EvalynOS

[hazy saddle]

Well yes

But people have gotten mad at me for using it in 64 bit contexts

Even though I think vector is a much better term regatdless

[fiery tangle]

:wahhgone:

bru

[kindred stirrup]

I use this term when talking about the number of the interrupt

[fiery tangle]

its not in thes sver 😔

[hazy saddle]

Yeah interrupt is a general term and vector is the number

[hazy saddle]

Ha ha my GDB stub works!!!

The IRQLs make it work so nice

I just have to add a bit more

And it’s ready to push

[hazy saddle]

Almost as if a global mask of all interupts is stupid

And being able to mask everything except a few things

Is like very good

[hazy saddle]

Holy crap

Like genuinely

IRQLs make the debugger smooth as shit

Maybe because serial actualy has proper priority

[peak cloak]

smooth operator

also did i somehow cause you to use

[hazy saddle]

I’ve always used

[peak cloak]

ehhhhh

[hazy saddle]

Hmmm

It appears everything works

except

When a breakpoint is hit

Single stepping fails

[peak cloak]

RIP

[hazy saddle]

it works perfectly

ONLY

if i do si

and it breaks inside printf vargs

for normal stepping

??????

[peak cloak]

what???

https://tenor.com/view/folk-druski-sona-instagram-instagram-meme-gif-18271404859547050805

[hazy saddle]

yes!!!

[peak cloak]

diarrhea

what does si do again

[hazy saddle]

single instruction step

vs multi instrution step

[peak cloak]

oh

do you en/disable that with an instruction

[peak cloak]

does gdb allow you to print text into the debugger

it better allow

[hazy saddle]

No that’s just a normal printf

[peak cloak]

oh

[hazy saddle]

[peak cloak]

i still wanna ask tho

[hazy saddle]

no idea

dont think so

[peak cloak]

oh

[hazy saddle]

Okay so

GDB is now asking for straight up bogus memory

This is what I got so far

This combines two commits

Tomorrow I’m gonna try to clean it up

And push what I have

And mark it “unstable”

Or whatever

[peak cloak]

gnu debugga(ss)

[hazy saddle]

I may of found it

I don’t support

Changing registers

[peak cloak]

What

why would you need to change registers from debugger

[cold barn]

so you can see what happens if you get fucked up

or I guess like

if you notice that a register value is bad, you can change it to verify that execution would be correct with the expected value

[peak cloak]

ohok

[hazy saddle]

GDB is issuing requests on its own to do that lmao

And to read and write memory

And it keeps asking for BS

[hazy saddle]

I have a theory

https://tenor.com/view/game-theory-game-theory-its-just-a-theory-game-theorists-gif-20883020

I literally explicitly told GDB

I do not support software breakpoints

But it keeps fucking sending them

XD

So

I need to implement them

I should also try lldb

Because gdb may not like llvm compiled code

bruh

[floral kettle]

idk why wouldn't gdb like llvm compiled code, it works just fine most of the times

[hazy saddle]

I do know whenever I use the wrong addr2line it complains about dwarf errors

 evalyn@EvalynPC  ~/Documents/Programming/evalynOS/evalynOS  ↱ dev ±✚  addr2line -e ./bin-x86_64/kernel.elf ffffffff8000ca4f
addr2line: DWARF error: mangled line number section (bad file number)
addr2line: DWARF error: mangled line number section (bad file number)
addr2line: DWARF error: mangled line number section (bad file number)
...
addr2line: DWARF error: mangled line number section (bad file number)
addr2line: DWARF error: mangled line number section (bad file number)
addr2line: DWARF error: mangled line number section (bad file number)
addr2line: DWARF error: mangled line number section (bad file number)
addr2line: DWARF error: mangled line number section (bad file number)
/home/evalyn/Documents/Programming/evalynOS/evalynOS/src/kernel/libc/stdio.c:48

This is a lot more than normal

I don’t normally addr2line inside of printf/va args

[floral kettle]

I think that is a dwarf 5 issue, with -gdwarf-4 it works fine

[hazy saddle]

i wonder if its also due to my use of clang spesific things tripping up gdb? i was not able to run addr2line with that sadly

though whats strange is lldb cannot connect over serial

https://github.com/llvm/llvm-project/commit/4a7b4beac759ad9001671a61846ee2bfc9076eec
though i found this old commit

and this blog post from a website that no longer exists lmfao https://web.archive.org/web/20211108134514/https://www.moritz.systems/blog/lldb-serial-port-communication-support/

[floral kettle]

did you compile your kernel with that flag?

