Nyaux

[edgy pilot]

iopb btw

[silver yarrow]

yeah

mb

like if u set offset to zero, arent u saying that the 104 bytes of tss are actually iopb

[silver yarrow]

yeah thats what I do

basically saying there is zero bytes in iopb

because the limit is 104

[surreal path]

okay but

i dont get how this relates to all the UB

going on

[silver yarrow]

it probably doesnt but im intrigued

[surreal path]

FAIR LOL

whoosp caps lock

[edgy pilot]

a valid tss wouldn't be a cause for a bug

[surreal path]

i know

but adding it causes this

🤷‍♂️

[edgy pilot]

then something else is the problem

[surreal path]

most likely

[silver yarrow]

iobp base == 0 would be a bit of a security issue no? because u are prolly allowing some portio in userspace by accident

[surreal path]

its fine

[edgy pilot]

yeah that's what I thought too
I just didn't touch it at all before

[silver yarrow]

fair

ok ill get back to debugging @surreal path

[surreal path]

ill try to also debug as well

thanks for the help btw

[silver yarrow]

wait

where do u even load the tss

[edgy pilot]

https://tenor.com/view/programmer-developer-debugging-code-testing-fixing-gif-7663766

[surreal path]

gdte.asm

[silver yarrow]

ah

thats... undocumented

lol

[surreal path]

lol

[flat nymph]

uint16_t sizeoftss = sizeof(struct TSS) - 1;?

[surreal path]

i mean

it fits in a uint16

[silver yarrow]

lol

[surreal path]

also - 1

yea

[silver yarrow]

its the limit

[surreal path]

so what lol

[surreal path]

yea

[silver yarrow]

whats the last commit that works

[surreal path]

uhhhh that one

[silver yarrow]

lmao aight

[flat nymph]

I give it sizeof(TSS)

Without -1

[silver yarrow]

dont do that

oh wait

no I do that

now I gotta figure out whos right

sec

[flat nymph]

What does SDM say on the matter

sizeof(TSS) -1 doesn't crash my kernel

But I'm setting granularity bit

Yeah it's working

uint64_t constructedlow = ((uint64_t)base_low << 16) | ((uint64_t)base_mid << 32) | ((uint64_t)base_midhi << 56) | ((uint64_t)access_byte << 40) | sizeoftss;

Base mid

why << 32

Ok it's right nvm

[surreal path]

because

[flat nymph]

Somehow everything looks right to me

[surreal path]

lmao

[silver yarrow]

I am still trying to figure out whether the limit is -1

lol

I assume it isnt

[flat nymph]

Have you tried aligning struct TSS?

To 4

Not sure it will change anything

since it's after lapic_id

[surreal path]

i dont think it will

tbf

[flat nymph]

Now

When does it crash

In init_gdt()?

[surreal path]

i told u its random each boot

[flat nymph]

Are you sure it's because of TSS?

[surreal path]

i dont know

genuinely

[flat nymph]

How do I build this thing

[silver yarrow]

do u have nix

[flat nymph]

error: 'cc-runtime' is not a Git repository

[flat nymph]

No

[surreal path]

remember

[silver yarrow]

L

this is consitently the fail

with one cpu

[surreal path]

this is

cd into cc-runtime

mkdir .git

[silver yarrow]

I am pretty convinced this has absolutely nothing to do with the tss

[surreal path]

yea same

[surreal path]

huh interesting

[silver yarrow]

OK @surreal path WTF

[surreal path]

WHATS GOING ON

[flat nymph]

0xd

[silver yarrow]

what in the fuck is this

[flat nymph]

That's what I told him to change

[kind root]

[surreal path]

its unfinsihed

[finite summit]

What in the actual fuck is this

[finite summit]

[surreal path]

is this the problem wux?

[flat nymph]

