EvalynOS

[hazy saddle]

Rather pre fill them in?

[cyan bison]

idk...

i allocate them for the init process and that's it

i leave the rest to handle cloning done by fork or god knows what

[hazy saddle]

So pid1 gets them

And then fork() and whatever does that

Does bash need clone?

[cyan bison]

no

[hazy saddle]

Or only fork

[cyan bison]

fork() and exec()

[hazy saddle]

I wonder if it will need source modification

Same with core utils

[cyan bison]

if u use mlibc u shouldnt have to patch bash

[hazy saddle]

[cyan bison]

except for config.sub

[hazy saddle]

What’s that

[cyan bison]

-# (just ignore the 4 patches i have for bash to get it to compile)

[cyan bison]

youll learn

[hazy saddle]

[cyan bison]

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

config.sub is autotools tomfuckery

[hazy saddle]

I do wonder if the mods would want this in #curated-resources ?

It’s pretty top quality but I’d have to reword the message

Also when I get my fucking 8tb hard drives

[cyan bison]

also TCP is funny because you find out what a RST packet and it's like "oh hehe it's like the computer is telling the other to go fuck itself"

[hazy saddle]

Im gonna mirror more bigger stuff

Like the gcc tarballs and stuff

[cyan bison]

i was gonna make a joke about what you should mirror but nvm

[hazy saddle]

https://tenor.com/view/muscle-man-you-know-who-else-my-mom-regular-show-funny-gif-9101122073397073314

[cyan bison]

"yo i got some data for you"
"kys"

pov tcp

hjwiusjofchjdskfhjnsdp

[hazy saddle]

[hazy saddle]

Honestly osdev.cc should have a spot here too

https://osdev.cc/

[rotund furnace]

[osdev.cc plug btw]

I love osdev.cc

the qemu log analyzer is still broken as fuck because I haven't bothered to fix it

[hazy saddle]

Did you make that site?

[rotund furnace]

no

wux did

I just did the qemu log shit

[hazy saddle]

Well rb trees from openBSD are really good

Atleast at memory efficiency

So

After a talk with @cyan bison

I’m going to use two of them

One for free vmemory

And one for used

So allocations are fast etc

And same for finding used regions

Because an O(n) method is bad

[sterile plank]

how did u get cool role

[cyan bison]

she implemented FRED

[hazy saddle]

https://files.evalyngoemer.com/share/manuals/fred9.pdf
@sterile plank its this btw

[sterile plank]

hmm

[hazy saddle]

I was the first to do it

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

[peak cloak]

5 or 6 nights at freddy

[hazy saddle]

ppc and aarch64 added

and the other sections have gotton plenty more manuals added

[rancid viper]

copyparty ftw

[urban mulch]

Damn you support a lot of arches

Oh this is not your os

[hazy saddle]

nope this is my manual archive

[peak cloak]

(i gave the powerpc manuals )

[hazy saddle]

Though I will be supporting them one day 😇

[rotund furnace]

I'll add arm support this time trust me

(I haven't touched arm support since paging)

[hazy saddle]

Add powerPC support

[rotund furnace]

fuck no

fuck powerpc

[hazy saddle]

@peak cloak

What do you think about this opinion

[peak cloak]

https://tenor.com/view/putin-walk-russian-memes-russian-walking-putin-wide-putin-gif-17176268809602528999

@Joe Biden send launch codes

[hazy saddle]

for virtual memory allocations

with POSIX mmap

can i clamp them down to page sizes?

is there a rule for clamping them etc

[magic saffron]

If addr is NULL, then the kernel chooses the (page-aligned) address at which to create the mapping

If addr is not NULL, then the kernel takes it as a hint about where to place the mapping; on Linux, the kernel will pick a nearby page boundary

from ⁨man 2 mmap⁩ on my system

[hazy saddle]

For fixed mappings that require overriding? Wait wouldn’t that trigger an error because it enforces such

Ahh yeah it does

So my VMA can safely just store ranges of address of the allocation and not worry about them being funky and overlapping

Il make ends of ranges be -1 of a page?

So 0x1000 to 0x2000

Is stored as 0x1FFF?

Or is my brain fucking up?

That way range 0x2000 to 0x3000 can be sorted cleanly?

[rotund furnace]

make sure you can't map_fixed over the kernel :P

[hazy saddle]

Well no shit :p

I’d ensure it’s a user pointer

Banning the last page before non canonical

To prevent sysret bug

[rotund furnace]

wdym

why the fuck would you allow the user to alloc up there anyway

[hazy saddle]

If the program can’t ever map that last page

They can’t trigger the bug

[hazy saddle]

The user program can do whatever the fuck it wants

It just can’t use the final page

[rotund furnace]

they shouldn't beable to map in the kernel memory anyway

[hazy saddle]

Not that edge of memory

The upper page of user memory

That’s where the bug can happen

If you have code on the end of that page

And syscall

RCX becomes non canonical

That’s the silicon bug etc

[rotund furnace]

wdym?

