EvalynOS

[gentle quest]

Like its great for generating solutions to tiny limited scope problems, but anything requiring greater context is gg

[hazy saddle]

Yeah, that’s why I only ever really use the web panels and paste relavent shit in and then see what it shits out for me to fix / if it did the thing right

90% of the time it falls flat on its face 💀

But somtimes it makes something workable

[gentle quest]

Yep

[urban mulch]

With the non-reasoning models? By good question.

That said

I just tried claude opus (I think it was called that?) and if you gently steer it in the right direction it can really help you build faster

Doesn't work if you don't already know how to do the thing you're doing though

[hazy saddle]

Yeah I only dump shit I to free models and have it shit out some garbage 💀

[urban mulch]

yea so

the paid shit is actually good if you know how to use it

[hazy saddle]

Fair

Good thing I’m never paying for it

[urban mulch]

lol

[hazy saddle]

I’m only on it to be a freeloader to be lazy when not wanting to make 10 lines for a calloc function because I have malloc already

[urban mulch]

if in the distant future you can afford it though, I'd definitely say it's worth trying

[hazy saddle]

I’d rather still keep having fun programming on my projects tbh

[signal plaza]

I'd like my OS to be AI optimized but not by using ai

[urban mulch]

well point is you have it do the boring tedious gruntwork

[signal plaza]

That's what the Good first bug tag is for

[urban mulch]

lmao as if your hobby kernel is gonna have more than 0.05 external contributors on average

[signal plaza]

If I can actually get good python support I will be the target OS for all beginning system developers everywhere

[hushed oracle]

I mean I wanted a microkernel so that you could write drivers in python...

[soft snow]

🙏 What is python gonna be used for though

[signal plaza]

Drivers, kernel modules, user space programs, libraries, etcetera. It's a design philosophy, not a scientifically decided choice.

[hazy saddle]

well its now only ~0.01 to ~0.02 seconds slower than gunzip!

just gonna do a bit more tweakings

then get the docs made and release it

[hazy saddle]

https://git.sr.ht/~evalyn/kgz

library is done

I already have it implemented into the old kernel

I’m gonna clean that up a bit and push it since having it support it is nice

[hazy saddle]

It’s also down to be a bit more optimized

There may be more that can be done

But for the tiny library it is, it’s pretty damn good

It compiles under C99 rn and C89 and K&R / unixv4? Support may be added later with the optimizations though unixv4 and K&R may have to be a seperate branch

[neon karma]

I'd use the one in limine

it's considerably faster than anything that's available

[hazy saddle]

What library do they use, I wana see if I can run benchmarks compared to it

[neon karma]

kamila wrote her own

[hazy saddle]

Ah

Well where is it

I know mine was approaching gunzip in userspace tests

[neon karma]

see the limine repo

it's in the readme iirc

[hazy saddle]

I did not see it there or in third party

[neon karma]

bruh

[hazy saddle]

Nvm it’s in the decompressor folder

💀

[neon karma]

https://github.com/Limine-Bootloader/Limine/blob/trunk/3RDPARTY.md

it's literally there

[hazy saddle]

But only on the noLLM branch

[hazy saddle]

How tf did I miss that

[neon karma]

skrill tissue

[signal plaza]

Time to steal some paging code

[hazy saddle]

also im micro optimising the FRED path

By shamelessly stealing from the compiler codegen of course

What’s a movezx?

No fucking clue

But it works

[haughty acorn]

move zero extend

[hazy saddle]

Just looked it up lol

Makes sense

I don’t need that on the IDT path since it pushes a full word for the vector

Now this wont work if I use extended features of FRED

Like NMI source reporting etc

But this does work for now

[hazy saddle]

I do wonder. Is zigs inline assembly powerful enough to properly align a function to a page and make another function at an exact offset from it?

[haughty acorn]

i will let you know in a bit, im going to be trying that in a minute lmao

[hazy saddle]

[hazy saddle]

I think il make a comment about inlining that function and later when I add support for advanced stuff that needs extra decoding I can make it a compile option?

[haughty acorn]

those will be packed in with the cs and ss in the frame anyway so dispatch_interrupt can detect it later presumably

[hazy saddle]

Oh yeah true

And then I can push things back into one unified frame?

And make FRED push a dummy vector

[haughty acorn]

yeah

and just you can switch on the flag for fred later to use the extra stuff in cs and ss or ignore it otherwise

thats legit what im doing

[hazy saddle]

Then if fred == on { decode the vector and fill it in on the stack and continue }

[haughty acorn]

yep

[hazy saddle]

Why didn’t I think of that

[haughty acorn]

(except im using inline asm in the fred thunk to do the decode and fill bit hopefully, its just a shr and a movzx)

[hazy saddle]

I can probably also make it a lower so it can stay in a register

[hazy saddle]

That code up there is doing the decoding btw

[haughty acorn]

mhm

[hazy saddle]

Without a shr

It just grabs the byte directly

[haughty acorn]

oh true you can stack offset

[hazy saddle]

Since its aligned

[haughty acorn]

well the aligned doesnt matter

its a byte

well ig you mean bit aligned

yeah

[hazy saddle]

Yes bit aligned

Because they have a possibility for extentions to go beyond 256 vectors

So that may need a bit of extra decoding

[haughty acorn]

now the only annoying bit is detecting syscall stuff

since otherwise those are hitting vectors 1 and 2

[hazy saddle]

its just this

[haughty acorn]

mhm

oh wait thats bit aligned too huh

[hazy saddle]

Yes

The Intel engineers were pretty smart

Let me go do this rework

oh i can even push a diffrent value based if it was a CPL3 vs a CPL0 from fred

this is a mess

decoding it in the asm is prolly best

or with a C shim?

actually

this isnt too bad

Good use of Goto aswell?

