootable iso

that should crash because it failed to load init

[dense vigil]

cp: -r not specified; omitting directory 'out/init' this stuff broke

dont make me fix it

im gonna have to fix this

[flint idol]

why did u create a directory called out/init

[dense vigil]

it asked for it

[flint idol]

doesnt mean its a directory lol

its the init program

which doesnt get compiled with clang enabled

[dense vigil]

wait what

[flint idol]

what

[dense vigil]

what am I doin

[flint idol]

the script worked

dw

[dense vigil]

🔥

this is enough to work with

[flint idol]

fucking hell

this is off master?

[dense vigil]

yes

[flint idol]

uhh I'll fix that

[dense vigil]

unless i fucked that up

[flint idol]

no thats a bug in the kernel on master

[dense vigil]

ahhh

[flint idol]

also r u gonna test it on amd

if so

[dense vigil]

i have amd and intel machines

[flint idol]

then that wont work most likely

[dense vigil]

what wont?

btw how tf did you get build IDs to work?

they always fuck up on my kernel

[flint idol]

amd

[dense vigil]

what messes up on AMD?

[flint idol]

like the git commit

[flint idol]

no clue, someone told me about a bug on amd cpus a while back

[flint idol]

?

[dense vigil]

is it with systemcalls

[flint idol]

no

[dense vigil]

this

[flint idol]

it just doesnt boot

[dense vigil]

ahh then idk what it is

oh

[flint idol]

I just get cmake to run a command with git that defines a macro called GIT_HASH or something

idk

[dense vigil]

broken .gitignore btw

[flint idol]

oh well

[dense vigil]

care if i add it into there

or just ignire it

[flint idol]

do it if u want

[dense vigil]

where do you init flanterm

[flint idol]

entry.c

also like

flanterm works

[flint idol]

^

[dense vigil]

huh

ahh

[flint idol]

im debugging that rn

if u want to try out obos u can use xbps to install packages btw

repo is http://obos-dev.ddns.net/current

[dense vigil]

@flint idol

Works on my machine

Let me try and

AMD

Yeah it really dosnt like AMD

[flint idol]

damn

[flint idol]

the fact that this is crashing because of that is weird

what clang version are you on?

[dense vigil]

My AMD 5700x hangs at startup on flanterm

[dense vigil]

Should be the latest

I’m on cachyOS

[flint idol]

not a number

[dense vigil]

[flint idol]

what clang version

[dense vigil]

Let me reboot

[flint idol]

I got no clue

[dense vigil]

Same lmao

 evalyn@EvalynPC  ~  clang --version
clang version 21.1.6
Target: x86_64-pc-linux-gnu
Thread model: posix
InstalledDir: /usr/bin

[flint idol]

thx

[flint idol]

and you used this command right

im gonna try and reproduce this

[dense vigil]

yes

btw it still works on my laptop without the flanterm patch

just the same callback bug

[flint idol]

im sorry to tell you this but your flanterm patch does nothing

on obos

[dense vigil]

Giving more memory fixed it then?

[flint idol]

on my pc, yes

[dense vigil]

Though wouldn’t stock flanterm use less memory than this idk?

Lmk if you need me to inspect anything else on my hardwares

[flint idol]

the amount of memory it uses scales with the display resolution tmk

[stark sigil]

magnificent

[flint idol]

and my display resolution is pretty massive

[dense vigil]

Stock flanterm has it fixed and it’s seperate from display resolution

It dosnt memory per pixel but somthing else iirc?

[flint idol]

well flanterm does double buffer

[dense vigil]

The default buffer 100% could not fit a 1080p screen

[dense vigil]

Yeah but it stores it as charaters or other representation no?

[flint idol]

I have no clue why giving it more memory fixes it

[flint idol]

but it does

[flint idol]

idk

im not mintsuki

[stark sigil]

maybe the memory you gave it at first wasnt enough???

[flint idol]

wow really

[stark sigil]

[dense vigil]

They are using a seperate bump allocator seperate from flanterms built in one that’s backed by an empty file module from limine

Which I mean either way is fucking cursed as shit rather than just using the built in one imo

And idk how it interacts with other things but

[flint idol]

how

flanterm init has the malloc and free parameters for a reason

[dense vigil]

Loading an empty file. And then binding it to a bump allocator for flanterm. Isn’t cursed?

Why not use standard malloc and frees?