OH

nvm

I'm stupid

[hazy saddle]

Lmfao

Or I could allow mapping that page even

[magic saffron]

https://fuchsia.dev/fuchsia-src/concepts/kernel/sysret_problem

[hazy saddle]

Just ensuring it’s NX always

[rotund furnace]

I forgot 0x0000800000000000 is also "non canonical"

[hazy saddle]

Though that’s risky imo

[rotund furnace]

and you just broke some random 40 year old application that tries to allocate memory up there

[hazy saddle]

That program can go fuck its self

And il send patches

[rotund furnace]

you can't

that program is under government nda

(this program may or may not be an nsa exploit ngl)

[hazy saddle]

Then they can pay me 100 thousand dollars for a military support contract version of my kernel with that feature enabled

[rotund furnace]

make it a cmdline option

:P

[hazy saddle]

No

I want 100 thousand dollars

[hazy saddle]

Okay so I’m making the VMA today

Two RB trees

One for free memory

One for used memory

To allocate you take from the free one and insert it into the used memory

And you do the reverse for freeing memory

And I’m not going to do fixed mappings right now

And if the VMA allocate can’t get a large enough node at the part it will just goto the next node

And traverse it like a free list

But this allows fixed mappings to be good and fast because I can go right to the right node

[hazy saddle]

chat dont kill me

[hazy saddle]

i actually do

for elf loading

i have to suport them

[peak cloak]

[hollow crater]

ooooooo
well how does that work o_O
like
what do the tree nodes describe here
pages or regions or whar

[peak cloak]

vmm freelist

[hazy saddle]

they describe ranges

and then the sort function that i 100% DID NOT STEAL FROM @cyan bison

sorts ir properly

[peak cloak]

sorting algorithm moment

[hollow crater]

oo

[hollow crater]

but wait u sort them by size or by address

[cyan bison]

it wasn't sort it was compare

[hollow crater]

[hazy saddle]

both

[hollow crater]

o

[cyan bison]

wrong, they are sorted by address

[hazy saddle]

im dunb

[hollow crater]

😭

[peak cloak]

mods, electrocute everyone who answered wrong

[hollow crater]

I do something similar for an early boot allocator
tho using lists
but now I do vmem
cuz constant time >>>> logarithmic

[hazy saddle]

but this is vmem no

or somthing

idk

does proper vmem make things easier for fixes mappings @hollow crater

pls tell me

also what is vmem and how do you do it

because O(1) i need

[hollow crater]

http://usenix.org/publications/library/proceedings/usenix01/full_papers/bonwick/bonwick.pdf

its
THE
VMEM
idk whether its been superseded by more sophisticated stuff
well ig it did
afaik linux has 74748346356272 mechanisms for managing memory or something rn
that linux slub allocator is another hot shit out there

lads had slab, slob and slub
not even joking 🥀

vmem isnt an object allocator tho its just a general purpose range allocator

[hazy saddle]

im scared mommy

[hollow crater]

tho its kinda similar
or idk not really
well basically its like buddy allocator but instead of strict sizes it relies on lists of more sparingly sized blocks

[hollow crater]

its literally a nothing burger
well to some extent

dw

[hazy saddle]

but for this i need to alloacte huge ass regions of virtual memory

so idk if struct sizes works

[hollow crater]

mm I dont get what u mean here

[hazy saddle]

like a program MMAP's 2gb ram

i need to use this to allocate that

[hollow crater]

mm yes..

[hazy saddle]

well allocate the virtual bit

also this O{1}

as long as the size the prgram wants

is at the front

and its per program / address space

so as long as the program isnt goin buck wild

it should be fine

[hollow crater]

mmmm

[rancid viper]

what are you saying, you would allocate 2gb from a slab?

because no

[hazy saddle]

yeah no

of course not

im making my VMA

not a heap or whatever

[hollow crater]

im dumb :P
like
shouldnt u have somewhere to allocate kernel memory from
which is separate from managing the rest of the address space
then have the rest be managed by an external allocator for memory objects
be it a program segment or per subsystem stuff

[hazy saddle]

I do

It’s called my heap

But I need to allocate address spaces

For like MMIO

[rancid viper]

address spaces for mmio? do you mean like regions or something

[hazy saddle]

So I can get a free address in virtual space? To map to

[rancid viper]

one address space is generally like

the entire address space is what people mean by it

so one pml4/5 == one address space

[hazy saddle]

I’m mistyping

I meant ranges

[rancid viper]

yeah

[hazy saddle]

Just fucking hell

Idk if there is a better way to do this

other than a free list

or some such

[rancid viper]

wdym

[hazy saddle]

and then im gonna remove it split it and then put it back in

and then also put it in the used list

[rancid viper]

seems fine

[hollow crater]

ye it will work

eh ig astral survived with its vmm being based on a list of regions or somethign so basically like
🔥

or so I heard

[rancid viper]

btree of regions (and holes like u do) is a pretty common way to do it

totally fine

[hazy saddle]

Mine is an rb tree

Is there even really a better way to handle this?

