Nyaux

[edgy pilot]

inshallah we shall make having a girfriend halal

[finite summit]

huh

[edgy pilot]

-# ||(ˢᵒʳʳʸ)||

[elder shoal]

real osdevs dont have partners

for they are mere distractions from the road to success

[finite summit]

factorio also has no end, as Mathewnd found out the hard way

[worldly condor]

indeed

real men have no partners

[finite summit]

is this some sorta sigma alpha male bs

If you want to be rich, women will only slow you down.
Invest many monies into my company called "BigScamDepressedLosers"

[elder shoal]

hey I commited like 4 lines of code yesterday to astral

[finite summit]

you did, or infy did?

[elder shoal]

the alpha leads the beta follows but the sigma walks alone awrooooooo

[elder shoal]

me

[finite summit]

soruce-recipes/libtool: fix build

[elder shoal]

it was a jinx script but it was code

[worldly condor]

what about lambdas?

[elder shoal]

no way I typod

[finite summit]

git commit --amend

I should add working set and swap to astral

[elder shoal]

nah I wanna tackle that myself some day

[finite summit]

someday, aka never
Mathewnd's routine:

• wake up
• factorio
• go to work
• factorio (with a side of work)
• go back home
• factorio
• sleep
• repeat

[elder shoal]

that sounds gay

[elder shoal]

wake up
go to university
go to work
the facotyyry must grow
go to sleep
repeat

[finite summit]

where are you right now?

the facotyyry must grow

?

[elder shoal]

my friend just told me to hop on factorio so yes

[finite summit]

in your routine ofc

[finite summit]

and you're not at work or uni, right

(right??)

[elder shoal]

though my physical location is right behind you

in your blind spot

[finite summit]

holy

SEND HELP

ETGHE

I live at 123 seasme street

someone broke in

/j

or at least, I hope is a joke

[surreal path]

isnt it not in the quran tbf

oh yeah btw i think i figured out the issue with isr_stub 32

so

😎

[finite summit]

You can't just deny the opinion of practically every modern scholar of islam

[surreal path]

lets notttt argue about this

[surreal path]

why arent any of the other lapic timers firing

tf????

and get lapic id clearly works because well

it worked fine in the print logs

but when used in an interrupt

it shits itself

huh?

like wtfff

anyone know whats going on?

[silver yarrow]

prolly need more context

[edgy pilot]

Give it proper food and it might not anymore

[surreal path]

okay

this is what the schedule function looks like

void *sched(struct StackFrame *frame) {
  kprintf("CPU %d Says: MY LAPIC TICKED HAHAHAHA\n", get_lapic_id());
  send_eoi();
  return frame;
}

void bootstrap(struct limine_smp_info *info) {
  kprintf("bootstrap(): Hello from CPU %d\n", get_lapic_id());
  init_gdt();
  init_idt();
  per_cpu_vmm_init();
  init_lapic();
  hcf();
}
void init_smp() {
  for (uint64_t i = 0; i < smp_request.response->cpu_count; i++) {
    struct limine_smp_info *cpu = smp_request.response->cpus[i];
    cpu->goto_address = bootstrap;
    kprintf("init_smp(): Found CPU %d\n", cpu->lapic_id);
  }
  // stall_with_hpetclk(100);
  // init_lapic();
}

nothing gets printed on the other cpus

only the bootstrap cpu has it tick???

its the exact same function

to which i read the msr

to get the lapic address per cpu

sti is run on all cpus

????

😭 ???

[surreal path]

everythings on github btw

[silver yarrow]

you might want to invest some time into organization

[surreal path]

organization

huh?

whats that

also i still cannot figure this out

[silver yarrow]

I gave up trying to figure out what parts of ur code did

[flat nymph]

What do you do with stall_with_hpetclk()?

And why?

[silver yarrow]

all the hardcoded & unannotated constants make it very hard to follow without pulling up the spec side by side

[flat nymph]

Oh it's calibration for 10 ms

But why...

All APICs should in theory work at the same frequency

[thorn bramble]

especially without messing with power management they do + inside qemu

[surreal path]

eh still

either way

dont get why only ONE lapic

wants to work

[flat nymph]

What are you doing with hpet?

[surreal path]

calibration for 10 ms

i prefer to use the hpet

as its easy

[flat nymph]

How are you doing it I mean

[surreal path]

wdym

[flat nymph]

Do you just look at how much time has elapsed?

