OBOS (not vibecoded) | Operating System Development | Page 35

flint idol Jan 19, 2025, 6:52 PM

#

ah well

flint idol Jan 19, 2025, 7:10 PM

#

why is linux sending this when I do ping on that address

#

when I get UDP reception working

#

I want to see if I can make stdout/stderr/stdin point to a "socket"

#

and access bash through that

short mortar Jan 19, 2025, 8:20 PM

#

why not shoot for ssh

flint idol Jan 19, 2025, 8:20 PM

#

TODO

frank flare Jan 19, 2025, 8:32 PM

#

ssh would be neat

#

gl

flint idol Jan 19, 2025, 8:46 PM

#

Calm down lol I don't even have tcp

#

Yet

frank flare Jan 19, 2025, 8:50 PM

#

yeah I know I've been keeping up

#

but it'd be neat still

flint idol Jan 19, 2025, 8:50 PM

#

Ye

flint idol Jan 19, 2025, 9:38 PM

#

ok so after exactly 1019 ARP packets transmitted through arping

#

the kernel kinda just disappears

#

yeah idk where the kernel went

#

didn't crash

#

it just hangs after around that many packets

#

maybe the driver has so many packets backlogged it appears to hang

#

something similar happens when I make arping run faster

#

but with much less packets

#

suspiciously

#

it stopped at 1024 packets

#

which is the maximum amount of descriptors in one ring

#

so maybe I'm running out of descriptors

#

because I am incorrectly releasing them

#

I'm testing this once again

#

but with OBOS_ENSURE(~dev->isr & RxOverflow); in r8169_rx

#

nope

#

never gets hit

#

but I still hang after ~1024 packets sent

flint idol Jan 19, 2025, 10:19 PM

#

I think I remember experiencing a deadlock a long time ago that was caused during high memory usage which I never solved

#

maybe that has something to do with this

#

yeah no

#

no immediate deadlocks found

flint idol Jan 19, 2025, 10:46 PM

#

yeah it's not a deadlock it seems

#

what if I register a fixed event handler for the power button

#

and in the handler

#

I print a stack trace

short mortar Jan 19, 2025, 10:50 PM

#

flint idol Calm down lol I don't even have tcp

tomorrow? trl

flint idol Jan 19, 2025, 10:55 PM

#

soon™️

#

I finally gave up on kate

flint idol Jan 19, 2025, 11:07 PM

#

flint idol I print a stack trace

haven't found anything

#

my stack traces aren't comprehensive enough

#

because each time it's just in CPU 0's idle thread

#

I think it's a bug with transmission

#

because data is indeed received

#

after exactly 1024 packets transmitted

#

the r8169 driver gives up

#

I think I did at least fix the memory leak

#

since the ping time stays around ~1 ms the entire time

proper trellis Jan 19, 2025, 11:41 PM

#

did you try qemu -trace?

flint idol Jan 19, 2025, 11:41 PM

#

unfortunately, real hardware does not have -trace

proper trellis Jan 19, 2025, 11:42 PM

#

is there no qemu emulation option for that dev?

flint idol Jan 19, 2025, 11:42 PM

#

not for the r8169

#

it has the r8139

#

idk how similar that is

flint idol Jan 19, 2025, 11:52 PM

#

flint idol the r8169 driver gives up

I think it was because I was accidentally clearing the EOR bit on the last descriptor KEKW

#

causing the r8169 to wander off into space

#

(whoops)

flint idol Jan 20, 2025, 1:39 AM

#

1038 packets transmitted, 1024 packets received,   1% unanswered (0 extra)
rtt min/avg/max/std-dev = 0.222/1.432/4.178/0.614 ms```
OBOS' replies to ARP seem to be pretty fast

thick jolt Jan 20, 2025, 1:51 AM

#

trl

flint idol Jan 20, 2025, 2:09 AM

#

yeah so as soon as it rolls back to the first descriptor

#

in the transmit function

#

it ceases to work

flint idol Jan 20, 2025, 2:32 AM

#

fixed the bug

#

-        desc->command = tx_frame->sz & TX_PACKET_LEN_MASK;
+        desc->command = (desc->command & EOR) | (tx_frame->sz & TX_PACKET_LEN_MASK);```

#

8816 packets transmitted, 8803 packets received,   0% unanswered (0 extra)
rtt min/avg/max/std-dev = 0.004/1.828/7.871/0.600 ms```

#

sudo arping 192.168.100.2 -i enp0s25 -W 0.004```

#

anyway: obos' stress test is now arping

#

--- 192.168.100.2 statistics ---
102 packets transmitted, 0 packets received, 100% unanswered (0 extra)```
ah shit it stopped working ![KEKW](https://cdn.discordapp.com/emojis/913708372329644042.webp?size=128 "KEKW")

timid elk Jan 20, 2025, 11:31 AM

#

802.11 when?

short mortar Jan 20, 2025, 11:46 AM

#

Nah, no one gets that right

white mulch Jan 20, 2025, 11:53 AM

#

its actually not bad, making a wifi card driver is the harder part lol

timid elk Jan 20, 2025, 11:58 AM

#

i watched the 38c3 talk on esp32 wifi liberation and it seemed like a simple driver is not that hard

#

they did it in like a year, not as a full time job but as a hobby

#

and they were reverse engineering

#

implementing a driver for a well documented one wouldn't be too hard

white mulch Jan 20, 2025, 12:23 PM

#

ig it depends on whether the card has proper specs and whatever yeah, I have only looked at some linux realtek drivers and they don't look fun to implement lol

short mortar Jan 20, 2025, 12:30 PM

#

also I'd imagine esp32's wifi is probably very simple

haughty abyss Jan 20, 2025, 12:31 PM

#

timid elk they did it in like a year, not as a full time job but as a hobby

to bring it up

#

not to do any packets

#

just poweron

#

iirc

flint idol Jan 20, 2025, 12:32 PM

#

I'll pass

haughty abyss Jan 20, 2025, 12:32 PM

#

yeah same

flint idol Jan 20, 2025, 12:33 PM

#

If I ever do that, it would probably be in a portable library

#

Because like wifi is cool

#

But not cool enough to sell my soul to

haughty abyss Jan 20, 2025, 12:34 PM

#

yeah

#

tbf esp32 is pretty hard

short mortar Jan 20, 2025, 12:39 PM

#

uWiFi?

haughty abyss Jan 20, 2025, 12:40 PM

#

u80211

timid elk Jan 20, 2025, 1:19 PM

#

haughty abyss to bring it up

no STA mode works

#

they're lacking AP and AWDL and mesh networking

#

and AWDL is an extra it's not actually supported by the proprietary driver in the SDK

flint idol Jan 20, 2025, 1:21 PM

#

timid elk they're lacking AP and AWDL and mesh networking

Idk what this all means

timid elk Jan 20, 2025, 1:22 PM

#

short mortar also I'd imagine esp32's wifi is probably very simple

it's "simple" but needs to be reverse engineered and almost entirely undocumented

short mortar Jan 20, 2025, 1:22 PM

#

as all wifi controllers

timid elk Jan 20, 2025, 1:22 PM

#

they only had function names because expressif didn't strip it

flint idol Jan 20, 2025, 1:23 PM

#

timid elk they only had function names because expressif didn't strip it

That's like saying fuck you you're not getting this code in a whole other way

timid elk Jan 20, 2025, 1:23 PM

#

flint idol Idk what this all means

STA is station mode, AP is access point mode, mesh networking is mesh networking and AWDL is AirDrop, AirPlay, etc.

timid elk Jan 20, 2025, 1:23 PM

#

flint idol That's like saying fuck you you're not getting this code in a whole other way

I mean apparently they're extremely cool about it and they would release the source if not under NDA

#

the binaries are apache licensed

#

oh and they don't have the esp32-specific long range wifi thingy (which is very cool because 1km wifi!!)

#

intel or someone else wouldn't be this cool about it

#

wifi and bt are pretty much the only closed source thingies in the esp32 sdk

#

because they're not espressif's and it's licensed from riviera waves i think

flint idol Jan 21, 2025, 3:02 PM

#

After school, I hope to implement IPv4 reception

#

And UDP reception

#

ARP is cool n shit but not quite enough

flint idol Jan 21, 2025, 3:04 PM

#

flint idol After school, I hope to implement IPv4 reception

And also ICMP to go with that, as the two seem related

#

Also iirc IPv4 also requires packet forwarding

haughty abyss Jan 21, 2025, 3:04 PM

#

do you have arp?

flint idol Jan 21, 2025, 3:04 PM

#

Yes

haughty abyss Jan 21, 2025, 3:04 PM

#

icmp is separate from ipv4 iirc

flint idol Jan 21, 2025, 3:04 PM

#

Yeah but ICMP is for the error reporting

#

I think

haughty abyss Jan 21, 2025, 3:05 PM

#

you do not need it

flint idol Jan 21, 2025, 3:05 PM

#

Okay

haughty abyss Jan 21, 2025, 3:05 PM

#

ngl i'd do ipv6 first, because its the cooler thing

#

and also modern and cool

flint idol Jan 21, 2025, 3:05 PM

#

I want to do that later

haughty abyss Jan 21, 2025, 3:05 PM

#

ipv6 is pretty simple ime

flint idol Jan 21, 2025, 3:05 PM

#

I will do it, don't worry

haughty abyss Jan 21, 2025, 3:05 PM

#

cool

flint idol Jan 21, 2025, 3:06 PM

#

I also need to see if a bug fix for my r8169 driver worked

#

Basically just a deadlock

#

#

#

Also I made a mistake saying the kernel dispatches DPCs when lowering to dispatch IRQL or lower

#

I meant to say IRQL passive

flint idol Jan 21, 2025, 3:59 PM

#

I need to fix my pagecache soon

#

It's shitty

#

None of the pages ever really go on standby/dirty

#

And speaking of that

#

I need to fix that too

flint idol Jan 21, 2025, 7:41 PM

#

I'm going to revive the obos m68k port today

#

Uhh so time to build gcc

#

For the billionth time this year

real pecan Jan 21, 2025, 7:48 PM

#

astral lore

flint idol Jan 21, 2025, 7:50 PM

#

flint idol I'm going to revive the obos m68k port today

I will also see if I can get mlibc ported too

flint idol Jan 21, 2025, 7:52 PM

#

flint idol I'm going to revive the obos m68k port today

It should hopefully be straightforward*
*obos m68k last compiled 5-6 months ago

#

Waiting for gcc to compile:

#

https://tenor.com/view/mr-bean-waiting-still-waiting-gif-13052487

Tenor

devout niche Jan 21, 2025, 7:56 PM

#

flint idol I will also see if I can get mlibc ported too

should be doable

#

mlibc/m68k-linux is fully test passing

flint idol Jan 21, 2025, 7:57 PM

#

I only need the crt begin I assume?