0: v=0d e=0048 i=0 cpl=0 IP=0028:ffffffff80087878 pc=ffffffff80087878 SP=0030:ffff800007b0efb8 env->regs[R_EAX]=0000000000000048
RAX=0000000000000048 RBX=0000000000000000 RCX=ffffffff800efc60 RDX=01000099802c0000
RSI=0000000000000000 RDI=ffffffff800efc00 RBP=ffff800007b0efe0 RSP=ffff800007b0efb8
R8 =0000000000000000 R9 =0000000000000000 R10=0000000000000000 R11=0000000000000000
R12=0000000000000000 R13=0000000000000000 R14=0000000000000000 R15=0000000000000000
RIP=ffffffff80087878 RFL=00000006 [-----P-] CPL=0 II=0 A20=1 SMM=0 HLT=0
ES =0030 0000000000000000 00000000 00009300 DPL=0 DS   [-WA]
CS =0028 0000000000000000 00000000 00209b00 DPL=0 CS64 [-RA]
SS =0030 0000000000000000 00000000 00009300 DPL=0 DS   [-WA]
DS =0030 0000000000000000 00000000 00009300 DPL=0 DS   [-WA]
FS =0030 0000000000000000 00000000 00009300 DPL=0 DS   [-WA]
GS =0030 ffff800001998018 00000000 00009300 DPL=0 DS   [-WA]
LDT=0000 0000000000000000 00000000 00008200 DPL=0 LDT
TR =0000 0000000000000000 0000ffff 00008b00 DPL=0 TSS64-busy
GDT=     ffffffff800efc10 00000057
IDT=     0000000000000000 00000000
CR0=80010011 CR2=0000000000000000 CR3=000000000005a000 CR4=00000020
DR0=0000000000000000 DR1=0000000000000000 DR2=0000000000000000 DR3=0000000000000000
DR6=00000000ffff0ff0 DR7=0000000000000400
CCS=00000001ffff0000 CCD=00000000ffff8000 CCO=SARQ
EFER=0000000000000d00
check_exception old: 0xd new 0xd
     1: v=08 e=0000 i=0 cpl=0 IP=0028:ffffffff80087878 pc=ffffffff80087878 SP=0030:ffff800007b0efb8 env->regs[R_EAX]=0000000000000048
RAX=0000000000000048 RBX=0000000000000000 RCX=ffffffff800efc60 RDX=01000099802c0000
RSI=0000000000000000 RDI=ffffffff800efc00 RBP=ffff800007b0efe0 RSP=ffff800007b0efb8
R8 =0000000000000000 R9 =0000000000000000 R10=0000000000000000 R11=0000000000000000
R12=0000000000000000 R13=0000000000000000 R14=0000000000000000 R15=0000000000000000
RIP=ffffffff80087878 RFL=00000006 [-----P-] CPL=0 II=0 A20=1 SMM=0 HLT=0
ES =0030 0000000000000000 00000000 00009300 DPL=0 DS   [-WA]
CS =0028 0000000000000000 00000000 00209b00 DPL=0 CS64 [-RA]
SS =0030 0000000000000000 00000000 00009300 DPL=0 DS   [-WA]
DS =0030 0000000000000000 00000000 00009300 DPL=0 DS   [-WA]
FS =0030 0000000000000000 00000000 00009300 DPL=0 DS   [-WA]
GS =0030 ffff800001998018 00000000 00009300 DPL=0 DS   [-WA]
LDT=0000 0000000000000000 00000000 00008200 DPL=0 LDT
TR =0000 0000000000000000 0000ffff 00008b00 DPL=0 TSS64-busy
GDT=     ffffffff800efc10 00000057
IDT=     0000000000000000 00000000
CR0=80010011 CR2=0000000000000000 CR3=000000000005a000 CR4=00000020
DR0=0000000000000000 DR1=0000000000000000 DR2=0000000000000000 DR3=0000000000000000
DR6=00000000ffff0ff0 DR7=0000000000000400
CCS=00000001ffff0000 CCD=00000000ffff8000 CCO=SARQ
EFER=0000000000000d00
check_exception old: 0x8 new 0xd

[kind root]

Most stable part of nyaux

[surreal path]

lmao

[silver yarrow]

surprise surprise when I commented out the unfinished part it works

[surreal path]

does removing that line solve anything

insane

[silver yarrow]

that and this

[surreal path]

i still need to load the kernel registers into a kstack

and then load the userspace registers

[flat nymph]

It's crashing in ltrw still

[surreal path]

oh

[flat nymph]

In here

[surreal path]

💥

*

[flat nymph]

No

We are fixing TSS

sorry not sorry

[silver yarrow]

oh wait

I did another fix

let me see if I remoev that

[surreal path]

what was it

[silver yarrow]

hm

no it still works for me

[surreal path]

huh

weird

[silver yarrow]

@flat nymph

does it work if u just run one core

[flat nymph]

Btw this

TR =0000 0000000000000000 0000ffff 00008b00 DPL=0 TSS64-busy
seems half-sus

[flat nymph]

It's crashing in late-init

When he's loading other cores

I guess

[silver yarrow]

lmao

alright lemme put in a couple more cores

ah yeap

[flat nymph]

How do I change how many cores it has

[surreal path]

[silver yarrow]

makefile

the top level one

[flat nymph]

Oh

SMP 20

[silver yarrow]

yeap

very debug friendly

[flat nymph]

It page faults

[surreal path]

fun

[silver yarrow]

a lot

[surreal path]

more UB

[flat nymph]

In memcpy

[silver yarrow]

per_cpu_data is NULL

so thats

eh

[surreal path]

it shouldnt be

[silver yarrow]

OH REALLY?

[surreal path]

yea

[finite summit]

Lmfao

[flat nymph]

In create kentry

[finite summit]

Stop blaming it on UB

[silver yarrow]

well ubsan is screaming

[finite summit]

You've no one but yourself to blame

[silver yarrow]

so he's not all that wrong

though he definitely created the ub

[finite summit]

Just come work on obos with me

[surreal path]

i mean yeah i created the UB

[kind root]

Come unfuck the obos ec driver

[surreal path]

im not giving up on my kernel

[surreal path]

exact;y

[silver yarrow]

on the BSP

[finite summit]

When uUSB

[silver yarrow]

do u even create the local data

[finite summit]

uEC

[silver yarrow]

I cant pinpoint where u do

[surreal path]

its statically allocated

[kind root]

Already wip by other people

[surreal path]

check sched.c

[finite summit]

Whom

[kind root]

Nooby

[silver yarrow]

no the fuck its not

where

[surreal path]

yes the fuck it is

[silver yarrow]

lol

where

[surreal path]

look

[silver yarrow]

look WHERE

[surreal path]

sched/sched.c

[finite summit]

That looks like per cpu data to me

[silver yarrow]

oh wtf

jesus

[surreal path]

wdym

@flat nymph told me to do it

for the bsdp

bsp*

[silver yarrow]

no its just in between a bunch of funcs

so where exactly do you write it into gs

[surreal path]

hold on

for creating it

for other cpus

that arent the bsdp

bsp*

[silver yarrow]

yes ik

[finite summit]

Whatever happened to normal variable names...