[hazy saddle]

ahh its a compile flag

[floral kettle]

tbh most of the times you don't really need that as you have better ways to communicate

like tcp (or in qemu -serial tcp where the guest can use serial but the host can connect using tcp)

[hazy saddle]

it seems lldb is working now

and i got it over serial

and it properly steps through

[floral kettle]

nice

[hazy saddle]

i am gonna try and rework my stuff to compile with gcc

it was being strange before

triple fauling before even being able to flanterm

process connect --plugin gdb-remote serial:///dev/ttyUSB0?baud=115200

[hazy saddle]

yup adding this works

for addr2line

gdb is still angry

[hazy saddle]

it seems somwhere on here its trying to debug the stub its self

cusuing an issue

i wonder

is it giving bad data making the stub call printf?

causing issues?

though i did remove that lock temp

somthing inside irql is toching IRQL??????

or some other interupt is coming in??

so it seems to lock up before this spot.

then if i press s it manages to get back in

lmfao the debugger debugging its self being useful

I’m actualy getting a stack frame of it dying

It’s going serial ISR -> debug ISR -> raise IRQL -> panic

It seems somthing within printf is fucking up?

Or it’s getting a serial ISR when it shouldn’t

AHA

nanoprintf must not like printf being recussirve??

when i process a packet its going inside of npf

Not for a printf but to format some data

I’m not sure if nanoprintfs internals are thread safe for this?

Somthing is triggering a dbg exception inside of it atleast

Wait what if

The debugger is putting a breakpoint inside of there

I could do a basic recursion check

If you enter it after raising IRQL

It will set a flag

If the flag is set it will not attempt another exception try

[clear bison]

Bro's getting IRQL_NOT_LESS_OR_EQUAL

[hazy saddle]

Well yes my kernel has IRQLs now

this fixed it

who could of guessed this code was bad

It works now!!!

Now

Time for real hardware

After I clean this up a bit more

[hazy saddle]

Should I make this as the panic code lmfao

As a reference to windows

[clear bison]

Maybe idk

[hazy saddle]

Somthing is dereferencing a null pointer

Only on real hardware

And it’s after serial setup it seems

im gonna presume my build env was wonky

because i did not copy everything to other laptop

and im gonna harden serial driver

good thing my smb server is fast

Okay it’s working now

Just looks like it got bad symbols ?

It was reading addresses wrong

But like

Not fully

AHA

I set the wrong parameters

I needed to set no parity bits and 1 stop bit

LLDB seemed to be picking other things as default?

It’s working!!!

A working GDB remote protocol server over serial (using LLDB because GDB was being stupid )

[gentle quest]

Thats cool

[hazy saddle]

Thank you :3

I will have to add and refactor a bit more later

Esp once I get SMP

But it shouldn’t be too bad

[cyan bison]

just do it™

trademark owned by Nike pls dont sue me nike

[hazy saddle]

Nah it was GDB just being dumb

Idk why but LLDB works fine

Talk is cheap send patches

[cyan bison]

me when setting one byte to int3 is too difficult:

[hazy saddle]

Me when sending patches is too difficult:

[cyan bison]

now its not my code now is it

[hazy saddle]

But your talking about it

[cyan bison]

then I wont

[hazy saddle]

I’m just messing with ya

Software breakpoints are on my todo

But I’m content with the 4 hardware ones

[cyan bison]

hmhm

im not

[hazy saddle]

Il add them later

The file is already 400loc

I wana refactor it to make it cleaner before adding anymore

• I pre allocate all buffers so il have to do some magic for this potentially

[hazy saddle]

This way I don’t allocate inside of IRQ contexts

[hazy saddle]

What is strange is the PS2 driver is now kind of wonky inside QEMU

But fine on my desktop

Good thing I was gonna rework it anyways

[peak cloak]

yeah qemu cant really emulate a PS2 very well

[hazy saddle]

Nah it’s also slightly buggy in real hardware

Considerably less so though

[peak cloak]

that means your emulator is buggy

you should find a better PS2 emulator

theres a bunch of open source ones i hope

playstation jokes aside good luck

[hazy saddle]

I mean the driver lowk needs a rework anyways so

it legit has a fucking lock inside of it

[peak cloak]

😭

[hazy saddle]

the commit™ is all ready

[hazy saddle]

Can confirm a build off the main branch works perfectly still

I honestly cannot believe it’s working this well now

[magic saffron]

https://klipy.com/gifs/charlie-day-4

[hazy saddle]

It turned out to just be GDB freaking out lmfao

It was trying to read and write to bogus memory

Whole LLDB was just fine XD

[soft snow]

safe_user_funcs.c best file name'

[hazy saddle]

It is what it says on the tin

Though I may change it to sage funcs

[hazy saddle]

VFS is tomorrow

I wokeup at 1pm today and I am unfit to work

But I may submit the ps2 driver patches

[hazy saddle]

tried to update my mlibc patch just a little

uhhhhhhh

(trying again to port bash)

well the other patch is fine idk

the diff for reference

[hazy saddle]

I wonder if it attempted to dynamic linking when I 100% compile it as static??

[clear bison]

😭 🙏

CR4 instead of CR0

[hazy saddle]

I was copying my CR4 code there

And I fucking forgot

🥀

😭

[clear bison]

Biggest fuckup of mine of this kind has been cr2 instead of cr3

[hazy saddle]

Damn 😭

[hazy saddle]

dawg the tiny ass bugs im finding

how is this not breaking real machines XD

[kindred stirrup]

Doesnt that throw a #UD

isn't cr2 read-only

[clear bison]

It's not afaik

[floral kettle]

yeah you can write anything you want to there

[kindred stirrup]

lol thats crazy

i deliberately chose not to restore cr2 in my interrupt handlers

[hazy saddle]

I’m sending EOIs here when I shouldn’t

Because with IRQLs I send them right away etc

[hazy saddle]

You still need to CPL0 no?