[stark sigil]

terminal library's memory manager works better for terminal than other memory manager idk

[stark sigil]

what is this this is worse than i thought

[flint idol]

because that is supposed to work

before anything is initialized?

[dense vigil]

So will the built in bump allocator 😭

[flint idol]

its not empty, its all zeroes

[dense vigil]

This wastes disk space

[flint idol]

ok cool

[dense vigil]

It’s not worth dying on this hill

Do what you want XD

[stark sigil]

if you're fine with your fart wasting space then why aint you fine with mintsuki's almighty memory manager 9000

[flint idol]

some1 get this guy outta here 🥀

[stark sigil]

bro what

mods, get this guy out

[dense vigil]

I wonder how mint would react to this kind of usage of flanterm and limine

[stark sigil]

its not like he made the operating system

he aint important anyway

[flint idol]

cc @deep ember

[stark sigil]

oh no

i mean they will surely react to that

probably

[flint idol]

me and mint go waaay back

(not really)

[dense vigil]

Mint is on my friends list technically

[flint idol]

ANYWAY the kernel hangs sending an IPI

[flint idol]

pretty sure I already told mint about this anyway

I told someone thats for sure

[dense vigil]

How did they react

[stark sigil]

for some reason this feels like a kindergartner saying they're friends with cthulu (admins are scary)

[dense vigil]

Did they say bravo good idea this is super good and intelligent

I lowkey just wana know why you don’t think this is cursed? And how it is better

Like what’s the reasoning

[flint idol]

it isnt cursed because of its origins

[stark sigil]

monkey want, monkey do

[flint idol]

the bootloader obos actually uses is hyper, which lets you have bootloader modules backed by memory

and the reason I ever overrode the flanterm bump allocator was because in some cases

[flint idol]

flanterm would run out of memory

[stark sigil]

oh

[flint idol]

and it is a non solution to patch flanterm to change the bump allocator size

[dense vigil]

I’ve never had that happen though i don’t use large monitors

What is your monitor size?

[flint idol]

so my solution: a memory backed bootloader module so that the user can configure it to their needs

[stark sigil]

32 kilobytes

[flint idol]

idr what caused this bug

i think it was a normal monitor

but a different font

[stark sigil]

was the font smaller

[flint idol]

no clue

I wrote this code

[dense vigil]

Yeah that’s why i needed my patch

[flint idol]

months ago

[dense vigil]

Can you try my kernel which uses the bump allocator

And see if it also crashes on you?

[flint idol]

if you have an iso download sure

[dense vigil]

https://files.evalyngoemer.com/share/evalynOS Builds/

Make sure to grab latest

[flint idol]

damn it now I actually have to do it

[dense vigil]

[stark sigil]

bad apple is everywhere

[dense vigil]

My kernel best kernel

Best VFS ever too

[stark sigil]

every time i see bad apple in the top right of the screen i get ptsd

[dense vigil]

https://tenor.com/view/minecraft-movie-chicken-jockey-bad-apple-quackki-touhou-gif-10957867186139056787

[stark sigil]

https://tenor.com/view/rocket-greenscreen-gif-5649293510147604779

[dense vigil]

Is my kernel just known as the bad apple kernel lmao

[stark sigil]

by me yes

[flint idol]

no

[dense vigil]

Next kernel i make is gonna be called green apple

[flint idol]

serial: failed to init

[dense vigil]

And the kernel will be called bad apple

[dense vigil]

That’s normal

[stark sigil]

good apple

[dense vigil]

You don’t have a serial port

Show full screenshot of logs

[stark sigil]

and then you're going to make a kernel for the apple mac

[dense vigil]

I should make it a powerPC kernel

[deep ember]

what am i supposed to react to lol

[stark sigil]

too late i did it first

[stark sigil]

all of it

[dense vigil]

This

[flint idol]

^

[deep ember]

wait why would you do that

[dense vigil]

Idk ask @flint idol

[stark sigil]

also please beware it's so horribly difficult you WILL hate your life

[dense vigil]

[stark sigil]

the docs are PAINFULLY hard to find

[dense vigil]

Isn’t that what OSdev is about

[flint idol]

there's ur log

[dense vigil]

What

That’s meant to gracefully fail

[stark sigil]

nono its genuinely horribler than what ANY limine or even mb2 kernel ends up being

it has been a week and i have found no actual way to get a hello world or a framebuffer

[flint idol]

because flanterm would run out of memory and I didnt want to patch it to increase bump allocator size