[haughty acorn]

id just use an else block personally

but ig it works

[hazy saddle]

then the indentation gets a bit messy

[haughty acorn]

yeah ig

[hazy saddle]

oh! this will also be nice

because in the non FRED path i can check if its from CPL3 and swapGS

[neon karma]

missing giant TODOs in the comments :P

makes it easy to search for them later

[hazy saddle]

[hazy saddle]

i dont need this but its prolly cheaper / better than
if(frame->cpu_frame.cs & 3)

[haughty acorn]

mhm

[hazy saddle]

• it means its not a wasted push

[hazy saddle]

Also @signal plaza I do just wana tell you now this is not the kind of thing to steal

Paging is a very fundamental concept and is meant to be a barrier to entry for getting into this hobby

As it makes you understand alot of concepts

[haughty acorn]

YES IT DOES

[hazy saddle]

That’s fucking insane

Once zig stabilizes I really should just try and get past it and learn

[haughty acorn]

did this and then bulk is a comptime string with the bulk of the handling

[hazy saddle]

I wonder if you can shove some constants in the free area and reuses them

Don’t waste a single byte

[haughty acorn]

maybe

[hazy saddle]

[haughty acorn]

oh and btw my offsets from rsp are computed by the compiler

[hazy saddle]

zamn

[signal plaza]

I know, I say that as a joke. I find it easier to understand what's happening by reading code and I usually read yours, so I joke that I steal ur code

[hazy saddle]

Ahh

I hope my new paging code is nice for you to read

It does have some limits to keep in mind that I left in due to performance stuff

[signal plaza]

It is, actually lol

[hazy saddle]

Eg it’s a pain to do merging and splitting of pages

[hazy saddle]

So I’ve elected to make it invalid and make some special things to make sure it will never need to happen

(They haven’t been made yet)

[haughty acorn]

the movl andl shrl andl blob might be wrong btw, the intent there is to put the fred vector in dx and (hold on i just found a mistake in there) the type field in bx, and then set bit 31 of the vector if its a syscall and put it in the vector field of the struct (and also leave in rdx for the dispatchy bit)

[hazy saddle]

Wouldn’t you want to put it into e(whatever) and have a zero extended byte move?

Because then it also zeros the upper part of the register

And if you use the 16 bit reg it may not work

[haughty acorn]

its actually in e_x and im anding it since you cant do a zero-extending write to memory in one insn and im using that upper bit to signal syscalls anyway

[hazy saddle]

Ahh

I wonder if doing the decoding in zig would be better?

And let the compiler optimize it heavily

[haughty acorn]

probably

but i cant replace only prolog and epilog so itd need to be its own function

and the compiler i dont think can inline into inline asm

so it wouldnt be inlined which i kinda want here

[hazy saddle]

i get away with this

i dont know if you can fully inline all the dispatch?

[haughty acorn]

so the way i do it

is i have a vtable for the exception vectors

[hazy saddle]

I was also planning a function pointer and syscall table, but having things still be in C is nice

[haughty acorn]

mhm

imo for this it isnt too bad

its basically just

• if bit 31 is set go to syscall entry
• if its less than 31 go to the vtable
• if its 0xDF go to the apic spurious handler
• otherwise go to the main irq dispatcher

[hazy saddle]

My thinking is if you add syscalls and the other various things you get into the realm where a higher level language is nice & the compiler can still make perfect optimizations in that case minus the extra function call (or two in syscalls spot but that is very minimal compared to the rest of a syscall)

[haughty acorn]

mhm

[hazy saddle]

Btw thank you @haughty acorn for letting me know I could just push an extra thing

And not need the jmp garbage

[haughty acorn]

np lol

it makes it so much easier yeah

[hazy saddle]

i think this is validated enough

il be pushing later™

and prolly work on either slab shit or printf shit

[haughty acorn]

yeah that looks good to me

[hazy saddle]

Because if I have a custom printf I can also make custom formatters for certain objects

It will be another combo single header & non single header library

So it can be integrated however you need

Though for this one I can imagine it being non single header is nice since I make it very easy to add custom formats

Only downside is I don’t wana parse IEEE crap

[haughty acorn]

see for this I just use zig std format and it's great and works on its own lol

[hazy saddle]

And this library will be made with no AI help (I somtimes use it for random debugging or whatever rarely) and I CBA to learn IEEE floating point shit

[haughty acorn]

yeah no for osdev you can probably ignore ieee

[hazy saddle]

Exactly

Il still add the hooks for it though if somebody wants to talk cheap send patches

And the format parsing fortune X.Y part

[haughty acorn]

mhm

[hazy saddle]

machine check exceptions are so fucking dumb omg

so it basicaly sends it to all cores by default i guess?

and i need to add a flag that says "hey im processing an #MCE just return"

i think this right now is fine though? since the parsing wouldnt fuckup other cores

intel

why

well this dosnt matter

since im not running this on i686 rn

[clear bison]

Good thing that my architecture will not have any of this bs

[hazy saddle]

welp its good enough for non SMP at least

Il just add an extra SMP flag so if it is mid #MCE handling other cores will just return and let the core that’s handling it right now decide to panic & broadcast etc

[hazy saddle]

🔥

[clear bison]

Btw if i get a C compiler running on my arch you might be able to make an port for it

[hazy saddle]

I will probably make a port for it after my loongarch64 one isn’t stub city 💀

[sterile plank]

meow

[hazy saddle]

Meow

[sterile plank]

I'll learn loongarch64 deeply

[hazy saddle]

This is working!

Well atleast good enough to not crash

Just by enabling it

Maybe I should add extra logging for stuff

[hazy saddle]

Welp it works on Intel and AMD

[haughty acorn]

well thats good

[hazy saddle]

I wonder if there is a way I can have this trigger an MCE on real hardware

[tender mantle]

as I said

[hazy saddle]

[tender mantle]

wiggle your ram a little

[hazy saddle]

I’m not wiggling my ram on this expensive of hardware

😭

Go wiggle your ram @tender mantle

[tender mantle]

I'll run it and unplug a ram stick from my old pc

I doubt it would mce

it's just funny

[hazy saddle]

Wiggle it first to try and get an MCE

Then yank it

[tender mantle]

wait

[tender mantle]

ram is WAY to expensive to be doing this shit rn

[hazy saddle]

It’s already been done

Your doing ts now

It’s DDR2 or 3 or some shit

That’s still cheap

And prices have been going down

@tender mantle lmk when your done

[hazy saddle]

I am unable to make a #MCE generate on hardware

Even using a special AMD only MSR

That lets you make a watchdog timer for instructions taking longer than 31us

To trigger a #MCE

🥀

And @tender mantle yanked ram and wiggled it

Nothing

[haughty acorn]

do you enable MCE in cr4?