[silver yarrow]

there u write that into the msr

where do u write the msr for the bsp

[surreal path]

i dont

[silver yarrow]

oh and we have arrived at the fucking ub

[surreal path]

have we????

[silver yarrow]

sec

[finite summit]

Nah

He returns &bsp if he's on the bsp

In get cpu local

[silver yarrow]

oh jesus

[flat nymph]

He does

[silver yarrow]

thats so cooked

[silver yarrow]

he doesnt

but its fine

[finite summit]

Until you access gs relative data on the bsp in assembly

[flat nymph]

In arch_create_per_cpu_data()

[silver yarrow]

doesnt get called for the bsp

[flat nymph]

It does

[silver yarrow]

doesnt

[surreal path]

yea ill fix it

but wheres the ub wux

u said u arrived

[flat nymph]

arch_late_init()

[silver yarrow]

that doesnt get called for the bsp

[surreal path]

check limine spec @flat nymph

[flat nymph]

It does in bootstrap()

[flat nymph]

What for

[surreal path]

->goto_address

its ignored for the bsp

[flat nymph]

Oh bsp

But still

It's never set

[finite summit]

Nyaux challenge: Don't have people get confused as shit when they read it's code, level impossible

[silver yarrow]

real

[finite summit]

Tbf reading obos' code can give brain cancer

At one point I started adding "Abandon all hope, ye who enter here" to the beginning of some files

[flat nymph]

void *him = get_some_shit()
void *that = him + 0x78;
x86_arch_specific_random_functionwithsomeungodlylongname(that - 1);

[silver yarrow]

speaking of

[finite summit]

lmao

[flat nymph]

Yes

[surreal path]

lol

[finite summit]

obos code is bad

but not THAT bad\

[surreal path]

but wux wheres the UBu said u found it?

??

[silver yarrow]

no I thought I did

[surreal path]

ohhh

okay

[silver yarrow]

apparently u do some cursed shit with the bsp local

not sure why it wouldnt work like every other cpu local but it is Nyaux so

[surreal path]

fair

[flat nymph]

It casually pagefaults in if (cpu->run_queue == NULL && cpu->start_of_queue == NULL)

[silver yarrow]

yes ubsan catches it

[flat nymph]

What if I actually set msr in BSP?

[flat nymph]

asserts everywhere really help with that btw

[surreal path]

it shouldnt be null tho]

that makes no sense

at all

[silver yarrow]

yes

[surreal path]

yes u agree with me

[silver yarrow]

sure do

[flat nymph]

Could rdmsr be casually broken?

It failed on all 20 cores

[silver yarrow]

LMAO

[flat nymph]

Although that makes no sense

[surreal path]

https://tenor.com/view/tom-meme-tom-and-jerry-gunshot-explode-gif-17823053

nyaux code

[flat nymph]

Or wrmsr

Because wtf

[silver yarrow]

nah they're fine

checked em

[flat nymph]

Because idk

He writes msr

I've printfed it

Then reads 0 back from it

Like how

[surreal path]

THAT SHOULD NOT HAPPEN

[flat nymph]

Your swapgs is broken

lemmy try something

[surreal path]

how

how are u so sure

[flat nymph]

je

I need bochs

or gdb

Nope

what the f

[surreal path]

on the foruth day of christmas nyaux gave to me

5 billion bugs

and the sixth day of christmas nyaux gave to me

[flat nymph]

Give up

[surreal path]

an extra memory bug under the tree

[surreal path]

are u telling me to give up

wtf

that is not nice

tf

[flat nymph]

I don't get the pattern

[surreal path]

what pattern

[flat nymph]

What is going on with your kernel

Some cores have cpu local data

Some don't

[surreal path]

ill solve it

but first

we need to solve these bugs

@silver yarrow i see everyone is struggling with the cursedness of nyaux lol

[surreal path]

oh

that is weird

that shouldnt happen

tf

[surreal path]

yea i think there may be severe memory corruption

cause otherwise none of this would happen

it has to be memory corruption or something

[finite summit]

obos only has minor memory corruption, come work on it instead

[surreal path]

some bug in the slab or vmm

[finite summit]

[surreal path]

please stop trying to get me to give up on my kernel

it will never happen

[finite summit]

It is joke

[surreal path]

ik

[finite summit]

Chill

[surreal path]

[flat nymph]

how does this even happen

[surreal path]

i know

u chill

[finite summit]

That knowledge is only with god

[kind root]

I just shilled for obos in lounge as well

Still waiting for suspend article on osdev wiki

[surreal path]

may allah help me figure out these bugs

i hope allah will help

[finite summit]

Yeah only prayer can help you now

[surreal path]

pretty much 😭

[kind root]

You know what else can help

[surreal path]

what

[kind root]

Working on obos instead

[surreal path]

lmao

@flat nymph can u maybe join vc

i wanna see the bugs for myself

if ur free ofc

[flat nymph]

Ok

[surreal path]

okay

[finite summit]

oberrow approved

Actually not oberrow approved, he will curse obos

you might be saying to yourself, "obos is already cursed though"

[kind root]

Or u can uncurse nyaux

[finite summit]

it pales in comparison to nyaux

[finite summit]

it's variable names:

• bruh
• hi
• shit
• dsfd
• stuff

no thanks

[kind root]

Lol

Mr beast should make a video about hiring every Microsoft engineer to work on nyaux

Can all ms engineers decurse nyaux

[finite summit]

no

they can rewrite it

[surreal path]

impossible task

[kind root]

The condition is they make do with existing code

No rewrites

[finite summit]

they would end up rewriting it in the end

the diff would be -10000 +10000

[kind root]

Lol

[finite summit]

speaking of diff

Can’t automatically merge.