[dense vigil]

[stark sigil]

the first thing you have to do is interface with PCI related crap, and also somehow figure out where your kernel is loaded because its actually in virtual memory

i want to run but i promised DO NOT DO IT

[dense vigil]

Try rebooting i think the TSC calibrated poorly because the ACPI timer sucks ass

I keep having it do that

[flint idol]

too much work

[stark sigil]

browhat

[dense vigil]

Atleast it dosnt match the frequency listen on Intels website for the base frequency

[deep ember]

you don't need to patch it, i am fairly sure you can increase the bump alloc size with a flag lol

[flint idol]

I have my own kernel to debug

[dense vigil]

They are giving it like 16+mb too

[flint idol]

a compiler flag?

[stark sigil]

understandable

[dense vigil]

Maybe later

@stark sigil why don’t you help me debug this

[stark sigil]

you didnt ask me to

[dense vigil]

(In the other Chanel)

[deep ember]

-D

a define

[flint idol]

I wanted it runtime configurable

[deep ember]

ah

well then no

[dense vigil]

I wonder if there could be a way for FT to bump alloc some logs to buffer until a proper malloc/free is hooked in

Or just buffer them yourself

[dense vigil]

Huh why is CR8 not zero? Do you set it to that?

[flint idol]

because my kernel uses IRQLs

implemented on x8664 through cr8

[dense vigil]

Ahh

[flint idol]

anyway I think the solution for my device will be to implement the x2apic

[dense vigil]

Oh yeah Intel arrow lake has no x1apic

[flint idol]

that explains it

you served us well, x1apic

you will never be forgotten

im gonna continue with my usb keyboard driver now

[flint idol]

because i need to "actually be careful with hotplugable devices" now

i am giving the usb keyboard struct a refcount

[flint idol]

ok well i added some boilerplate code to the usb hid driver

for submitting IRPs to it

and doing I/O on it

but i am yet to actually add code for the driver itself

[uncut narwhal]

of curiosity, how does obos do device hierarchy internally? do all devices have (optional) parents in a unified way or is it a per-driver thing if a device has some parent/bus device it references, or are buses as devices not even a thing in obos?

[flint idol]

it's a per-driver thing

and currently, no driver has a concept of device hierarchy

except, when I do USB hubs, such a concept will be added

[uncut narwhal]

none of that for pci-pci bridges?

[flint idol]

devices on pci-pci bridges do not have anything like pci_bus* parent_bus

[uncut narwhal]

👍

side note as i scroll through irq.c, whats with these gotos that do nothing

also shouldIgnoreObjectCapacity is always false so this is always an infinite loop if it cant find a vector

those gotos at least do something near line 304ish but the shouldIgnoreObjectCapacity is still never modified so itll still be an infinite loop if it cant find one

[flint idol]

no clue

(QOTD: "When I wrote this code, God and I knew what it does. Now, only God knows")

[flint idol]

good catch

[uncut narwhal]

i think the first gotos i linked was copying from the other function, removing the continue, and then just not removing all the useless stuff that existed to skip the continue sometimes lol

[flint idol]

who knows

wait that makes no sense

[uncut narwhal]

lol

[flint idol]

i need to fix a bug with xhci

when it reaches the end of a TRB queue

it fails to go back to the beginning

and so the HID driver hangs

nvm that's not the bug

but something in the xhci is breaking

causing the HID driver to hang

[flint idol]

perhaps it's because the dequeue pointer is literally for a whole different queue...

[flint idol]

fixed it

it turned out

i was trying to write the doorbell for the completely wrong endpoint

[flint idol]

well i fixed that

and also added a link TRB at the end of each TRB ring

but now it hangs after receiving 2 pages worth of events

or well it doesnt hang

but the dequeue pointer becomes zero for some reason

nvm it isn't

[flint idol]

the dequeue pointer somehow manages to become higher than the enqueue pointer

causing the trb queue function to return would_block

ok fixed that

now it hangs because it doesnt seem to find the link TRB anymore

fixed that

oh yeah

the HID driver can receive packets now

but it doesn't interpret any packets yet

which is a TODO for tomorrow

ill add mouse packet interpretation first because that is easier

then ill add keyboard packet processing code

and then ill test it ig

ill see if i can get a TTY working (in qemu) with it

then, i will implement usb hubs

and see if the HID+USB Hub+XHCI driver works on any of my devices

[blazing python]

are you planning to do boot protocol or hid

[flint idol]

boot protocol

[real pecan]

