AxeialOS, A Modern AMD64 POSIX Operating System

[sleek mason]

that's fine

but stil rust

1000th message!

[marble epoch]

[exotic ibex]

free advertisement is real

[bronze prawn]

marvin just watching all channels for the menix mentions

[neat sandal]

shameless plug

[sleek mason]

why only 2 stars here but 12 stars and 3 forks on repo

😭

so ded btw

well, gotta fix a random triple fault...

[sleek mason]

fixed that, now fucking fixing random GPF

[exotic ibex]

Lmao

[sleek mason]

LESS GOOOO

finally fixed the gpf

now a pagefault....

well Thats simple fix just a NULL deref

SO CLOSE to stability

and fixed the timer stuff

like the redundant function to setup APIC timer, like one for APs and one for BSP

[exotic ibex]

bruh

[sleek mason]

also i'm stress testing with like 100 different prio threads across 4-8 cpus

well that works

all threads get some schduling time... and wreaking fast

tho the loggers are MASSIVE bottleneck

[exotic ibex]

hmm, lets test my kernel then

[sleek mason]

well it was surprisingly fast, so it mean my kernel is efficient

[exotic ibex]

right

hmm my kernel cant handle too much threads

havent touched for a while though

[sleek mason]

insane

you kernel must atleast handle 100 - 500 threads. MINIMUM

ig make it less overcomplicated

your kernel is just...

[exotic ibex]

or maybe it cant handle creating

threads

[sleek mason]

fast?

[exotic ibex]

no, creating is slow asf

[sleek mason]

well how many thread it can handle before it blows up?

[sleek mason]

how...

[exotic ibex]

MAX_CPUS * 2

which is 128

hmm

[exotic ibex]

im so confused

[sleek mason]

your max CPUs is 64?

[exotic ibex]

yeah

[sleek mason]

i use 256

i'm not sure, i only created 100 threads stress test, maybe 1000?

[exotic ibex]

ohwait

creating is not slow

it halts somewhere i think

[sleek mason]

that's way weirder

how 😭

[exotic ibex]

should be around here

[sleek mason]

wtf, you release spinlock, where do you even aquire it in first place?

Oh nvm

i'm blind 😭

[exotic ibex]

gotta touch some grass

me too also

[sleek mason]

fr

it could be activate task

if it also uses spinlocks

and it acquiring the same lock, it may just deadlock

[exotic ibex]

this is the most unsafe code in my kernel

struct task_struct *spawn_user_process_raw(void *data, size_t len, const char *name) {
    struct task_struct *curr = get_current();
    struct task_struct *p = copy_process(0, 0, curr);
    if (!p) return NULL;

    strncpy(p->comm, name, sizeof(p->comm));
    p->flags &= ~PF_KTHREAD;

    p->mm = mm_create();
    if (!p->mm) {
        free_task(p);
        return NULL;
    }
    p->active_mm = p->mm;

    uint64_t code_addr = 0x400000; // Standard base for simple ELFs/bins
    if (mm_populate_user_range(p->mm, code_addr, len, VM_READ | VM_WRITE | VM_EXEC | VM_USER, data, len) != 0) {
        free_task(p);
        return NULL;
    }

    uint64_t stack_top = vmm_get_max_user_address() - PAGE_SIZE; // Dynamic canonical stack top
    uint64_t stack_size = PAGE_SIZE * 16;
    uint64_t stack_base = stack_top - stack_size;
    if (mm_populate_user_range(p->mm, stack_base, stack_size, VM_READ | VM_WRITE | VM_USER | VM_STACK, NULL, 0) != 0) {
        free_task(p);
        return NULL;
    }

    cpu_regs *regs = (cpu_regs *)((uint8_t *)p->stack + (PAGE_SIZE * 4) - sizeof(cpu_regs));
    memset(regs, 0, sizeof(cpu_regs));

    regs->rip = code_addr;
    regs->rsp = stack_top - 8; // Align stack
    regs->cs = USER_CODE_SELECTOR | 3;
    regs->ss = USER_DATA_SELECTOR | 3;
    regs->rflags = 0x202; // IF=1, bit 1 is reserved and must be 1

    regs->ds = regs->es = regs->fs = regs->gs = (USER_DATA_SELECTOR | 3);

    uint64_t *sp = (uint64_t *)regs;
    *(--sp) = (uint64_t)ret_from_user_thread;
    *(--sp) = 0; // rbx
    *(--sp) = 0; // rbp
    *(--sp) = 0; // r12
    *(--sp) = 0; // r13
    *(--sp) = 0; // r14
    *(--sp) = 0; // r15

    p->thread.rsp = (uint64_t)sp;

    wake_up_new_task(p);
    return p;
}
EXPORT_SYMBOL(spawn_user_process_raw);

[exotic ibex]

fair point

[sleek mason]

i do see why.

[exotic ibex]

the name speaks

for itself

[sleek mason]

raw 😭

[exotic ibex]

basically

dw i just dont have VFS

[sleek mason]

well, i have question, why not isolate processes and threads?

[exotic ibex]

what

i do

its just implicit

task_struct->mm and task_struct->active_mm is the identifier

i use linux concept

[sleek mason]

no i mean, in my kernel, there is two separate structs, one is AxeThreads and AxeSchd

[exotic ibex]

i have the same abstraction

but only from a user perspective

[sleek mason]

so task_struct is thread right?