So like you shouldn’t need to really restore it unless you are using it as a GPR

Though wouldn’t that mean you can use CR2 as a cpu local?

Like that actually dosnt seem like a fully bad idea if that’s consistent across amd and intel

[kindred stirrup]

No, because cr2 would be overwritten on page fault

[hazy saddle]

Oh wait yeah

💀

I forgot that’s what CR2 was for

Tbf why don’t they do somthing like that with the unused control registers :p

Like yes there’s GSbase

But having extra can’t hurt

[magic saffron]

iirc there is a kernel here that briefly uses cr2 as storage

[magic saffron]

they probably have internal uses

[gentle quest]

Lmao

There is

[magic saffron]

if you know, you know

[gentle quest]

Revolutionary tech

[hazy saddle]

Is the registers used for Intel to figure out what feature to remove next

Ahh I miss read that reply

[hazy saddle]

why is my debug stub now like ever so slightly out of sync

let me try the physical cable again

and not qemus PTS thing

why is pressing b on that like making a breakpoint there??

ahh

it was -O2

Of course that’s gonna make debugging a PITA

Because it’s not gonna align with the source code as much

okay yeah the patch works as expected

and dosnt break the debugger

@magic saffron this should fix that TSS stuff and properly give #DB/#DB its own stack along with #DF (I just chuck #NMI into the generic unhandled IRQ stuff for me to deal with later™ and have it panic)

I am also considering making my flanterm lock not block IRQs

Because in real hardware the framebuffer can be slow in intensive sections

And I’d rather keep preemption going and allow tasks to run when not printing

Rather than having it spin a bit under contention which I can fix with better scheduling later

[hazy saddle]

Wait my scheduler takes spinlocks

😭

So much of my crap is cursed rn

Actualy not sure why that isn’t breaking my shit somewhat if external things can take it technically ;p

Same for the reaper queue

I’m not even sure a good solution to fix that

Well if it’s not breaking for now it’s fine

Even if it’s technically breaking IRQLs in some way?

Though since I’m at dispatch it technically won’t crash

Oh wait no I’m just dumb 🥀

It’s a spinlock

You can’t preempt durring that

It’s fine

The scheduler shouldn’t even need to take locks given it’s already at dispatch?

[magic saffron]

cool, nice to see 🙂

[hazy saddle]

its still triple faulting on @rotund furnace's old desktop lmfao tho

i even attempted to add some delays

[magic saffron]

just mark it as unsupported hardware 😛

[hazy saddle]

and then the delays caused it

💀

I do want to mark it as unsupported but his kernel can run just fine

[magic saffron]

hmm have you tried disabling features of the kernel until it stops faulting? that might give you some insight

[hazy saddle]

And the problem is so strange

[hazy saddle]

It faults on the wait for you to press enter lmao

And if you remove all waits it still crashes

It just gets angry at some point after printing everything

[magic saffron]

how narrow is this "some point"?

also I forget, do you have smp support?

[hazy saddle]

This is what happens right before reset

[hazy saddle]

Not yet

And this is the code minus stuff not being commented

though the good thing is i found a ton of teensy bugs from this

https://git.evalyngoemer.com/evalynOS/evalynOS/commit/4280b6ec744697b5bb4e716a25d3e6b08a6a123b#diff-c002b2cf587659a9b7054177b4fac3c56dd0d337

[magic saffron]

thats fine, its one less thing to worry about for now

[hazy saddle]

i was gonna add smp after i get everything for bash and get my major rewrite done to get the codebase up to standards

[rotund furnace]

until they add smp support and everything breaks :P

[hazy saddle]

Better to do it in a stable spot versus now :p

[magic saffron]

yeah ngl adding smp to a kernel this late is going to suck, but you'll learn a lot of things I'm sure

[rotund furnace]

not wrong

[magic saffron]

but its one less thing to diagnose now

[hazy saddle]

My initial scheduling things were so fucking unstable

I possibly would of given up if SMP was added in the mix then

Being able to make it better without SMP is quite useful

[magic saffron]

I think thats an entirely reasonable decision, better that than giving up 🙂

[hazy saddle]

Yup :3

I also will try and go for my own 16 bit trampoline

Verses using the linine request

[magic saffron]

linine

[hazy saddle]

[urban vale]

prob going to inevitably have a worse outcome than if smp had been there from the start

[hazy saddle]

Oh btw did I tell you about the time I tried to compile the limine test kernel and it ended up trying to copy my / drive into it 💀

[hazy saddle]

Pros and cons

And atleast now I am way more competent at this so

[magic saffron]

yeah probably

[hazy saddle]

Not like it’s impossible to end up working out

It will work eventualt

Heck my kernel has already had 2 major rewrites

What’s another

Oh and then an extra for architecture ports

[urban vale]

by worse outcome i mean worse smp support since like

[hazy saddle]

Plus heck if I have to do a rewrite if I got to a decent point with bash and no SMP good chance I can just steal a fuck ton from the current codebase

[urban vale]

youll be more tempted to use huge locks and stuff

that cover vast amounts of things

rather than be forced to redesign tons of code you already wrote

to make it work with fine grained locking

[hazy saddle]

I already have a global PMM lock

And I was planning to rewrite that anyways

And I have to redo my heap

And my VMA could use some more work after my initial revision

I want to rework a lot anyways eventually so :p

[hazy saddle]

Well actualy make my own heap because the one I stole is dookie

I even wanted to attempt some lock free algorithms

[hazy saddle]

i still cannot figure out how to build it too

[magic saffron]

isnt it just a make target?