#

And the other crt*.S files

#

Because my syscalls have the same numbers across architectures

#

I would just need to implement a syscall trap handler

#

iirc the trap instruction on m68k is for syscalls

frank flare Jan 21, 2025, 8:00 PM

#

flint idol https://tenor.com/view/mr-bean-waiting-still-waiting-gif-13052487

me when trying to build astral

#

I didn't even know I had "so little" disk space until then

flint idol Jan 21, 2025, 8:02 PM

#

An astral minimal build for me took 16G

frank flare Jan 21, 2025, 8:17 PM

#

and how long did it take?

#

cause I made a PR for it and I thought to test it before publishing the PR
and I let it build for ~1h and it still wasn't finished and I said aight, let's make the PR anyways

flint idol Jan 21, 2025, 9:49 PM

#

ah damn it

#

obos m68k crashed first thing

#

wait not first thing

#

it got to vfs initialization

flint idol Jan 21, 2025, 11:00 PM

#

bruh what the fuck

#

Mm_VirtualMemoryAlloc had the brilliant idea of overwriting a kernel code mapping

#

#

[  LOG  ] Arch_KernelEntry: Done early boot.
[  LOG  ] Currently at 8796 KiB of committed memory (36 KiB pageable), 0 KiB paged out, and 8760 KiB non-paged, and 0 KiB uncommitted. Page faulted 47 times (0 hard, 47 soft).```

#

got the m68k port to boot again

#

!!

#

ngl the m68k port is so boring on the qemu virt machine

#

no devices to support

#

no PCI

#

not even ACPI

#

and the audacity for it to use that much memory

#

but still do nothing

#

[ DEBUG ] Arch_KernelEntry: Initializing allocator.
[ DEBUG ] Arch_KernelEntry: Initialize kernel process.
[ DEBUG ] Arch_KernelEntry: Initializing IRQ interface.
[ DEBUG ] Arch_KernelEntry: Initializing VMM.
[  LOG  ] Initialized MM.
Working set capacity: 4096 KiB.
4 pageable pages.
2187 committed pages.
Using AGING PRA.
[ DEBUG ] Arch_KernelEntry: Initializing timer interface.
[ DEBUG ] Arch_KernelEntry: Initializing scheduler timer.
[ DEBUG ] Arch_KernelEntry: Loading kernel symbol table.
[ DEBUG ] Arch_KernelEntry: Loading InitRD driver.
[  LOG  ] Drv_LoadDriver: Loaded driver 'Initial Ramdisk (InitRD) Driver' at 0xc08ba000.
[ DEBUG ] Arch_KernelEntry: Initializing VFS.
[  LOG  ] Arch_KernelEntry: Done early boot.
[  LOG  ] Currently at 8796 KiB of committed memory (36 KiB pageable), 0 KiB paged out, and 8760 KiB non-paged, and 0 KiB uncommitted. Page faulted 49 times (0 hard, 49 soft).```

#

8748 KiB of that memory is used by the kernel binary itself

#

ok great

#

with optimizations enabled

#

/home/oberrow/Code/obos-pkgs-m68k/pkgs/lib/gcc/m68k-obos/14.2.0/../../../../m68k-obos/bin/ld: src/oboskrnl/CMakeFiles/oboskrnl.dir/vfs/pipe.c.obj: in function `read_sync':
pipe.c:(.text+0x7c): undefined reference to `__atomic_load_8'

#

that happens

#

maybe it helps to include stdatomic KEKW

#

nvm it doesn't

#

ah great

#

it doesn't even use stdatomic

#

I decided to change a member of a struct to _Atomic(uoff_t) to _Atomic(uintptr_t)

#

and it compiles

#

ok it boots

#

time to see if x86_64 isn't broken

flint idol Jan 21, 2025, 11:23 PM

#

flint idol ``` /home/oberrow/Code/obos-pkgs-m68k/pkgs/lib/gcc/m68k-obos/14.2.0/../../../../...

wait I actually think I know the problem

#

linking with libatomic doesn't work

#

I want to implement execve for m68k next

#

and the hand off to init function

#

then I will merge the branch

flint idol Jan 21, 2025, 11:28 PM

#

flint idol I want to implement execve for m68k next

surely that won't be too hard

#

it's not even that I need execve

#

I just need to code to hand off to a process after exexce

#

OBOS_NORETURN OBOS_WEAK void OBOSS_HandControlTo(struct context* ctx, struct exec_aux_values* aux);
OBOS_NORETURN OBOS_WEAK void OBOSS_HandOffToInit(struct exec_aux_values* info);

#

the latter might as well be implemented in a non-architecture specific way

#

void OBOSS_HandOffToInit(struct exec_aux_values* aux)
{
    OBOS_OpenStandardFDs(OBOS_CurrentHandleTable());

    irql oldIrql = Core_RaiseIrql(IRQL_DISPATCH);
    OBOS_UNUSED(oldIrql);
    context* ctx = CoreS_GetCPULocalPtr()->currentContext;

    Core_GetCurrentThread()->context.stackBase = Mm_VirtualMemoryAlloc(ctx, nullptr, 4*1024*1024, OBOS_PROTECTION_USER_PAGE, VMA_FLAGS_GUARD_PAGE, nullptr, nullptr);
    Core_GetCurrentThread()->context.stackSize = 4*1024*1024;

    CoreS_SetThreadPageTable(&Core_GetCurrentThread()->context, ctx->pt);

    OBOSS_HandControlTo(ctx, aux);
}```

#

x86_64 does this

#

all of those APIs except for the stack things

#

are not x86-64 specific

#

then I will also implement syscall traps

#

and then I will merge

#

that branch

#

and continue on my network stack

flint idol Jan 22, 2025, 1:04 AM

#

ah damn it I also need to add signal support to the m68k port

#

and a TLS syscall

#

For the subset of the floating-point instructions that generate exception traps, the FPU
loads the 32-bit FPIAR with the logical address of the instruction before executing the
instruction. Because the IU can execute instructions while the FPU executes floating-point
instructions and, the FPU can concurrently execute two floating-point instructions the PC
value stacked by the MC68040 in response to a floating-point exception handler cannot
point to the offending instruction.

#

cursed

#

it seems like I need to manually save FPE registers on the m68040

#

https://github.com/Keyronex/Keyronex/blob/master/kernel/kern/m68k/thread.c#L28-L32

GitHub

Keyronex/kernel/kern/m68k/thread.c at master · Keyronex/Keyronex

The Keyronex Operating System. Contribute to Keyronex/Keyronex development by creating an account on GitHub.

#

bruh

thick jolt Jan 22, 2025, 1:22 AM

#

obos making good progress

#

m68k funny archiecture

flint idol Jan 22, 2025, 1:24 AM

#

I forgor how m68k assembly worked

#

if only I commented this code...

thick jolt Jan 22, 2025, 1:25 AM

#

bro

#

bro writes a port

#

to m68k

#

then forgets how the assembly works

#

💀

flint idol Jan 22, 2025, 1:25 AM

#

I wrote this port 6 months ago

thick jolt Jan 22, 2025, 1:25 AM

#

why weren't you working on it along side x86

#

instead of just

#

forgetting about it

flint idol Jan 22, 2025, 1:25 AM

#

flint idol I wrote this port 6 months ago

*4 motnhs

flint idol Jan 22, 2025, 1:25 AM

#

flint idol *4 motnhs

*months

flint idol Jan 22, 2025, 1:25 AM

#

thick jolt why weren't you working on it along side x86

because

#

at one point I broke a bunch of stuff

#

didn't want to look at m68k docs

#

https://github.com/OBOS-dev/obos/pull/18

GitHub

m68k port by oberrow · Pull Request #18 · OBOS-dev/obos

Added support for the Motorola 68040 CPU, on the qemu virt platform.
This should not break compatibility with x86-64.

thick jolt Jan 22, 2025, 1:26 AM

#

fair

flint idol Jan 22, 2025, 1:26 AM

#

move.l (+4,%sp), %a0```

#

what dose htis does

#

moves

#

move source, destination

#

I think was the syntax

#

a0 = *(sp+4)

#

ah yes

thick jolt Jan 22, 2025, 1:30 AM

#

ah yes

#

chad syntax

#

okay but

#

source, destination

#

😡

#

destionation, source

#

chad

frank flare Jan 22, 2025, 5:24 AM

#

I think of src, dest as "put src in dest" and dest, src as "dest = src"

src, dest makes more sense because if I take a mov %rax, %rbx for example, it comes natural to say it means "move rax to rbx"

so src, dest is superior

#

jk it's just an opinion

frank flare Jan 22, 2025, 5:51 AM

#

I think Intel tried to give that feel of a higher level lang where you do a = b; to put b in a

main girder Jan 22, 2025, 6:15 AM

#

frank flare I think Intel tried to give that feel of a higher level lang where you do a = b;...

or

#

yk

#

math

flint idol Jan 22, 2025, 1:13 PM

#

frank flare I think of src, dest as "put src in dest" and dest, src as "dest = src" src, d...

you're all brainwashed into thinking that anything related to unix and GNU is good, when i see AT&T assembly i start vomiting and that effects my mental & physical health. Have you ever thought about that, dummy? like dude who thought "ah yes we will put a % before the register name" this is so fucking retarded man. I dont know how you people live with this, I would rather write userland programs for windows than use AT&T syntax for an hour. You can also shove your q,l,w,b suffixes into your asses. Nobody can be more retarded than the guy who thought movq was okay. Also shove your $ up your ass. That thing used to make me vomit, man. Imagine having to put $ for immediates. This is like the most retarded thing ever invented, i dont get why people use this syntax. intel syntax is waaay better. instead of littering the world with your fucked up assembly syntax you should go learn intel syntax assembly. i want to fucking die when i see AT&T assembly man. Like one time i was trying my best to read some AT&T assembly and it took me 4 days to figure out how jmp worked, man. 4 days man. I don't get what's wrong with you people, you keep saying "belwewbewb at&t syntax good barwebqrebrewr" and it's so fucking annoying. When you say that i want to gouge your eyes out and put them in front of the intel assembly so you can stare at it and understand its beauty. Oh btw no one fucking cares about your random .cfi_def_cfa_offset or your fucking section ".note.GNU-stack","",@progbits like dude nobody fucking cares if you want to do shit that nobody understands.