[hazy saddle]

Is that better or worse than mine

[hollow crater]

well idk about the specifics of it but urs should be better in principle I think O_o

[hazy saddle]

And for fixed mappings

I can go right to the range

And split it there

Or deny it if it’s not there for no replace mappings

And then I will use the used tree @hollow crater

So I can free all the mappings

By digging the PTEs for them etc

[hollow crater]

aha
well sounds good actually
now that im thinking about it i might keep my used ranges in an rbtree too lel
idk if it'll be an overkill I was gonna do a hashmap :P

[hazy saddle]

Honestly rbtree is better tbh

Because you can go directly to a used region to free it

And iterate easily

And you don’t have to deal with buckets

Just steal the openBSD one too @hollow crater

[hazy saddle]

Look how data efficient these are too lmao

[hollow crater]

mm fair

well once im done with the pmm rewrite first

[hazy saddle]

Buddy allocator

[hazy saddle]

Think with a hashmap you have to do the hash function and then a % operator to pick the bucket. You have to store every bucket and a linked list for it and stuff

[hollow crater]

well yea but I meant that for only the structure holding the used nodes
where i wouldnt need to do any coalescing or anything
so it would work no worse than a tree o_O

[rancid viper]

I am failing to see how a hashmap would work well for this

worst case for a hashmap is O(n)

I suppose if you have a good hash function u might be able to get some good perf for some amount of ranges

[hollow crater]

I had buddy from the start but it was the way i implemented it that was wild
like
super wild
super super super wild
but was efficient
sad it had to go
now its still efficient but eh
less beautiful
much faster tho

[hazy saddle]

Yeah rbtree is o log n and o(n) for finding a free range when using it as a free list

[rancid viper]

well u cant really get away from the O(n)

[hazy saddle]

Idk maybe I could do an rbtree sorted by size and have three trees

[hollow crater]

tbh its not a stance I can reason about
I just didnt have a reason to do anything else :P

[hazy saddle]

So for allocation I go by the one sorted by size

[rancid viper]

yeah but even then worst case is always O(n)

actually no its not

mb

lol

[hazy saddle]

But then I can just instantly find one big enough o(1)

At the cost of more ram usage

I will consider that for later

[rancid viper]

well not O(1)

O(log n)

[hazy saddle]

If it becomes an issue*

[hazy saddle]

Finding the first node though?

I always would just need the first node that’s the largest

[rancid viper]

well then there is no reason to do an rbtree

[hazy saddle]

[rancid viper]

lol

[hollow crater]

like
for just keeping track of descriptors of allocated regions i cant see why a hashmap would be problematic

[hazy saddle]

I could have a varrible for pinning the largest known allocation?

[rancid viper]

didnt say it was problematic, just I dont see how its better

[hazy saddle]

Nah it would still be o logn

Because I have to find the node in the original tree anyways

[rancid viper]

I dont know the fancy math but worst case with a hashmap is all the entries fall into the same bucket and u just have a linked list

and when u can achieve binary search because the key is just a number that seems like the clear choice

[hollow crater]

but a good hash function and large enough buckets wouldnt have pathological patterns being common?
and for an address space ig a very large number of buckets is justified
welp idk
like I might as well just place the load of keeping track of allocated regions on the callers anyway
like
return a pointer to the bookkeeping structure instead of a pointer to the start of the region
simce the use of that layer is internal anyway
a vm_object could as well hold references to its own region

idk 🤷‍♂️

[hazy saddle]

The rbtree does hold book keeping objects tho @hollow crater

[hazy saddle]

You can put anything else you want in here

[hollow crater]

yea im aware
but its like
im not sure if traversing that used regions structure (regardless of what it is) is actually necessary
especially since unlike object allocations where it'd be totally unreasonable to have the callers hold pointers for their allocations other than just yk, the pointers
vm regions are usually allocated for a much longer time and arent used directly

afaik reference counting for like
shared objects isnt done at that level either

aight its 1am already ig im hallucinating rn

[hazy saddle]

@hollow crater POSIX requires that you are able to map fixed regions and when unmapping it can be in the middle of a region

[hollow crater]

mmmmmmmmmmmm
welp thats crazy 🥀
I mean
I can still do it that way

[hazy saddle]

just use an rbtree

[hollow crater]

fair lmao

I basically thought i can force callers to store the handles as list heads instead to get around this
and therefore unmapping a partial region will have the list head passed to vmm
vmm would split the region and gut out the part being unmapped and thus the region (well list of subregions regions) would be returned

[hazy saddle]

just use an rbtree

🙏

[hollow crater]

😭

okay I concede

[hollow crater]

beaware from fragmentation tho

[hazy saddle]

there is 128TiB of virtal space

if it cannot find memory

kill it

[hollow crater]

last time I checked chrome uses half as much for a single about blank tab

[hazy saddle]

well then chrome can go suck a lemon

[hollow crater]

LMFOA

[hazy saddle]

hows this look

[hollow crater]

well looks good

[hazy saddle]

this is better lol

[hollow crater]

idk if variable alignment >= page_size is a requirement for anything but for a starter it looks fine to me

[hazy saddle]

wuh?

it has to be page aligned

so i ensure all allocations happen on that

[hollow crater]

I was wondering if there's anything that may require allocations being a multiple of that
for some reason
but i'll take this as no

[hazy saddle]

page size is the minimum

so like

it has to be rounded

[hollow crater]

fair

[hazy saddle]

i uhh also now need to just put a chunk of free memory in there

and let her rip

atleast for the kenrel

i need to implement these after

i thinkin for the kernel

i could put 8 terrabytes after 16 terrabytes offset of the HHDM?