[hazy saddle]

i tried the make file part of it and its missing tons of env stuff

and i did not see it in ./configure

and i did a full all target build

even mkdir needs a ENV 💀

[hazy saddle]

btw talk cheap send patches. did you try the older iso :p

This kernel used to be held up with so much duct tape I’m shocked I even got to this point

Well it still is but a good bit of that tape is now super super glue

Very good improvement

It will become good at some point

Atleast I have fun along the way tinkering on this stuff

[magic saffron]

did you run ./bootstrap?

[hazy saddle]

yes

limine the bootloader builds

not the test kernel

[magic saffron]

I just tried a fresh clone of limine, then ./bootstrap and configure --enable-uefi-x86-64 and make test-uefi-x86-64 and it just worked (tm)

I did have to install autoconf though 😔

hmm ok

[hazy saddle]

./configure --enable-all
make -j16

this is from my fuckery of attempts lol

it managed to yoink that much crap

[magic saffron]

maybe start again, that looks fucked

[hazy saddle]

I don’t wana risk it again

And I CBA to find all the ENV things I set again

It was @rotund furnace Who needed it to test on the fuck ass PC

Because APs cannot start until you press a key on PS2

💀

[magic saffron]

ok interesting

[hazy saddle]

also my UBSAN did this

[rotund furnace]

this could be my fault ngl

but like

my code will get to the "waiting for aps"

and then just nothing

[hazy saddle]

I’m personally willing to bet ventoy is fucking with my kernel tbh :p

[rotund furnace]

until I press a key 🥀

[hazy saddle]

🥀

[rotund furnace]

it happened when I dded it directly

[hazy saddle]

[rotund furnace]

it also works fine in qemu and on modern hardware

[magic saffron]

yeah ventoy is known to be a big pile of hacks (or it was), have you (or Water bottle) tried without it?

[rotund furnace]

so like

idk

[magic saffron]

ah ok

[rotund furnace]

blame amt?

[hazy saddle]

it is a dual core pentium

and old as shit

who knows that kind of hardware issues it could have

[rotund furnace]

it works fine in windows and linux :P

[hazy saddle]

both kernels that have years of work put into them with tons of possible quirks tables :p

whats the exact CPU model again

i wana look this up

@rotund furnace this?

[rotund furnace]

maybe

no

iirc it's the 5400

[hazy saddle]

@rotund furnace what timing device did you end up using?

for your kernel on that system

[rotund furnace]

wdym

[hazy saddle]

like PIT / HPET / ACPI

[rotund furnace]

my kernel only supports the lapic timer

oh for calibration

[hazy saddle]

but to calibrate lapic

[rotund furnace]

my kernel only supports the pit chip for that 🥀

because I was lazy

[hazy saddle]

becaus somthing has to be fucky with my timing system?

thats where it was faultin

mmm i dont use delays on the uhh put thing

we did try and force PIT?

and that failed semingly

@rotund furnace wana try this?

[hazy saddle]

this build also forces PIT

Lowk I should make that a CLI arg :p

ctimer=pit/hpet/acpi

[opaque marten]

ram is too expensive, use cr2 register for more memory

[opaque marten]

ey i have one of these in an optiplex

i also have the e5200 in another machine

[hazy saddle]

Wana test if my kernel breaks on there to see if it’s just a bad PC?

Or some other hardware fault fucking up

[opaque marten]

sure

[hazy saddle]

This is the debugging iso

Though they may have an E5400

Should be close enough :p

[opaque marten]

yea

[hazy saddle]

Thank you

[opaque marten]

lemme whip this thing out rq

wait nooo i forgot i moved it to another house

don't have it rn

[hazy saddle]

fuuuuuuuuuckkk

[opaque marten]

i mean i do have my i5-2450M laptop if that helps but

yeah

[hazy saddle]

i mean its always worth a test lol

[opaque marten]

it got to the kernel shell prompt

i cant take a picture because my phone is being fucking stupid

[hazy saddle]

can you type badapple or doom

[opaque marten]

doom works although its half on one side of the screen and the other on the other half

but i think its because you put it to the right

[hazy saddle]

yeah that happens on low res displays

technicaly it dosnt meet the minium specs for display res

[opaque marten]

badapple also works and the pc speaker works too

[hazy saddle]

so yeah

idk why it wont work for @rotund furnace

[hazy saddle]

Why not play some cross eyed doom

[opaque marten]

lol

[hazy saddle]