[exotic ibex]

my kernel just treats everything as "tasks"

[exotic ibex]

no bruh

its user proc/thread and kernel proc/thread

[sleek mason]

i see, my kernel treats them differently

[exotic ibex]

thats why mine is implicit

[exotic ibex]

separation is good but its a bit more work when you update the scheduler

[sleek mason]

like for processes, i don't schedule in processes "context" i schedule in threads context and each process has its addr space and a main thread which is its entry point

because when i actually see videos talking about how processes and multitasking works in OSes

well treating them as "task" may centralize control

like unify

[exotic ibex]

i see

[sleek mason]

but yea, you gotta improve your kernel

[exotic ibex]

just the bugs

though still cooking my MM again

gotta make sure its solid

btw i freaking port linux maple tree xarray and radix tree in one go

lmao

[sleek mason]

well if yall weren't aware i have community on discord

https://discord.gg/dEPMKfktBt

[sleek mason]

most of progress is here

[sleek mason]

still 3 stars ⭐

[sleek mason]

yay 5 stars

[sleek mason]

stresstesting 100 threads across 4 CPUs, on REAL HARDWARE, super fast

[exotic ibex]

are you willing to test my kernel on your hw? 🥰

[exotic ibex]

thats is pretty crazy

[sleek mason]

i didn't though my kernel CAN so that fast

ig very efficient and optimized

[crimson raft]

What are the loops doing

Make them malloc somthing calculate some numbers of pi and then free it

And maybe some other system calls

[exotic ibex]

good suggestion

[crimson raft]

If you’re doing dynamic memory management shit it’s gonna stress out your system. Make sure it’s large enough to make mmap unmap aswell too

And make sure you don’t have any deadlocks or race conditions there

That’s the main place I’d suspect them imo

Atleast where I suspect them happening to me

Also make the tasks last for 5 minutes and see how far they all got to check how fair the schedular is

[sleek mason]

printing only

also they are kernel threads

[crimson raft]

Oh

You should still make them do a bit of more work to properly stress test it?

[sleek mason]

i may do like inconsistent loop calls to KMalloc and KFree

to see if my MMs blow up

[crimson raft]

Calculate prime numbers

Malloc on composte

Free on prime

[sleek mason]

yea, lemme do that

[crimson raft]

And if you hit a limit free all of them anyways

Like num % 256 == 0 do a free on them all

[sleek mason]

whats more impressive is that I'm PRINTING in EVERY 10000 Loops, NOT EVERY LOOP

just imagine HOW wreaking fast that is

[sleek mason]

yes

[tepid burrow]

Not really.

Yes its fast.

But printing is the expensive part.

Not the loops

I bet if you printed every loop it would be a lot slower.

[tepid burrow]

Not that fast lol

But as always, its 100% faster than a emulator

[crimson raft]

Loops are basically free especially with modern branch predictors

They are gonna be nailing it every loop

And basically make it one cycle

[tepid burrow]

yea

[crimson raft]

Possibly less than a cycle depending on throughput too?

I’d have to check the uops list tho

[sleek mason]

indeed, i'm aware of that fact

also my KrnPrintfs loggers ARE HEAVY, very very HEAVY

Update: doing final stability touch ups

today and complete this shit

[daring mortar]

you should port the fireworks test from my OS: https://github.com/iProgramMC/Boron/blob/master/drivers/test/source/fworktst.c

[sleek mason]

the name is pretty self-explanatory of what it does

[daring mortar]

each firework particle (1 pixel) is controlled by a thread

so you can have hundreds of threads all using up a little bit of scheduling time

[sleek mason]

hmm... worth the try honestly

[sleek mason]

how do i port it?, as a kernel module?

[daring mortar]

however you see fit

[sleek mason]

hmm, doing that

[sleek mason]

fixing some posixfd stuff

my fds broke and printf doesnt work for libc (newlib)

also thats my final thing to fix

[exotic ibex]

Bro picked the worst possible libc

[sleek mason]

ngl was the easiest to port, as i have ported and tried it before, mlibc gave me stroke
-# or i'm just a noob

[exotic ibex]

Sysdep moments

I'm just gonna use musl

[sleek mason]

musl is also good

[exotic ibex]

Or llvm libc if I want to be truly GNU free

[sleek mason]

do yall think i need to add documentation? for my kernel?

because it's almost complete

also doing some more error handling overhaul

[exotic ibex]

obviously

fact, the xv6 kernel (MIT teaching kernel) have more docs LOC than the kernel itself

just the old xv6 though, newer revisions (like the 2022 course) is much bigger

[marble epoch]

use mlibc

it's way more portable

llvm libc is incomplete and hard to port

[tepid burrow]

I'm unsure why but a lot of hobby OS's all of a sudden started porting newlib for some reason.

[tepid burrow]

What made it easy to port

[exotic ibex]

fair point

[marble epoch]

i don't get what's so hard about mlibc tbh

there's literally a demo of how to do it

[tepid burrow]

toolchain ig

[marble epoch]

????

[tepid burrow]

it isnt better with newlib though lol

[sleek mason]

the problems with headers and sysdeps

[marble epoch]

https://github.com/managarm/mlibc/tree/master/book/src

this should explain how everything works

[tepid burrow]

huh

never had any issue with that

[tepid burrow]

Suprisinigly since that thing is riscv so that didnt make it any easier for me personally.

[sleek mason]

like static assert problems and stuff, making it compile my sysdeps was also hard

[crimson raft]

Yeah GCC/binutils need diffrent patches for x86

[marble epoch]

but you need these patches regardless

[tepid burrow]

its really easy.

[marble epoch]

like, using newlib doesn't magically fix this issue

[tepid burrow]

or well maybe its only me

[tepid burrow]

thats what im thinkinh

it wont be better with newlib

[crimson raft]

Mlibc is just a plain better libc too in general with atleast decent documentation and examples for getting started

[marble epoch]

i might be biased, but mlibc is literally piss easy to port. you just implement the sysdeps and add it to the build system

and you have people here that actually care about it and can support you

[exotic ibex]

so whats the choice then, musl or mlibc, what do you guys think

[marble epoch]

are you doing linux binary compatibility

[exotic ibex]

yes

[marble epoch]

then musl works fine

mlibc is for when you want POSIX/linux source compatibility

[sleek mason]

i had like problems with headers and just refusing to compile, like tons of static assert kind of problems

[marble epoch]

in mlibc?