[hazy saddle]

yes ofc

im gonna try this to slow the cpu down so bad

the watchdog may trigger

[haughty acorn]

gl

[hazy saddle]

this crap is so cursed as its only on AMD

#define MSR_AMD_MCE_WATCHDOG 0xC0010074
the MSR btw

[gentle quest]

oh nice

why do u need the wathcdog?

[hazy saddle]

its the only way i can trigger an #MCE without hardware damage 💀

[gentle quest]

oh i thought u were doing more interesting things lmfao

[hazy saddle]

if an instuction takes longer than 31us to complete it should give an MCE

XD

i am also making some MCE handling stuff

[gentle quest]

usually a watchdog is used to detect hard lockups

[hazy saddle]

technicaly it is but on an instruction level basis

[normal turret]

what's an MCE?

[gentle quest]

how can an instruction take longer than 31us?

[hazy saddle]

machine check exception

means your CPU is toast

[gentle quest]

nah

[normal turret]

ah

[gentle quest]

not necessarily

[hazy saddle]

well its somtimes recoverable

[gentle quest]

generic interrupt used for notifying a cpu of urgent things

its not masked with cli

[hazy saddle]

its just the minimum timeout

[gentle quest]

yeah but i think u misinterpreted it

[hazy saddle]

Hm?

[gentle quest]

its not tied to how long an instruction takes

its just a timeout after which the watchdoing is going to fire

and you're expected to prime it before it fires

[hazy saddle]

There is no prime MSR

💀

[gentle quest]

(if doing lockup detection)

[gentle quest]

prime just means rewrite the same value into it

[hazy saddle]

This and the other screenshot is all the docs in the AMD manual

[gentle quest]

its like a oneshot lapic timer

but you're supposed to reset it before it fires

and if you dont u consider yourself stuck

[hazy saddle]

So where do I need to write this

[gentle quest]

usually in the periodic interrupt for example

if it doesnt fire that means u have cli for too long

[hazy saddle]

thats not existing for this watchdog

[gentle quest]

am i really that bad at explaining

void timer_tick()
{
    touch_watchdog(); // rewrites the same value into the MSR
    sched_tick();
{

[hazy saddle]

no? this is a special watchdog. not with LAPIC oneshot. i know how that works

[gentle quest]

...

[hazy saddle]

What 😭

[gentle quest]

you have a periodic timer in the kernel right

for preempting threads

[hazy saddle]

This is my rewrite, I haven’t gotten yo that yet. I ended up on a side quest with machine check exceptions and I’m trying to test them on real hardware

[gentle quest]

yeah but you will have one right

eventually

[hazy saddle]

This is a special that can specifically generate a #MCE

Yes ofc

[gentle quest]

yes

just forget about the watchdog for now

so

you have a periodic timer

you can consider that timer a heartbeat of your kernel

but without a watchdog

if you skip a heartbeat

or flatline

there's nothing to catch that

[hazy saddle]

Yes

[gentle quest]

if you arm a watchdog, you're guaranteed that not generating a heartbeat (that does wrmsr(watchdog, fire_in_1ms)) will fire the watchdog MCE

and if that does get fired

you realise that a heartbeat was skipped

because no one rearmed the watchdog

so it expired and fired

does that make sense?

[hazy saddle]

Then why isn’t it triggering at all?

Yes it does make sense

[gentle quest]

wdym by that?

[hazy saddle]

And for some reason this specialty is talking about instruction execution?

i do set a thing here

[gentle quest]

can u link that spec btw

[hazy saddle]

i CBA to find the link again

search for CPU Watchdog Timer

or PDF page 762

[gentle quest]

cant grep this msr in linux

dont think they even use this

maybe it is specifically a hw debug msr, not the software watchdog

[hazy saddle]

Yeah it’s some really strange thing I found when trying to see what can generate an MCE

The Intel manual partly mentioned it but dosnt refer to the MSR at all

and it just #GPF's under KVM

as well it dosnt emulate it

[gentle quest]

@hazy saddle perf events seems to be the way to do NMIs on x86

and its how linux does a watchdog on x86

[hazy saddle]

ahh yeah

but the main thing im trying to do is see what happens at all when i get an #MCE since they are so uncommon

[haughty acorn]

I mean nmis are easy since you can just set delivery mode on apic things, mce is harder

[gentle quest]

cant you just IPI yourself on that vector?

[hazy saddle]

maybe? but i dont think thats considered a "true" #MCE with stuff i can parse etc?

[gentle quest]

u want like literal hw bugs?

[hazy saddle]

yes lol

[gentle quest]

the easiest way is to get a broken mobo where half of pci transactions fail

lol

[hazy saddle]

i sadly dont have one

[haughty acorn]

what's the vector? cause iirc the apic won't trigger vectors 0-16 or something like that

[hazy saddle]

0x12

[haughty acorn]

ok so it's above 0x10 so apic can ipi that vector

[gentle quest]

what info do u currently parse on mce

[hazy saddle]

weather its fatal & to panic or to log it and move on

[gentle quest]

i hope your printf is reentrant lol

[hazy saddle]

it takes a lock on putc which is why i am gonna make a proper log buffer later™

[gentle quest]

make a log ring like all of us lol

[hazy saddle]

thats what i meant

[gentle quest]

ye

maybe simics has a way to generate mces

or bochs

[hazy saddle]

i can generate them in QEMU, and they work as expected

was just wanting to verify real hardware

but idk if i can for now atleast

[gentle quest]

do they set all the msrs and stuff

[hazy saddle]

yup!