[opaque marten]

laptop's a packard bell from ~2012

i5-2450M, 4gb of ram, 1366x768 screen

[hazy saddle]

Idk why the 2008 PC won’t work them

Like my kernel is meant to run on the oldest of hardware

I specifically avoid features that would break older systems

Without fallbacks*

[opaque marten]

i have a few old machines at my mom's house but i'm not there rn

i have an amd sempron le-1250 machine, an e5200 and an e5300 machine

[hazy saddle]

I’d love to see things tested if possible

I hope they have a working serial port

Maybe we can get into LLDB and my stub

[opaque marten]

they should

[hazy saddle]

The one water has dosnt even have a working serial header

[opaque marten]

i also have a serial to usb converter that i can use on my pc

[hazy saddle]

Ooo hell yeah

Just don’t use gdb

[opaque marten]

aight

[hazy saddle]

Because gdb cannot work with my gdb stub

💀

[opaque marten]

bru

[hazy saddle]

It asks for fucking invalid ass memory

And it cannot single step right at all

[opaque marten]

mfw

[hazy saddle]

LLDB works perfectly on -O0

-O2 works strange because of how some things are inlined and removed etc

Not with the stub atleast tho

[opaque marten]

yeah

[opaque marten]

anyways i booted alpine linux on serial and it worked and after fixing my serial code i was able to use bentobox with it too

although the machine was an i7-4790 machine

[hazy saddle]

Ooo nice nice

I hope my kernel can start getting some progress

I keep taking random side quests

Like old ass hardware and gdb stubs

Instead of the glaring crapping VFS XD

Though this has gotten me to fix like 7 various bugs in drivers

Which is nice

[opaque marten]

lol

[hazy saddle]

Oooo nice

I also want to try and compile bash but the build system keeps being annoying

And I tried to add some stuff to mlibc and it ended up breaking it

By just adding more headers and stuff

💀

Not even changing my logic

[opaque marten]

mfw

[hazy saddle]

It is a 4 month old commit

I should probably update it

[opaque marten]

prolly

[hazy saddle]

After VFS

[hazy saddle]

It legit made every app just page fault on 0

💀

[opaque marten]

bruh

[hazy saddle]

I have the changes I made somewhere up here

Let me find it

[hazy saddle]

Here

[opaque marten]

i see

[hazy saddle]

Is there anything in there that should break

[opaque marten]

no clue

[hazy saddle]

im willing to be its just an old commit that i got the short end of the stick on

[sterile plank]

https://klipy.com/gifs/what-65727

[opaque marten]

what

[clear bison]

The logo

[hazy saddle]

working on updating my mlibc

[hazy saddle]

guess who found another bug with IRQLs

and the TPR being set to 1

because i had a header file wrong

[hazy saddle]

wow woulndt ya know

the ps2 driver i stole

is what somehow

was causing triple faults

😭

[wanton grove]

how

[hazy saddle]

idk

ask @rotund furnace

they used the goat claude code to make a patch

which im refactoring

:p

[hazy saddle]

well

the ps2 code works

my shit deadlocks right away on real hardware now :/

But all I did was change the ps2 code

[peak cloak]

https://tenor.com/view/skeleton-gif-26826812

[hazy saddle]

Oh and I made a correctness change

On the spinlocks

So it won’t lower always to IRQL1

On the scheduler

Soooooo

I dunno that may be causing issues

But I can’t repro in VM

So all I can do is use my gdb stub

[peak cloak]

maybe it was useful after all

[hazy saddle]

True

Yeah if I undo that change it works

But like

I sort of need that change

[hazy saddle]

So I should not be taking spinlocks like this

Or rather

[warped ermine]

alright so the allocator uses 1 page every 4MiB for metadata

about

itll go up by 1 byte extra for every next power of two but not bad

[hazy saddle]

Why is this in here pewpew 💀

[warped ermine]

oh

bruh

i might be oopid

prodoundly oopid

[hazy saddle]

It’s fine lol

You just have to fix my deadlock isssues now :3

https://tenor.com/view/family-guy-joe-joe-taken-away-gif-17520860855595872089

[warped ermine]

AAAAAAAAAAAAAAAAAAAAAAAA

[hazy saddle]

here you gooo

apply to dev

lmk if you have issues ;3

[gentle quest]

I can send code like this

[hazy saddle]

I was making a joke because it’s a friend I know outside osdev lol :^)

I have a debugger on real hardware so il find the issue in like 10 seconds hopefully

[opaque marten]

my real hardware debugger is called spamming printf

[severe marlin]

<insert lecture about how printf has side effects that can mess with certain bugs that I always get every time I say that>

Real though, also commenting out half the code until it works

[hazy saddle]

It’s a deadlock in a random part of potentially the schedular

A GDBstub is IMO really nice here

[severe marlin]

Oh

Fun

[hazy saddle]

Which I have made

I can just attach it

Make it run

Wait for it to lock which takes 10 seconds

Stop it

Backtrace

[severe marlin]

Ahh yeah nice

That sounds very convenient ngl

[hazy saddle]