the heap starts here

[hollow crater]

well idk i just leave the higher half entirely for the kernel subsystems
barring the hhdm and the pfndb
with kernel non contiguous heap so that i dont have to care about the spacing

[hazy saddle]

my valloc needs the heap

[hollow crater]

same
for that I just let both manage their own structures internally

this's basically where I left that part and I havent fully figured it out yet tbh lol

this's the point where everything fell apart that I had to rewrite my pmm kekw

[hazy saddle]

well i can presume the HHDM wont be over 8 TiB right

atleast for 4 level paging systems

[hollow crater]

well ig u can make it as small as u want then have it fall back to temporary mappings and shit for accessing out of its bounds
or even shifting hddm coverage on demand
tho that'd be crazy lel

[hazy saddle]

i mean what system will have more than 8 tib ram

[hollow crater]

[hazy saddle]

esp one without 4 level paging

there i could do diffrent values

later™

[hollow crater]

well
guess it cant
or wait idk
is there any cpu out there with 48bit pa width that lacks 57bit address space

[hazy saddle]

[hollow crater]

anyways imma just
eep
😴

[hazy saddle]

wait one sec

[hollow crater]

[hazy saddle]

HHDM   0xffff000000000000 - 0xffff080000000000 (08TB)
VALLOC 0xffff080400000000 - 0xffff480400000000 (64TB)
HEAP   0xffff810000000000 - .................. (????)
KRNL   0xffffffff80000000 - 0xffffffffffffffff (02GB)

@hollow crater hows this look

for the mem map

[hollow crater]

o
well the heap doesnt need too much vm space does it
like
ig even some 256mb would be an overkill

[hazy saddle]

how about

HHDM   0xffff000000000000 - 0xffff080000000000 (08TB)
VALLOC 0xffff080000000000 - 0xffff480000000000 (64TB)
HEAP   0xffff480000000000 - .................. (????)
KRNL   0xffffffff80000000 - 0xffffffffffffffff (02GB)

uhh err

that heap GPFs

wtf

[hollow crater]

real
0xffff8

[hazy saddle]

errm what the sigma

[hollow crater]

lmao

well that one is legal tho

[hazy saddle]

all i did was
#define KERNEL_HEAP_START 0xFFFF810000000000
to
#define KERNEL_HEAP_START 0xffff480000000000

[peak cloak]

what is being mapped

[peak cloak]

booo lowercase hex

[hazy saddle]

im not even using my valoc

[peak cloak]

shame on thee

[hazy saddle]

maybe its corupting shit

[hollow crater]

bc
the latter isnt canonical

[hazy saddle]

fuck you mean

[peak cloak]

calculator

[urban mulch]

There's a reason kernel heaps are typically in the HHDM

[hazy saddle]

fuck im gonna have to redo my god damn heap

because the person i took this from

fucking made this mess

[hollow crater]

well then uhh idk :P

[hazy saddle]

So uhhh I just allocate some free list nodes from the PMM and HHDM it

🥀

And then that’s enough to bootstrap

Or some crap

[peak cloak]

is hhdm higher half mapping or something

[hazy saddle]

Yes

[peak cloak]

oh ok

insane crap

[hollow crater]

real

[peak cloak]

real

[hazy saddle]

How should I do this fucking heap 🥀

[peak cloak]

whats the heap for

[hazy saddle]

Been meaning to redo it anyways

[hazy saddle]

malloc()

[peak cloak]

oh

uhhh crap

no idea

[hazy saddle]

@hollow crater any tips 🥀

[peak cloak]

pmm allocate it, make it a linked list or something idk

[hollow crater]

well waiaiaiaiiaiiat a sec
I think yea
its not canonical

sign extension here should match the last pml4 bit

[hazy saddle]

Oh 🥀

[peak cloak]

is the address signed???

[hazy saddle]

https://tenor.com/view/waah-waa-sad-wah-waaah-gif-25875771

[hazy saddle]

Yes

Kinda

[peak cloak]

w h a t

literally psychotic behavior

[hazy saddle]

There is an upper half and a lower half

And in the middle is nothing

[hollow crater]

well as I told u I haven't fully figured out the cyclic dependence part 🥀 (I still belive i can do it WITH dependance)
I do slab + magazines tho

[hazy saddle]

I mean 🥀

[peak cloak]

[hazy saddle]

I could use a

Bump allocator

To bootstrap it

🥀

To get the starting freelist nodes

[hollow crater]

ohhh
I do that

[peak cloak]

yall's virtual memory structures look like a total mess ngl

[hollow crater]

gummi does that too

everybody does that

u should as well

[hazy saddle]

I could also just use the PMM to allocate one free list node

Wait

FUCK

I have to make a buddy allocator for this too

[peak cloak]

realization

[hazy saddle]

Don’t I

[peak cloak]

bro what

what buddy allocator

[hazy saddle]

https://tenor.com/view/20-twenty-minutes-rick-and-morty-gif-12928450

https://tenor.com/view/rick-and-morty-crying-gif-9607254

[peak cloak]

if this is for the kernel then why does your kernel need a malloc, i mean its memory efficient but in that case just dont allocate a lot

[hazy saddle]

Osdev summed up

[hazy saddle]

Every kernel needs a malloc

Linux has a malloc

[peak cloak]

but why

[hazy saddle]