[gentle quest]

cool

[hazy saddle]

.

[gentle quest]

but yeah the easiest way is to get some throwaway hw i guess

and try pulling a pci device at runtime or something lol

[hazy saddle]

@tender mantle tried doing it with ram

but not pci

[gentle quest]

i doubt ram would generate an mce

btw does qemu allow u to specify fatal/non fatal?

[hazy saddle]

you set the flags register which says fatal/non

[gentle quest]

ah

[hazy saddle]

i do mce 0 0 0xB100000000000001 0x5 0 0 for a non fatal

and mce 0 0 0xB100000000000001 0x0 0 0 for fatal

[gentle quest]

but yeah, enable PCI MCEs, and read a pci device config space in a loop,

then just pull it out

that should produce a read timeout MCE

[hazy saddle]

iirc all MCEs are enabled by my code

[gentle quest]

then all u need is a testbed lol

[hazy saddle]

good thing luna has one

iirc this enables them all?

[gentle quest]

who knows

[hazy saddle]

welp its good enough

If I ever find hardware that wants to spit them out il enhance the handling

But this will atleast catch them and ignore non fatal one and behave good

[gentle quest]

yeah

[pseudo wedge]

Meowing dragon lol

[hazy saddle]

hey dragons arent too far off cats

rawr :3

[pseudo wedge]

Faxxxx

Rawrrrr :3

[hazy saddle]

https://git.sr.ht/~evalyn/EvalynOS/commit/44f33db

pushed

I think this code is good

I think I’m gonna add logging helpers macros next to cleanup some crap

And then maybe work on mm or printf ✨

[hazy saddle]

I definitely need to make logging macro

I’m probaly gonna have
LOG()
LOG_TAGGED(ansi code for the tag, “tag”, “text”)
LOG_TAGGED_PASS/FAIL to append a tag for passing and failing which should just be able to also call the LOG_TAGGED macro

[hazy saddle]

how do we all like this?

using this macro soup

eg
printf(ANSI_BYELLOW "[ARCH EARLY INIT]" ANSI_RESET " GDT INIT " ANSI_BGREEN "[OK]\n" ANSI_RESET);
becomes
LOG_TAGGED_OK("ARCH EARLY INIT", ANSI_BYELLOW, "GDT INIT")

[hazy saddle]

i think i got them into all the proper spots

[restive spire]

i like how this looks, do the colours have a meaning

e.g. have you chosen a colour for a particular subsystem

[hazy saddle]

just randomly picked per section

The logging has always been like this

Just uhh

I manually fucking did all the ANSI and spacing shit

Without macros

Btw I do wonder, does retro rocket of any support for the machine check exception / architecture?

[restive spire]

you should change your font

just to make it different and unique

that way w hen someone sees a screenshot they can tell its EvalynOS not just generic VGA font OS

[hazy saddle]

My old font was an 8x8 and honestly borderline unreadable 💀

Once I get userspace I will probably have that load a .psf file for the non earlycon console

I do want to get microcode updates supported soonish tbh too

As aparently it’s not too bad

You just have to get the file loaded into memory and then do some MSR magic

Which limine can do for me (the file loading)

[wanton grove]

now commit to limine to use this library

[hazy saddle]

No

[hazy saddle]

got KASLR working

[hazy saddle]

also fixed up the build system a bit so you can do KASLR=false make run and debugging builds where it waits for gdb always has KASLR off

Not doing this to be ulta secure but it’s easy enough to support it

And it makes me be not a dumbass and hard code values in my paging crap

[quiet rose]

wow when did you become a mod

[hazy saddle]

Today

Also you doesn’t have to post about it here lol

[quiet rose]

um did the admin just ramdonly asked you about becomming a mod? or did you apply or something

[quiet rose]

wdym

[urban mulch]

it is the mintocracy

their majesty personally chooses who gets to be court jester

[hazy saddle]

I mean you don’t have to come to this thread there was a bit of talk about it in lounge0

[quiet rose]

oh rly

[hazy saddle]

Mint said they wanted to make me a mod when I turned 18

I turned 18 today

[quiet rose]

i just opened this thread and saw your name in red

[hazy saddle]

Ah lmao

Speaking of my kernel

[haughty acorn]

congrats btw

[hazy saddle]

I need to stop being lazy and make my allocator lol

[quiet rose]

nice

[hazy saddle]

Thank you ^^

[quiet rose]

what allocator

how do you not have an allocator

[hazy saddle]

Memory allocator and virtual memory allocator

[hazy saddle]

I’m doing a rewrite

I finished the new paging code recently

I lowkey burnt myself out a lot snd rushed too much

So it rewriting and not doing that

[quiet rose]

maybe you should implement some shiny userspace apps

[hazy saddle]

🥀

I still need heap and vmem

I had a bash port and doom etc before

[quiet rose]

i mean

something even fancier

like xorg

[hazy saddle]

I will

But like

I have a lot of stuff to do?

[quiet rose]

btw i have weston and other stuff running and dont evne have smp and a usb driver

[hazy saddle]

I need a memory allocator

😭

[quiet rose]

plus a lot of stuff that im probaly supposed to have

[hazy saddle]

I also want to have SMP in my kernel

Which is another reason for this

[quiet rose]

isnt smp easy with like limine

[hazy saddle]

And portability to other architectures

[hazy saddle]

Not when your kernel isn’t really designed for it lol

Me with 7000 thread safety issues even in one core

[quiet rose]

luckly my new kernel is sort of designed for it

[hazy saddle]

The main thing is just to add it as soon as you can

Which I plan to do after I get the schedular and stuff up

[quiet rose]

well im not so sure, by that i mean just putting stuff on cpu specific things

[hazy saddle]

Yeah but

If your kernel already has a ton of stuff

It’s gonna be a pain

And it’s not just “make it CPU local”

[quiet rose]

pretty sure rust already protected me before hand with Send + Sync and mutexes

[hazy saddle]

And your done

Even fishix had issues

[quiet rose]

the only thing that ill have to worry about is prob race conditions

[hazy saddle]

And they use rust lol

[quiet rose]

isnt it in cpp

[hazy saddle]

No.