And because I have interrupt priority it should just work ™

Yes yes

My code should be pretty portable too if you have IRQLs or atleast TPR too btw

And given you use a serial port

[severe marlin]

"it should just work" - famous last words

[hazy saddle]

https://git.evalyngoemer.com/evalynOS/evalynOS/src/branch/dev/src/kernel/drivers/dbgstub/dbgstub.c

If you are curious on what it looks like

[severe marlin]

I am curious but I also have this shitty uni assignment (java and we have to use inheritance/oop) I gotta do :(

[hazy saddle]

Damn

Il be here later when your free ^^

[severe marlin]

hmm not sure I'll get it finished today but maybe 🤷

[hazy saddle]

tbf as long as nothing gets trapped into a cli section its not a "true" deadlock here as it can still get high prioirty things

but we shall see

[hazy saddle]

okay uhh

funn

somthing is getting into userspace

with IRQL2+ i presume

which means it cannot preempt anything properly

and its trapped inside userspace

good thing is

the debugger worked perfectly

so thank you past me

and also screw you past me for adding IRQLs

okay so if somthing holds a lock. then preempts its self it could break?

i do see OBOS has the locks commmented out of the schedular

wich technicaly should be fine

well no

because it can be called seperatly

// This should be assumed to be called with the current thread's context saved.
// It does NOT do that on it's own.
spinlock Core_SchedulerLock;
static bool suspendSched = false;
static _Atomic(size_t) nSuspended = 0;
attribute((no_instrument_function)) void Core_Schedule()
- obos

well yeah i dont have a context save / ISR on self preempt ig?

okay so

i think i may be mucking up register state?

or some other context

whenever schedule isnt called from an IRQ

uhhhh

nothing can preempt

[hazy saddle]

so any sleep that happens simply fails

because the timer never gets to run

so it thinks zero time passes

which means well

💥

i am just gonna pass this through aguments?

though maybe there is a cleaner way

eg CPU local

later tho™

i can just grab it from CPU local for the old IRQ to restore and check if its needed

i think this is fixed now?

It’s working

[hazy saddle]

though hardcoding this to zero may be technicaly valid too?

because
A. you should never hold a {spin} lock when being preempted

B. this is called from IRQL2 {or zero} which should always lower back to zero

which is also what happened before just by mistake

which seems to be right

(As that’s not triggering)

though the ps2 driver seems to be inconsistent?

as in somtimes i just dont get ant IRQs

i should add asserts soon

this seems sane enough

    int lock1r = spinlock_lock(&scheduler_spinlock);
    // Make sure IRQL is set to zero when unlocking as this
    // can be called from an ISR which will not return and
    // lower the IRQL properly. This is safe because this can
    // only be called from contexts where IRQL should become
    // zero after. EG the timer IRQ or places where you
    // are not holding onto a spinlock
    lock1r = 0;

[cyan bison]

lmao oops

[hazy saddle]

lmfaooo

[hazy saddle]

bte does this look correct @cyan bison?

for IRQL stuff

[cyan bison]

yes

[hazy saddle]

ty

[hazy saddle]

Why am I getting GPFs

When I disable LTO

What the fuck

And the stack trace is useless

Only happens with bad apple??

this patches it

if it diddnt like having this?

so it decided GPF

only without lto

[hazy saddle]

Okay so current list of todo

Update mlibc to trunk

Then I’m gonna swap my makefile out for meson + a shell script to make it one command to use

Maybe throw in jinx too

Then then

I will work on the VFS

[hazy saddle]

And I will support MMAP under tmpfs

Given it’s all already inside ram

And my tmpFS should be just adding vnodes when needed to the VFS instead of its own full advanced layer?

• implementing the right functions for tmpfs ops

[peak cloak]

anything but vfs

[hazy saddle]

To be fair the build system is probably worse than the current VFS

[peak cloak]

https://klipy.com/gifs/family-guy-car-crash

then get a better BS build system

[hazy saddle]

Yeah 💀

I have like way too many random bash and python scripts

The libc script is good though that’s staying

[peak cloak]

also if you still want to use borscht then i dont think its gonna help the random bash script situation

[peak cloak]

fair

[hazy saddle]

I’ve done my PMM rewrite for the second time

I need to clean it up a bit more

But now it won’t call back into the fucking heap

💀

Also I may be gettting a slab allocator soon

So the memory system should be a good less bit shit soon

I may still want to redo paging

And make that cleaner

Again

And the VMA could use a bit of abstractions

And cleanup

But that can wait because it’s functional

[hazy saddle]

also a good time to clean this function up

💀

this should be fully safe now

atleast from crashing ze kernel

[hazy saddle]

https://git.evalyngoemer.com/evalynOS/evalynOS/pulls/2

[hazy saddle]

Okay so

I’m probably

Gonna have to do the VFS soon

funn

[warped ermine]

500000000 lines of code

[hazy saddle]

Nah it’s not that bad

[urban mulch]

yoooooooooooooooooooooo

[hazy saddle]

The slab is so fucking based

It also stress tested my VMA

And it looks to be working mostly fine

[normal turret]

[hazy saddle]

The fragmentation looks to not cause issues too bad as long as it can reuse nodes