[sleek mason]

maybe i'm a noob

[sleek mason]

yea

[marble epoch]

please drop the errors, might help us in the future too

[sleek mason]

sure, if i would switch mlibc, and ofc i would retry, newlib is pretty stripped

and kind of bad

like long sysdeps list

[marble epoch]

the mlibc sysdep list is pretty short actually

https://github.com/managarm/mlibc/blob/master/options/posix/include/mlibc/posix-sysdeps.hpp

and since they're weak functions, you can just leave out the ones you don't have

[exotic ibex]

interesting to know that mlibc is a bit larger than musl

[sleek mason]

yea

[marble epoch]

it's also a bit more correct :P

most notably, musl only targets linux, but managarm needs a lot of userspace code in libc to work

[exotic ibex]

hmm, maybe mlibc is a better fit?

though again, i forgot, i make kernel

so maybe my distro will use mlibc

[marble epoch]

note that not all programs support mlibc upstream (because of messed up includes, aka not posix compliant programs)

so you might need few patches for some programs

i recently upstreamed a fix for fastfetch

and xbps

[sleek mason]

Should i change name of "AxeialOS?"

it's too random

and was random

i just thought "Axe" and "Axial Loads" would be nice

[daring mortar]

you would have to build a cross compiler (or hack your way around the fact that your host compiler compiles executable for your host OS) on any libc

so thats not really a problem

mlibc is actually super portable, like to start with you need like 10 whole mlibc::sys_X implementations

(as you port more programs you implement more sys_X callbacks)

its really neat

im not sure how the arch portability is but they did get a 32-bit x86 port going so

[tepid burrow]

Firstly no, for mlibc you need to build a full toolchain.

[tepid burrow]

I know, I've already ported mlibc lol

[sleek mason]

please, suggest a new name for AxeialOS

should we change?

[sleek mason]

PushError("Ayo, Exception #GPF", -Unknown)

[bronze prawn]

10 words??????

[sleek mason]

Well the constrains are all optional

forgot the asterisk

[bronze prawn]

Okay but still

How do you even name an os more then like 4 words

Without it sounding ass

[sleek mason]

It should be creative

Swiftix?

[tepid burrow]

If it's not built with swift, which ik it isnt no.

[sleek mason]

It is pretty optimized and efficient

just not stable

Lemme cook

[tepid burrow]

what does that have to do with the thing?

[sleek mason]

Swift = fast

[tepid burrow]

https://www.swift.org/

💀

[sleek mason]

Fuck, lets think of others

[tepid burrow]

Personally I would just think it's a OS built in swift

[sleek mason]

already taken

[tepid burrow]

https://swiftos.dev/ saw that

[sleek mason]

Yea

IxyBox?

very random

also taken

[sleek mason]

64IX?

[tepid burrow]

really weird

[sleek mason]

Ik

random

lemme cook again

[cinder otter]

Axis?

[sleek mason]

it's taken maybe

[sleek mason]

what 'bout "Velonix"?

"Optix"?

wait that's nice, "Optimize" + "Unix/Posix" + "Optics"

meaning: fast + standard + clarity(open source)

hmm

[sleek mason]

what do you think about this @tepid burrow

[exotic ibex]

Optix might have issues with Blender Optix and nvidia Optix (ray tracing)

[sleek mason]

fuck, that was actually a great name

lemme think something else

[exotic ibex]

Yeah, Optix is very sleek tbh

[sleek mason]

also matches lot of my OS

[exotic ibex]

Your os have anything special in mind

Like design considerations

Flexible

Etc.

[sleek mason]

It would be modular and ultra optimized while being lightweight

with modernness

i had another "Velix" or "Velonix"

[exotic ibex]

likely velonix here

velix is a bit unclear

[sleek mason]

I see

[exotic ibex]

and velonix clearly express velocity

the velo-

[sleek mason]

lemme check if it's ALSO taken

[exotic ibex]

my kernel name is partly taken

theres already some Aero kernels

[sleek mason]

FUCK

[exotic ibex]

AI BIKE ☠️☠️ ☠️☠️ 🔥🔥🔥🔥

[sleek mason]

fr its software would also be written by AI

[exotic ibex]

shit

but you could use that though

[sleek mason]

Huh

[exotic ibex]

its not copyrighted as long as the industry differs

i think

[sleek mason]

Imagine if it's trademarked

[exotic ibex]

☠️

[sleek mason]

that doesn't matter if the name is trademarked

AxeialOS is obviously not a taken name

so that's the reason for AxeialOS

what about ShitBox

totally not stolen name from Busybox

[exotic ibex]

hmm, what if i use GPT-5.2-Codex-Max-High or Claude 4.5 Opus-Max to think of a name 🔥

[bronze prawn]

5.2-codex eh

opus-max you go broke 🔥

I hope you enjoy your lighter wallet

[exotic ibex]

btw, got some news on some random guy making a PR to redis (the database) deleting 4000 lines of a float library in C++ and replaces it with vibecoded raw C impl within 350 lines written with claude and reviewed with gpt

[exotic ibex]

im already broke

using those make me broker ☠️

[bronze prawn]

how much do you like debt to our ai overlords

[exotic ibex]

wat

like how much i spend

if so, i spend none, since i just do docs using agents

and i can wait for it to renew

what is not ok is my token usage, though just input token

[sleek mason]

speedrunning how get

[exotic ibex]

Droid and opencode

You get 1m token free and 10m token if you add a cc

Got this for free

[sleek mason]

fire

[exotic ibex]

PS/2 best port ever

Just the fact that you need to restart to change keyboards

[sleek mason]

porting Play Station 2 🔥
-# \jk

[sleek mason]

the reason why error handling in kernel is so superior in my OS

[sleek mason]

we got 3 stars in a row TODAY

on the github

ttl: 15 stargazers

[tepid burrow]

huh what do you do different

it just looks like normal error handling

[sleek mason]

Its part of my second Error and Errno handling overhaul

[exotic ibex]

lmao, some intel chips does not suppor windows

[sleek mason]

[sleek mason]

Huh a Debug Exception, not intentional? huh?

the TF is set in RFLAGS? hmm...

also adding support to Syscall/Sysret instructions

as i used Int 0x80

But i'll support both of them

Int0x80 and Syscall/Sysret

[bronze prawn]

hm tf set?

that's strange

what's your IA32_SFMASK msr set to?