[quiet rose]

huh

[hazy saddle]

Wait?

No it was c++

They mentioned a rust require

Rewrite*

[neon karma]

tbh it kinda is

[hazy saddle]

You may be fine then?

[quiet rose]

true

niceee

[neon karma]

if you want it to be good, that's another matter

[quiet rose]

tho the bootloader crate dont natively support smp, so maybe ill make it my self after 10 years

[haughty acorn]

I had smp before scheduler even lol

[hazy saddle]

I want my SMP to atleast be properly stable and for threads to migrate and stuff lol

[quiet rose]

or steal it from my original ai slop filled os

[quiet rose]

unlilke limine

[hazy saddle]

I plan to have every core pick N buddy cores

And then only share with those

So it evens out naturally

And should work well with NUMA as I can make a seperate class of migrations

[quiet rose]

btw does C programmer usually make things them self, and only use libraries for huge things

like idk, ext4 driver?

[hazy saddle]

And not connect cross NUMA for standard buddy migrations

[quiet rose]

and code other stuffs like datatypes them self

[hazy saddle]

Yes I would make my own ext2 driver

[quiet rose]

(becuase i usually just use crates everywhere)

[hazy saddle]

And I generally make my own data types etc

Crates are a bad practice in general lol

[quiet rose]

why 😭

[hazy saddle]

I wrote my own UART driver even and GDT IDT and all of that

[quiet rose]

ig it makes you learn less

[hazy saddle]

CC: @urban mulch because I’m not experienced enough in rust

But i can say a few crates i have seen are just bad

Like the x86 one

[quiet rose]

x86_64?

[hazy saddle]

The x86 rust crate

[quiet rose]

never heard of it

and also my new kernel (the no ai one) actually uses the latest version of uart

[hazy saddle]

https://docs.rs/x86_64/latest/src/x86_64/registers/control.rs.html#82

[urban mulch]

Also Rust is not the be-all end-all of safety

[hazy saddle]

This one atleast

[urban mulch]

You can get pretty close with just linters

[hazy saddle]

This is stuff you should make

This is straight up missing crap

[urban mulch]

The real reason I use Rust is because I like its type system

You use unsafe so often that the notion of safe rust becomes almost an oxymoron

[hazy saddle]

The only libraries I use are flanterm and nanoprintf because I haven’t NIHd them yet

[quiet rose]

why would you use unsafe tho

[urban mulch]

You have to

[quiet rose]

yea but

[urban mulch]

It's either that or unsound

[quiet rose]

mostly

you dont

[urban mulch]

You are very very wrong

[hazy saddle]

You have to use unsafe for linked posts in rust right?

[quiet rose]

unless you hate your self or something

1sec lemme check howmany unsafes are in my code

[urban mulch]

IDK if you're writing in Rust but if you don't agree you need unsafe all over the place your kernel is full of unsoundness

[hazy saddle]

They have an AI make a kernel in rust

[urban mulch]

Ah

[haughty acorn]

that's why I go zig lol, safe rust is annoying and won't be that useful for anything osdev but zig also has a good type system and is meant for low level stuff like this

[urban mulch]

So the AI has been making unsound code as usual

[quiet rose]

i mean my no-ai kernel lol

[hazy saddle]

https://git.evalyngoemer.com/evalynOS/evalynOS/src/branch/main/kernel/src/arch/x86_64/drivers/16550uart.c

Why would you use a crate

For something you can make yourself

[urban mulch]

Also

[quiet rose]

well because

[hazy saddle]

And you get way more functionality and it generally will be better

[urban mulch]

any driver is unsafe to construct at the very least

[quiet rose]

why would you make it your self

when you can use a crate

[urban mulch]

Because what if the device does DMA?

[quiet rose]

:/

[hazy saddle]

https://git.evalyngoemer.com/evalynOS/evalynOS/src/branch/main/kernel/src/arch/x86_64/drivers/16550uart.h
this is also relavent

[quiet rose]

tbh if i dont like some parts of a crate, i usually just fork it

[hazy saddle]

At that point

Just sit down

Use your brain

Read some docs

And make a full driver

The UART crate is really not that good lol

It enabled IRQs

But provided no way of using them

[quiet rose]

if it works, and it prints characters, and doesnt crash your whole system, and is reasonable fast

then im fine with it

[hazy saddle]

That’s not what UART is only for

It’s also an input device

And you may want a proper TTY layer over it

And it took a lot of iteration to even make my driver work and be stable enough for real machines

I can tell you now the versions of the rust UART crate would fail on lots of my hardware

It works in a VM sure

But you have E9 for that

Plus a lot of the fun in making a kernel is making the drivers yourself

You just add an ext4 crate

Your done

You learnt nothing

And didn’t go through any of the fun

Also it’s gonna be harder to integrate into other systems of the kernel potentially

https://tenor.com/view/willy-wonka-and-the-chocolate-factory-gene-wilder-willy-wonka-you-get-nothing-nothing-gif-5432076

[quiet rose]

like, no?

i hate making drivers

[hazy saddle]

Then why are you making a kernel?

Like an overwhelming amount of Linux code is drivers

[quiet rose]

because its cool

[hazy saddle]

So you do it because it makes you look cool instead of doing it to learn and program low level things and hone your programming skills and stuff

[quiet rose]

basically yes

[neon karma]

drivers are one of the most boring parts tbh

it's just specs

[quiet rose]

i like shiny userspace apps

[haughty acorn]

kernels as a rule are either 90% drivers or theyre small and 50% of the code is to load drivers as external modules

[hazy saddle]

Using a crate is even more boring

[neon karma]

true

kernel internals are the most interesting parts

[quiet rose]

maybe, but it lets you focus on the thing that you actually wants to focus

[hazy saddle]

Using a crate skips alot of that too

[quiet rose]

true

[quiet rose]

how?