Which it will as long as nothing malicious abuses it

Like how I allocate tons of 64kb stacks etc

If it gets fragmented it will eventually find its way though to working

Or just carve off a new chunk etc

And later I can redo it a bit more to handle merging of chunks better

[urban mulch]

slabs are based

[hazy saddle]

And well the metadata is always smaller than the describing region so

It should play out fine for a while at the very least

[hazy saddle]

Ya know maybe I could benchmark this

And track the size of the tree and how many nodes are in it

[hazy saddle]

Or I can just make a command that iterates through the entire thing too tbh and be lazy instead of counters

(Which for a basic sanity check isn’t horrid)

For the first part of the VFS I am going to have to make a generic bitmap allocator tbh

Spesificaly for FDs

I’m thinking of holding a block of 64 bytes worth of bitmaps with a prev and next pointer

Plus an extra uint64 for metadata if needed

Then if it needs to expand and shrink it can

And then the allocator will find the lowest free bit using C compiler intrinsics and other things

And I will store in the metadata the pointer to the block that was last used to allocate from or free from

So that way I can get the lowest always pretty fast

Good thing is this allocator just needs to track free indexes so i shouldn’t need any macro fuckery for the most part

I will use a linked list for FDs but I will store 8 or 16 (pointers) per node to reduce allocation and other overhead along with cache locality

Il try and work on this today perhaps

See I’m finally thinking about this ahead of time and planning instead of just putting some bullshit down and moving on

https://tenor.com/view/tiktok-meme-i-just-made-some-bullshit-gif-15690632352507439130

I could also make the chunked linked list generic with void* stuff too

Which would be good

Maybe I can even make it dynamically sizable with flexible strict arrays

That way anything that needs a chunked linked list can just use it

[hazy saddle]

Then a macro that just auto casts can be made where needed to interface with these functions per thingy?

Maybe not that

[wanton grove]

when disk loading

[hazy saddle]

later

i wana get bash first

and the big code cleanup

[wanton grove]

you want to have linked list for fds right

[hazy saddle]

i also mentioned that

[wanton grove]

then why you need bitmap

like you can just pick lowest free fd from list or make new

[hazy saddle]

I always need the lowest and a bitmap is cheaper than checking the entire linked list

[hazy saddle]

ngl i should also expand on my abortable functions stuff

and add rdmsr/wrmsr and stuff

i may also want generic read & writes of diffrent widths

and i can make error handling a bit better

by returning a 128 bit value for things

(rax rdx pair)

[hazy saddle]

honestly i think i should do CPU local first

esp for performance things

and i did end up testing the removal of the timer preempting

now it dos this at the end of an IRQ

which is good!

and the IRQ just sets a flag

[hazy saddle]

okay so i think the todo list should be

• meson & jinx
• cpu local
• unite prremption from schedular (which is pretty much already the case but i dont wana push my half assed messy patches)
• make generic linked list helpers & support circular ones too
• make the chunked linked list helper funcs ontop
• implement a generic bitmap allocator for things
• implement the VFS

i have a few more random fixed to commit before i get started on this work tomorrow or two days from now

[hazy saddle]

im gonna keep makefiles

and just add jinx

i may use meson later once things are less tangled

but i dont wana learn TWO new buildsystems

also im finnaly gonna update flanterm to latest

il do limine revision 5 later tho

[marble ridge]

makefile > meson so good choice :3

[hazy saddle]

:3

[hazy saddle]

welp time to wait for debootsrap

[sterile plank]

https://klipy.com/gifs/sogga-serval-8

[hazy saddle]

wait i kind of need a sub module

or

well

i cant do in tree builds?

but then it asks for

./jinx: line 1358: cd: /home/evalyn/Documents/Programming/evalynOS/evalynOS/src/recipes: No such file or directory

there we go sub module works

:3

Looking like a good stat tho

Oh crap

I could integrate borcht into this

With jinx

And have it as a host recipe

./jinx: line 1358: cd: /home/evalyn/Documents/Programming/evalynOS/evalynOS/src/recipes: No such file or directory

again???

* done
* building package: kernel
./jinx: line 1120: [: too many arguments
./jinx: line 1130: [: too many arguments
./jinx: eval: line 242: syntax error: unexpected end of file from `if' command on line 239

what did i do

ahh

looks like i can only give one from source

sources seem to go into Cloning into '/base_dir/sources/flanterm'... here tho

idk what xbps-create: ERROR: invalid pkgver! got kernel-0.0-git_1' expected foo-1.0_1' is

[hazy saddle]

I think im cooking here

Things are going quite well

I now have a much more stripped down make file for building the kernel too

Which just make the executable

I’m gonna have to reorganize this project a bit more to make it clean

But it’s going good

Mainly I am going to bring out /src/kernel into /kernel/src

So the stuff needed to build the kernel is in /kernel

Because there are some extra generation steps that needs to invoke

Also I think this may be cursed but

I could use nodeJS for generation scripts

I think bunJS TECHNICSLY has faster startup times than Python?