[kind root]

Commit message: fix all memory corruption bugs in obos

[surreal path]

we found the issue

its the asm

in load and save ctx

and the thing for kstack

[finite summit]

https://github.com/OBOS-dev/obos/commit/027f2d4e7990a5bb4594f7857f4eeeb4af71c4ba

[kind root]

Lol

[finite summit]

master hasn't had code pushed to it since september

[surreal path]

nvm

we didnt find it

still some UB

this is mega duper super duper cringe

[edgy pilot]

UB is the nyauxmaster equivalent of gremlins

[surreal path]

its 1am

ive been at this for nearly 6hours

maybe ill js sleep

okay so instead of just sleeping

im gonna debug nyaux all night

maybe ill find something

[surreal path]

yea bug my bad bug thats undefined behavior that was caused by me

at this point there has to be some memory corruption going o

n

i want to mcmcmc

end my life

[finite summit]

don't

[surreal path]

so

someone on the forums js told me this

what if i

add a lock on memory

[finite summit]

did you not before

[surreal path]

no

wait

did adding a lock make the behavior consisent

it has!!!

[finite summit]

I wonder why

@kind root this man didn't lock his malloc

[surreal path]

lmao

im still having page faults

but the behavior is consisent now

ubsan is telling me this

so thats weird

BUT HEY

BEHAVIOR IS CONSISENT NOWW!!!

WE GOT CLOSER!!!

[surreal path]

infy is asleep lmao

[finite summit]

I know

[surreal path]

@flat nymph sleep in peace

ive made the behavior consisent

by putting a lock on kmalloc

[flat nymph]

Race condition

[surreal path]

yep

[flat nymph]

Nice

[surreal path]

smp is pain

anyways we still have this issue

[flat nymph]

The shittest thing to debug

[surreal path]

literarly

[surreal path]

to which im unsure is being caused by

well now we know nyaux code is not impossiblely cursed

just

yk

always put a lock on ur allocators guys!!!

if any function is gonna be access and written to by multiple cpus

put a lock

please

[flat nymph]

Lock free structures are fun

[surreal path]

The Rules Of Nyaux:

1. Zero Out Your Pages.
2. Put locks on things that will be accessed and written to by multiple cpus.
3. thy variable names shall make zero sense
4. have the worst most cursed formatting possible

anyways

[surreal path]

time to debug this

[cinder plinth]

so whats happening now

did you fix it

[surreal path]

no

my brain is absolutely fried

from no sleep

no my problem solving is probs shit rn

ngl

[cinder plinth]

time for redbull

[surreal path]

lol

[cinder plinth]

when I have a bug thats hard to resolve and it's late i just give up and spend the next day fixing it

[cinder plinth]

that just means your per-cpu data wasnt set btw

[surreal path]

it was tho

something funky is going on

but at least its consisent for once

[silver yarrow]

ello

sorry something came up

where we at

[surreal path]

ALL GOOD

WUX WE HAVE GOODNEWS

AND SOME ISH NEWS

GOOD NEWS WE FIXED THE WEIRD INCONSISENTCY

IT WAS THE FACT WE DIDNT HAVE A LOCK ON KMALLOC

BAD NEWS STILL HAVING ISSUES WITH PER CPU DATA BEING NULL

[silver yarrow]

latest in gh?

[surreal path]

yep

[silver yarrow]

alright

oh motherfucker

99% sure I figured it out

yeap I got it @surreal path

[finite summit]

Bros doing miracles

That is indeed your problem

[cinder plinth]

yeah I knew it

[surreal path]

remove that i suppose

[finite summit]

I have had this kind of problem before

[cinder plinth]

I just didnt want to spoonfeed

[surreal path]

bro im not getting spoonfed

stop with these jokes not funny

[cinder plinth]

I'm trolling lol

[surreal path]

about me getting spoonfed

ik

[cinder plinth]

well

would you have fixed it on your own is the question

but yeah I'll stop

[surreal path]

no because my problem solving is shit ren

plus i wouldnt have thought of that

[finite summit]

If it were my kernel I would have thought of it

Because I fixed a problem like that one day ago

[silver yarrow]

to be clear the only data segment u need to load is ss

[surreal path]

thanks

[silver yarrow]

ds can be zero

[surreal path]

yea we get it ur much more experienced

[silver yarrow]

fs, gs u never touch

[finite summit]

I am not tho

I only have a year and a half

[cinder plinth]

like me with women 😔

[surreal path]

so @silver yarrow

just ss and ds

or only ds?

[cinder plinth]

no need for ds

ds is useless

[silver yarrow]

just ss

[surreal path]

okay

[finite summit]

Don't forget cs

[silver yarrow]

zero in ds and es

[cinder plinth]

idk why you would call it reload_CS

[surreal path]

okay put zero in ds and cs

[cinder plinth]

if you dont do CS stuff there

no

not CS

CS is important

[surreal path]

es*

sorry

[silver yarrow]

its the label for far jmp

[finite summit]

He far jumps there

[cinder plinth]

ah

[finite summit]

Or rather far return

But whatever

[surreal path]

wait its fine

i dont load cs

so its fine

dunno why i call it .reloadcs

[silver yarrow]

u literally do

[finite summit]

The people at AMD decided it would be a good idea to remove far jumps

[silver yarrow]

the retfq

[surreal path]

yea

it does reload the cs

my bad its nearly 4am

[silver yarrow]

lol

[surreal path]

so this

[silver yarrow]

im in the states for christmas rn so only 7pm for me

[finite summit]

I'm in Canada for the living

So it's 10 30 pm

[silver yarrow]

eh sure, u could load ds and es somewhere much earlier and forget about it but that works

[surreal path]

does it matter?