[surreal path]

mhm

[flat nymph]

If you're programming it from several cores at the same time it might be a race condition

[surreal path]

no cause

im just reading the hpet main counter\

im not messing anything else

[flat nymph]

Also this

[surreal path]

ik

[flat nymph]

Idk

I've had problems with periodic mode a long time ago

[surreal path]

either way this isnt the issue

[flat nymph]

And just gave up and used one shot for everything

[surreal path]

currently

just the damn other cores hate me

😡

[flat nymph]

Code quality issues

[surreal path]

NO

growls at you

[surreal path]

i mean all that matters is just i know what its doign right

[flat nymph]

If it isn't working then you're probably not doing it right though

[surreal path]

yea but it works on ONE core

SOOOO

it works PARTIALLY???

like

huh

it should work on all cores

[flat nymph]

Doesn't have to?

[surreal path]

i mean its the exact same code lol

[flat nymph]

If you're forgetting to initialize something

[surreal path]

in this case what would i have to initialize, lapic should work out of the box right???

js set the vectors and plug in play

RIGHT????

https://tenor.com/view/miku-hatsune-miku-anime-dance-silly-gif-7439677361449276795

yeah im debugging this tmrw i need sleep

if i cant debug this, ill ask the osdev forums if i cannot figure it out but im sure ill figure it out

like im not stupid lmao

probs js overlooking something

also @silver yarrow maybe ur right about organizing the code

i might js comment anything

everything*

yeah it shows that the code quality sucks but

im not rewriting the entire codebase

again

im js gonna comment things that dont make sense at first glance

[surreal path]

wait is the default behavior for the other cores to have the local apic disabled?

[flat nymph]

Idk I don't remember

[surreal path]

*software disabled

[flat nymph]

But I'm setting this bit explicitly

[surreal path]

okay js in case ill have it forcefully enabled

yeah

[flat nymph]

I'm doing this https://gitlab.com/mishakov/pmos/-/blob/main/kernel/arch/x86_64/interrupts/apic.cc?ref_type=heads

(call enable_apic() per core)

And it seems to have more steps than you

[surreal path]

0x100 is for the software apic enable bit correct?

too lazy to open the SDM

rn

well not but my ram is like

90%

[flat nymph]

Then there's this

[surreal path]

oh bit 11

thanks

[flat nymph]

This is what osdev wiki says

[surreal path]

@flat nymph still the same issue

[flat nymph]

Idk

[surreal path]

*sur_interrupt = 34 | (*sur_interrupt & ~0xFF) |
(1 << 11);

[flat nymph]

I was planning to be sleeping one hour ago

[surreal path]

bro sleep

😭

@flat nymph

go to sleep.

now.

[flat nymph]

No

Bit 8

Bit 11 is the MSR

[surreal path]

ohhh

why'd u give me the msr version 😭

i dont use xapic

[flat nymph]

But I'd suppose it would be set

[surreal path]

of

or

x2apic

[flat nymph]

LAPIC base?

[surreal path]

how do i get my lapic base?

reading the msr

its solved

@flat nymph

thank you

[flat nymph]

Yes, and that same MSR has the bit to disable APIC

[surreal path]

i could do that

but i use the sur register instead

also

look guys

[surreal path]

🥳

[flat nymph]

Yeah

[surreal path]

pin this

chat

this is really cool

proper multicore

PROPER

i didnt get this working on other nyaux rewrites i think

so

this is rlly cool

[kind root]

Congrats

[surreal path]

hi chat

im sorry i got caught up with appointments yesterday

[silver yarrow]

You don't need to apologize, there are no expectations

[surreal path]

i put expectations on myself tho

[silver yarrow]

You got all the time in the world 😃

[surreal path]

still caught up with irl stuff

we gonna get some work done later today

dw

[surreal path]

hi chat my bad, still stuff going on

[surreal path]

GUESS WAHT

CHAT

ITS TIME

okay so

we need to

how would i like

seperate different archictures

would i have an arch_init?

function

[cinder plinth]

https://tenor.com/view/the-return-of-the-king-lord-of-the-rings-ro-tk-title-card-gif-22191928

[surreal path]

and some macros to check which architecture it is

[cinder plinth]

id just have a per-architecture entry

that calls into main

[surreal path]

wdym?

[surreal path]

im confused

[cinder plinth]

how is that confusing

you have one seperate entry point for every architecture

[surreal path]

oh

yea that makes sense

but if u booted on lets say an arm archiecture somehow via limine right

and i just went into entry point

do i check the cpu archiecture?

then call into that per architecture entry?

@cinder plinth

[cinder plinth]

no

the entry point is your architecture specific code

[surreal path]

yea but this entry point is gonna be the entry point all cpus will go into tho..