#

just do it the normal way, will ya? if we were in an arena i would beat the piss out of y'all because of how annoying you all are. the worst part is you fucking got your shitty asssembly syntax into inline assembly like dude wtf the syntax already sucks why make it suck more?? Every time i see inline assembly i cringe because of the fucked up AT&T syntax thing you managed to put there. I dont even know how you read your assembly and actually be able to understand it. You need to win the "person with the strongest stomach" award man. I dont last five minutes when i look at your fucking assembly syntax. Now go use intel assembly or shove your AT&T up your ass. Peace.

#

Anyway, after these network stack shenanigans and m68k stuff

#

I might port ncurses

#

Or curses idk what the difference is

#

Maybe one makes obos cursed and the other doesn't

leaden carbon Jan 22, 2025, 1:21 PM

#

at the end of the day it all sucks

#

I think intel syntax is easier to read and write but its slightly less explicit

short mortar Jan 22, 2025, 1:28 PM

#

I've just been taught AT&T at university, and don't know Intel syntax

#

So to me people unsing Intel are anoying

leaden carbon Jan 22, 2025, 1:29 PM

#

I write at&t inline and wanted to drop the extra nasm dependency so just rewrote my asm in at&t

flint idol Jan 22, 2025, 1:29 PM

#

Bruh my thread

#

Is gonna get taken over

leaden carbon Jan 22, 2025, 1:29 PM

#

for my bootloader I kept using nasm cuz too much assembly

flint idol Jan 22, 2025, 1:29 PM

#

flint idol ``` move.l (+4,%sp), %a0```

Note to self: never comment on assembly syntax in your progress thread

#

You guys can go to uacpi thread aka lounge 2

short mortar Jan 22, 2025, 1:31 PM

#

flint idol Note to self: never comment on assembly syntax in your progress thread

You've got to compete with Nyaux thread somehow

flint idol Jan 22, 2025, 1:31 PM

#

No

#

That's 12k messages

#

I'd need crazy progress on obos

short mortar Jan 22, 2025, 1:31 PM

#

Couple of posts about assembly

#

C vs C++ vs Rust

flint idol Jan 22, 2025, 1:32 PM

#

C

#

Argument done

short mortar Jan 22, 2025, 1:32 PM

#

risc-v is gonna take over

flint idol Jan 22, 2025, 1:32 PM

#

limine opinion

short mortar Jan 22, 2025, 1:32 PM

#

gcc vs clang vs msvc

flint idol Jan 22, 2025, 1:32 PM

#

Clang

short mortar Jan 22, 2025, 1:33 PM

#

flint idol Clang

at least I agree on that

flint idol Jan 22, 2025, 1:33 PM

#

Although I am scared of compiling llvm

short mortar Jan 22, 2025, 1:33 PM

#

why

#

it's fine

flint idol Jan 22, 2025, 1:33 PM

#

https://tenor.com/view/laptop-smoking-fire-burning-lag-gif-19373925

Tenor

#

Also I don't feel like making an os specific tool chain for clang/llvm

short mortar Jan 22, 2025, 1:34 PM

#

it's ez

flint idol Jan 22, 2025, 1:34 PM

#

Yeah but I don't care enough

short mortar Jan 22, 2025, 1:34 PM

#

and much better than gcc madness

#

/binutils

flint idol Jan 22, 2025, 1:35 PM

#

Idk porting gcc and binutils wasn't too hard

#

Just needed to compile gcc 10+ times

#

In a day

short mortar Jan 22, 2025, 1:35 PM

#

I've ported libgo...

#

Once you want more than just c/c++, it's getting obnoxious

short mortar Jan 22, 2025, 1:36 PM

#

flint idol Just needed to compile gcc 10+ times

At some point I gave up and bought 64GB of RAM

frank flare Jan 22, 2025, 1:38 PM

#

flint idol you're all brainwashed into thinking that anything related to unix and GNU is go...

I said to myself that I'll take a break from homework to relax, only to come up to this tuxdrat

#

lemme actually read it

leaden carbon Jan 22, 2025, 1:40 PM

#

not sure u wanna read it trl

frank flare Jan 22, 2025, 1:41 PM

#

I would rather write userland programs for windows than use AT&T syntax for an hour
the syntax already sucks why make it suck more??
comedy gold

frank flare Jan 22, 2025, 1:41 PM

#

leaden carbon not sure u wanna read it<:trl:829316641191886888>

nah I got a decent laugh out of it

flint idol Jan 22, 2025, 1:41 PM

#

Source is dreamingwarlord

#

Used to be on this server

#

He left though

frank flare Jan 22, 2025, 1:42 PM

#

so is it copypasta

flint idol Jan 22, 2025, 1:42 PM

#

Yes

flint idol Jan 22, 2025, 1:42 PM

#

flint idol Source is dreamingwarlord

He can be found in the mintia project server

frank flare Jan 22, 2025, 1:43 PM

#

honestly I agree about $ and % being stupid
I just like src, dest so I tolerate all of the rest chad

flint idol Jan 22, 2025, 1:43 PM

#

noprefix

#

.att_syntax noprefix

#

trl

frank flare Jan 22, 2025, 1:44 PM

#

well yea

short mortar Jan 22, 2025, 1:53 PM

#

frank flare honestly I agree about $ and % being stupid I just like src, dest so I tolerate ...

nah, you get used to it after a couple of page faults because of movl 0, %eax

frank flare Jan 22, 2025, 2:20 PM

#

I had add %rsp, $16 which is even worse

timid elk Jan 22, 2025, 2:53 PM

#

flint idol Just needed to compile gcc 10+ times

i think my record is like 16 gcc compiles in a day

#

including failures tho

#

and i wasn't even doing proper osdev, i was just working on mlibc in nixpkgs

#

and yes mlibc in nixpkgs is almost ready it just needs some cleanup and reviewing

#

and a couple bugs in mlibc to be squashed

white mulch Jan 22, 2025, 2:56 PM

#

my record is like 5 llvm + clang + libc++ compiles yesterday KEKW only two of them would actually have been required but I were trying to compile an aarch64 toolchain + libc++ using the same sysroot folder as what I used for x86_64 and that created weird errors that I tried to resolve in different ways until I realized that

celest acorn Jan 22, 2025, 3:09 PM

#

flint idol Also I don't feel like making an os specific tool chain for clang/llvm

the only benefit of porting llvm is llvm-pipe really

frigid tide Jan 22, 2025, 3:53 PM

#

flint idol you're all brainwashed into thinking that anything related to unix and GNU is go...

yeah what the fuck is atnt syntax it gives me more cancer than windows

#

fuck intel but damn

frank flare Jan 22, 2025, 4:36 PM

#

jmp *%rax trl

timid elk Jan 22, 2025, 5:46 PM

#

white mulch my record is like 5 llvm + clang + libc++ compiles yesterday <:KEKW:913708372329...

llvm is just another level

#

once i was compiling llvm under wsl i think

#

it took around 16 hours or so

white mulch Jan 22, 2025, 5:49 PM

#

lol

#

at least for me its not that bad its like 20 min or whatever

timid elk Jan 22, 2025, 6:12 PM

#

yea 20 mins is gcc for me

sharp pike Jan 22, 2025, 6:30 PM

#

bro get a better machine lmao

#

smhmh

timid elk Jan 22, 2025, 6:34 PM

#

this is the best machine i ever touched 6 core ryzen laptop

sharp pike Jan 22, 2025, 6:36 PM

#

ayymd

timid elk Jan 22, 2025, 6:36 PM

#

what does ayymd mean

sharp pike Jan 22, 2025, 6:41 PM

#

ayymd means ayymd

#

https://tenor.com/bDIjl.gif

Tenor

timid elk Jan 22, 2025, 7:13 PM

#

https://tenor.com/view/lisa-su-リサ-amd-radeon-gpu-gif-21819113

Tenor

flint idol Jan 22, 2025, 8:03 PM

#

Bah stupid debian maintainers

#

Changed repositories to require signing

flint idol Jan 22, 2025, 8:30 PM

#

?

#

Do you mean the dumb hexdump clone?

#

Ah that

#

It's called speedcrunch if you didn't know trl

flint idol Jan 22, 2025, 9:44 PM

#


apt-listchanges: News
---------------------

apt (2.9.24) unstable; urgency=medium

  /etc/apt/trusted.gpg is no longer trusted. Setting the Dir::Etc::trusted
  option manually continues to work for some more time.

  sources.list(5) entries without the Signed-By field are deprecated;
  migrate any legacy entries to the deb822 .sources format. See the
  apt-secure(7) manual page for best practices for signer configuration.
  This deprecates the /etc/apt/trusted.gpg.d directory.

 -- Julian Andres Klode <[email protected]>  Tue, 21 Jan 2025 12:17:36 +0100

#

fuck u

flint idol Jan 22, 2025, 10:24 PM

#

after spending way too much time figuring that out

#

I will continue

#

my shenanigans

thick jolt Jan 23, 2025, 6:27 AM

#

flint idol you're all brainwashed into thinking that anything related to unix and GNU is go...