[sleek mason]

here is that setup:

        uint64_t Efer = ReadMsr(0xC0000080); // EFER MSR
        Efer |= (1 << 0);                    // Set SCE (System Call Enable)
        WriteMsr(0xC0000080, Efer);
        uint64_t Star = ((uint64_t)0x1B << 48) | ((uint64_t)0x08 << 32);
        WriteMsr(0xC0000081, Star);                               // STAR MSR
        WriteMsr(0xC0000082, (uint64_t)SysEntASMSys /*Syscall*/); // LSTAR MSR
        WriteMsr(0xC0000084, ~0x2);                               // SFMASK MSR

#DB happens on Sysret

which is weird

[bronze prawn]

hm

looks fine

[slim bane]

SFMASK is for syscall

for sysret you'll want to look into whatevers in rcx/r11 and see how that value is getting there

[sleek mason]

Hmm...

[sleek mason]

But i do handle them and don't touch them

[bronze prawn]

oh shit

yeah you're right

[slim bane]

well you put something into r11 before using sysret, what is it and where does it come from?

https://tenor.com/view/arnold-schwarzenegger-whos-your-daddy-gif-11193918

[sleek mason]

Well i dont?

The above one is the kernel's handler and the bottom is the userspace caller

Also i just don't know where the fuck CS = 0x2B coming from, ik its a valid usercode selector but i don't set it?

[slim bane]

you should look up how sysret works (especially wrt to populating CS and SS)

and right... so you dont set r11, so it could contain anything

yeah. R11 is caller preserved, those functions you call are free to trash it.

[sleek mason]

yes... wait a min, lemme try something

[bronze prawn]

oof yeah

the first time you implement sysret is when you learn sysret doesn't load selectors how you assume it would

:P

[sleek mason]

:p

I guess i fixed the DPF by saving the R11 in the kernel stub on the stack

and restoring during sysret

[sleek mason]

Wait i understood this, thanks for the help

[slim bane]

no problems 🙂

also note the rest of what sysret does with CS/SS, it uses fixed values for the hidden parts of the those registers. You'll want to ensure this matches your GDT otherwise you may into an issue on the next iret, or anywhere else that checks those values.

[bronze prawn]

not checking the values on sysret still pisses me off

[slim bane]

this is also one of those edge cases where intel and AMD behaviour differ, but I forget the specifics

check the manuals for that

[slim bane]

yeah, but it is what it is

[bronze prawn]

intel does cs |= 3 on sysret and (cs|ss) &= fffc on syscall (same for ss)

amd does not

[sleek mason]

YAY Syscall/Sysret work! now i support both!

[cinder otter]

If you don't wish to support compatibility mode, make sure to write a empty stub in CSTAR register

[bronze prawn]

do you need to write a stub tho?

if you don't support cmode then you will never get to cmode

meaning you will never get a cmode syscall

unless it allows you to do cmode syscalls from lmode

which is stupid

[cinder otter]

Well it's better to safe

I've learned to never trust the user

Some malicious program might trigger some compatibility mode syscall

[bronze prawn]

intel what?

wait what the fuck

wtf is the point??????

[cinder otter]

sysenter

[cinder otter]

Compatibility with AMD's x64 implementation?

[bronze prawn]

no

not to that

for fucking intel

yeah

amd does it correctly

strange that intel just didn't bother

[bronze prawn]

that's something completely different tho

sysenter and sysexit were made by intel

and don't work in lmode

the sysenter_esp and sysenter_eip were never updated to 64 bit

[cinder otter]

Mhmmm

[bronze prawn]

I have a feeling intel isn't telling the whole truth

"is not recognized in compat mode"

🤔

they always say something like "If Mode != 64-Bit.`

and omg apparently sysenter does work in 32 bit

bro what the fuck is intel on

RSP := IA32_SYSENTER_ESP;
RIP := IA32_SYSENTER_EIP;```

yeah okay intel

oh my days

intel extended them to 64 bit

😭

nonono let's not bother implementing the cmode versions of syscall/sysret

but let's extend sysenter & sysexit because we made them so we can feel good about ourselves

okay so tldr
intel doesn't allow 32 bit user apps to use syscall
amd doesn't allow 64 bit kernels to support sysenter

sooooo

[crimson raft]

https://tenor.com/view/were-sorry-tony-hayward-south-park-s14e11-coon2rise-of-captain-hindsight-gif-22864475

Intel rn ^^

[bronze prawn]

you wanna know the most brain dead fucked part about sysenter

IT DOESN'T SAVE A STACK OR THE RIP

[crimson raft]

It

What

It dosnt

Save

[bronze prawn]

nope

[crimson raft]

The RIP

[bronze prawn]

nope

[crimson raft]

What the fuck

[bronze prawn]

yep

[crimson raft]

Are you meant

To do

https://tenor.com/view/deltarune-tenna-mr-tenna-freaking-out-freak-out-gif-9208790555435211630

[bronze prawn]

or rflags

(linux/arch/x86/entry/entry_64_compat.S)

for flags you can just pushfq and be fine* (the if flag is cleared on entry <3)

[crimson raft]

Intel

WHY

[bronze prawn]

at least amd puts them in rcx and r11 for syscall

two caller saved regs

[crimson raft]

Intel making AMD look sane

[sleek mason]

what goin' on here

[bronze prawn]

linux seems to just say okay ebp is the stack

I have no fucking clue where it puts the return address tho

[crimson raft]

More shitting on Intel

Apologies

[sleek mason]

lemme join in

well, i guess my kernel is too massive

god

i need python script to configure and auto uncomment-comment #define

[crimson raft]

@bronze prawn I’m pretty sure Linux still uses interupts for 32bit syscalls no

[crimson raft]

Fuck you mean

Use ifdef

And check for arch or whatever

[bronze prawn]

you can't support 32 bit without supporting sysenter

you would just fault on the instruction in kernel mode

very smart!

[crimson raft]

Fuck

Why

Why Intel

What the fuck is wrong with you Intel

[bronze prawn]

on amd this is a non issue

sysenter doesn't work.