[silver yarrow]

no

[finite summit]

No

[surreal path]

okay not worrying abt it

[silver yarrow]

does it matter to my ocd brain? yeap

[surreal path]

yes this solved the issue

thanks chat

[silver yarrow]

but ur entire kernel makes my brain want to shutdown for a year to begin with

[surreal path]

😭

[finite summit]

I was here to insult motivate

So yw

Now when nyaux on real he

*wh

**hw

Damn it I can't type on me phone

[silver yarrow]

skill issue

[surreal path]

i will run

but after i fix up the scheduler with yk

before that however im gonna nap

sleep*

goodnight

[silver yarrow]

night

[finite summit]

I'm sleeping too

[cinder plinth]

it's only 10:30pm

come on

you gotta grind

[surreal path]

😭

[cinder plinth]

https://tenor.com/view/goggins-david-goggins-work-get-back-to-work-back-to-work-gif-27698662

[finite summit]

If this was directed toward me, I am too tired to do stuff anymore

Brain is shutting down

Debugging EC drivers does that to a person

Especially when you do it from 7 am to 9 pm

Drives you nuts

[surreal path]

tried impling this

not so great time

thats load ctx into kstack

which runs fine

[surreal path]

do_savekstackandloadkstack: ; rdi has thread old and rsi has thread new
    push rbp
    push rax
    push rbx
    push rcx
    push rdx
    push rsi
    push rdi
    push r8
    push r9
    push r10
    push r11
    push r12
    push r13
    push r14
    push r15
    test rdi, rdi
    jz .skipstore
    mov [rdi + 0xB0], rsp ; store into kernel stack ptr the value of rsp
    .skipstore
    mov rsp, [rsi + 0xB0] ; load from kernel stack ptr into rsp
    mov rdi, rsp
    call change_rsp0 
    pop r15
    pop r14
    pop r13
    pop r12
    pop r11
    pop r10
    pop r9
    pop r8
    pop rdi
    pop rsi
    pop rdx
    pop rcx
    pop rbx
    pop rax
    pop rbp
    ret

thats my do_savekstack_andloadkstack

dunno whats going on i know im supposed to be asleep ngl

its 4am

i cannot slerep

😭

[flat nymph]

You only have to save the saved registers

[flat nymph]

And you throw this away

or maybe not

You have to push the return address into there

[surreal path]

morning

[surreal path]

no i dont

[surreal path]

userspace stack already has its rip set to the threads entry

[flat nymph]

No

You need to set kernel stack return address as well

[surreal path]

to what exactly

and does that mean i have to pop those registers out as well??

[flat nymph]

Add label right after call syscall whatever

[surreal path]

i dont have syscalls yet

again

[flat nymph]

In syscall instruction handler

[surreal path]

i dont have a syscall instruction handler

lol

[flat nymph]

Or some interrupt handler

[surreal path]

explain further

why do i need to add a label

also thats not how u spell label

https://tenor.com/view/spelling-mistake-minor-spelling-mistake-meme-gif-23650588

[flat nymph]

How do your interrupts handlers look

[surreal path]

isr_stub_%1:
    ; Push a dummy error code if a 2nd parameter has been passed to the macro
%if %0 == 2
    push %2                  ; Push dummy error code
%endif
    test byte [rsp + 16], 0x3; Are lower 2 bits(priv lvl) of CS selector 0?
    jz .skipswapgs1          ; If they are 0 (kernel mode) then skip swap
    swapgs                   ; Otherwise swapgs in user mode
.skipswapgs1:
    push rbp
    push rax
    push rbx
    push rcx
    push rdx
    push rsi
    push rdi
    push r8
    push r9
    push r10
    push r11
    push r12
    push r13
    push r14
    push r15

    ; push gs                  ; Save previous state of segment registers
    ; push fs


    push %1                  ; Push the interrupt number
    mov rdi, rsp ; put value of stack pointer into paramter 1 of c interrupt handler
    mov rax, [idt_handlers + %1 * 8]
                             ; Get the registered handler to call
    cld                      ; Required by the 64-bit System V ABI
    call rax
    mov rsp, rax
    add rsp, 8               ; skip int number
              ; Restore previous state of segment registers
    ; pop fs
    ; pop gs

    pop r15
    pop r14
    pop r13
    pop r12
    pop r11
    pop r10
    pop r9
    pop r8
    pop rdi
    pop rsi
    pop rdx
    pop rcx
    pop rbx
    pop rax
    pop rbp

    test byte [rsp + 16], 0x3; Are lower 2 bits(priv lvl) of CS selector 0?
    jz .skipswapgs2          ; If they are 0 (kernel mode) then skip swap
    swapgs                   ; Otherwise swapgs in user mode
.skipswapgs2:
    add rsp, 8               ; Skip error code
    iretq
%endmacro

like this]

and also i realized why it page faults

because theres still values that wern't popped off the kstack

this

[flat nymph]

No

[surreal path]

yea

[flat nymph]

Because you don't push return address

[surreal path]

the stack is therefore corrupted

[surreal path]

i do

[flat nymph]

Where

[surreal path]

arch_create_frame

[finite summit]

iretq?