W take

#

tho u sound a big crazy and u sound like ur losing it a little

thick jolt Jan 23, 2025, 6:28 AM

#

flint idol Source is dreamingwarlord

ohhh

#

i thought u were going crazy

#

💀

frank flare Jan 23, 2025, 7:29 AM

#

actually I remember dreamingwarlord from a few years ago

#

but it seems like they got banned lol

ornate ginkgo Jan 23, 2025, 7:30 AM

#

nah there was a mass unbanning a while ago

#

they rejoined but decided not to stay

flint idol Jan 23, 2025, 1:51 PM

#

I want to change my vfs mount code

#

It's quite inefficient

#

Or maybe the problem is my fat fs code

#

Either way it's very slow

main girder Jan 23, 2025, 1:55 PM

#

what is it doing

flint idol Jan 23, 2025, 1:56 PM

#

Slow shit

#

Also my pagecache is le dogshite

#

I need to fix that

#

In a hurry*

#

*sometime in the next 4 months

flint idol Jan 23, 2025, 1:58 PM

#

flint idol Also my pagecache is le dogshite

I made it when I didn't understand what standby and dirty lists did

flint idol Jan 23, 2025, 8:28 PM

#

According to obos

#

It's 2021

#

I made CMOS code

honest citrus Jan 23, 2025, 9:22 PM

#

flint idol you're all brainwashed into thinking that anything related to unix and GNU is go...

L take
I could say the same about Intel syntax, all of the things you're complaining about make it so much more readable and obvious what its actually doing

flint idol Jan 23, 2025, 9:27 PM

#

Omfg

#

This conversation ended

#

1 day ago

flint idol Jan 23, 2025, 9:29 PM

#

flint idol It's 2021

Idk why it thinks it's 2021

#

Kinda weird

short mortar Jan 23, 2025, 9:30 PM

#

probably memory corruption KEKW

flint idol Jan 23, 2025, 9:30 PM

#

No

#

Considering all other fields were correct

#

The struct was a stack buffer

#

And stuff

#

Also 2021 is way to close to 2025

#

For this to be that

flint idol Jan 23, 2025, 10:10 PM

#

It seems as if when I run obos on my laptop

#

The same thing happens

flint idol Jan 23, 2025, 10:36 PM

#

-    time->year = read_cmos8(CMOS_REGISTER_YEAR);
+    time->year = read_cmos8(CMOS_REGISTER_YEAR) + 4;```

#

trl

#

k just merged that branch

frank flare Jan 24, 2025, 9:29 AM

#

flint idol ```diff - time->year = read_cmos8(CMOS_REGISTER_YEAR); + time->year = read...

trolland

thick jolt Jan 24, 2025, 9:09 PM

#

flint idol ```diff - time->year = read_cmos8(CMOS_REGISTER_YEAR); + time->year = read...

wait u can read the cmos???

#

HOW

flint idol Jan 24, 2025, 9:09 PM

#

For the time

#

Yes

white mulch Jan 24, 2025, 9:09 PM

#

@flint idol I tested obos on the laptop, it looks to at least somewhat work as it gets stuck after it says "Could not make device wake capable" and two uacpi reads

flint idol Jan 24, 2025, 9:09 PM

#

thick jolt HOW

Using port io

flint idol Jan 24, 2025, 9:10 PM

#

white mulch <@841762704766140486> I tested obos on the laptop, it looks to at least somewhat...

There are many ISOs

#

Here

#

One of them hangs after uacpi init

#

Iirc

thick jolt Jan 24, 2025, 9:10 PM

#

flint idol Using port io

is there a cmos on every archiecture or just x86

flint idol Jan 24, 2025, 9:10 PM

#

thick jolt is there a cmos on every archiecture or just x86

Idk

thick jolt Jan 24, 2025, 9:10 PM

#

flint idol Idk

fair ill just keep it on x86

white mulch Jan 24, 2025, 9:10 PM

#

no there isn't on all architectures

thick jolt Jan 24, 2025, 9:10 PM

#

i see

flint idol Jan 24, 2025, 9:11 PM

#

white mulch <@841762704766140486> I tested obos on the laptop, it looks to at least somewhat...

What do you mean by uacpi reads

white mulch Jan 24, 2025, 9:11 PM

#

it just had two uacpi trace lines for some reads after that and nothing more

flint idol Jan 24, 2025, 9:11 PM

#

Did you find this iso in the uacpi thread?

white mulch Jan 24, 2025, 9:12 PM

#

and this iso was some that I downloaded like 10 days ago when I tested it on the other laptop where it triple faulted

flint idol Jan 24, 2025, 9:12 PM

#

I am in the process of fixing a bug with the IOAPIC

#

When I fix that bug I can send you an iso

#

Because that is the only thing that I know that could cause obos to hang in uacpi init

#

The error you got from the wake gpe was just firmware devs being too stoobid for obos to handle

#

Out of curiosity what was the status it printed

#

@white mulch

white mulch Jan 24, 2025, 9:15 PM

#

rn_image_picker_lib_temp_fc0386e2-695d-4006-9df0-617d50209835.jpg

flint idol Jan 24, 2025, 9:16 PM

#

Status mismatch which is interesting

#

It means the _S3D and _S3W values contradicted each other

#

    if ((snd != UINT64_MAX && snw != UINT64_MAX) && snw >= snd)
    {
        if (status)
            *status = OBOS_STATUS_MISMATCH;
        return DSTATE_INVALID;
    }```

flint idol Jan 25, 2025, 12:14 AM

#

I'm implementing a gettimeofday syscall for obos

#

I'm definitely not stealing anything from mlibc

#

static int days_from_civil(int y, unsigned m, unsigned d) noexcept {
    y -= m <= 2;
    const int era = (y >= 0 ? y : y - 399) / 400;
    const unsigned yoe = static_cast<unsigned>(y - era * 400); // [0, 399]
    const unsigned doy = (153 * (m > 2 ? m - 3 : m + 9) + 2) / 5 + d - 1; // [0, 365]
    const unsigned doe = yoe * 365 + yoe / 4 - yoe / 100 + doy; // [0, 146096]
    return era * 146097 + static_cast<int>(doe) - 719468;
}```

#

trl

flint idol Jan 25, 2025, 1:29 AM

#

[23/37] Linking C executable /home/oberrow/Code/obos-patch/out/powerctl
FAILED: /home/oberrow/Code/obos-patch/out/powerctl 
: && /home/oberrow/Code/obos-pkgs/pkgs/bin/x86_64-obos-gcc -O2 -g -DNDEBUG -fsanitize=undefined src/user_examples/CMakeFiles/powerctl.dir/powerctl_example.c.obj -o /home/oberrow/Code/obos-patch/out/powerctl   && :
/home/oberrow/Code/obos-pkgs/pkgs/lib/gcc/x86_64-obos/14.2.0/../../../../x86_64-obos/bin/ld: cannot find -lubsan: No such file or directory

#

bruh

flint idol Jan 25, 2025, 2:05 AM

#

flint idol Jan 25, 2025, 2:45 AM

#

bruh the IOAPIC is being stupid

#

it registers the GSI for GPEs as level triggered and active high

#

as the MADT entry says GSI9 (the one for the GPE) should be

#

it hangs as if the opposite was set

#

the values I set are correct in my ioapic code

#

as in, I don't set the registers to anything invalid

uncut narwhal Jan 25, 2025, 4:10 AM

#

there isn't a dumb isa irq redirect is there?

#

or some other thing changing it like that

flint idol Jan 25, 2025, 4:11 AM

#

uncut narwhal there isn't a dumb isa irq redirect is there?

I print the values after processing the Isa redirect stuff

#

In the madt

uncut narwhal Jan 25, 2025, 4:11 AM

#

ok cool

#

and you set interrupt mode in uacpi? other than that I got nothing

flint idol Jan 25, 2025, 4:12 AM

#

I think so

#

Wait I do that after gpe finalize

#

Hmm

uncut narwhal Jan 25, 2025, 4:13 AM

#

gpe finalize is when it hangs right?

#

if so maybe that's it

#

idk when it's supposed to be called

flint idol Jan 25, 2025, 4:17 AM

#

uncut narwhal gpe finalize is when it hangs right?

Ye

real pecan Jan 25, 2025, 9:36 AM

#

flint idol Wait I do that after gpe finalize

Wtf

#

Irq installation happens during early init

#

By that point you must have configured overrides

real pecan Jan 25, 2025, 9:37 AM

#

real pecan Irq installation happens during early init

Or well, during uacpi_initialize I think

flint idol Jan 25, 2025, 11:59 AM

#

real pecan Wtf

Overrides are configured

#

The acpl interrupt model thing isn't set until then

real pecan Jan 25, 2025, 12:00 PM

#

yeah but _PIC configures the firmware to actually do the thing

flint idol Jan 25, 2025, 12:00 PM

#

*acpi

real pecan Jan 25, 2025, 12:00 PM

#

so that could be why it hangs

#

either that or

#

your redirections are not very good

flint idol Jan 25, 2025, 1:05 PM

#

bruh

#

it still hangs

real pecan Jan 25, 2025, 1:06 PM

#

try applying different polarity and triggering

#

just willy nilly

#

whichever one works

flint idol Jan 25, 2025, 1:07 PM

#

that feels wrong

#

just setting random polarity/trigger

real pecan Jan 25, 2025, 1:07 PM

#

if that works we'll know ur just not parsing it correctly

flint idol Jan 25, 2025, 1:07 PM

#

I made sure I am

#

using logs

real pecan Jan 25, 2025, 1:07 PM

#

yes but

#

who knows what other bugs u have that might mess with that

flint idol Jan 25, 2025, 1:08 PM

#

the log is right before I set the polarity/trigger mdoe

#

*mode

#

in the I/O APIC code

#

    // Set vector info.
    if (vector)
    {
        printf("GSI %d is of polarity %s and trigger mode %s\n", gsi, polarity == PolarityActiveHigh ? "active-high" : "active-low", tm == TriggerModeEdgeSensitive ? "edge-sensitive" : "level-sensitive");
        entry |= (vector&0xff);
        set_bit_to(&entry, 13, polarity);
        set_bit_to(&entry, 15, tm);
    }```