L

[flint idol]

well im gonna add the other protocol I just want smth basic to work first

this driver is currently only as a test for the xhci driver

[flint idol]

obos hit 70k loc

ive no clue what all those LoC do, but yknow

[flint idol]

anyway there is a bug

with the thing

if u detach an hid device after having submitted an IRP to it

then the kernel dies

use-after-free

i see the bug

[flint idol]

well the mouse driver works

(barely)

input is really choppy for some weird reason

worse than the ps/2 mouse

[dense vigil]

I wonder what the polling rate is?

[flint idol]

who knows

ok i fixed the choppiness

the driver wasn't polling the mouse fast enough

now for USB keyboards

[flint idol]

ok well whoever designed this should kts

[stark sigil]

sudoku

[flint idol]

ok so im writing an algorithm to see what keys are released given the previous state of the keys and the new state

#define BITFIELD_DIFF(bitfield1, bitfield2, bitfield_out) \
do {\
    for (int i = 0; i < (104/8); i++)\
        bitfield_out[i] = ~bitfield1[i] & bitfield2[i];\
} while(0);
// ...
BITFIELD_DIFF(currently_pressed_keys, dev->pressed_keys, released_keys);

i think this works

[flint idol]

there are some slight bugs

[flint idol]

time to implement usb hubs

i think i just get the hub descriptor and tell the xhci controller "probe these ports pls"?

this is schizophrenic

https://tenor.com/view/war-vietnam-ptsd-shell-shock-moment-gif-3022747568394546158

how do usb hubs work and interact with xhci

[stark sigil]

idk id assume usb hubs are just more usb ports

[flint idol]

i mean that's objectively wrong

[stark sigil]

how they literally are

at least physically

companies vs making the most reasonable solution for a problem

[flint idol]

who the actual fuck designed this

i am currently reading a pirated usb book

that explains how usb works

and i am reading the usb hub section

and What The Fuck

actually

it's not that bad

[flint idol]

this actually makes perfect sense

when a hub driver detects a device connected:

call enable slot to get a new slot
configure that slot to be associated with the port by setting the route string appropriately
profit

[flint idol]

ok usb2 hubs work now

https://tenor.com/view/catjam-gif-10178102486865202192

rn it goes: root hub->hub->hub->keyboard

hold on there is a bug

xhci: input slot context: 08000011 00010000 00000103 00000000

hm

this is supposed to represent 1.1.2

[flint idol]

i removed the recursive hubs to make it easier to debug

but anyway

xhci: input slot context: 08000001 00010000 00000201 00000000
xhci: port 1.1 already assigned to slot 2

[flint idol]

it's hacky

but it works

i think ive ironed out the last bug with the route strings and such

[flint idol]

for tomorrow: finish usb hubs (implement the interrupt endpoint thingy), get root hub port enumeration to work on real hardware

the goal before i merge this branch is to get the xhci driver working as well as possible on my test subjects

and to have USB3 hubs implemented

and to make sure that hotplugging doesn't cause the kernel to shit itself

and after, ill add a usb block device driver

[stark sigil]

yo usb man how bad is usb to impelement (i think i have to implement that )

[stark sigil]

fire

[flint idol]

(also todo: implement isochronous trbs)

[flint idol]

as long as you read the spec really carefully then you might be fine

[stark sigil]

fair

i think i have to do 1.0 for now

[flint idol]

hint: start with xhci

USB1?

[stark sigil]

cant

[flint idol]

why, 2011 device?

[stark sigil]

2001 ibook

[flint idol]

bruh

[stark sigil]

[flint idol]

start with xhci anyway, test on qemu

i feel like xhci gave me a better way to grasp the concept of usb

[stark sigil]

cant do the second one either

[flint idol]

you can't use qemu?

[stark sigil]

it cant emulate powermacs correctly

im gonna port my os to x86 but thats later

[flint idol]

oh so this is a non-x86 os

yeah gl with that icl

[stark sigil]

yehs

[stark sigil]

yeah understandable

it has like the worst firmware ever too

[flint idol]

ehci made no sense to me, and you're going to be using uhci which is apparently even more schizo

[stark sigil]

oh no

im already a schizoid with picking osdev for powerpc i think ill be fine

[flint idol]

k

im gonna sleep now so gn ig

[stark sigil]

ight

good luck with your evil usb rituals

[flint idol]

thx

[flint idol]

todo: fix this and make it use xor

[flint idol]

nvm

I tested the logic, it doesnt make sense for this case

[flint idol]

with xor, i found that if a key is pressed for one keystroke then it is reported as being released instantly

which actually might be correct behavior?