Dynamic memory allocation

[magic saffron]

what you were doing was fine, just pick a canonical address next time 😛

[peak cloak]

obviously

[hazy saddle]

But I want to rewrite anyways

[peak cloak]

actually whatever every kernel needs a malloc

allegedly

[hazy saddle]

And make my PMM a buddy allocator anyways

So like

[hollow crater]

for physical memory
so yea 3 allocators for memory
pmm
vmm
and then one for heap
thank fucking u paging
all hail good old segmentation

[peak cloak]

scary

[hollow crater]

real

ANYWAY IMMA EEP

😴

[peak cloak]

https://klipy.com/gifs/jcsoos-2

[hollow crater]

[peak cloak]

now do a 3 hour OSdev session

[hazy saddle]

im taking a break today

[hollow crater]

already done

[peak cloak]

ok return to sleep

please beware that there's maintenance performed on the houses of evil monsters

they chose to sleep under your bed if thats fine

👍

[hollow crater]

I just cant
skewl is starting again next week and all in all I havent done shit in this break so im sprinting in hope that I can achieve something before im back into the grind 💔 🥀

[hollow crater]

nah i dont mind them monsters 🙏

theyre

welcome

[peak cloak]

WHAT CAN I SAY EXCE-

[hollow crater]

https://tenor.com/view/the-stanley-parable-stanley-parable-stanley-the-narrator-gif-27539740

[severe marlin]

oH mAn YoUr gOnNa LoOk So StUpId In a BaJilLioN yEArs wHeN AI FiXeS tHe RaM sHoRtAgE

~ some dumbass probably

[hazy saddle]

to that dumbass im doing 16TIb

checkmate

[severe marlin]

Well damn

ram shortage solved everyone

Time to pack it up

Openai can have all their ram now

Everyone say thank you Evalyn for solving economics

[hazy saddle]

HHDM   0xFFFF800000000000 - 0xFFFF900000000000 (16TB)
VALLOC 0xFFFFA00000000000 - 0xFFFFE00000000000 (64TB)
HEAP   0xFFFFF00000000000 - .................. (????)
KRNL   0xFFFFFFFF80000000 - 0xFFFFFFFFFFFFFFFF (02GB)

this mem map should be correct

[severe marlin]

Lol

What's the limit for ram with canonical addressing anyway?

It can't be too much more

[hazy saddle]

512TiB total for 4 level paging

256 for each region

so i got room to shmoove this

if needed

[severe marlin]

I was just thinking if you support 8tib physical ram there's only so much more you can get before you need a CPU that doesn't even exist

[hazy saddle]

tf you mean lmao?

thats HHDM

so if it dosnt exist it wont be mapped

its just the range im reserving for it

it wont need it

infact i support features that only exist on CPUs that came out last month

[severe marlin]

Oh yeah no just thinking about reasonable maximum memory limits

[hazy saddle]

"16TiB ought to be enough for anybody.”

[severe marlin]

I mean realistically 8mb ought to be enough for anybody but then someone invented emacs

[hazy saddle]

I mean realistically 4gb ought to be enough for anybody but then someone invented chrome

[severe marlin]

😭 I use chrome on my work laptop and rn the average tab ram usage is like 500MB per tab

[hazy saddle]

i think it may be time to retire this tbh

i need a smaller panic logo

and maybe switch to a better font

(il make it a compile option for the legacy one)

• i may make a compile option for these

as its bloating the binary a little

or maybe have it in the initramfs?

god i love overloading

now clang just add templates to C

and then its perfect

you can do NASTY things with them

and i wana do it

thowback to the time i put C++ shared/raw pointers into lua

[rotund furnace]

I do wonder what these names are in the object

[hazy saddle]

wdym

[rancid viper]

how do they get mangled

is it 1:1 cpp

[hazy saddle]

iirc it is

https://clang.llvm.org/docs/AttributeReference.html#overloadable

Clang provides support for C++ function overloading in C. Function overloading in C is introduced using the overloadable attribute.

[rotund furnace]

well obv you can't have 2 symbols with the same name

[hazy saddle]

Given these declarations, one can call tgsin with a float value to receive a float result, with a double to receive a double result, etc. Function overloading in C follows the rules of C++ function overloading to pick the best overload given the call arguments, with a few C-specific semantics:

[rotund furnace]

yeah they are just c++ named

[hazy saddle]

this also means no gcc for my kernel

[rancid viper]

btw why are u using that old nasty attribute syntax

[rotund furnace]

⁨⁨```c

int attribute((overloadable)) test(int a) {
return x;
}

float attribute((overloadable)) test(float x) {
return x;
}⁩⁩ resolve to ⁨⁨
_Z4testi
_Z4testf

[rancid viper]

when ⁨⁨[[clang::overloadable]]⁩⁩ exists

[hazy saddle]

dosnt work

[rancid viper]

yes it does

[rancid viper]

u are placing it in the wrong place

it needs to be before the function

[hazy saddle]

ahh

wondefull

[rancid viper]

: )

[hazy saddle]

thank u

[rancid viper]

no worries

I despise the gcc attribute syntax

and since C23 has a well defined attribute syntax might as well

[hazy saddle]

i fucking hate GCC

it tripple faults my kernel

before flanterm even opens

[severe marlin]