[sleek mason]

Not that, my kernel have configurations which are controlled via tons of macros and these statements

[crimson raft]

Genuinely what kind of crack were you smoking

And can I have some

Intel cmon

[sleek mason]

and i trun them on/off via uncommenting/commenting respectively

[crimson raft]

And you can disable syscall/stsret in EFER

[sleek mason]

their #define statements

[bronze prawn]

the wiki states

CPU registers

These must be set by the application, or the C library wrapper

    ECX: Ring 3 Stack pointer for SYSEXIT
    EDX: Ring 3 Return address

Operation

When SYSENTER is called, CS is set to the value in IA32_SYSENTER_CS. SS is set to IA32_SYSENTER_CS + 8. EIP is loaded from IA32_SYSENTER_EIP and ESP is loaded from IA32_SYSENTER_ESP. The CPU is now in ring 0, with EFLAGS.IF=0, EFLAGS.VM=0, EFLAGS.RF=0.

When SYSEXIT is called, CS is set to IA32_SYSENTER_CS+16. EIP is set to EDX. SS is set to IA32_SYSENTER_CS+24, and ESP is set to ECX. ```

but I'm not sure if this is true

we prob could move to #x86

or just somewhere else

[crimson raft]

I don’t even wana argue about this tbh

It’s a dead horse

The instructions suck dookie shit

[bronze prawn]

you can't disable sysenter tho?

oh

I know what you mean

[bronze prawn]

oh and not even talking about the funny intel sysret silicon bug

(if a non cannonical address is in rcx when executing sysret, a gpf will be taken in cpl0 with a cpl3 stack)

[crimson raft]

😭

How do you prevent that one

[bronze prawn]

just check it's cannon before sysretting

do intels job for them

[crimson raft]

[sleek mason]

i'll push teh changes soon after fixing some... stuff

-# i broke the driver manager database

[raven lotus]

So, if i got it right from the SDM

SYSENTER does indeed not save anything, and SYSEXIT uses ECX and EDX for SP and IP respectively

[bronze prawn]

I have no fucking idea I'm not gonna lie

[raven lotus]

so you as an user program must give SYSENTER a return address and stack, linux uses ECX and EDX

But you could just make it EAX and ESI if you wanted

[sleek mason]

yay a kernel config file

[sleek mason]

yay a python script

[exotic ibex]

Oh cool

[marble epoch]

why not just kconfig

also these config options are way too granular

[exotic ibex]

@sleek mason told ya

[sleek mason]

that's the goal

give as much granular control i can

[raven lotus]

i mean these seem mostly to be for development builds

[marble epoch]

yikes

[raven lotus]

like if they have a problem in x thing they can just enable logging in it

[sleek mason]

that's way easier to trace than having everything enabled at once

[sleek mason]

hmm...

driver manager freaking out

[sleek mason]

well nvm fixed it (problem was a inverted conditional)

should i push the changes?

i'm delaying this shit too much 😭

[marble epoch]

why

you can just log everything and filter for relevant logs

i generally don't understand the point of hiding logs

you want a lot printed on the console

[sleek mason]

They can affect performance, so hidden the better

[marble epoch]

how?

are you logging in a hot loop?

rather, why not just have a loglevel command line option

then you can limit the logs to debug/info/etc

[crimson raft]

Logging every page your VMM maps and every page your PMM gives

[marble epoch]

💀 what's the point

[crimson raft]

I had a bug with my VMM and I had to find where it was fucking up

[sleek mason]

my kernel prints like HUMANGOUS amount of log

[crimson raft]

So I had to sit there and wait a good 20 minutes for it to HHDM

💀

[marble epoch]

but you remove those after

printf debugging

[sleek mason]

it takes too much time to filter out

[marble epoch]

you don't leave the prints in

[crimson raft]

True

[crimson raft]

Log to an in memory buffer and then have a dmesg tool and grep it

[marble epoch]

how

ctrl + f

💀

you already print the lines of where stuff is happening

[sleek mason]

i'll do that

[sleek mason]

lessgooo , 16 stargazers on our repo

https://github.com/VOXIDEVOSTRO/AxeialOS

And ONCE AGAIN something broke this time the PosixFds

[exotic ibex]

lmao

[cinder otter]

And the cycle continues...

[sleek mason]

just because of bad logic or inverted logic...

[sleek mason]

i'm just hoping this is the last thing i have to fix...

after the fix i'll push the changes because i've delayed too much

[exotic ibex]

I just pushed this morning

[sleek mason]

Very nice

[exotic ibex]

The delay is like a few days

Same as yours, i have a staging branch

[sleek mason]

i've delayed a week

[exotic ibex]

And the main branch is just a mirror or the latest release

[sleek mason]

Yes it's nice

[exotic ibex]

So we donf pollute the main while having the sources online

[sleek mason]

i guess after the push i'll merge the pre-main branch to main

because it's too outdated now

[exotic ibex]

I just do it when an update is worth a release

[sleek mason]

yes, that works

[exotic ibex]

Current diff state

[sleek mason]

but yeah the changes ARE ALOT in my kernel

[sleek mason]

33K additions

[exotic ibex]

Agreed

[exotic ibex]

Right

[sleek mason]

you've seen it in my live stream from yesterday

[exotic ibex]

Still wonder why so much

[sleek mason]

ig you lost track off it, relatable

[exotic ibex]

Right

[sleek mason]

ig there will be more deletions than addition in my push diff 😭

because cleaned up lot of clutter

[exotic ibex]

Oh wow last relase is already 3 weeks

And 2 more weeks is lunar new year

[sleek mason]

it was suppose to be finished in the first week...

[exotic ibex]

I got 2 week off

[exotic ibex]

Yes

Then the second

And third

And im still not planning r0c2 till next weekend

[sleek mason]

my commits are inconsistent

[sleek mason]

stablizing?

[exotic ibex]

Pretty much the same

[sleek mason]

this maybe my most delayed commit

[exotic ibex]

Id say that im trying to equalize the work put into subsystems

In a fair way so they can grow gradually

[sleek mason]

3 weeks ago was the last major commit

[exotic ibex]

You planning any release, versioning model yet?