[cinder plinth]

so?

you need a per-cpu initialization anyway

[surreal path]

void entryallcpusgointo() {
  // wtf do i do?
  #ifdef __x86_64__
  // do stuff here
  #else ifdef yadada
  // do stuff here
  #endif
}

something like this?

@cinder plinth

[cinder plinth]

no

use your build system

have a per-architecture entry file

[surreal path]

i use make

files

[cinder plinth]

should be easy then

[surreal path]

i dont know how to do that

😭 😭

im just gonna have an arch_init() function in main

and do my shit yada

hi fadanoid!!!

nice to see you lurking nyaux's thread

lol

[surreal path]

okay im working on it

[kind root]

Update uacpi for new fun printouts

[surreal path]

i did with git submodule update

i dont know if its properly updated

but the kernel api looks the same to yk

[kind root]

Does it print out the installed tables

[surreal path]

i didnt even run it 😭

[kind root]

Run it then

[surreal path]

i still have to make a generic abstraction

for the arch

// heres what im thinking
void output_to_port(uint64_t data, uint8_t bit_width) {
#if defined(__x86_64__)
  // check for the correct instruction for the bit width
  outb()
#else defined(riscv or whatever tf)
  // idfk there instructions
#endif
}

[kind root]

x86/main.c

Substitute x86 with current arch

Have x86 main call generic_main once its done with arch stuff

[surreal path]

i already figured out an abstraction

so im doing that currently

okay nearly done

i just need to make the generic interface now

fixed all the errors

[molten grotto]

port io is pretty much specific to x86 (along with some other smaller arches)

[surreal path]

how should it be done then

[cinder plinth]

you only use port io in x86 code

[molten grotto]

^

[cinder plinth]

or you have some linux iomem kind of thing

[surreal path]

then

what do u do on other archiectures

[kind root]

Aarch pci

[surreal path]

here

[cinder plinth]

you use mmio maybe?

[kind root]

Yes its not port io but same abstraction applies

[cinder plinth]

idk arm

[surreal path]

okay then

ill do mmio in this function otherwise

if its another archiecture

idfk arm either

[cinder plinth]

Macros can get messy

[surreal path]

its fine

[cinder plinth]

Just implement it in an arch file

[surreal path]

ima take a break for 2 hours

cause im tired

so im gonna nap a little bit

maybe eat

cause i havent eaten yet lol

[flat nymph]

I have a generic limine function that's jumped to from small assembly function (since I don't use limine's stack)

[surreal path]

oh

my

god

after seperating the archiecture through an abstraction

this happens?

wth

null deref anyway

seems to be happening here

but its going so fast

i cannout

find it

the root cause

oh right

im stupid

@kind root wheres the cool tables u promised

beautiful isnt it

[edgy pilot]

maybe it needs pci api implemented

the ops per second thingie

[surreal path]

idk

[plush hearth]

I don't think that is enough to update it

I remember having issues in the past related to that

[surreal path]

i did --recursive

[plush hearth]

i remember having to go into the directory and do git pull

[surreal path]

ill try

[kind root]

nah

hes just still on old revision

[surreal path]

damn

[plush hearth]

did it work?

[surreal path]

no

theres no .git in uacpi?

tf

[plush hearth]

empty the directory and reclone it?

[surreal path]

yep

time to do that

oh

fun

[finite summit]

No it needs nanoseconds since booy

*booy

[surreal path]

how tf do i get that

[finite summit]

**boot

[finite summit]

It's a kernel api

For uacpi

You just ask your timekeeper for that

[surreal path]

ill js give it whatever the hpet main counter is

🤷

[plush hearth]

HPET counter is femtoseconds iirc

[surreal path]

ik

[finite summit]

Depends

Afaik

[finite summit]

That's gonna break the counter?

Ops/s counter

Which is the only reason why you are implementing it

[surreal path]

ill convert it to nanoseconds

[finite summit]

Kk

[plush hearth]

is the hpet main counter time since boot or could it have been reset?