Based take, gcc has so many outdated design choices

And stallman can eat a drain pipe

[true oak]

who even cleans up memory?

i just set magic to 0 so the picker wont pick the thread

[hazy saddle]

goin to sleep in a sec

anyone got a moment to code review as a sanity check

[hazy saddle]

i need to free memory to have a working system

then you will OOM

super fast

[hazy saddle]

Just wana make sure I’m not gonna shoot my self in the foot right as I try and use this lmfao

[hazy saddle]

well i diddnt sleep

and im open sourcing my riscv emulator finnaly

il test the other code today after i sleep

[hazy saddle]

God honestly once this is done along with the proper VFS and fork and exec and a few others

And then il do a final pass of code quality and do the last two rewrites I’ve been meaning (buddy allocator and SLAB)

Il be so happy to have beta1-rc1 working

But then

I have to make SMP work

[true oak]

So Much Processors™

[hazy saddle]

So many problems*

[true oak]

im fighting the urge to say the other p word

[mortal bane]

that's gotta be POSIX

[true oak]

nah powerpixel

pp

cool acronym

[cyan bison]

just use swap lol

is it really a good OS if it cant survive a fork bomb?

[mild glacier]

what OS can survive a fork bobm

[cyan bison]

one of them

[mild glacier]

name one

i wanna know

[cyan bison]

I forgot which one

survive here means it doesnt crash btw

[mild glacier]

ohh

[cyan bison]

not that it actually survives

[cyan bison]

and is usable

[mild glacier]

tbh if u can detect what executable a process is then u can detect if there are more than some limit of processes of the same executable and if so kill all copies

[mortal bane]

Linux

[true oak]

windows 95

[hazy saddle]

lets fuckin gooo

[hazy saddle]

huh wtf

trying to do 1gb mappings

hats not right

and its the exact same with less

[rotund furnace]

hm?

[hazy saddle]

im testing by making mappings to see if it works

its doin somthing

not right

[hazy saddle]

im doin the paging like this now

it was working a bit ago

but REALLY slow

and then crashing

[rotund furnace]

well that's not right :P

[hazy saddle]

whats not

[rotund furnace]

it being slow then crashing

normally it just crashes

[hazy saddle]

yeah :p

and the framebuffer was nuked

so theres that

but E9 worked

[hazy saddle]

https://tenor.com/view/destroy-pc-destroy-pc-bash-the-pc-gif-14915322778488621950

[hazy saddle]

i fixed it!! @rotund furnace @cyan bison

thus is the cycle of osdev

[hazy saddle]

NEVERMIND

2mb pages

are severely FUCKED

[hazy saddle]

It’s working now!

I’m queen of the world

And it’s 2:30am

And I should sleep

But also it seems program loading times are heavily reduced using 1GB pages

[mortal bane]

then sleep

[hazy saddle]

(For HHDM)

(Because I load programs using the HHDM)

[mortal bane]

ayo what

no user virtual base?

[hazy saddle]

I get a physical page and I use the HHDM to write to it

Because it may be read only in user mappings

• it reduces TLB misses with way using the big ass pages

Which I may call internally big CHUNGUS pages

[mortal bane]

https://tenor.com/view/chungus-gif-21555962

[hazy saddle]

I see you like my naming of 1GB pages

https://tenor.com/view/megamind-villain-chair-spin-suave-cool-gif-4559758

[mortal bane]

[peak cloak]

clangers

[hazy saddle]

What

Also I ended up making it look better

[peak cloak]

talk is expensive, send bugs

[hazy saddle]

What bugs do you see

[peak cloak]

i dont see any bugs thats the thing

add bugs

anyways the code looks fine

from my understanding

if it doesnt work then im sorry im tired

(as in my brain doesnt work straight up)

[severe marlin]

put a ladybird emoji or something in a comment

if you put all the bugs in yourself intentionally there won't be space for any other bugs

better a bug in the hand than two in the bush or something

[hazy saddle]

okay so

i need to get this commit ready

and then work on exit()

which should just be stop it from sheduling

and have a repear thread clean the thing up

go through the free list and clear all thoes nodes

go through the used list and the PTEs and free thoes pages

then free all the nodes

then free the last few structures and cleanup the shceudling linked list

https://tenor.com/view/gif-gif-9403556598434283568

then tasks wont be
while (1) {} ing

then il do a leak test

and PRAY i dont have a leak

💀

then its time to FORK

or well rather

il need to fix up how i spawn processes

THEN fork

(and then exec which will do a partial exit() and then reload the process but keeping the needed things)

then i can get bash

cleanup some code

ans then its uhh

well ofc x11

but thats a bit off from just bash + coreutils

[hazy saddle]

Sorry VFS and tmpFS

THEN fork

[peak cloak]

https://klipy.com/gifs/skinenr

[hazy saddle]

What

[hazy saddle]

time to push this

then get the reaper thread and exit() workin nicely

maybe il even make my schedular aware of a "work buffer" type where it will check if it has work to have it never get scheduled if it has nothing to do?

but that should fix all leaks

which means VFS time and tmpfs

and then

[true oak]

threek

[hazy saddle]

im doing a real fork()

[hollow crater]

im stealing this

[hazy saddle]

https://cdn.discordapp.com/attachments/1420566657541214320/1468828977706832179/ezgif-306d93eac6eb2f4e.gif

Thank you

I made it

[hollow crater]

well ive stolen that other one too already

[severe marlin]

oh I already stole it lol

[hollow crater]

the internet is a lawless wasteland smh

[mild glacier]

twok

[soft snow]

well x11 requires it to be pretty darn stable.