Currently uses this https://github.com/assembler-0/AeroSync/blob/dev/include/aerosync/version.h.in

[sleek mason]

nope

still considering

well i will have the first release if i at least ported bash shell and terminal

[sleek mason]

[exotic ibex]

peak

[sleek mason]

again... fucking again... something broke , user stack is misaligned during schduling and user stack's argv and envp points to NOTHING

[sleek mason]

gotta fix the argc, envp and argc...

[sleek mason]

ig i fucked up the crt0.asm

lemme fix

[exotic ibex]

got to userspace already??

veri cool

[sleek mason]

perfecting some stuff

[sleek mason]

almost fixed the argv, envp and auxv

was tiny problems in the SetStack function

well... the push order was wrong... lmao

LESSSGOOOO

[crimson raft]

Yoooo 🗣️ 🔥

🗣️ 🔥

KERNEL=1

[sleek mason]

magnificent test string

[crimson raft]

I don’t even do argc or argv or envp or auxv

[sleek mason]

i do

[crimson raft]

When I fix my userspace app spawning

I will add it

[sleek mason]

i guess that's enough to port a decent bash

[crimson raft]

Do you uhh

Have fork

And exec

[sleek mason]

I do.

[sleek mason]

Yes

well fork is pretty shit

[crimson raft]

How do you have fork before userspace :o

I’m putting fork off for now

[sleek mason]

because my main focus is on kernel

[crimson raft]

And I’m gonna copy on write

[crimson raft]

[sleek mason]

i don't use COW

[crimson raft]

🥀

[sleek mason]

[crimson raft]

You should use copy on write

It’s not that hard

[sleek mason]

well as real programmer, i marked it as

/*TODO: Add COW*/

[crimson raft]

It’s just a reference count and making pages read only if it’s > 1 ref count

🥀

And then handling the fault by allocating one 4kb page and copying the data and changing the PTE

[sleek mason]

well i guess i would overhaul my fork and execve

[crimson raft]

https://tenor.com/view/thesimpsons-bart-simpson-dont-have-a-cow-don't-have-a-cow-dont-have-a-cow-man-gif-269547880223598654

Something like that

Plus CoW also lets you do on demand allocations later as it’s a similar framework

With allocating things and copying/zeroing on the fly

[sleek mason]

i guess i must invest my time to add CoW

should i make a insult randomizor for my exceptions?

like different insults to driver devs and me (kernel) with function called StrTroll

[sleek mason]

fuck, i can't create new processes via execve if in userspace

fork is breaking stuff

as well as execve is fucked

AND ONCE AGAIN! BROKE SOMETHING AS ALWAYS...

🥀 💔 😭

why am i speedrunning in breaking stuff

well on the brightside atleast it would make my kernel "safer" and "usable"

without crashing now and then

[cinder otter]

Never celebrate too early - A lesson I learned on chess.com

[marble epoch]

bruh

[sleek mason]

ig i'll just overhaul the process stuff

and especially add CoW

[crimson raft]

I personally do pop culture references to stuff I like for the most part and a few other random things

[sleek mason]

hmm... i just noticed that the BSP Cpu's APIC timer never interrupts...

interesting...

well nevermind

i've fixed it

problem was the BSP conditional which keeped the APIC masked lol

[sleek mason]

yay execve works (as well as segmentation fault).

[exotic ibex]

cool!

whats your philosophy?

*for contributors

im cooking up my guidelines

currently looks sth like this

looking for inspiration

[sleek mason]

i don't?

[exotic ibex]

thought you like optimized stuff

thats a philosophy

[sleek mason]

well i dont have a contributor.md nor a "active working" contributor

[exotic ibex]

if you are not too strict

[exotic ibex]

well same here

but im doing it in advance

though for sure no ones gonna care

[sleek mason]

fr

well i didn't made one because obviously i thought this project won't get popular so i just made my mind to do it later

[exotic ibex]

i see

i probably have to do this carefully

since stuff like definition issues

etc..

people can exploit those

like this example

[sleek mason]

well ig i'm fine to be solo, like if any one wants to contribute their changes just fork and open PR for consideration

[exotic ibex]

thats 1990 workflow

[sleek mason]

i'll tell them the philosophy later when they had already opened a PR

like changes and etc

[exotic ibex]

bruh

[exotic ibex]

very inefficient then

though have a look at this

[sleek mason]

well i'm not expecting any active working contributor anytime soon

[sleek mason]

mellos

[bronze prawn]

🥀

but I wanna use it for killer drones :(

[exotic ibex]

the issue is, you are threatening]

and that is the worst possible thing to do

in a legal document

plus, it have unclear claims

[sleek mason]

ig it's nice, atleast it's being clear

[exotic ibex]

"you're free to do whatever you like with it (free as in freedom!)."

and "However, there are a bunch of things we would prefer you avoid to do with this Software"

[bronze prawn]

"do whatever. also here is a list of shit we don't want you to do with it"

like

dawg

[exotic ibex]

literally conflicting words

[bronze prawn]

I mean

not really tho

you can do whatever you want

but we would rather you not do these things

[exotic ibex]

company wants (very) clear intentions

[bronze prawn]

doesn't conflict

at all

legally you are free to do whatever

[exotic ibex]

then you shoudnt say that in the first place

that you can do whatever you like

[bronze prawn]

we ask you not to do this

legally > personal asks

but yeah

I agree

[exotic ibex]

it should be rephrased

[bronze prawn]

wdym?

it perfectly fine

"You can do whatever. but we would like you not to do this"

[sleek mason]

well i don't see anything wrong with it.