[hazy saddle]

Because you don’t end up designing a lot of the systems you would have to integrate it all together

[quiet rose]

??????

[hazy saddle]

You’re just calling external code? That does all of it internally and exposes some public API

[haughty acorn]

my drivers are like 30-50% integration with the rest of the kernel

[quiet rose]

um

[hazy saddle]

Certainly a lot less flexible and you miss out on alot of it

[quiet rose]

what exaclty does "system" mean

do you mean like

scheduler

or something like that

[hazy saddle]

The kernel

[quiet rose]

or like memory allocator

[hazy saddle]

Like the kernel as a whole

All of the subsystems

And making it all work together

[quiet rose]

um

[haughty acorn]

yeah all of the above lol

[quiet rose]

i dont think theres any crate that will help you

make other crate work togeather

[haughty acorn]

exactly

[hazy saddle]

Maybe it’s a difference of opinion, but you still miss out on a chunk of the work and fun

Your given a clean public API for things

[neon karma]

kiss out

[hazy saddle]

?

[neon karma]

but you still kiss out on a chunk of the work and fun

[hazy saddle]

Crap

[haughty acorn]

lmao

[quiet rose]

lol

[quiet rose]

well you still get the fun, except you dont have to work, which is probably ebtter

imo the fun of developing any software is to make subsystems work to geather

[hazy saddle]

I mean I get not wanting to make ACPI or ext4 drivers, but a lot of crates are really dumb lol

[hazy saddle]

If you have a clean public API the dev already did all that work for you

My KGZ library does all the porting work for you and provides macros and a single function to use

If I had made this for my kernel specifically

It would have been way more integrated

[hazy saddle]

I feel like you’re missing alot of my points here. I’m not saying it’s piss easy to integrate it. But it’s certainly a different thing to integrate crates together than your own code

[quiet rose]

idk i feel like C developors just like making everything by them self

[hazy saddle]

Rust and zig people also do it too

[haughty acorn]

yeah

[quiet rose]

maybe zig but i dont think rust people does it

i mean theres crates.io

[hazy saddle]

@urban mulch

[haughty acorn]

theres a lot of rust people that do it all themself

not all maybe

[quiet rose]

definitely not as much as C people

[haughty acorn]

or even definitely not all

but a bunch

[hazy saddle]

I mean Emma certainly does it themselves

[quiet rose]

or zig people

[hazy saddle]

I’m pretty sure you ended up NIH more stuff than me @haughty acorn

[neon karma]

don't use uint64_max crates for a 100 loc project challenge: impossible

[hazy saddle]

I mean like I also use external deps in C projects

[quiet rose]

whats that

[hazy saddle]

But I like to keep it minimal

[haughty acorn]

c you also cant often take as many other libs easily because theyll depend on stdlib stuff that you dont have

[neon karma]

what's what

[hazy saddle]

I sort of make my kernel have a standard libc because I use standard header and function names

Makes some things a little less annoying to reuse

[haughty acorn]

im not sure if this means imported more libs or less libs than you lol. my only library dependency is uACPI

[hazy saddle]

uACPI is special because doing AML fucking sucks

[haughty acorn]

yeah but you wont have standard libc file io lol

[hazy saddle]

What stuff am I gonna try and port into my kernel that needs that though

I’m taking random stackoverflow shit that wants like stroull or whatever BS

Etc etc

[haughty acorn]

my kernel has external deps on:

• zig
• an emulator to run it on
• uACPI (fetched automatically by zig)
• standalone/freestanding libc type headers for uACPI (autofetch)
• dimmer to make the disk image (autofetch)
  • dimmer depends on a zig library called zig-args (autofetch)
  • dimmer depends on zfat (autofetch)
    • zfat depends on a standalone C fatfs impl (autofetch)

[hazy saddle]

Who needs an emulator
Just dual boot and reboot every time you wana test jt

[haughty acorn]

my only runtime dep is zig and uacpi tho lol

[hazy saddle]

(When I first did OSDev years ago that’s what I did)

Well I actually flashed it to a USB by manually copying files and using gparted to change the ESP flag

So it was even slower

😭

[haughty acorn]

OOF

[hazy saddle]

I think I got up to exiting UEFI boot services and getting some PS/2. input and then I quit

[haughty acorn]

yeah thatll do it lmao

[hazy saddle]

I don’t know how the fuck I did it with my ADHD

💀

[haughty acorn]

idk lol

[hazy saddle]

My old kernel actual originally was a remake of that old project

Since I was bored one day

[haughty acorn]

some part of my brain keeps going what if i made an acpi library in pure zig with full support and used that instead and then i squash that part of my brain because FUCK NO I AINT DOIN THAT

[hazy saddle]

-# do it

[haughty acorn]

someday lol

i havent had the ability to even do any dev work in over 10 days now because of irl shit i def dont have time to make an acpi library when uacpi exists

[full fern]

What if I reinvent the wheel but make it a sphere instead so it can go in all directions

[haughty acorn]

someone already did that twice

theres omni wheels and mecanum wheels

[hazy saddle]

Let’s make it out of concrete

[urban mulch]

Rust devs absolutely make too much shit themselves

That's not a language specific thing

Most programmers like reinventing the wheel

[haughty acorn]

yep lmao

[hazy saddle]

It’s fun too somtimes

[kindred stirrup]

im not sure x86_64 machines have the ns16550 uart controller do they

[hazy saddle]

Well it’s a compatable chip or whatever

The interface

So it’s still a “16550” driver

[kindred stirrup]

nvm then

[clear bison]

It'll be "fun" once you realize the spec is shitty and reality doesn't follows it

[soft snow]

Congrats on mod.

[hazy saddle]

Thank you

[soft snow]

You did a lot for this community

Btw how is your decompressing library?