[surreal path]

isr stubs

[finite summit]

wut

[flat nymph]

What do those functions doo

[finite summit]

you literally pop those

[surreal path]

here

[surreal path]

no cause i also push kstack

[finite summit]

bru

[surreal path]

#if defined(__x86_64__)
struct StackFrame arch_create_frame(bool usermode, uint64_t entry_func, uint64_t stack)
{
    if (usermode)
    {
        struct StackFrame meow = {.rip = entry_func,
                                  .rsp = stack,
                                  .cs = 0x40 | (3),       // USER CODE
                                  .ss = 0x38 | (3),       // USER DATA
                                  .rbp = stack,
                                  .rflags = 0x202};
        return meow;
    }
    else
    {
        struct StackFrame meow = {.rip = entry_func,
                                  .rsp = stack,
                                  .cs = 0x28,     // USER CODE
                                  .ss = 0x30,     // USER DATA
                                  .rbp = stack,
                                  .rflags = 0x202};
        return meow;
    }
}
#endif

[flat nymph]

Why

[flat nymph]

Where do you push retrurn address

[surreal path]

.rip is setting the stackframes shi

rip

its entire frame is then pushed

[flat nymph]

That's not it

[surreal path]

wdym

[flat nymph]

You have isr stub

It calls to some functions

[surreal path]

yea

[flat nymph]

Which calls to function which I can't find

To switch tasks

[surreal path]

yea

[flat nymph]

This

What do you have on stack?

• userspace rip, rsp, cs, ss, rflags
• userspace registers (rax - r15)
  ======
• a bunch of C shit which calls do_savekstackandloadkstack
  ======
• return address which called do_savekstackandloadkstack
• kernel registers which you save

[flat nymph]

(again, this only needs to save the saved registers, but that's besides the points)

[surreal path]

calls in order

[surreal path]

yes

[flat nymph]

What you push:

• userspace registers
• return address? no
• kernel registers

[flat nymph]

Where are you expecting this to return in ret?

[surreal path]

into sched

[flat nymph]

Wtf is mov %rsp?

[surreal path]

debug purposes

[flat nymph]

First ```x86asm
do_savekstackandloadkstack: ; rdi has thread old and rsi has thread new
; do_savekstackandloadkstack(Thread *old, Thread *new)
push rbx
push rbp
push r12
push r13
push r14
push r15
mov [rdi + 0xB0], rsp ; store into kernel stack ptr the value of rsp
mov rsp, [rsi + 0xB0] ; load from kernel stack ptr into rsp
pop r15
pop r14
pop r13
pop r12
pop rbp
pop rbx
ret

[surreal path]

what about it

lol;

[thorn bramble]

that looks okay tho

assuming 0xb0 is the address of the kernel stack

[surreal path]

mhm

[flat nymph]

Second ```c
const size_t saved_stack_size = sizeof(long)*6;

void load_ctx_into_kstack(struct thread_t t, struct StackFrame userctx)
{
t->kernel_stack_ptr -= sizeof(struct StackFrame);
(struct StackFrame)t->kernel_stack_ptr = user_ctx;
t->kernel_stack_ptr -= sizeof(void);
(void)t->kernel_stack_ptr = return_from_kernel_in_new_thread;
t->kernel_stack_ptr -= saved_stack_size
}```

[surreal path]

mishakov

[flat nymph]

third ```x86asm
return_from_kernel_in_new_thread:
pop r15
pop r14
pop r13
pop r12
pop r11
pop r10
pop r9
pop r8
pop rdi
pop rsi
pop rdx
pop rcx
pop rbx
pop rax
pop rbp

test byte [rsp + 16], 0x3; Are lower 2 bits(priv lvl) of CS selector 0?
jz .skipswapgs2          ; If they are 0 (kernel mode) then skip swap
swapgs                   ; Otherwise swapgs in user mode

.skipswapgs2:
add rsp, 8 ; Skip error code
iretq

[surreal path]

this is the defination of stackframe

[flat nymph]

Yes

[surreal path]

do u see stuff like error code --- ss

those values

[surreal path]

i see what ur trying to do

[flat nymph]

Well, add it

[surreal path]

return_from_kernel_in_new_thread is that function address?

[flat nymph]

This is a general idea of what you should have

Yes

[flat nymph]

Also, you have to swap the kernel stack

But you can do it from C

[surreal path]

so ur not exactly saving the kernel stack garbage in load_ctx_into_kstack

[flat nymph]

What

[surreal path]

as in\

[surreal path]

krnctx

[flat nymph]

https://wiki.osdev.org/Calling_Conventions

[surreal path]

i know this

im saying something else

[surreal path]

wdym, we are clearly swapping stacks here

[flat nymph]

I mean in TSS

[surreal path]

yea ofc

[surreal path]

i mean we are not doing anything about krnctx or whatever tf

[flat nymph]

Idk

[surreal path]

what about the userspace registers

[surreal path]

oh nvm

[flat nymph]

You load them here

[surreal path]

so when we ret, we go into return_from_kernel_in_new_thread

that makes sense

hold on while i implement this stuff

[surreal path]

what is saved stack size

and why are we
t->kernel_stack_ptr -= sizeof(void*);

[flat nymph]

Because you push a pointer

[surreal path]

oh yesh makes sensee

but why saved stack size

whats that purpose for

[flat nymph]

For 6 registers that you push

[surreal path]

why just those 6 registers

why not more

yk what im not gonna question

[flat nymph]