[surreal path]

actually i cant

cause i use uacpi to scan for the table

[plush hearth]

tables are valid before the namespace is loaded

iirc it maps them in uacpi_init

[surreal path]

okay

[plush hearth]

so you just have to init uacpi in 2 stages

[cinder plinth]

you can reset it

i think

or else just use limine's time at boot feature

nah not really that isnt good for timekeeping

[surreal path]

to turn fentoseconds into nanoseconds

whats the math again

i forgot

[edgy pilot]

yes but I thought that maybe pci not being implemented somehow made it exit prematurely before printing the stats

[kind root]

they dont have to be literally since boot lol

its just a growing nanosecond counter

[surreal path]

yea this is wrong

ME WHEN I CANT GET MATH RIGHT

😭

hpet counter ticks every 10 nanoseconds

i have no fucking idea how to convert the counter to nanoseconds ughhhhh

[edgy pilot]

if a single tick is 10 nanoseconds just ticks * 10

[surreal path]

okay so its ticks * ctr_clock_period

yet doing that...

nvm this is right

[edgy pilot]

12114 ms isn't right

[surreal path]

yeahhh

idk why

this is correct

[kind root]

maybe thats how long it takes in nyaux

[surreal path]

💀

[kind root]

does it take 12 seconds to boot?

[surreal path]

it did not take 12 seconds

more like

OH

FROM BOOT?

[kind root]

no

just that part of uacpi init

[surreal path]

yeah it did take 12 seconds

nvm

its right

💀

im running tcg tho

tbf

[kind root]

astral gets 40k on TCG

how are u two orders of magnitude behind

[surreal path]

idk 😭

nested virtualization maybe?

im running via wsl

wait no that wouldnt

nvm

idk

[kind root]

bruh

[surreal path]

i mean nyaux has always been slow

so eh

idrc computers are fast nowadays

it runs very fast on kvm/native hardware

[edgy pilot]

this is a little trick for converting ticks to nanoseconds without overflowing that qookie taught me

u64 freq = 10^15 / clk
u64 p = log2(freq)
u64 n = (1'000'000'000ull << p) / freq

// when reading the timer
u64 nanos = (u128(ticks) * n) >> p

[surreal path]

anyways changes pushed ima sleep now

night night

[kind root]

did u see the table printout at least?

[surreal path]

mhm

[kind root]

thoughts?

[surreal path]

love it

[kind root]

thanks

[surreal path]

np

[kind root]

good night

[surreal path]

goodnight

[surreal path]

good morning chat

so

i did some changes and now smp also works with the archiecture abstraction

now whats next is well

the scheduler itself

but we encounter a problem

we need a stackframe and different archiectures have different stackframes

how exactly do we

abstract that

[orchid dawn]

#ifdef ARCH_X86_64
typedef struct {
  /* stuff */
} stackframe;
#elif ARCH_ARM64
typedef struct {
  /* other stuff */
} stackframe;
#elif ARCH_RISCV
/*[etc.....]*/
#endif

[plush hearth]

What timezone are you?

[surreal path]

thanks

[surreal path]

im in irish time

[plush hearth]

I didn't think it was morning there

Isn't it like 10:30pm?

[surreal path]

9:28pm

[edgy pilot]

good morning

[plush hearth]

Ah ok. I'm only asking because its 7:28am here in Australia, so you confused me

[surreal path]

😭

[finite summit]

Good evening

[surreal path]

good evening

[surreal path]

hi chat, no work for today BECAUSE AT 12AM TODAY THERES A RUMOR HALF LIFE 3 WILL BE ANNOUNCED?

ive been waiting all day

i need

this is a need not a want

[elder shoal]

Every year theres a rumor

If it gets announced I will write a driver for nyaux

[cinder plinth]

if it gets announced I will write a gpu driver

[surreal path]

okay sure :)

[steady flume]

lol it's 10AM for me

[finite summit]

It's 4 56 pm here

[plush hearth]

It's 8:08am for me

[kind root]

here goes every timezone

[cinder plinth]

it's 25:03 here

[elder shoal]

nothing ever happens

[surreal path]

u were right!!

[tender gorge]

My disappointment is immeasurable and my day is ruined

At least Helldivers 2 update is very good

[surreal path]

yea :c

[surreal path]

true

[tender gorge]

And the new IP that was announced at the very end also looks sick

But still

No half life 3

[surreal path]

[tender gorge]

And no valve deckard

🥺

[surreal path]

😔 😔

idk what valve is doing chat

they setting on they lazy butts

just

ranking in steam money

[kind root]

they dont have to do anything at this point

its just passive money generation

[tender gorge]

Lord gaben knows what is best for us

[surreal path]

true

[surreal path]

amen!

https://tenor.com/view/whatsapp-hl2-halflife2-gordonfreeman-gif-4476561939831071209

[tender gorge]

    Who art in Valve, Hallowed by thy name.
    Thy Half-Life 3 come, thy puchase will be done,
    Online as it is in offline.
    Give us this day our daily deals,
    and forgive us our guest passes,
    as we forgive those who guest pass against us.
    And lead us not into temptation,
    But deliver us from Origin.
    For thine is the kingdom,
    the power and the glory,
    forever and ever.