Is it faster than miniz now

[hazy saddle]

No idea I haven’t benched it (other than in userspace vs gunzip as a ballpark)

It was mostly made for fun tbf

It is certainly fast enough to not make me upset though™

[soft snow]

🙏

Any new features in the kernel

[hazy saddle]

I’m working on a rewrite

[soft snow]

Again lol

[hazy saddle]

Gonna make the memory allocator stuff soon

No not again

First rewrite

[soft snow]

Not your memory allocator

[soft snow]

No you rewrote it in C++ i think

Or was it rust?

[hazy saddle]

Oh after C++ I went back to the original project

I actually started anew her proper by myself

https://git.sr.ht/~evalyn/EvalynOS

I should possibly update the post™

But it’s 2am and I’m going back to bed in a moment

[soft snow]

Feel free

Sorry for walking you up.

[hazy saddle]

Nah you didn’t wake me up lol

[cyan bison]

isn't that the broken uart

tmk the pc16550dn is the newer version that should actually be used

[hazy saddle]

There was like one or two models with some issues

It’s just always called the 16550 by most things

[cyan bison]

my thing uses the pc16550dn

[hazy saddle]

Did any models with the broken FIFO even end up in any PCs?

[cyan bison]

beats me

[hushed oracle]

Huh

It's a ns16550

[clear bison]

I don't think so? However the ns16550 had some predecessors and those lacked a FIFO and other stuff

[restive spire]

happy belated birthday! 🎂

[hazy saddle]

Thank you ^^

[hazy saddle]

i think i have finished my early memory allocator

I may tweak it a bit more

But it’s good enough™ for now

After a bunch of work on it

And defer is working on clang22 nicely aswell

[neon karma]

my earliest allocator is bump

[hazy saddle]

I need this one to support free() for vmem stuff later

And I didn’t wana make statically allocated objects

This one is able to return memory to the PMM too so there is nothing wasted either

[neon karma]

my early allocator is for pfndb

[hazy saddle]

Ahh

No that’s probably fine

Though I use a freelist for that and I plan to just not free() anything before it’s setup so I can track things properly™

And I can flush out whatever is partially filled from the memory map into a more advanced PMM after pfndb init

[hazy saddle]

I also hope nobody sees this and goes bitmap!!!!! Bitmaps must be good for everything because I use them

I have had like 3 people referring to my frankly garbage code in some cases

And I am very concerned

[hazy saddle]

One of them even stole my code and made a PR somewhere 💀

While completely disregarding the license

After I get vmem and a proper slab allocator going though

Il work on SMP shit

Some timer crap

And try and get a mlibc port going for hello world

And then get the VFS squared away and /dev/fb0 /dev/ps2 and /dev/pcspkr crap so I can port doom and bad apple

But in a way more proper way

Then bash probably should work fine but I’d need to throw in fork()

Which should hopefully just work?

Oh yeah il also need the page table clone funcs and scheduler things

[sterile plank]

when will you get networking done

[hazy saddle]

But it should hopefully just work™ after I finish it

[hazy saddle]

After I get x11 and nvme il look into some network drivers

[sterile plank]

good good

also get some browser ported

[hazy saddle]

Il do a CLI browser 💀

I ain’t porting Firefox for a while

[viral bison]

my next thing is prob nvme or usb

lol

[hazy saddle]

Unless a png is acceptable

Wayland/DRM may also be fun?

Iirc zinnia did that

And that would be cool af

[sterile plank]

i want serchat on evalynos

[hazy saddle]

If you get a CLI client in C I will gladly have a port of it

[sterile plank]

boring

[hazy saddle]

Or atleast a light weight enough JS runtime

But I cannot run a full browser anytime soon 😭

[sterile plank]

why

nothing is impossible

[hazy saddle]

Yeah but

It’s really hard lol

[sterile plank]

meow

[hazy saddle]

Mrrp

[tender mantle]

Meow

[tender mantle]

Give it a few years I’m sure she’ll get there

[hazy saddle]

I mean I could just do Linux binary compat :x

But that’s lame as shit

[tender mantle]

Boring

[hazy saddle]

Though if I get enough ports having some compat layer in userspace could be cool though

[tender mantle]

Do something interesting like nt binary compat

[hazy saddle]

https://tenor.com/view/aww-hell-nahh-aww-hell-aww-hell-nahh-gif-9659330954843262630

I’m not a fucking insane as reactOS people

If I can I would port wine though

[sterile plank]

its eassyyy

git clone wine

[hazy saddle]

😭

im pretty sure @full fern had to do a fuck ton of shit to make it work right

[hushed oracle]

Don't they basically also reuse a lot from Wine?

[hazy saddle]

they share alot back and forth iirc

[round kestrel]

what do u use to allocate the linked list node structure?

on llist_push

[hazy saddle]

It’s an intrusively linked list

typedef struct spalloc_page_header {
    llist_node_t llnode;
    uint64_t bmap[2];
} spalloc_page_header_t;

it just has the node in the header

and push just connects pointers

void llist_push_front(llist_t* llist, llist_node_t* node) {
    if(llist->head == nullptr) {
        llist_initial_node(llist, node);
        return;
    }

    llist_node_prepend(llist, llist->head, node);
}

[round kestrel]

ohh huh

thats cool

[hazy saddle]

And I use a macro to get the header from a node

With type checking and shit

[round kestrel]

damn thats nice

[hazy saddle]

Intrusive structures are quite nice

oh and push and pop are aliases

[[gnu::alias("llist_push_front")]] void llist_push(llist_t* llist, llist_node_t* node);
[[gnu::alias("llist_pop_front")]] llist_node_t* llist_pop(llist_t* llist);