Because of calling conventions

[surreal path]

yea cause we need those still are scratch registers or wtv

[surreal path]

@flat nymph what about syscalls

[flat nymph]

Change it to be the same as what you have in your structs

[surreal path]

ik i did

[flat nymph]

Like of you push interrupt number and error code, pop those

[surreal path]

but

what about syscalls

[flat nymph]

What about them

[surreal path]

if we are iretq'ing

are we iretqing back into a new thread

[flat nymph]

Yes

What do you want to do

[surreal path]

okay

sorry

nvm

[flat nymph]

You can return to C which could do some extra shit

[surreal path]

no its okay

all good

dw

okay im running it

it died

when calling change_rsp0???

no nvm its not even consisent

this is my asm code

global return_from_kernel_in_new_thread
return_from_kernel_in_new_thread:
    add rsp, 8 ; skip int num
    pop r15
    pop r14
    pop r13
    pop r12
    pop r11
    pop r10
    pop r9
    pop r8
    pop rdi
    pop rsi
    pop rdx
    pop rcx
    pop rbx
    pop rax
    pop rbp

    test byte [rsp + 16], 0x3; Are lower 2 bits(priv lvl) of CS selector 0?
    jz .skipswapgs3          ; If they are 0 (kernel mode) then skip swap
    swapgs                   ; Otherwise swapgs in user mode
.skipswapgs3:
    add rsp, 8               ; Skip error code
    iretq
do_savekstackandloadkstack: ; rdi has thread old and rsi has thread new
    push rbx
    push rbp
    push r12
    push r13
    push r14
    push r15 ; dont ask why these 6 registers only, mr miskakov never bothered explaining
    test rdi, rdi
    jz .skipstore
    mov [rdi + 0xB0], rsp ; store into kernel stack ptr the value of rsp
    .skipstore
    mov rsp, [rsi + 0xB0] ; load from kernel stack ptr into rsp
    mov rdi, rsp
    call change_rsp0 
    pop r15
    pop r14
    pop r13
    pop r12
    pop rbp
    pop rbx
    ret

my schedd code

g

it wont let me post the code snipped

cringe

thats load ctx

that is schedd

i dont get why its dying ngl

im doing everything right?

😭

uhh any idea @flat nymph ?

i got none

[flat nymph]

What's the point of skipstore?

[surreal path]

if old thread is null

but thats besides the point

seems to be failing somewhere

[flat nymph]

Your kernel_stack_ptr should probably point to the top of the stack

Not to the stored location

Same in rsp0

[surreal path]

wdym

[flat nymph]

Is 0xB0 the right offset?

[surreal path]

where the stack pointer is rn?

[surreal path]

yes i counted

[flat nymph]

No

[surreal path]

where it isnt?

[flat nymph]

kstack + 262144

[surreal path]

makes sense

ill have to make 2 pointers in my thread

one for the base

one for the current ptr for yk

[flat nymph]

Broken rsp?

[flat nymph]

Yes

[surreal path]

yea

[flat nymph]

or you can put thread on stack

[surreal path]

okay so for rsp0 it will have to be base ptr of stack

hold on

[flat nymph]

and for syscall

[surreal path]

yeayea

thats 0xC0

and also

its not 0xb0

its 0xb8

[flat nymph]

You can also probably do call change_rsp0 from C

[surreal path]

its good

okay im testing now

IT FUCKING WORKS

IT FUCKINS WORKS @flat nymph

AND IT WORKS PERFECTLY??

YOOOO

[flat nymph]

Why wouldn't it

[surreal path]

WELL THANK YOU MISHKAVO

NOW I HAVE A PROPER

YK

PROPER SCHEDULER IG

OLD NYAUX NEVER HAD THIS BTW

AND THAT RAN BASH

SO

😎

also

even tho i got my voice back

i still havent talked

should i talk

it feels so weird to talk

i havent in like 4 days now

[surreal path]

@kind root @finite summit 😎 proper saving and loading context that the very first nyaux kernel (which ran bash btw never had)

ANYWAYS

vfs time chat?

the fun part now

oh wait

before that

interrupt controller like APIC

i need to get working

i also might need a hashmap

i will port a library

somehow

[kind root]

What does that mean

[surreal path]

okay so

in all previous nyaux schedulers

there was no actual switching out of rsp0, or saving kernel registers and stuff. it was just *stackframe = thread->frame

now we actually have proper shit

as all schedulers have

[finite summit]

with every new nyaux feature comes 50 bugs

[surreal path]

lmao

[finite summit]

(per each line of code written)

[surreal path]

real

[finite summit]

and 5 days of debugging

[surreal path]

lmao fr

[finite summit]

(per single bug)

and +1000 messages in the nyaux thread

[surreal path]

💀

[finite summit]

just for some nerd to tell him he did it wrong

and it should be done this way

and not that way

[surreal path]

anyways

i will port a hashmap library after i play some roblox

because is christmas

[finite summit]

you can copy+paste the one I copy+pasted

if you want

[surreal path]

https://tenor.com/view/christmas-holiday-cheer-festive-xmas-meme-gif-9616242526166840657

[finite summit]

but why does the lapic need a hashmap??

[surreal path]

apic*

but

i have a way of doing isa -> GSI

using hashmaps

[finite summit]

ioapic or lapic?

[surreal path]

ioapic

i already have lapic

lmao

[finite summit]

oh

[finite summit]

why not just have a linked list

[surreal path]

how else am i getting my timer interrupts??