    Gaben.```

[surreal path]

🙏

[elder shoal]

Nothing ever happens

[cinder plinth]

No gta6 either

[surreal path]

yep

[frigid cliff]

Amen

[surreal path]

welp i just realized one big

thing

before writing the scheduler

THE FUCKING PAGE TABLES ARE NOT ARCHIECTURELLY ABSTRACTED

thats fun

im js gonna do something rlly stupid

do any other architectures have 2mb pages (risc v, arm etc)

?

[cinder plinth]

Yea

Arm has 64k pages too

And bigger ones too

[kind root]

U dont need this information

To abstract away paging

[molten grotto]

that's 32-bit arm specifically tho, 64-bit supports 4kb, 2mb and 1gb

[unkempt relic]

he meant the 4/16/64k basic page size

or granule as they call it for some reason

[molten grotto]

ah like kernel page size

oh wait there are actually different translation granules, I didn't remember that

[tawdry mirage]

yeah aarch64 has 4k, 16k, 64k base page size, and from that you get 512*size and 512^2*size and 512^3*size blocks

er, for larger page sizes it's not 512 but you get the point

[kind root]

Yeah arm has some of the more flexible mmu configuration I've seen on any arch

[unkempt relic]

you ain't seen nothing yet then

[kind root]

Lol

[kind root]

Which are more flexible

[kind root]

Wtf is this lol

[unkempt relic]

m68030

you may configure the number of tables up to 5

[kind root]

Damn

[unkempt relic]

pick your own page table geometry

it has large pages as well ("early termination page descriptors") and you can have a PTE point to another PTE to be used instead

[hexed mango]

Damn!!!

[surreal path]

memory issues

[hexed mango]

Me too, me too

[surreal path]

kill me

genuinely

why is one damn pnode refering to 0xffffffff

wtf

[kind root]

Bruh

[surreal path]

and how did uacpi not notice this

[hexed mango]

You probably aren't clearing that field in the insertion or removal functions

Wait

What are pnodes

[surreal path]

my own type

linked list basically

[hexed mango]

Ok, then ya you're def not clearing that field

[surreal path]

i memset before inserting though?

[hexed mango]

Do any other functions touch the pnode list, though?

[surreal path]

yep

on alloc

one sec

ill pull it up

[surreal path]

and on creation too

[molten grotto]

no newlines

[hexed mango]

first->next is uninit here, I think

Could that end up causing this issue?

[surreal path]

lets test

nope

[hexed mango]

Hm

Not entirely sure then, nothing stuck out to me

[surreal path]

maybe u'd have full context with the entire file

[hexed mango]

I have a party all weekend, so I'll be busy
But if you still have this contact me sunday night and I can try to help

[surreal path]

okay

[hexed mango]

Good luck bestie

[surreal path]

i have a very slight suspection

its because of free_unused_slabs

im gonna remove the call to it

and see what changes

never fucking mind

okay heres what ill do

NVM

i have no ideas

fucking hell

and wheres my fucking stacktrace

hold on

well

now we have more context

okay i see a calloc

i dont see

a kernel_free() being called

@kind root is the call being called somewhere else?

WHY IS MY LAPTOP LAGGING

okay

um

Page Fault! CR2 0xffffffff
RIP is 0xffffffff80006935
-> Function: slabfree() -- 0xffffffff800068e0
-> Function: uacpi_execute_control_method() -- 0xffffffff80011120
-> Function: uacpi_execute_table() -- 0xffffffff800132e0
-> Function: uacpi_table_load_with_cause() -- 0xffffffff8001a1b0
-> Function: uacpi_namespace_load() -- 0xffffffff8001bf50
-> Function: init_acpi() -- 0xffffffff80001000
-> Function: kmain() -- 0xffffffff80005d70
-> Function: none -- 0x0

that is the full stack trace

i dont know if uacpi is doing some macro shanngains to expand the macro into calling the kernel free

obviously its my code thats at fault here

i just dont know why

[kind root]

Context release gets inlined perhaps?

[surreal path]

okay hold up lemme check

yea i see

yea this is a triggy bug

TRICKY

OMG I CANNOT SPELL

😭

[surreal path]

okay we know when it is allocated and deallocated

but thats still not the source of the issue

because the slab we free ->next field is ferring to 0xffffffff

and i tried to impl ksan

i couldnt get it working

its still in the source code

for some reason not working

🤷‍♂️

[surreal path]

tried using gdb

got nowhere

[haughty kite]

skill issue

okay memory debug 101

take your free function

and replace it with memset(memory ptr, 0xaa, malloc size)

then insert memset(memory ptr, 0xcc, malloc size) right at the end of your alloc function

[surreal path]

ur the skill issue 😡

/j

[haughty kite]

keep coping

my malloc worked second try

[cinder plinth]

mine worked first try

L

(no)

[finite summit]

Yes.