real pecan Jan 25, 2025, 1:08 PM

#

idk if this bit setting is correct too

#

try inverting just for fun

flint idol Jan 25, 2025, 1:09 PM

#

before that, it sets polarity and tm are set to the proper values in the I/O APIC code

flint idol Jan 25, 2025, 1:09 PM

#

real pecan idk if this bit setting is correct too

static void set_bit_to(uint64_t* val, uint8_t bit, bool to)
{
    if (to)
        *val |= BIT(bit);
    else
        *val &= ~BIT(bit);
}```

real pecan Jan 25, 2025, 1:09 PM

#

no like

flint idol Jan 25, 2025, 1:09 PM

#

the value of the bits

real pecan Jan 25, 2025, 1:09 PM

#

if polarity == PolarityActiveHigh is what it will be seen as by ioapic

flint idol Jan 25, 2025, 1:10 PM

#

typedef enum ioapic_trigger_mode
{
    TriggerModeEdgeSensitive = 0,
    TriggerModeLevelSensitive = 1,
} ioapic_trigger_mode;
typedef enum ioapic_polarity
{
    PolarityActiveHigh,
    PolarityActiveLow,
} ioapic_polarity;```

real pecan Jan 25, 2025, 1:10 PM

#

the first entry of an enum is not guaranteed to be 0

#

unless u make it so

flint idol Jan 25, 2025, 1:11 PM

#

I highly doubt that has anything to do with this but ok I'll try

real pecan Jan 25, 2025, 1:11 PM

#

nah i dont think so as well

#

but better safe than sorry

flint idol Jan 25, 2025, 1:11 PM

#

I mean if clangd says so

#

OBOS_STATIC_ASSERT(PolarityActiveHigh == 0, "tf?");

this static assertion doesn't fail

#

and when I check in qemu with info pic

#

the values match

real pecan Jan 25, 2025, 1:15 PM

#

yeah idk

#

but u have the same hang on multiple pieces of hw

#

where e.g. pmos doesnt hang

#

so that makes me suspicious

flint idol Jan 25, 2025, 1:16 PM

#

maybe my ioapic code is good, but my ISR is somehow bad

real pecan Jan 25, 2025, 1:16 PM

#

maybe but other interrupts work

flint idol Jan 25, 2025, 1:21 PM

#

well that's weird

#

I only ever get one GPE IRQ

real pecan Jan 25, 2025, 1:22 PM

#

in qemu?

flint idol Jan 25, 2025, 1:22 PM

#

no, on real hw

#

before it hangs

real pecan Jan 25, 2025, 1:22 PM

#

didnt it hang before u could get even 1

flint idol Jan 25, 2025, 1:23 PM

#

no

#

it hung right after a specific GPE was enabled

real pecan Jan 25, 2025, 1:23 PM

#

what did you change?

flint idol Jan 25, 2025, 1:23 PM

#

nothing

real pecan Jan 25, 2025, 1:24 PM

#

im confused

flint idol Jan 25, 2025, 1:24 PM

#

same

real pecan Jan 25, 2025, 1:24 PM

#

how do u know u get a gpe irq

flint idol Jan 25, 2025, 1:24 PM

#

logs

real pecan Jan 25, 2025, 1:24 PM

#

so u get a gpe irq

flint idol Jan 25, 2025, 1:24 PM

#

yes

real pecan Jan 25, 2025, 1:24 PM

#

then u enable this other gpe

#

and it hangs?

flint idol Jan 25, 2025, 1:25 PM

#

GPE enabled
GPE IRQ
hang

real pecan Jan 25, 2025, 1:25 PM

#

wait

#

so u do get it before hanging?

flint idol Jan 25, 2025, 1:25 PM

#

yes

real pecan Jan 25, 2025, 1:25 PM

#

previously u said u didnt

#

then just log where exactly it hangs in the isr

flint idol Jan 25, 2025, 1:26 PM

#

Ye

#

I think it's a recursive lock

#

in the ISR handler

real pecan Jan 25, 2025, 1:38 PM

#

of what

flint idol Jan 25, 2025, 1:38 PM

#

some random ass spinlock idk

real pecan Jan 25, 2025, 1:38 PM

#

bruh

flint idol Jan 25, 2025, 1:38 PM

#

I added some logs related to that

#

to confirm

#

it's a spinlock which is allocated on the kernel heap

#

so probably uacpi's

#

but it might not be

real pecan Jan 25, 2025, 1:39 PM

#

write_gpe happens under a spinlock

#

so if you're calling into uacpi with uacpi spinlocks held its your bug

flint idol Jan 25, 2025, 1:40 PM

#

well it's a fucking GPE

real pecan Jan 25, 2025, 1:40 PM

#

spinlocks are supposed to disable interrutps

flint idol Jan 25, 2025, 1:40 PM

#

I need to call into uacpi don't I

flint idol Jan 25, 2025, 1:40 PM

#

real pecan spinlocks are supposed to disable interrutps

well for me, they raise IRQL to IRQL_DISPATCH

#

which is < IRQL_GPE

real pecan Jan 25, 2025, 1:40 PM

#

what does this mean

flint idol Jan 25, 2025, 1:41 PM

#

you can't get preempted by the scheduler with a spinlock held

real pecan Jan 25, 2025, 1:41 PM

#

your irq handler can call into uacpi even when uacpi locks a spinlock?

#

thats a bug ofc

flint idol Jan 25, 2025, 1:41 PM

#

flint idol you can't get preempted by the scheduler with a spinlock held

but you can get interrupted by GPEs for example

#

since IRQL_GPE > IRQL_DISPATCH

real pecan Jan 25, 2025, 1:41 PM

#

what did u expect lol

flint idol Jan 25, 2025, 1:41 PM

#

shi idrk

real pecan Jan 25, 2025, 1:41 PM

#

uacpi does

lock_spinlock(idk)
enable_gpe(); < u get a GPE irq here
unlock_spinlock()

#

ofc this wont work

flint idol Jan 25, 2025, 1:42 PM

#

maybe the reason it worked before for me was because it was before uacpi's thread safety thing

real pecan Jan 25, 2025, 1:42 PM

#

this spinlock is very new

#

so yeah

#

it was added with thread safety

short mortar Jan 25, 2025, 1:43 PM

#

flint idol ```c // Set vector info. if (vector) { printf("GSI %d is of ...

are you reading and writing as uint32_t's?

flint idol Jan 25, 2025, 1:43 PM

#

short mortar are you reading and writing as uint32_t's?

you kinda have to, so yes

real pecan Jan 25, 2025, 1:43 PM

#

nah nvm we figured it out, his ioapic is correct

flint idol Jan 25, 2025, 1:43 PM

#

also scroll down

#

the solution is to make IRQL_GPE == IRQL_DISPATCH

real pecan Jan 25, 2025, 1:43 PM

#

GPE should be relatively low priority

short mortar Jan 25, 2025, 1:44 PM

#

short mortar are you reading and writing as uint32_t's?

(asking just to be sure)

flint idol Jan 25, 2025, 1:44 PM

#

there is no other way lol

real pecan Jan 25, 2025, 1:44 PM

#

write it as u128

flint idol Jan 25, 2025, 1:45 PM

#

funny how modifying one header irq/irql.h causes 107 files to be rebuilt

#

and modifying int.h causes the entire project to be rebuilt

real pecan Jan 25, 2025, 1:46 PM

#

if this works we should test on qwincis laptop too

#

because his suspend breaks there

flint idol Jan 25, 2025, 1:46 PM

#

ok

real pecan Jan 25, 2025, 1:46 PM

#

we should see if obos suspend is superior

flint idol Jan 25, 2025, 1:46 PM

#

ok it boots now

real pecan Jan 25, 2025, 1:47 PM

#

LETSFUCKINGGOOOOOO

#

does it suspend etc etc

flint idol Jan 25, 2025, 1:47 PM

#

I'm going to test that in a bit

real pecan Jan 25, 2025, 1:47 PM

#

i'm am

flint idol Jan 25, 2025, 1:48 PM

#

a warning log caught my eye

real pecan Jan 25, 2025, 1:48 PM

#

whats the warning

flint idol Jan 25, 2025, 1:48 PM

#

something in the IRQ interface

#

[ WARN  ] register_irq_vector: IRQ object 0xwhatever (IRQ Vector 1) does not have a move callback, and was moved```

real pecan Jan 25, 2025, 1:49 PM

#

what does that even mean

flint idol Jan 25, 2025, 1:49 PM

#

it's something in my IRQ abstraction

#

IRQ Vector 1 is for the GPE IRQ

real pecan Jan 25, 2025, 1:49 PM

#

why was it moved

#

and what does that even mean

flint idol Jan 25, 2025, 1:50 PM

#

real pecan why was it moved

it was moved to a different IRQ vector

real pecan Jan 25, 2025, 1:50 PM

#

skeleking

flint idol Jan 25, 2025, 1:50 PM

#

there is a callback in the IRQ object that is called when that happens

real pecan Jan 25, 2025, 1:50 PM

#

no but why

flint idol Jan 25, 2025, 1:51 PM

#

idk I wrote this code 7 months ago

real pecan Jan 25, 2025, 1:51 PM

#

under which conditions do u move it

flint idol Jan 25, 2025, 1:52 PM

#

iirc it's when a new IRQ object is made, that disallows the IRQ to be shared, requests a specific vector to be allocated for that IRQ object

#

if that makes sense

real pecan Jan 25, 2025, 1:57 PM

#

yes

flint idol Jan 25, 2025, 2:00 PM

#

@white mulch could you test this obos iso, it might fix the hang you experienced

📎 obos.iso

#

also @proper trellis if you could test it, since you were also seeing a crash

real pecan Jan 25, 2025, 2:01 PM

#

flint idol <@379250891838193667> could you test this obos iso, it might fix the hang you ex...