[finite summit]

idfk

[surreal path]

cause funny and fast

[finite summit]

of madt ioapic redirection entries

[finite summit]

also stupid and un needed

you register a GSI once in a blue moon

[surreal path]

is funny

anyways i go play roblox, ugh the smell of pizza CAN SOMEONE JS TAKE MY PIZZA LIKE SERIOUSLY

WHY DOES MY DAD WANT IT IN MY EROOM

IF IM NOT GONNA EAT IT TIL TMRW

💀

bro i get hes trying to get me to eat more but

come on

[finite summit]

btw you mean madt ioapic redirection entries, right?

[surreal path]

yea

ofc

[finite summit]

good

[surreal path]

what else am i talking about??

lmao

[finite summit]

what you said

ISA->GSI IRQs

thought you were gonna do some cursed thing

[surreal path]

no????

lmao

[finite summit]

you're nyauxmaster, you do the most cursed stuff sometimes

[surreal path]

fair

nyaux will one day run portal 2 co-op

[finite summit]

aim high

[surreal path]

so omar will be forced to play portal 2 co-op with me

because obos will also be able to run it

[finite summit]

I will be dead by then

[surreal path]

i think half life 2's source code is leaked

so u can probs compile that and somehow port that

maybe one day

[molten grotto]

*an array, there are at max like 16 or whatever the number of isa irqs is

[surreal path]

and when will portal 2's source code get leaked

so i can port it

😡

[finite summit]

smh

[molten grotto]

just use a glibc compatible libc

[finite summit]

you need to get 1-1 binary compatibility

[molten grotto]

like hzlibc

[surreal path]

mlibc

[molten grotto]

its not binary compatible with glibc tho

[finite summit]

with linux

[surreal path]

fair

[finite summit]

you need to make nyaux linux

[surreal path]

i am not using some obsecure libc tho

[molten grotto]

hzlibc

[finite summit]

*obscure

[surreal path]

[surreal path]

😡

[finite summit]

https://tenor.com/view/minor-spelling-mistake-gif-21179057

[finite summit]

port glibc then

you won't

[surreal path]

no i wont

that is torture chamber

port glibc to obos

[molten grotto]

it even runs factorio (though you cant share saves between factorio ran on glibc)

[finite summit]

lies

[surreal path]

prove it

[molten grotto]

anyway you could port wine

[surreal path]

probs

[finite summit]

with hzlibc's plentiful 20 commits

over a period of 5 months

[surreal path]

lol

[molten grotto]

it does run factorio tho

[finite summit]

proof?

There are a lot of functions still to be implemented

[surreal path]

okay qwinci

make hzlibc more active

[molten grotto]

I have sent scs of it somewhere

[surreal path]

make it easy to port

[surreal path]

and ill use it

thats not happening tho

[molten grotto]

yeah likely no with the amount of sysdeps it requires

[finite summit]

so if I port it to obos, I get factorio?

[surreal path]

yea

[finite summit]

[surreal path]

who is this guy

in the white shirt

not me

bro thinks hes pro

[finite summit]

does it run bash

https://tenor.com/view/sus-cat-cat-stare-stare-bruh-cat-really-n-gif-25597234

[molten grotto]

yes

[surreal path]

fake not real

[molten grotto]

about stuff like xorg I am not sure, I should do LFS with it again and get that far

[surreal path]

u could probs get factrio on mlibc

ngl thop

someone get linux on mlibc

to test this

do lfs with mlibc

[molten grotto]

yes you can and I have done that (though it needs a preloaded compat lib)

[surreal path]

preloaded compat lib?

wdym

[cinder plinth]

already exists

[surreal path]

lol

[cinder plinth]

chromium ran on mlibc linux first

[surreal path]

yes!!!

[molten grotto]

well some mlibc stuff is incompatible with glibc (like the pthread struct sizes) so you have to override them with ones that are https://github.com/qwinci/mlibc-glibc-compat

[surreal path]

peak roblox gameplay

[surreal path]

maybe ill have that working with nyaux

[finite summit]

these sysdeps look like they were taken straight from linux' syscalls

[surreal path]

to get portal 2

[finite summit]

shit wrong reply

[finite summit]

meant to reply to this

[molten grotto]

in the compat layer?

or hzlibc

[finite summit]

yes

hzlibc

sysdeps

[molten grotto]

I mean some of them are for implementing linux specific libc fns

but then there is just posix stuff that mlibc also has sysdeps for

[cinder plinth]

I will try porting glibc stuff eventually

through linux compat

imagine factorio on your OS

[surreal path]

on nyaux

it will be

pro

[finite summit]

I would get ~1 fps

[surreal path]

then @elder shoal can waste his time doing factrio in nyaux instead of doing astral work

[molten grotto]

managarm got like 20 or smth idr

with llvmpipe

[haughty kite]

oh damn

thats barely playable but still

waiting for i915 tho

[surreal path]

imagineu

udrm

on nyaux with factrio

[cinder plinth]

I wonder if the bottleneck is llvmpipe or managarm

[molten grotto]

you can def get more out of llvmpipe, on linux I got like double that (I don't exactly remember how much tho, might also have been slightly more than that)

[haughty kite]

lmfao

managarm moment

[molten grotto]

nvm it was 9 on managarm, I think on linux it was like 40

[haughty kite]

is the perf still shitty if you turn off vsync

[silver yarrow]

Nyax with factorio before gta 6

[haughty kite]

also, what do the profiles look like

[surreal path]

nyax

:sk

💀