Would need to work on stability.

and port more drivers

[hazy saddle]

Well of course

[peak cloak]

fork bomb

[true oak]

when you accidentally decrease the instruction pointer on fork

[hazy saddle]

Okay so. I’m going to need to add processes. I’m thinking it’s a structure with of course the FDs and thier information and the other stuff. Then they contain pointers to all the threads they contain which have the actual stuff the scheduler works on.

And then processes have pointers to the parent and the children

Which would be a liked list of “siblings” and a linked list of “children”

[hazy saddle]

Or I can use a non intrusive linked list

And then do it properly

https://tenor.com/view/jgmm-cat-meme-ragebait-cat-think-cat-ponder-gif-7034802297606826390

Okay so question is

How long can I avoid this rework

I only need it for waitpid

And fork()

I think I can do the exit() and VFS shit without this

And then just make exit() zombie you later

Which would be really easy

[hazy saddle]

I think I’m gonna take a break today

To goto sleep on time

And be able to work proper tomorrow

Gonna try and get exit() working properly and not leaking anything

Gonna do a test of spawning threads and having them die and see etc

Then fix the VFS

And then it’s just get fork() exec() and waitpid()

Then bash and core utils

This should work

[hazy saddle]

Then cleanup the code and then start working on SMP

[hazy saddle]

so for reclaiming all memory in a processes page tables i am going to dig for all pages in the lower levels and iterate through the entire thing skipping zeroed entires

and on seeing a valid page i will check the PFNdb and if its ref count is = 0 (non CoW) or its = 1 (last referer) free it otherwise decrement the ref count and move on

should be a basic ahh function

[hazy saddle]

lookin good so far

just have to properly reclaim & track all memory

But the system call is getting gotten

And the schedular notices said thread needs to be reaped

And then sends it over

This is really fucking thread unsafe

But I think I may implement lock free algorithms with 128 bit CAS pointers

Which means I have to drop support for CPUs from 2006

[hazy saddle]

the fuck

this line of all things?

and it goes away if i remove this free

the fuck

[hazy saddle]

okay this one is a race condition

and i need to get proper locks added to shit

[hazy saddle]

this is coprution when sending things to the reaper

the fuck

this causes the crasf faster

just acting as if it were asleep

[hazy saddle]

why the FUCK is this causing it

still

i may or may not be dumb

I never initialized any of the RBtrees

For user processes

I uhhh

Would think that’s very undefined behavior

No wonder the kernel is shitting it’s pants

When it’s fucking with unintalized shit

[hazy saddle]

So when I tried to ya know allocate from them

With nothing to allocate from at all

That’s uhhh

Not really good exactly

[hazy saddle]

okay that. wasnt it

😭

okay well

somthing is corupting the RB trees

[hazy saddle]

Okay so I’ve narrowed it down to my schedular

nvm??

idk

[hazy saddle]

Well they were corrupted

But then I tried to fix it

And then somthing else is blowing the fuck up

I have no fucking idea

[cyan bison]

ohh mb

that was me

also like

if it ain't locked

idk why u expect it to work in a multithreaded environment

[hazy saddle]

i put a ton of stis and clis and crap to test that

[hazy saddle]

i should add some spinlocks

lmao

[cyan bison]

well u at least have spinlocks implemented right

[hazy saddle]

im makin a minimum repdrofuction

[hazy saddle]

[cyan bison]

bro...

this is bad

[hazy saddle]

its lock free /s

as in

i have zero locks

anywehere

[cyan bison]

cro is not smp safe 🥀

[hazy saddle]

its not even single core safe

[cyan bison]

trust me ur gonna die when u try to add smp

[hazy saddle]

im gonna add locks laterrr

dontttt worryyyy

[cyan bison]

Not a good mentality.

[hazy saddle]

true

but

https://tenor.com/view/fuck-it-we-ball-cat-dna-ballin-gif-2085555445816338978

@cyan bison know what the issue was

[cyan bison]

no

[hazy saddle]

I wasn’t zeroing the memory

Of the RBtrees

I have now added zalloc() to my kernel

So I can just use that

And not need to manually zero the shit

what do you know

undefined behavior

makes shit crash

[hazy saddle]

well lets go

gonna leave this on

and hope it wont OOM

But I think my reaper is doing its job

Will need some extra work for CoW to not delete those pages etc

Maybe some optimizations too

BUT

Oh and I will have to turn them into zombies before a full clear

But good enough for now

[hazy saddle]

100k runs

no crash

this is with 64mb total ram and ~47mb detected by the kernel as usable

well

now its time to fix the VFS

and then

https://cdn.discordapp.com/attachments/1420566657541214320/1468828977706832179/ezgif-306d93eac6eb2f4e.gif

and then bash

[true oak]

thread reaper

[hazy saddle]

Yeah it’s the grim reaper

When a thread exists it’s put into a reaping state

If the scheduler attempts to run it it will instead put it into the reaping queue

And then all of its memory gets reclaimed

[true oak]

exists?