does suspend and stuff work now?

flint idol Jan 25, 2025, 2:02 PM

#

Testing rn

#

bruh

#

I accidentally access nullptr during suspend

real pecan Jan 25, 2025, 2:04 PM

#

this post is sponsored by your daily obos regression

proper trellis Jan 25, 2025, 2:04 PM

#

cursor on screen
key press
blank screen
key press
panic cant find /usr/bin/bash

flint idol Jan 25, 2025, 2:05 PM

#

proper trellis cursor on screen key press blank screen key press panic cant find /usr/bin/bash

that means it worked

#

I think

#

it's definitely supposed to panic at the end because I didn't put bash in this initrd

#

and I have the init program set to bash trl

real pecan Jan 25, 2025, 2:05 PM

#

is this with uacpi 1.0?

flint idol Jan 25, 2025, 2:05 PM

#

yes

real pecan Jan 25, 2025, 2:05 PM

#

does it have power button hooked up

#

or will that not work because panic

flint idol Jan 25, 2025, 2:06 PM

#

won't work because panic

real pecan Jan 25, 2025, 2:06 PM

#

sad

flint idol Jan 25, 2025, 2:06 PM

#

it's accessing nullptr in the acpi nvs saving code

#

            if (entry->type == ULTRA_MEMORY_TYPE_NVS)
            {
                saved_nvs[nvs_index].region_address = Arch_MapToHHDM(entry->physical_address);
                saved_nvs[nvs_index].size = entry->size;
                saved_nvs[nvs_index].saved_region = OBOS_NonPagedPoolAllocator->Allocate(OBOS_NonPagedPoolAllocator,
                                                                                         saved_nvs[nvs_index].size,
                                                                                         nullptr);
                memcpy(saved_nvs[nvs_index].saved_region, saved_nvs[nvs_index].region_address, saved_nvs[nvs_index].size);
                nvs_index++;
            }

#

uh oh

#

it PFs in that memcpy

real pecan Jan 25, 2025, 2:07 PM

#

noo

flint idol Jan 25, 2025, 2:07 PM

#

and rdi is zero on page fault

real pecan Jan 25, 2025, 2:08 PM

#

allocator bug?

flint idol Jan 25, 2025, 2:08 PM

#

and rdi is the allocated memory...

flint idol Jan 25, 2025, 2:08 PM

#

real pecan allocator bug?

likely OOM

real pecan Jan 25, 2025, 2:08 PM

#

Bruh

flint idol Jan 25, 2025, 2:11 PM

#

ok it was an allocator bug

#

very subtle

#

the allocator can only allocate up to 4G

#

I accidentally checked nBytes > (4*1024*1024)

#

instead of nBytes > (4*1024*1024*1024)

#

ah shit

#

wake from suspend broke

#

nooo

real pecan Jan 25, 2025, 2:12 PM

#

flint idol instead of `nBytes > (4*1024*1024*1024)`

💀

real pecan Jan 25, 2025, 2:12 PM

#

flint idol wake from suspend broke

Bruh not another regression

#

What happens btw

flint idol Jan 25, 2025, 2:13 PM

#

-wakes from suspend
-mysterious hang

#

hang could mean anything

real pecan Jan 25, 2025, 2:13 PM

#

Time for triple fault debugging

flint idol Jan 25, 2025, 2:13 PM

#

from panic to actually hanging

#

nooo

#

literally me rn:

#

https://cdn.discordapp.com/emojis/816178972672131083.gif?size=256

real pecan Jan 25, 2025, 2:14 PM

#

Yeah unlucky

#

And it works on the laptop right

flint idol Jan 25, 2025, 2:14 PM

#

yes

real pecan Jan 25, 2025, 2:14 PM

#

F

flint idol Jan 25, 2025, 2:14 PM

#

let me make sure it still works on the laptop

real pecan Jan 25, 2025, 2:15 PM

#

Inb4 regressed

flint idol Jan 25, 2025, 2:17 PM

#

Bruh my stoobid ahci driver

#

Regressed

real pecan Jan 25, 2025, 2:17 PM

#

Is that where it dies

flint idol Jan 25, 2025, 2:17 PM

#

I mean it has been broken for a long time

flint idol Jan 25, 2025, 2:18 PM

#

flint idol Regressed

Long before suspend it hangs

real pecan Jan 25, 2025, 2:18 PM

#

F

flint idol Jan 25, 2025, 2:21 PM

#

Suspend works on the laptop

white mulch Jan 25, 2025, 2:43 PM

#

flint idol <@379250891838193667> could you test this obos iso, it might fix the hang you ex...

Ill try

flint idol Jan 25, 2025, 2:43 PM

#

thanks

white mulch Jan 25, 2025, 2:46 PM

#

rn_image_picker_lib_temp_bd722354-999b-4578-a9c9-3a71a32a4cbc.jpg

flint idol Jan 25, 2025, 2:46 PM

#

ok thanks

real pecan Jan 25, 2025, 2:46 PM

#

why is that path invalid lol

#

looks valid to me

white mulch Jan 25, 2025, 2:49 PM

#

obos tries to do ecdt init before namespace load apparently

real pecan Jan 25, 2025, 2:49 PM

#

💀

white mulch Jan 25, 2025, 2:49 PM

#

so yeah that explains it, there is nothing in the namespace at that point except the predefined ones

real pecan Jan 25, 2025, 2:50 PM

#

Oberrow what were you thinking

flint idol Jan 25, 2025, 2:52 PM

#

silly me

#

@white mulch can you try this

📎 obos.iso

#

it will probably still hang

#

but it at least will initialize the EC properly with the ECDT

white mulch Jan 25, 2025, 3:03 PM

#

rn_image_picker_lib_temp_8455a0bb-14e4-48c3-b383-1d5032c4ac09.jpg

flint idol Jan 25, 2025, 3:03 PM

#

ah shit

real pecan Jan 25, 2025, 3:03 PM

#

trl

flint idol Jan 25, 2025, 3:03 PM

#

ok thanks for testing

#

I'll be back when I fix the pf

#

0xffffffff800a7367: Mm_AgingPRA at /home/oberrow/Code/obos/src/oboskrnl/mm/aging.c:89 (discriminator 7)```

#

fuck this shit

#

REMOVE_WORKINGSET_PAGE_NODE(ctx->referenced, node);```

#

meanwhile line 89

#

which means memory corruption

white mulch Jan 25, 2025, 3:05 PM

#

I got a good idea to speed up testing on this laptop, maybe I can set up my raspi5 in device mode and make it emulate an usb stick out of a folder shared from my main pc or smth

flint idol Jan 25, 2025, 3:06 PM

#

you can do that

#

@white mulch could you test this

📎 obos.iso

#

lol

#

like soundboards but in text

white mulch Jan 25, 2025, 3:16 PM

#

white mulch I got a good idea to speed up testing on this laptop, maybe I can set up my rasp...

actually I don't think I can nooo the pi needs power over usbc and I don't have any kind of other power supply that could supply 5v 3A over gpio

white mulch Jan 25, 2025, 3:16 PM

#

flint idol <@379250891838193667> could you test this

sure

flint idol Jan 25, 2025, 3:17 PM

#

thanks

hollow geyser Jan 25, 2025, 3:17 PM

#

flint idol

woah dude thats so cool

flint idol Jan 25, 2025, 3:21 PM

#

qwinci is typing...

white mulch Jan 25, 2025, 3:21 PM

#

it doesn't fault anymore but it hangs after the "could not make device wake-capable"

#

no gpe enable messages or anything

flint idol Jan 25, 2025, 3:21 PM

#

did it log anything like "lock is contended"

white mulch Jan 25, 2025, 3:21 PM

#

no

flint idol Jan 25, 2025, 3:21 PM

#

oh nothing at all

#

fuck

real pecan Jan 25, 2025, 3:26 PM

#

flint idol which means memory corruption

did u figure it out

flint idol Jan 25, 2025, 3:29 PM

#

Uhh I think I made it worse

flint idol Jan 25, 2025, 8:17 PM

#

anyway where was I last

#

fixing bugs with obos is getting boring

#

I'm sure qwinci's bug will end up fixing itself eventually

real pecan Jan 25, 2025, 8:18 PM

#

yes but

#

fix resume from suspend

flint idol Jan 25, 2025, 8:18 PM

#

I couldn't be bothered to do that atm

real pecan Jan 25, 2025, 8:19 PM

#

hmm

#

then port mesa

flint idol Jan 25, 2025, 8:19 PM

#

okay

#

just let me go to the future when I actually port mesa and take my code from then

#

so I don't have to do it now

real pecan Jan 25, 2025, 8:21 PM

#

Just become astral

flint idol Jan 25, 2025, 8:22 PM

#

obos will become astral fork

real pecan Jan 25, 2025, 8:23 PM

#

Letsgo

flint idol Jan 25, 2025, 8:24 PM

#

wtf is wrong with u virtualbox

white mulch Jan 25, 2025, 8:24 PM

#

sudo rmmod kvm_intel (or whatever the one for amd is if you have amd)

flint idol Jan 25, 2025, 8:25 PM

#

what if I want kvm back

white mulch Jan 25, 2025, 8:25 PM

#

then you modprobe kvm_intel

flint idol Jan 25, 2025, 8:25 PM

#

sudo modprobe kvm_intel

white mulch Jan 25, 2025, 8:25 PM

#

I am not sure why it suddenly started complaining about that tho, I think it worked just fine in 6.11 even with the kvm module loaded

#

not at the same time but like if you had no kvm vm's running it would work

flint idol Jan 25, 2025, 8:27 PM

#

what if I made an ip address have no father

#

ip addr .. nodad

flint idol Jan 25, 2025, 9:24 PM

#

--- 192.168.100.2 statistics ---
16699 packets transmitted, 16618 packets received,   0% unanswered (0 extra)
rtt min/avg/max/std-dev = 0.205/1.792/5.565/0.632 ms```
I have fixed a bug with the r8169 driver