I'll test it after school ig

[flint idol]

https://github.com/OBOS-dev/mlibc/blob/upstream-merge/sysdeps%2Fobos%2Fgeneric%2Fsysdeps.cpp#L347

I need to fix this bug in my mlibc sysdep

[flint idol]

yeah the xhci driver doesn't work on real hardware

i decided to try and submit a NOP COMMAND trb

and it just hangs

without so much as an irq

i will note that this device seems to use msi

for the xhci controller

[flint idol]

reminder to self: fix the bug on line 602 of xhci.c

[blazing python]

what's bug

[flint idol]

its an if statement with the wrong condition

while the code is not technically wrong it makes the kernel have to do unnecessary stuff

oh and I also need to fix a bug on line 967 of xhci.c and line 975

[flint idol]

chat

should I join

[dense vigil]

wait is that THE waterloo?'

also yes you should join may be fun

[devout niche]

not the waterloo of napoleonic (and abba) fame

[dense vigil]

I mean the same Waterloo that did this to stufents

[flint idol]

yes

[blazing python]

https://tenor.com/view/seonggihun-serious-transition-change-gif-14441801055391863257

[dense vigil]

Did Waterloo do the same to you?

I do gotta know

[flint idol]

I dont go

[dense vigil]

Ohhh

[flint idol]

im still in high school

[dense vigil]

Ahhh makes sense

[flint idol]

its a contest for high schoolers nstuff

[dense vigil]

Honestly I bet you have a good shot at winning

Your pretty damn smart

[flint idol]

thx

im gonna see that teacher sometime soon™

sign up

[dense vigil]

I’m gonna be asking you about USB btw later

Few months probably

[lean glen]

definitely

Ive wanted to participate before but I've never been to a school with a programming club or whatever that organizes these kinds of things

and you cant enter as an individual

[lean glen]

yeah uwaterloo is one of the top CS/Eng unis

[lean glen]

this is the first step to IOI btw

then there's CCO (canadian olympiads)

then if u win all of that you go to IOI

[flint idol]

aight so either tomorrow or after exams im gonna find the teacher and register for this

[flint idol]

i fixed the bug with TRBs on real hardware

fuckass xhci spec

"oh yeah on top of having to set the CCS bit in op_regs.crcr, you also have to set the cycle bit in the first trb because fuck you"

[flint idol]

usb's a bitch

[flint idol]

https://tenor.com/view/war-vietnam-ptsd-shell-shock-moment-gif-3022747568394546158

xhci is fucking insane

[blazing python]

https://tenor.com/view/plastic-man-real-real-powder-powder-that-makes-you-say-real-animated-comic-plastic-man-gif-7493651428267836743

[flint idol]

[blazing python]

when i was doing xhci i had pain with real hw too

[flint idol]

i just want the CCS bit to be set man

January 20th, 2026, at 2311: I am debugging XHCI, it is driving me insane. I can hear voices when I sleep telling me about TRBs and xhci stuff and I think I am going insane.
(diary entry end)

ok it is kind of almost getting there

wtf

i managed to make it try to initialize a disconnected port on qemu

i should git reset my changes and try again icl

like i managed to generate the bit out of thin air on qemu

so like

it hangs on waiting for the ADDRESS_DEVICE to finish

and one thing i noticed is that the portsc has the port link state as "compliance mode"

which seems suspicious

[flint idol]

TODO for whenever i get back to this: rewrite the port attach and reset logic

after that i think ill have a working xhci implementation on real hardware

because TRBs work

and well that's pretty much all that matters

anyway gn yall

tomorrow im studying for an exam probably so

wont work on this most likely

and thursday im studying for my other exam on friday because well procrastination at its finest ig

and i will have time on the weekend

and on monday i will study for my exam on tuesday

then im done

and i will be free once again

god willing

[blazing python]

send iso i can test it on my hw

[flint idol]

when i actually get correct code, sure

[flint idol]

fucking hate

xhci

and usb

WHY WONT YOU JUST REPORT A GODDAMN CONENCTION ON THE ROOT HUB

JOB HAD ONE BRO

AND BRO FAILED JOB

wtf is wrong with my hardware

there's an xhci controller

for no goddamn reason

i booted into linux

and the usb3 root hub is completely unused

wtf is wrong with hardware, who designed this shit

nvm

it's just a USB2 hub on the xhci controller

with all my devices connected on it

kms

what have i done to deserve this

all ports are in RxDetect

two ports never even get the reset bit cleared

sdfgsdjoikgfhdsfjkgdsfhjgkfdshngfd

[stark sigil]

i think there was no good luck with the usb rituals

[flint idol]

i dont think so no

[stark sigil]

osdev

[flint idol]