[bronze prawn]

they can't enforce those legally it's just suggestions on what not to do

ig

[exotic ibex]

do whatever you like is actually very strong

[bronze prawn]

well

it's not whatever

[exotic ibex]

in friendly terms it is fine

[bronze prawn]

it's whatever gpl3 allows

[exotic ibex]

but for legal documents its not that straightfoward

[exotic ibex]

ik

[sleek mason]

what's more straight forward is to attach this statement in the readme "read the LICENSE.md for legal details"

[exotic ibex]

yeah

what they just do is add another layer of implicit declaration

[bronze prawn]

yeah

[exotic ibex]

wait

[bronze prawn]

I agree with that

while not legally binding nor legally enforceable it does make it less clear to a normal person

[exotic ibex]

it is

[bronze prawn]

"prefer you avoid doing" is not legally enforceable tho

[sleek mason]

ig they are just joking around,
-# also the fact it's not a commercial OS for commercial use

[bronze prawn]

it's not written in the licence

and therefor they cannot stop me

[exotic ibex]

idk, internally, they discuss it very seriously

i was there the whole time

[sleek mason]

well obviously, we follow license, not readme declarations for uses

[exotic ibex]

they also mentioned about not for military uses

aka blowing people up

[exotic ibex]

cant agree more

[neat sandal]

"memory violation" sounds like someone breaking/snapping RAM while inserting it 😭
but yay segfault

[exotic ibex]

bro got a point there

[cinder otter]

Must have a heartbeat

And preferably not be a bot with human front

[exotic ibex]

wdym?

[cinder otter]

Tho in all seriousness, Include writing tests as a requirement

[sleek mason]

les fancier way of saying "Segmentation fault"

[neat sandal]

yea ik

[raven lotus]

Is this from LinkedIn 😭?

[sleek mason]

https://tenor.com/view/miau-gif-11651142449495197453

[raven lotus]

guau

[sleek mason]

gnau

[cinder otter]

yup

[sleek mason]

my kernel love segmentation fault

also AxeOS is the smaller nickname for AxeialOS

like axeos

[gray crescent]

and axe body spray is the official axeialOS body spray made for real men

[sleek mason]

lol

17 stars on my repo

insanity

well still fixing my execve and fork

execve has problem where i successfully load the elf and create the process, but returning to the reused thread causes this segmentation fault

[sleek mason]

^which is this^

fork has the problem where the forked child is executed but just never proceeds to run at all

does my kernel have lot of features?

lemme list em all

[slim bane]

Wow that's a short list

[sleek mason]

List of features my shitty kernel has:

• VMM
• PMM
• Kernel Heap (Slab free list allocator)
• ModMemMgr (manages memory for kernel modules) well partially deprecated*
• Kernel module linker
• Driver manager database.
• Device manager
• Plug-n-Play (well unsure if it works well, untested)
• MLFQ Scheduler
• SMP
• Unix-Like VFS
• POSIX processes
• Threads
• Procfs
• Devfs
• POSIX Signals
• POSIX FDs
• POSIX Syscalls
• Syscall instruction
• Int 0x80 Syscalls (legacy is supported)
• IDT
• GDT
• TSS
• CoW for fork
• Timer support for APIC
• Timer support for HPET (well its TODO for ignore this in list)
• Timer support for PIT
• kernel config
• Testing stuff
• PCI Bus system
• RAMFS
• Initrd/Initramfs (my BootImg)
• Firmware blob management stuff (request firmware and etc stuff)
• a EarlyBootConsole (WHICH STARTS WAY BEFORE EVEN GDT/IDT/TSS, every first thing to start in the kernel)
• UART (serial port debugging)
• kernel symbol exports (KExports for drivers and kernel api)
• Mutexes
• Spinlocks
• Semaphores
• BusyWait (Blocking and Non-Blocking)
• another massive debugging infrastructure (my printf loggers and per-file loggers).
• handle user faults
• module record manager (handles all loaded module because driver manager is dependent on this as its older ancestor)
• block devices(devfs)
• char devices(devfs)
• parses memory map ofc
• massive and overcomplicated error handling structure and logging (Errnos)

insane wall of text

[exotic ibex]

EarlyBootConsole getting a shoutout

the first thing that starts in my kernel is the panic handler

the prinkt (s1) the when modules are loaded, s2 kicks in

[sleek mason]

do you think my kernel has not enough features, from the list?

[exotic ibex]

ig its pretty featureful

'enough' vastly depends

[sleek mason]

its a GP kernel so yeah

https://tenor.com/view/krnl-roblox-exploit-exploit-robux-kernel-gif-19597348

[exotic ibex]

then nothing is enough (imo)

a GP os has to be up-to-date with whatever the world is like

[sleek mason]

but yeah that's way more than enough to handle some sophisticated applications

[exotic ibex]

hmm, idk what to focus on now

[sleek mason]

your kernel is also a GP right?

[exotic ibex]

drivers stack, an fs, perfecting my sched

[exotic ibex]

yeah

[sleek mason]

i guess have a proper driver and device abstraction

like DAM in my kernel

[exotic ibex]

thats what i have been making

with UDM

but yeah

[sleek mason]

nice

[exotic ibex]

good point

[sleek mason]

also make it good enough to work as PnP

[exotic ibex]

PnP support is hard

because of the diversity

of USB is infinite

[sleek mason]

indeed

[sleek mason]

well my kernel maybe can handle PnP

still i have only a PCI bus

[exotic ibex]

very cool then

[exotic ibex]

thats enough for now

people say pci is dead

in reality, it is still there powering everything

AHCI, xHCI controllers are on pci bus, PCIe still needs PCI, and NVMe itself is just PCIe

[sleek mason]

because of my probemgr and device abstraction makes it easy to load specfic drivers when needed at run time and because hardware events exist

[exotic ibex]

im thinking of adding real modules some where in the future

i will make a distro using my kernel

[sleek mason]

and my kernel has a function called CheckForHardware which calls the probemgr to trigger lots of probes and check for any hardware changes like a new driver and load the driver and put into the device tree

[exotic ibex]

maybe in the late 2030s ngl

[exotic ibex]

i see

[sleek mason]

lol

so indeed my kernel has PnP in it's minimalist form

[sleek mason]

nah i think the name "AxeialOS" is fine

I'm too used to it now

and changing it is a bad idea

[gray crescent]

the only good instance of renaming a project applies to windows, they should rename it to "if EA made an operating system (episode 1)"