[hazy saddle]

okay i think every bug is zapped

i also have more asserts here to catch bugs

i also am temp tracking the PMM pages to make sure its not leaking

and i think it all works now

and the clankers arent complaining so

i think ive got this all properly fixed

any issues should be from severe miss use of the allocator at the very least

which im not particularly interested in hardening beyond the basics

I’m gonna let this clanker test to run overnight

This should catch anything

Then I can fix any issues or whatever

And then have a finished nice allocator

To be clear I did not generate the allocator with AI

Just the tests because tests are fucking annoying to make

im having this run overnight on two computers*

@tender mantle is gonna run it too :3

ty btw

[tender mantle]

all good :3

[hazy saddle]

Looks like things worked over night

Though I had to change the test and code to return false on invalid configs

Because the test was of course dumb as shit 💀

[jade olive]

haiii :3
continuing the bad ram test chain from #when-your-os-goes-crazy

[hazy saddle]

Yup

I also may wana test managarm on this

Since I may work on some #MCE stuff for them

[jade olive]

well when i get a functional nonfunctional ram to boot you can throw anything you want at me :3

[hazy saddle]

🔥

[jade olive]

ok so limine got invalid opcode with perfectly fine ram as well
so i'm pulling out the "modern" server

[hazy saddle]

May wana ask mint about that #UD?

[jade olive]

i think the #UD handler is definitely part of the ILO
cuz i keep seeing that when trying random bs

[hazy saddle]

i do also hope we dont hit this deadlock 😭

[jade olive]

eh

surely not

[hazy saddle]

Printf would only touch the stack and the framebuffer

So it should trigger mid log for this

Shouldn’t*

oh also on panic you get a ralsei

[jade olive]

yooooooo

worth breaking my equipment just to get that art :3

[hazy saddle]

🔥

Hopefully the terminal emulator can render those right

Also my stuff shouldn’t break things

Well atleast not until I add microcode updates

Which I don’t wana do

Because AMD fucked it up SO FUCKING MUCH

To the point Linux just sha256s every single fucking ucode for AMD for a lot of CPUs

Atleast on Intel it seems mostly sane?

[jade olive]

no shot
this server actually boots up within 1 working day

[hazy saddle]

Yoooo

😭

[jade olive]

on the bright side this server actually has serial bios log and the adapter seems to work

i really don't want to pull up my pentium III laptop just for serial

okies
got link local IPv6
very relevant

[hazy saddle]

I paid 20 bucks for an overpriced cable just for an FTDI chip lol

Just because I couldn’t be added to have something not work

[jade olive]

well

you ain't gonna believe this

illegal opcode 😭

i'm not even joking

this has to be a limine thing

i'm doin grub-mkrescue
f this noise

or is it the limine protocol?

[hazy saddle]

💀

What happens if you just make a custom limine install on a USB

Then add my conf and kernel binary

(Remove the initramfs from the conf)

And my kernel only boots off limine

[jade olive]

wanna see the error message up close?
like the faulting IP?

[hazy saddle]

What is the IP?

And sure

But idk why it shows 32 bit registers

💀

Because my kernel is very much x86-64 and supports VERY modern features

(Though not required)

If anything it should fault in my kernel its self?

[jade olive]

no clue to be hones

trying the limine usb thing now

kinda checking everything along the way

[jade olive]

and i meant instruction pointer

[hazy saddle]

I know

lol

[jade olive]

also sry if not responding that often i literally cannot hear anything over the server lmao

[hazy saddle]

ITS FINE

[hushed oracle]

what did I miss?

[jade olive]

my server actually doing its job unfortunately for us

OMG FINALLY

[hushed oracle]

How do you even handle MCEs?

[hazy saddle]

You can check my code

[jade olive]

CMONNNNN

[hazy saddle]

https://git.evalyngoemer.com/evalynOS/evalynOS/src/branch/main/kernel/src/arch/x86_64/drivers/mce/mce.c

[hazy saddle]

Try pkill-9 minicom

Also maybe picocom will behave better?

[hazy saddle]

@hushed oracle

[jade olive]

nahh
bustin out the pentium-II box
i5-1345U skill diff obv

[hazy saddle]

Is that pentium 2 32bit iirc?

[jade olive]

yep

[clear bison]

Does evalyn's kernel even support 32-bit?

[hazy saddle]

Nope

But that’s just for serial

[jade olive]

nonono
you misunderstood

yess

for the serial

[clear bison]

Ah

[hazy saddle]

If I ever do 32bit it will be a source compatible kernel with a very similar syscall interface

[hushed oracle]

My kernel supports it if it's i686

[clear bison]

Idk you should just use parallel, it's a simpler protocol

[hushed oracle]

I wanted to port mlibc and gotten depressed

[clear bison]

Pentium II is i686 yeah

[hushed oracle]

But I can try implementing MCE handling real quick

[clear bison]

i686 also has MCEs iirc

[hushed oracle]

And my kernel supports compat mode

[hazy saddle]

It may be the legacy variants though depending on how old it is

[hushed oracle]

But which server is failing?

[hazy saddle]

Though I support that too

[hazy saddle]

The one running my kernel

Limine was dying

[hushed oracle]

So x86_64?

[hazy saddle]

Yes

[hushed oracle]

My kernel boots with hyper as well

[hazy saddle]

My kernel isn’t running on x86-32

[hushed oracle]

We can check hyper

[hazy saddle]

I don’t wana make a bootloader abstraction thing rn

[hushed oracle]

I mean I can give an image (though I don't handle MCEs)

[hushed oracle]

I don't have that...

[hazy saddle]

Maybe later I will make functions that can grab things genericly

[hazy saddle]

Did I say you did?

[clear bison]

Idk I've got this on the pentium pro manuals

[hushed oracle]

I support 2 bootloaders, but it's not very abstracted

[hazy saddle]

MCA is the new one, you can lack MCA and just have normal MCEs which are always fatal

[hushed oracle]

Idk, it all came from supporting RISC-V

[hazy saddle]

MCA lets you see if it’s non fatal

But it’s still a #MCE for both

[clear bison]

This is MCA

[hazy saddle]

[clear bison]

[hazy saddle]

I know.

I was just saying that

[clear bison]

MCE exists since the pentium days iirc