So I can just JavaScript as part of the build process

[hazy saddle]

Jinx is now building the kernel

Time to just get the ports building with it

Like the gcc and user apps and doom

And then it just has to generate the initramfs

And then build the ISO

[hazy saddle]

binutils now compiles under jinx

Let’s go

[hazy saddle]

Now all my ports and mlibc etc compile under jinx

I just have to combine all the packages into an initamfs

Then make the iso

[warped ermine]

You should add KConfig to the build system

[hazy saddle]

No im gonna use borcht

From @peak cloak

[warped ermine]

I love make menuconfig rahh

[hazy saddle]

It’s just a menuconfig ripoff

[peak cloak]

Thanks for using Borscht™, please do not die!
-# Copyright (c) 2026, jastahooman, licensed under the 3-clause BSD license plz do not infringe

[peak cloak]

as stated in the instruction manual

literally

[hazy saddle]

I may use borcht later to pick what programs to compile

[peak cloak]

why are they in one repo

[hazy saddle]

Because it’s a mono repo

Everything in one repo

[peak cloak]

questionable behavior

fair

[hazy saddle]

I think the BSDs do this

[peak cloak]

the linux hybridmicrokernel () is the peak example of a monorepo

it has drivers for everything in ts

[hazy saddle]

It’s better than having to download two seperate repos and glue them together

[hazy saddle]

I mean userspace too

So like you can pick weather or not to compile doom

The one repo to rule them all

The one repo to compile it all

https://tenor.com/view/lord-of-the-rings-looking-reading-ring-gif-23305215

[peak cloak]

fair

[hazy saddle]

Oh BTW

Jinx gives you a package manager

All this shit is in packages

[hazy saddle]

just gotta wrap up implementing these

Oh and then test

I already have the shell script to execute these etc too

[hazy saddle]

I think it’s all properly working

https://tenor.com/view/family-guy-cartermiroquai-moving-furniture-carter-gif-24925834

This entire repo has just been shuffled around lmfao

I’m possibly gonna go more reorganization for the kernel though

Just testing the commands and scripts but it should just work now

[hazy saddle]

It works!

After fixing two typos

[hazy saddle]

https://git.evalyngoemer.com/evalynOS/evalynOS/commit/b476bbcd7d84df59e36016833f5c4ecd2c66946e

Let’s go I did actual important work

https://tenor.com/view/dancing-family-guy-peter-gif-20560406

[hazy saddle]

Tomorrow is gonna be CPU local

https://klipy.com/gifs/shannon-sharpe-v1

Or mlibc update

Maybe both

Who knows

[hazy saddle]

So for CPU local I just need to update my linker script with a section to be page aligned and to have start and end markers

And I will make sure that I have a special fixed section at the top with the linker

That way I can add things to that section and calculate the offset from GS

And then on SMP init I just have to allocate a new memory region for it for each core

[rancid viper]

and copy the original sections data into it

that way if u init static variables they will be initialized for all the cpus

are u planning on using __seg_gs?

[hazy saddle]

Yes

[hazy saddle]

What is this cool thing?

[rancid viper]

attribute

tells the compiler to access the variable via gs:

in clang its an alias for a specific address_space attribute

[hazy saddle]

How do you make that work properly

And tell it the offset?

[rancid viper]

it doesn't need an offset.. it uses gs

xD

the only issue is that for cpu local accesses you need to guarantee atomicity so u would need to use __atomic* accessors or something with it

[hazy saddle]

But how does it know the offset from gs to use

[rancid viper]

...

[hazy saddle]

What

[rancid viper]

it accesses the variable with gs:

[hazy saddle]

Yes?

[rancid viper]

I am not understanding

[hazy saddle]

gs:??

[rancid viper]

lol

[hazy saddle]

Like in assembly you go gs:64 or whatever. How does the compiler know what to generate for that. And how can I know how to allocate it right for other CPUs and copy it

[rancid viper]

oh

[hazy saddle]

I’ve tried to lookup clangs documentation on address spaces before but always came up empty handed for x86-64 stuff

[rancid viper]

well so it still generates the variables into the data section or wherever u put it normally, and just assumes kind of that gs is zero

but what u can do then is put the vars into a section and at runtime allocate sections of memory at offsets from that original section

and put the offsets into gs

[rotund furnace]

that's nasty as fuck 🥀

[rancid viper]

not really

[hazy saddle]

So basically I say this variable is in section CPUlocal and then set GS to the start of it and assign the varribles to that address space?

And then it takes the section it’s in and guesses?

[rotund furnace]

no

[rancid viper]

ehh, well no so if gs is zero the accesses would point to the original section

[rotund furnace]

it just tells the compiler to use gs instead of ds for the access iirc

[hazy saddle]

So how does this work? For this?

Like???

So it ends up not being relative because gs has to be zero??????

[rancid viper]

no gs doesn't have to be zero

u allocate blocks of memory at runtime and set gs to an offset from the original section to the new data

anyway there are some caveats with this and not how I do it but it is an option

[hazy saddle]

So gs = original_section - new_address

Because it things gs is zero so I have to increase it for it to match with the loaded executable with the memory I allocated

I presume it would handle wrap around?

This seems like it should work fine and potentially be better for codegen