(THAT'S NOT EVEN SUPPOSED TO BE FUCKING POSSIBLE)

[stark sigil]

qemu problem?

[flint idol]

NO THIS IS ON REAL HARDWARE

[stark sigil]

it doesnt follow certain specifications correctly

[stark sigil]

oh

doesnt work on your machine 😔

[flint idol]

(xhci spec) Only a USB3 protocol port may fail the bus reset sequence. USB2 protocol ports
never fail the bus reset sequence.
resets USB2 port
reset fails
fsdfdsklfsd

[sterile sparrow]

And that is why I'm not even going to touch the spec in a looong time

[flint idol]

yknow what maybe the problem is with the bios handoff code

like maybe the bios still thinks it has control of xhci

and is doing evil shit

[sterile sparrow]

Never trust the bios

[flint idol]

(this is not the case)

im reading freebsd code

it has gotten to this point

haiku maybe?

nvm

https://cdn.discordapp.com/emojis/816178972672131083.gif?size=256

[dense vigil]

Before and after XHCI driver

https://tenor.com/view/squid-game-squidgame-squidgames-gif-2915153355997802308

How’s it going

[flint idol]

on qemu: amazingly
on real hardware: terribly

[flint idol]

ok well ive an English exam tomorrow

its like there's an essay question on there and like I dont wanna write it but like I have to because like its worth 80% of the exam mark

[flint idol]

so tomorrow after my exam: continue work on this, maybe ill try it on another device

(maybe I should study for my exam on Friday instead? hm)

[dense vigil]

I do remember seeing some youtube tutorials on XHCI that showd it worked on real hardware and how to get it working? Might be a long shot but maybe they have something?

[flint idol]

idk about that

anyway im studying some stuff i mightve forgotten for my exam to make sure I dont blank tomorrow

[flint idol]

im having absolutely no luck with this

so my solution: use the kernel gdb stub

this is my only hope

[flint idol]

fack the gdbstub doesn't work anymore

or ig it's that the network stack is being doodoo

and not the gdbstub

[white mulch]

maybe you could pass through the controller to qemu? though I guess it depends on what controller your keyboard/mouse are connected to if any, if they are connected to the same one then its probably not a good idea

[flint idol]

nah, i dont think this pc has passthrough

(as in the test subject)

so that's not an option

[white mulch]

it doesn't have an iommu?

[flint idol]

nope, not to my knowledge

i mean there's no harm in checking

oh there is an iommu

[flint idol]

usb hates me

i do passthrough, now instead of getting port reset errors

im getting trb completion code 4

"USB Transaction Error"

[white mulch]

that is a really annoying error due to how many different things it can be caused by

[flint idol]

oh fuck

i think i bricked my pc

(test subject pc)

nvm

[flint idol]

wait apparently im supposed to send ADDRESS_DEVICE twice

[blazing python]

twice ???

isnt it only one time

[flint idol]

idk

let me check the spec, because as i know that always helps me with my problems /s

For some legacy USB devices it may be necessary to communicate with the
device when it is in the Default state, before transitioning it to the Address state.
To accomplish this system software shall issue an Address Device Command with
the BSR flag set to ‘1’. Setting the BSR flag enables the operation of the Default
Control Endpoint for the Device Slot but blocks the xHC from issuing a
SET_ADDRESS request to the device, which would transition it to the Address
state.

A USB Transaction Error shall be generated if an error is detected on the USB
SET_ADDRESS request and the Device Slot shall not transition to the Addressed
state.

amazing, very useful

(what was the goddamn error)

[flint idol]

https://tenor.com/view/mercy-overwatch-angel-come-with-me-gif-14015043

https://forum.osdev.org/viewtopic.php?t=57621

[flint idol]

yo wtf is the xhci controller smoking