[hazy saddle]

When a program exits

Like the main function returns

And the the libc calls exit()

[true oak]

i got it you just said exists

[hazy saddle]

Everyone makes typos

[true oak]

im just evil and hate everybody ig

[hazy saddle]

much better

[true oak]

everyone who uses snake_case needs to be studied

[hazy saddle]

I use them all

I use every kind of case

Whenever I feel like it

[mild glacier]

snake react this guy he's asking for it!!

[mortal bane]

https://tenor.com/view/easy-steal-gif-6770495535008702461

[mild glacier]

https://tenor.com/view/easy-steal-gif-6770495535008702461

[hazy saddle]

Excellent my meme is spreading

[mortal bane]

i made my own

[mild glacier]

that last frame:

https://tenor.com/view/black-guy-shcoked-gif-27654754

[true oak]

10 gallons of water welln't spent

[hazy saddle]

Alright maybe we should get back on topic

[hazy saddle]

And don’t post AI stuff in my channel please

[mild glacier]

i did a false negative 😔
my dumbass thought it was just a recreation

[hazy saddle]

@cyan bison happy now

here even better

i fixed the attribute

[rotund furnace]

wdym happy now

where your spinlocks not atomic

[hazy saddle]

i hsd no spinlocks

at all before

[rotund furnace]

oh

[hazy saddle]

is ts better @rotund furnace

[rotund furnace]

I don't use stdatomic

or inline asm :P (cli, sti, and pause are all arch functions, (disable/enable)_interrupts and arch_pause)

[hazy saddle]

:p

Then what the fuck am I meant to do

Oberrow said to use stdatomic

[rotund furnace]

you can use stdatomic

it's just I don't

[hazy saddle]

Ah

[rotund furnace]

I use the builtin compiler intrinsics

[hazy saddle]

Well is this safe

[rotund furnace]

__atmoic_exchange_n & __atmoic_store_n

[hazy saddle]

Is there even compiler intrinsics for pause and sti/cli

[rotund furnace]

make sure you don't take an ipi while locking :P

well

it depends on how you implement ipis

my ipi code is terrible and has had deadlocks

[hazy saddle]

An IPI would be a normal maskable interupt no

And then once they are enabled

It will pop up back

[rotund furnace]

yeah

[hazy saddle]

This spinlock should be fine as long as I don’t hold it for long right

And it should work cross core

[rotund furnace]

I switched to only using async ipis because sync ipi caused too many problems

[hazy saddle]

[rotund furnace]

and since shit like forcing other cores to die and tlb flushes can be async it's fine

[hazy saddle]

Now where should I slap this lock

But this will make my kernel slower

It will no longer be lock free /s

Okay so @rotund furnace I should maybe add one to

• every VMM instance
• the PMM
• the heap
• the reaper queue

I think that’s everything that’s cross thread?

[mortal bane]

stdatomic <3

[hazy saddle]

Okay so

Shit is fucked

[mortal bane]

i can SMELL modularus is gonna do something as similar but to an extreme

and frequent level

[cyan bison]

mo

no

it should save the previous interrupt state and restore it on spinlock release

[hazy saddle]

Yeah I fixed that

Still
Fucked

Check the secret group chat

Lmfao

[wanton grove]

secret group chat 🤔

[wanton grove]

who is 4 people

[hazy saddle]

The chat was originally made because @rotund furnace Played 19 hours of rust straight

It kinda just stuck around lmao

And got renamed

A few times

[rotund furnace]

shut the fuck up

lemme use my free time how I want

[hazy saddle]

Hey I’m just sharing the backstory

Have fun playing your games

@wanton grove btw since your in here. Any idea why the fuck this is happening 💀

[wanton grove]

what you did

[hazy saddle]

https://tenor.com/view/tesla-and-einstein-luma-ai-gif-14475294631445060178

Yeah but what did I do

[wanton grove]

after what you get this

[hazy saddle]

😭

Honestly

I’m lowkey impressed it’s fucking up that bad

And adding locks made it worse

[wanton grove]

so are you running this with >1 core (with smp)

[hazy saddle]

1 core

No SMP

Just normal threading

[wanton grove]

what is tries to print this random memory

send src ill compile this

[hazy saddle]

@rotund furnace Do you have your fucked up patch

That’s based on my fucked up patch

The corrupted frame buffer is mainline dev though on the repo

and they have crashes on that

That I can’t reproduce

💀

https://git.evalyngoemer.com/evalynOS/evalynOS/src/branch/dev

LMK if you need the compiled userspace apps to drop into the initramfs folder if you don’t wana build GCC

[rotund furnace]

wdym broken patch

[wanton grove]

wait you have own .com domain

how much it costs

[rotund furnace]

there is no patch

It was off clean origin/dev

[hazy saddle]

10 15 bucks a year

They are cheap as shit if you don’t need a short or fancy one

[wanton grove]

rich

[wanton grove]

wdym

[hazy saddle]

My aunt pays it for me

[hazy saddle]

I have a userspace

And the build system compiles it

And stuff

And if you don’t wana spend 40 minutes compiling GCC and binutils

I can give pre compiled binaries

[wanton grove]

yes i dont