#

that's the result of arping

#

sudo arping 192.168.100.2 -I enp0s25 -W 0.005```

flint idol Jan 25, 2025, 9:36 PM

#

flint idol After school, I hope to implement IPv4 reception

ok time to do this

#

ig

#

oh btw how do I work with the "time to live" field of the ipv4 header

flint idol Jan 25, 2025, 10:06 PM

#

theoretically, IPv4 packet reception should be implemented now

#

I do some basic size checks on the frame

flint idol Jan 25, 2025, 11:51 PM

#

so at the UDP layer of my net stack

#

I kinda have no idea what to do with my newly received packet

#

obos_status Net_UDPReceiveFrame(frame* what, frame* raw_frame)
{
    OBOS_UNUSED(raw_frame);
    udp_header* hdr = (void*)what->base;
    // TODO: Checksum validation
    if (be16_to_host(hdr->length) > what->sz)
    {
        OBOS_Warning("On NIC %02x:%02x:%02x:%02x:%02x:%02x: Received UDP packet has invalid packet size (NOTE: Buffer overflow). Dropping packet.\n", 
            what->source_mac_address[0], what->source_mac_address[1], what->source_mac_address[2], 
            what->source_mac_address[3], what->source_mac_address[4], what->source_mac_address[5]
        );
        return OBOS_STATUS_INVALID_HEADER;
    }
    
    return OBOS_STATUS_SUCCESS;
}```

#

I'm wondering whether I should just put this into some sort of queue

#

then when someone wants to receive from port n

#

they look into the packet queue for port n, and pop the first entry

#

although, a packet isn't actually removed from the queue until it's refcount is zero

#

there will be a "base refcount" for each queue
this refcount is incremented on each bind() call

short mortar Jan 25, 2025, 11:56 PM

#

why do you need refcounts for packets?

#

this sounds strange

flint idol Jan 25, 2025, 11:56 PM

#

since you can have multiple programs

#

on one UDP port

#

ig

#

unless you can't

flint idol Jan 25, 2025, 11:56 PM

#

flint idol on one UDP port

if one recvs a packet, the other would miss that packet

empty kernel Jan 25, 2025, 11:57 PM

#

I always assumed that only 1 program can be accessing a single port

flint idol Jan 25, 2025, 11:57 PM

#

otherwise

#

through a simple test with nc

#

that seems not to be the case

#

at least with UDP

#

ah wait

#

nvm

uncut narwhal Jan 26, 2025, 12:02 AM

#

flint idol if one `recv`s a packet, the other would miss that packet

i think the norm for when you do multiple on one port is to split load among processes, no?

#

maybe im wrong idk

flint idol Jan 26, 2025, 12:02 AM

#

doesn't seem like it

#

in fact, in my test with nc

#

with two instances of nc -u 192.168.100.2 -l -p 32768

#

and on the other side

#

I had nc -u 192.168.100.0 32768

flint idol Jan 26, 2025, 12:03 AM

#

flint idol I had `nc -u 192.168.100.0 32768`

when I sent data here

flint idol Jan 26, 2025, 12:03 AM

#

flint idol with two instances of `nc -u 192.168.100.2 -l -p 32768`

linux would only show those packets to the first instance of this command started

#

until I closed that instance, then all future packets went to the other instance of nc

#

the other way around

#

if I had multiple nc -u 192.168.100.2 32768 opened

#

any other instances simply died if I tried sending anything

#

only the first instance worked

#

so basically, the first to bind a udp port is the one that "owns" it

flint idol Jan 26, 2025, 12:08 AM

#

flint idol any other instances simply died if I tried sending anything

it seems as if the host in this case sent destination unreachable

#

the first packet is from a 2nd instance of nc -u 192.168.100.2 32768

#

idk why but I feel like having:

udp_packet_queue Net_UDPQueues[0x10000];```

#

for each respective UDP port

#

sizeof(udp_packet_queue) will probably end up being pretty damn big

#

yeah that won't work I'm stupid

#

ah nvm it could

#

I keep forgetting how UDP works

#

there will be an RB-tree or hashmap of UDP queues

#

each udp_queue will have the actual queue in it, as well as an rw-lock, destination port number, and a struct process specifying the owner of the queue

#

I think it's safe to ignore source port in UDP packets

#

my kernel doesn't really care where it came from, as long is it was directed toward the kernel

flint idol Jan 26, 2025, 1:38 AM

#

I just wrote a long comment in one of obos' network stack's headers explaining how IP routing works

#

at least, what my understanding of it is

#

https://gist.github.com/oberrow/631ce3542f08dc97617f39554dc824c8

#

that's what I have so far

#

nic_data is just a struct

#

typedef struct ip_table_entry {
    ip_addr address;
    uint8_t subnet_mask;
    LIST_NODE(ip_table, struct ip_table_entry) node;
} ip_table_entry;
typedef LIST_HEAD(ip_table, ip_table_entry) ip_table;
LIST_PROTOTYPE(ip_table, ip_table_entry, node);
// ...
typedef struct address_table_entry {
    ip_addr addr;
    mac_address phys;
    RB_ENTRY(address_table_entry) node;
} address_table_entry;
typedef RB_HEAD(address_table, address_table_entry) address_table;
RB_PROTOTYPE(address_table, address_table_entry, node, cmp_address_table_entry);
// ...
typedef struct nic_data {
    ip_table table;
    address_table_entry* gateway_phys;
    ip_table_entry* gateway_entry;
    address_table address_to_phys;
    vnode *interface;
    mac_address physical_address;
} nic_data;```

#

address_table is a map of ip addresses to MAC addresses corresponding to that IP address

#

so when the kernel makes an ARP request on an interface, it caches the MAC address in the nic_data struct if it gets a reply

#

I'm going to need a way to remove an entry from that cache

#

some sort of syscall to do so

#

hopefully this all isn't overthought

#

  Time to Live:  8 bits

    This field indicates the maximum time the datagram is allowed to
    remain in the internet system.  If this field contains the value
    zero, then the datagram must be destroyed.  This field is modified
    in internet header processing.  The time is measured in units of
    seconds, but since every module that processes a datagram must
    decrease the TTL by at least one even if it process the datagram in
    less than a second, the TTL must be thought of only as an upper
    bound on the time a datagram may exist.  The intention is to cause
    undeliverable datagrams to be discarded, and to bound the maximum
    datagram lifetime.```

#

from the IPv4 spec thingy

#

basically when I receive an IPv4 packet:

record start timestamp
if I'm forwarding a packet, the end timestamp is set to right before I remake the IPv4 packet, but after I have resolved the destination MAC address. I do end-start, then convert that to seconds, and subtract that from time to live, discarding the packet if time to live <= 0. (Do I need to send an ICMP message here?)
otherwise I just process the packet normally because I couldn't be bothered to look at time to live otherwise

#

I was capturing my wifi interface

#

and I came across a restarted device

#

it sent an ARP request on each IP address on the subnet

weary hound Jan 26, 2025, 1:59 AM

#

flint idol basically when I receive an IPv4 packet: - record start timestamp - if I'm forwa...

afaics linux does not care about the time it took to process the packet and always decrements ttl by 1

flint idol Jan 26, 2025, 2:00 AM

#

interesting

#

netbsd probably does the right thing since it has net in the name y'know /j

weary hound Jan 26, 2025, 2:00 AM

#

https://elixir.bootlin.com/linux/v6.12.6/source/net/ipv4/ip_forward.c#L149

#

the comment in linux also says that you need to send an icmp control message if the ttl reaches 0 ```
/*
* According to the RFC, we must first decrease the TTL field. If
* that reaches zero, we must reply an ICMP control message telling
* that the packet's lifetime expired.
*/
if (ip_hdr(skb)->ttl <= 1)
goto too_many_hops;

flint idol Jan 26, 2025, 2:01 AM

#

good to know

#

since linux does it, it's probably correct (in a way?)

#

since linux is used in a lot of routers, afaik

weary hound Jan 26, 2025, 2:02 AM

#

weary hound afaics linux does not care about the time it took to process the packet and alwa...

i'd imagine it does this under the assumption that no packet takes >1 second to process

flint idol Jan 26, 2025, 2:04 AM

#

I wonder what "process" even means here
does it just mean to receive the packet, or does it include the time it takes to find out the destination MAC address if forwarding

#

the latter can technically take more than a second

weary hound Jan 26, 2025, 2:05 AM

#

i don't think finding the destination mac is gonna take >1 second

#

since that's always done on the local network

flint idol Jan 26, 2025, 2:05 AM

#

true

weary hound Jan 26, 2025, 2:05 AM

#

unless you have an extremely lossy link ig

flint idol Jan 26, 2025, 2:06 AM

#

8 bits/s internet

#

https://github.com/OBOS-dev/obos/blob/net-stack/src/oboskrnl/net/nic_data.h

GitHub

obos/src/oboskrnl/net/nic_data.h at net-stack · OBOS-dev/obos

I was bored... Contribute to OBOS-dev/obos development by creating an account on GitHub.