[sleek mason]

I gonna just push

too much delays

less goooohttps://github.com/VOXIDEVOSTRO/AxeialOS

[sleek mason]

@exotic ibex i've pushed the changes after a long time

[exotic ibex]

lets see whatve you been cooking

what?

why bad? invalid is the word

the patch is filled with atomics

where is the advertised configurator?

mostly a logging and return values consistency

[sleek mason]

indeed

[sleek mason]

it's not yet complete

[sleek mason]

there is the header file in ./Kernel/KrnlLibs/Includes/__AXEKCONF__.c

[sleek mason]

lots of fixes and features too, some SMP fixes, POSIX Fds fixes, Syscall Instruction, Fork with CoW, Timer has a BSP and AP Conditional, Error logs everywhere with tracebacks and errno codes, atomics, fixed the userstack argc, argv, envp and auxv

and other fixes i lost track off

[sleek mason]

https://tenor.com/view/8086-assembly-x86-x86-assembly-8086-assembly-gif-8095595086549488321

[exotic ibex]

no i mean the gui python you showcased

[exotic ibex]

yeah

[cinder otter]

GUI idea: Use CSS and Javascript to design the UI

(You need to technically write a web page renderer in userspace)

Maybe a js engine too

[exotic ibex]

which ui you are talking abt>

[cinder otter]

Userspace ui

Not the configurator

[exotic ibex]

web-based os?

[cinder otter]

Use CMAKE for that

[exotic ibex]

ccmake?

[cinder otter]

Yeah that

[exotic ibex]

dude, thats not a depedency engine

and it sucks

for kernel configuration

[cinder otter]

Nah, under the hood. It's your typical os, just the gui is designed using CSS and js

[exotic ibex]

here dependant is a thing

[exotic ibex]

interesting

[cinder otter]

Maybe

I haven't really explored granular cmake configuration with my kernel yet

[cinder otter]

Pro: GUI changes are easy
Cons: Typical web browsers contain like a few million lines of code (assume that ~30% of that is the renderer)

[crimson raft]

Windows does exactly this (well react native) for the start menu and it’s slow as shit

[cinder otter]

It does??

[crimson raft]

Yes

Lmfao

My friends even saw some of the OOBE was using html and CSS for the animations

This would be tearing a page right out of windows

[cinder otter]

Tbh, I got this idea after configuring gnome

[crimson raft]

Well you see I really hate gnome

But native UI stuff will be alot better in the end

Opening an electron app vs a native app for example

[exotic ibex]

number 1 reason why i delete rhel 10 after 2 hours (beside its nerfed dnf)

[sleek mason]

my kernel isn't stable to fricking support lol 😭 i dont even have my own and very first driver either

well my kernel can support that if i'm a miniac to port that

assuming i fixed everything in my kernel

i would still have to make some real sophisticated drivers to support that or even internet at all

well back to fixing my Processes (fork and execve)

i've been procrastinating alot

[exotic ibex]

do we use gnu extensions here @sleek mason ?

like gnu11, gnu23

[sleek mason]

no?

i dont think so

[exotic ibex]

you explicitly set -std?

if so then yeah, no extension

[sleek mason]

i'm on vacation guys

for the next 16 days

so this will be some standstill

tho i'm getting my GCC work on my arch linux laptop lol

[exotic ibex]

SIXTEEN

[sleek mason]

nah i will be active on github

and try a workaround to work on it

also my laptop battery lasts way longer on linux

for some reason

[neat sandal]

new funny number

[neat sandal]

hope you have a great next 2 weeks <3

[sleek mason]

developing AxeialOS on arch linux! for the entire span on vacation 🔥

[gray crescent]

windows does too much crap it basically halves the battery

[sleek mason]

this projects is not ded

[exotic ibex]

we will see

[cinder otter]

He hasn't posted in 5 hours

Looks ded to me

[neat sandal]

5 months is when it becomes ded

[cinder otter]

That's super-ded dude

https://tenor.com/view/death-familyguy-gif-5935081

[neat sandal]

yea but messages not being sent for 5 hours is just the time a channel is quiet pretty often
algorithms channel is a good example of what ded truly is

[cinder otter]

I should've added \s

[neat sandal]

yea but /s not \s (doesn't matter tho :P)

[cinder otter]

Nah, that's just super-sarcasm

[sleek mason]

reminding this PROJECT IS NOT DED

[sleek mason]

THIS PROJECT IS NOT DEAD, i will make this into distro for sure if #1467786251510743061 goes nicely

[graceful terrace]

just do what I do, take breaks of years in length and keep coming back and adding more to the os. people stopped assuming mine was dead the 3rd time I came back lol

[sleek mason]

well i'm on vacation, my main PC where i put my AxeialOS workspace is at home, i'm just with my arch linux laptop

[neat sandal]

that explains the inactivity
just the single word "vacation"

take a break o7

[sleek mason]

i will make a AxeialOS distro but from the Modularus kernel #1467786251510743061

[exotic ibex]

great idea

[sleek mason]

Liquid OS is on the go!

#1467786251510743061

[exotic ibex]

linux direct competitor

very ded thread

[foggy swallow]

we should clearly go back to the weird k&r c function definitions they were so good!!!

[cinder otter]

Except defining function arguments outside the parenthesis

[foggy swallow]

what why would you exclude the best part

[exotic ibex]

DED @sleek mason

[sleek mason]

The code is gonna be repurposed for the Liquix

[exotic ibex]

ig so

[sleek mason]

So AxeialOS is not ded, just that it will not use the axkrnl

but the Modularus

[exotic ibex]

this didnt last long

[exotic ibex]

your disappearance is still a mystery

[graceful terrace]

hmm when did I post here

ah months ago

[graceful terrace]

considering they're now a deleted user, their return is unlikely

[exotic ibex]

though just a bit sad, hes my frist friend in the osdev journey ngl

we cant do anything so yeah 👍

its just memories

[graceful terrace]

lost a good contributor to one of my projects that way, one day just gone

but that's just the way it goes