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

[orchid dawn]

wait, you guys actually got memory allocation working? (I gave up before ever actually making it work properly)

[surreal path]

yes cause ur brain is so big

and mine is so small

[surreal path]

tried, still found nothing im afraid

might ask the forums and irc chat for help cause i rlly cannot figure this out

[surreal path]

well what i found interesting

is

i do this function after uacpi inited

hold on

WAIT

nvm

this is not the problem

cur->next = (struct slab *)init_slab(mod->size);
  cur = (slab *)cur->next;
  void *guy = cur->freelist;
  cur->freelist = ((pnode *)cur->freelist)->next;
  memset(guy, 0, mod->size);
  return guy;

[surreal path]

HELP

and i tried software watchpoints

and it crashed qemu

[edgy pilot]

using a debugger

[surreal path]

💀

[edgy pilot]

cringe

just spam printf

[surreal path]

i cant

[edgy pilot]

why not

[surreal path]

i dont know where to put the printfs because i dont know where the memory corruption is happening

[edgy pilot]

shrimple: put it everywhere

[surreal path]

fine ill put it everywhere

js to prove u wrong

[edgy pilot]

now that's based

[edgy pilot]

what have I said that needs proving

[surreal path]

im gonna prove thats not gonna help

[edgy pilot]

https://tenor.com/view/skill-issue-urge-resist-cat-gif-25536160

[surreal path]

bru

[edgy pilot]

https://cdn.discordapp.com/emojis/1121477522958327819.webp?size=44&animated=true

[surreal path]

yo is this a valid address

wtf

he gives me reasonable addresses

then this???

huh

[frigid cliff]

Omg

Me me me me me

[surreal path]

i solved it

yea thats a vmm address

TURNS OUT IT WASNT MY ALLOCATOR

BUT JS

some stupid ah

shanngains

im gonna check what size uacpi gave

what lol

so lemme get this straight

uacpi tells me its 10 bytes

gives me a vmm region address

gives me garbage sizes

too

[edgy pilot]

did the printfs work

[surreal path]

we got closer yea

can i ask

why for the love of god

is uacpi giving me a bunch of gibberish for size

[edgy pilot]

it is not

[surreal path]

uhhh excuse me

that is gibberish

[edgy pilot]

kfree is called with gibberish

and you call kfree

not uacpi

[surreal path]

no

[edgy pilot]

yes

I don't see any kfree in uacpi source

[surreal path]

im js passing whatever size uacpi gave

and therefore

uacpi is giving me gibberish

look at the stacktrace

[edgy pilot]

what

what's your io_map

[surreal path]

hold on

thats all it is

[edgy pilot]

why do you allocate anything there

[surreal path]

why

shouldnt i not

[edgy pilot]

do you need it?

no

then no

just put base in out handle

copied from obos smh

[surreal path]

no cause i can check

anyways thats not the main issue it

is*

[edgy pilot]

maybe it is

[edgy pilot]

and what's the use for that

[surreal path]

how??

[surreal path]

to check if the offset goes over the len

[edgy pilot]

and when would that happen

[surreal path]

return uacpi_status_invalid_argument

[edgy pilot]

it's nyaux, anything is possible

[surreal path]

idk maybe infy makes a mistake

[edgy pilot]

it wouldn't be called with an incorrect size

[surreal path]

in future

[edgy pilot]

fix your mistakes first

worry about someone else's later

[surreal path]

fine

ill change it

[finite summit]

Infy tests sized frees

If that's what ur talking about

[surreal path]

still

the same

didnt solve shit

[edgy pilot]

did you remove the kfree?

[surreal path]

yep

i did

[edgy pilot]

trace your stack

[surreal path]

thats the stack trace

[edgy pilot]

when the value is garbage

[surreal path]

😭

look at the screenshot

[edgy pilot]

^

[edgy pilot]

yeah and is that useful at the end?

[surreal path]

i cant watch address ^

[edgy pilot]

trace the stack

[surreal path]

and im not manually gonna step through every allocation

[edgy pilot]

if kfree is called with a garbage value

[surreal path]

fine

[edgy pilot]

wdym fine I'm trying to help

is this against your will?

[surreal path]

no

im js

rlly tired

is all

my bad

[edgy pilot]

sleep then

[surreal path]

no thanks

garbage

trace

[edgy pilot]

there definitely something else between 0 and 1

[surreal path]

wha

[edgy pilot]

are the optimisations off

[surreal path]

no one

on*

[edgy pilot]

use your stack tracer

[surreal path]

i already have the stack trace

[surreal path]

^

[edgy pilot]

when the value is garbage

not when the exception happens

[surreal path]

and how am i supposed to trigger a fault

when that happens

ill add a stupid check in kfree then manually page fault or some shit

[edgy pilot]

???

[surreal path]

how else am i supposed to do it

genuinely

[edgy pilot]

if (size > some_random_garbage_millions) trace_stack()

in kfree

wait what

you don't need to page fault

just read rbp smh

[surreal path]

fine

its better if i trigger a fault im not gonna spaggathi code a bunch of garbage

into that file

id have to load symbols early

and a bunch of other shit

[edgy pilot]

dump the addresses and use addr2line

[surreal path]

its the same as the gdb stacktrace @edgy pilot

[edgy pilot]

print size in uacpi kernel free

[surreal path]

okay

see

uacpi_size is just a size_t

[edgy pilot]

what kind of printf do you use

[surreal path]

nanoprintf

[edgy pilot]

@kind root

sizes are all correct for me

[surreal path]

yea this is very weird

[kind root]

addr = 0

frees of 0 are no op

[surreal path]

yea i ignore those

[edgy pilot]

I totally missed that

[kind root]

well clearly not

[haughty kite]

what is that size as hex

[kind root]

thats not whats called here

[haughty kite]

which uacpi commit are you in?

[edgy pilot]

FFFFFFFF800149B0

looks like an address lol

[haughty kite]

yeah

addr2line it

[kind root]

nyaux heap corruption any%

[haughty kite]

which uacpi commit

lmfao

yes

this is why every os should have a hardened heap

[surreal path]

huh

[haughty kite]

which uacpi commit

[surreal path]

i dont know but i just updated it

lemme run

[haughty kite]

please run cd third_party/uACPI && git rev-parse HEAD

[edgy pilot]

are you returning null in some handle in some uacpi kernel api function

or something

[surreal path]

still the same

[haughty kite]

what if you remove the entire free path

[surreal path]

hold on

[haughty kite]

and add 64k to every allocation size

[surreal path]

what

[edgy pilot]

150ops/s will plummet

[haughty kite]

put return; right after that free check

[surreal path]

okay

[haughty kite]

obviously perf will suck the point is to check the heap isnt totally broken

[edgy pilot]

yes it was a joke

Sadge

[kind root]

💀

I wonder if u port uacpi to NES will it run faster than 150/s

[haughty kite]

probably lmfao

[surreal path]

its running

[haughty kite]

as in it just hangs?

[surreal path]

no

got here

same issue

and that address that was given to

slabfree

[haughty kite]

i see i see okay

[surreal path]

is a vmm region one

which is why it dies

[haughty kite]

wait how do you get to slabfree if you put a return right after this first check in kfree

this is what you should do

actually, memset(addr, 0xcc, size) first

[haughty kite]

NO

[surreal path]

okay then what should i do

😭

[surreal path]

do u want me to do this? where?

[edgy pilot]

nyaux rewrite 2^64-1

[surreal path]

im not

we are done with rewrites

[haughty kite]

void kfree(void* addr, uint64_t size) {
  if (size >> 63) { kprintf("kfree: memory corruption detected\n"); __builtin_trap(); }
  kprintf("kfree(%p, %#llx)\n", addr, size);
  memset(addr, 0xcc, size);
  return;
}

do this

[surreal path]

same

[haughty kite]

okay then

this is weird

very

[surreal path]

i told you

[haughty kite]

can you send the code?

[surreal path]

yep

https://github.com/rayanmargham/NyauxKC

[haughty kite]

cool

[edgy pilot]

https://tenor.com/view/its-quite-cool-gif-26287398

[haughty kite]

holy shit so many errors on clang

oh god

okay im never going to compile it

[surreal path]

please use

gc

c

[edgy pilot]

garbage collector compiler

[surreal path]

use gcc

[haughty kite]

id need to build gcc

and that sounds annoying and cbf

ok static analysis it is

[edgy pilot]

what distro are you on

[haughty kite]

fedora

on arm

[edgy pilot]

ah

aren't there any cross gcc s in repos

[haughty kite]

ah lol

[haughty kite]

okay i think you have way bigger memory corruption issues @surreal path

i tried building your shit

and it memory corrupts itself to death

[surreal path]

wtf is going on

[haughty kite]

idk it hits this #PF every time for me

[surreal path]

where

[haughty kite]

this time it was before uacpi too

[surreal path]

show me entire

cr2 is weird

wtf

[haughty kite]

cr2=pc=ffff8...117a

[surreal path]

this is fucking weird

[haughty kite]

im on gcc 14.2

[surreal path]

i dont know how to get further

[haughty kite]

this is probably the issue tbh

[surreal path]

the issue is memory corruption obviously

but tracking it down?

idk

this is difficult

[haughty kite]

initially you map this

[surreal path]

mhm

[haughty kite]

and then get_symbols does something

and bad shit happens

[surreal path]

i allocate a resource

[haughty kite]

tbh i blame your kmalloc

[surreal path]

obv it is

but where the issue is

idfk

this js makes it even more difficult

[edgy pilot]

I would suggest rewriting it

[surreal path]

i already printf'd everything

[edgy pilot]

plus 150 ops/s is very slow

[surreal path]

oh

i thouyght u said something else

[edgy pilot]

you need a better allocator

[surreal path]

i was reading the screen from an angle

[surreal path]

i think the vmm region stuff is fine

slab allocator? idfk

im not rewriting

[edgy pilot]

yes

[surreal path]

it wont solve the issue

[edgy pilot]

how are you so sure of that

[surreal path]

because ill write the same design

[edgy pilot]

rewrite means write something else

[surreal path]

rewrite a slab allocator how

[edgy pilot]

write a better one

there are docs on the internet

[surreal path]

i read those docs

and impl'd it

[edgy pilot]

read them again, read other ones

[surreal path]

and for some reason its shitting on me

🤦‍♂️

[edgy pilot]

-# ||shrimply a krill issue||

[surreal path]

the slab allocator isnt even

that much code

im geuinely unsure

how that little code is somehow corrupting the entire kernel

[edgy pilot]

why not

[surreal path]

• i have read the code for the slab allocator top to bottom i dont see an issue

[edgy pilot]

I'm sorry but clearly you did something wrong

[surreal path]

like

slab alloc is literarly

void *slab_alloc(cache *mod) {
  if (mod->slabs == NULL || mod->size == 0) {
    return NULL;
  }
  slab *cur = (slab *)mod->slabs;
  while (true) {
    if (cur->freelist == NULL) {
      if (cur->next != NULL) {
        cur = (slab *)cur->next;
        continue;
      }
      break;
    }
    void *guy = cur->freelist;
    cur->freelist = ((pnode *)cur->freelist)->next;
    memset(guy, 0, mod->size);
    return guy;
  };
  cur->next = (struct slab *)init_slab(mod->size);
  cur = (slab *)cur->next;
  void *guy = cur->freelist;
  cur->freelist = ((pnode *)cur->freelist)->next;
  memset(guy, 0, mod->size);
  return guy;
}

[edgy pilot]

no point in trying to find problems in some other things

[surreal path]

even pitust didnt understand whats going wrong from static analysis

[surreal path]

obviously

[edgy pilot]

what if you don't free anything

[surreal path]

ill run out of memory but sure

hold on

[edgy pilot]

a few minutes of just looking won't find every bug

[edgy pilot]

give it more memory if that's the case

[surreal path]

still

no kfree

still

[edgy pilot]

I see a kfree in there

[haughty kite]

okay i found the bug @surreal path

[surreal path]

oh hold on

[surreal path]

WHERE

[edgy pilot]

I retract my statement

[surreal path]

removing it runs fine

[haughty kite]

src/cc-runtime.c:2663 (__mulvXi3_25) and src/cc-runtime.c:2678 (__mulvsi3) are mutually recursive

leading to an infinite loop

[surreal path]

thats not even

my code

[haughty kite]

yeah lol

[surreal path]

thats something from mints c template

bug?

[haughty kite]

wait really

[surreal path]

yea

legit

[edgy pilot]

who needs freeing anyway
just download more ram

[surreal path]

@wicked loom

[edgy pilot]

unused ram is wasted ram

[edgy pilot]

that's not mint's code

[surreal path]

who is it then

who made the cc-runtime

[edgy pilot]

llvm or gnu

[surreal path]

because that was from the c template

[haughty kite]

mint

its repackaged from LLVM

[edgy pilot]

ping the llvm dragon

[haughty kite]

but i 100% bet the llvm people do build config stuff to make sure this isnt a problem