#

I just committed these changes

#

god willing, I will do some of this IP routing stuff tomorrow

#

and after that, UDP packet reception

#

which should be as shrimple as:

finding the proper ip_table_entry for the destination address
adding it to the received packet queue for the destination port

flint idol Jan 26, 2025, 2:21 AM

#

flint idol god willing, I will do some of this IP routing stuff tomorrow

which includes changing my ARP code a bit

#

before, it just assumed the IP address was in interface_vnode->data (data is a uint64_t)

#

which will become an invalid assumption

#

after I implement this

#

I would need to change the code such that it looks for the requested address in each ip_table_entry of the interface

#

I think I will have my own syscalls and programs to deal with configuring internet routing

#

which includes my own DHCP implementation, which will go in userland

#

hopefully DHCP won't be too hard

#

thankfully I chose the route of making my network stack, this is kinda fun

#

(until I stop ignoring TCP)

frank flare Jan 26, 2025, 8:05 AM

#

flint idol oh btw how do I work with the "time to live" field of the ipv4 header

~~for receiving packets, you can ignore it, since all its use is before it gets to the destination, so you as the receiving part can ignore it, if it's invalid the router will discard it anyways~~ nvm i was wrong
for sending, just pick a default value, linux chose 64 iirc, which seems sane

#

idk if you did look it up but if not, wikipedia is enough

#

well nvm it was actually discussed, i just didnt read the other messages

flint idol Jan 26, 2025, 12:29 PM

#

frank flare idk if you did look it up but if not, wikipedia is enough

I was exclusively looking at the spec

flint idol Jan 26, 2025, 4:21 PM

#

after a few hours

#

I got some code down

#

to route an IPv4 packet

flint idol Jan 26, 2025, 5:00 PM

#

the code to route packets seems to be working

flint idol Jan 26, 2025, 6:09 PM

#

flint idol which should be as shrimple as: - finding the proper `ip_table_entry` for the de...

    udp_queue *queue = NetH_GetUDPQueueForPort(ent_, hdr->dest_port, false);
    if (!queue)
    {
        // TODO: Send destination unreachable (port unreachable) on ICMP.
        NetH_ReleaseSharedBuffer(what->base);
        return OBOS_STATUS_NOT_FOUND;
    }
    Core_RwLockAcquire(&queue->lock, false);
    frame* new_frame = OBOS_KernelAllocator->ZeroAllocate(OBOS_KernelAllocator, 1, sizeof(frame), nullptr);
    *new_frame = *what;
    new_frame->buff += sizeof(*hdr);
    new_frame->sz = hdr->length;
    LIST_APPEND(frame_queue, &queue->queue, new_frame);```

#

it was pretty simple

#

oops just found a bug there (forgot to release the lock)

#

    udp_queue *queue = NetH_GetUDPQueueForPort(ent_, hdr->dest_port, false);
    if (!queue)
    {
        // TODO: Send destination unreachable (port unreachable) on ICMP.
        NetH_ReleaseSharedBuffer(what->base);
        return OBOS_STATUS_NOT_FOUND;
    }
    frame* new_frame = OBOS_KernelAllocator->ZeroAllocate(OBOS_KernelAllocator, 1, sizeof(frame), nullptr);
    *new_frame = *what;
    new_frame->buff += sizeof(*hdr);
    new_frame->sz = hdr->length;
    Core_RwLockAcquire(&queue->lock, false);
    LIST_APPEND(frame_queue, &queue->queue, new_frame);
    Core_RwLockRelease(&queue->lock, false);```

flint idol Jan 26, 2025, 6:31 PM

#

ok so theroetically receiving a UDP packet is as easy as binding yourself to the queue NetH_GetUDPQueueForPort(ip_table_entry, port, true);

#

then waiting on the event object that is set on each UDP packet's arrival

#

then you read however many packets you like

#

by popping from the head of the queue

#

then you use the frame from the queue and copy it into your buffer

#

then you call NetH_ReleaseSharedBuffer(frame->base);

#

and you're done

flint idol Jan 26, 2025, 7:55 PM

#

flint idol Jan 26, 2025, 8:14 PM

#

wtf is happening

#

the kernel has suddenly started switching IP packets and ARP packets

#

the host sends my laptop (the client) a UDP packet

#

it thinks it's ARP

#

both when I print it

#

and when it goes to dispatch it

#

and when I arping the laptop

#

it thinks it's getting an IPv4 packet

#

and this only happens sometimes- ah shit

#

memory corruption nooo nooo nooo

#

now it seems like it thinks all packets

#

are ARP

#

#

#

KEKW

real pecan Jan 26, 2025, 8:22 PM

#

ARPOS

flint idol Jan 26, 2025, 8:22 PM

#

other protocols were overrated anyway

#

ok I fixed it

flint idol Jan 26, 2025, 8:44 PM

#

local man forgets to convert UDP header field from big endian to host endian

honest citrus Jan 26, 2025, 8:45 PM

#

just run it on a big endian machine and then suddenly these problems go away meme

flint idol Jan 26, 2025, 8:45 PM

#

I do have an m68k port thinkong

#

but no NIC driver for it (yet)

#

hmm why is my kernel sending packets in the wrong direction

#

udp packet in question:

#

it's sending a packet to the current NIC's mac

#

and saying the source is the destination's source

flint idol Jan 26, 2025, 8:51 PM

#

flint idol and saying the source is the destination's source

that makes no sense

#

wait I'm stupid

#

bro wtf

#

my kernel is fucking

#

idefk

#

at one point

#

by some fucking miracle

#

it does a use after free

empty kernel Jan 26, 2025, 9:12 PM

#

Always back to memory bugs

flint idol Jan 26, 2025, 9:13 PM

#

all I know rn is that it's in the network stack

#

except the stack trace is doing NOTHING

#

it literally ends at a call to LIST_REMOVE

#

LIST_REMOVE(frame_queue, ..., ...);

#

and there are only two calls to that macro in the net stack

#

both of which

#

there cannot be a use-after-free

#

    LIST_REMOVE(frame_queue, &frames, data);
    OBOS_KernelAllocator->Free(OBOS_KernelAllocator, data, sizeof(*data));```

#

unless the compiler reorders these two calls KEKW

#

wtf

#

    LIST_REMOVE(frame_queue, &frames, data);
    asm ("" : : :"memory");
    OBOS_KernelAllocator->Free(OBOS_KernelAllocator, data, sizeof(*data));```

#

I put a compiler memory barrier

#

and it works

#

KEKW

uncut narwhal Jan 26, 2025, 9:34 PM

#

lmao

flint idol Jan 26, 2025, 9:39 PM

#

just fixed a bug where you could crash obos by sending it a packet that's 170 bytes long

#

(170 = 0xaa, I had an assert in free() checking for nBytes==0xaa to make sure no one does a double free of an object, it just backfired)

#

#

I made some dumb echo server in kernel-mode

#

to test UDP reception and transmission

#

and it works, almost

#

I have found a pattern

#

the kernel always echoes back 8 more bytes than there should be

flint idol Jan 26, 2025, 9:59 PM

#

flint idol my kernel doesn't really care where it came from, as long is it was directed tow...

this is false btw

#

if you're sending a reply to that host, you need to set reply.src_port to received.dest_port and reply.dest_port to received.src_port

#

the extra 8 bytes in the header are from the sizeof(udp_header)

#

so that's fine...

#

the problem is

#

I was forgetting to subtract that

#

in the received packet thing

#

     frame* new_frame = OBOS_KernelAllocator->ZeroAllocate(OBOS_KernelAllocator, 1, sizeof(frame), nullptr);
     *new_frame = *what;
     new_frame->buff += sizeof(*hdr);
-    new_frame->sz = be16_to_host(hdr->length);
+    new_frame->sz = be16_to_host(hdr->length) - sizeof(*hdr);```

#

flint idol Jan 26, 2025, 11:08 PM

#

I decided to test obos' network stack on a real network

#

to no avail

#

no packets are received

#

no packets are transmitted

#

arping the IP address I assigned it does nothing

#

perhaps I just need to do something in my router

#

anyway, I just pushed all the code I've written today

flint idol Jan 27, 2025, 12:55 AM

#

I was able to get a response over the network

#

but only if I disabled IPv4 packet forwarding

thick jolt Jan 27, 2025, 12:56 AM

#

hi

#

i

#

shouldnt be awake rn

flint idol Jan 27, 2025, 12:56 AM

#

hello

thick jolt Jan 27, 2025, 12:56 AM

#

agian

flint idol Jan 27, 2025, 12:56 AM

#

thick jolt shouldnt be awake rn

sleep

thick jolt Jan 27, 2025, 12:56 AM

#

i broke my sleep schedule again

#

nooo

flint idol Jan 27, 2025, 12:56 AM

#

sleep

thick jolt Jan 27, 2025, 12:56 AM

#

cant

flint idol Jan 27, 2025, 12:56 AM

#

go in your bed then

thick jolt Jan 27, 2025, 12:56 AM

#

i am

#

with my laptop

flint idol Jan 27, 2025, 12:56 AM

#

turn off your laptop

#

and phone

thick jolt Jan 27, 2025, 12:56 AM

#

flint idol

damn

#

lessgooo

flint idol Jan 27, 2025, 12:57 AM

#

this means that the kernel is stable enough to handle all these packets*

#

*after enough packets it starts to die

#

and by die I mean the latency becomes really high

thick jolt Jan 27, 2025, 12:59 AM

#

well

#

at least it works ish

#

im sure youll figure it out

flint idol Jan 27, 2025, 1:08 AM

#

#

Even worked on my phone

#

One problem I am observing is use after frees

#

I think I know why

#

After a few hundred packets received from the network

#

It crashes

#

Which isn't terrible

ornate ginkgo Jan 27, 2025, 1:11 AM

#

I mean who needs more than a few hundred packets anyway

#

the progress is cool though

flint idol Jan 27, 2025, 1:13 AM

#

flint idol After a few hundred packets received from the network

For reference this is after ~5 minutes

#

Of packets

#

And crashing after 5 minutes worth of packets because of memory corruption isn't optimal

#

however, I suspect this is not an allocator bug, and instead just plain memory corruption

#

in the network stack

flint idol Jan 27, 2025, 1:34 AM

#

flint idol One problem I am observing is use after frees

I suspect this is because I was forgetting to reference a buffer in the data_ready callback of the network stack

#

and sometimes multiple packets are referenced by this buffer

#

since data_ready can mean multiple packets are ready

#

or just one

#

so:

receives two frames, and reports two frames are ready in data_ready
first frame processed, but the buffer's refcount is zero on entry, so when the frame is finally finished processing, and the last call of NetH_ReleaseSharedBuffer happens in Net_IPReceiveFrame, the buffer is freed, even though it is still referenced
next frame: use after free, as we never referenced the shared buffer in 'data_ready'```

flint idol Jan 27, 2025, 2:08 AM

#

just found a bug related to certain page faults

#

in the network stack

#

basically, I would use the paged pool allocator for memory that could be used at IRQL > DISPATCH

#

which is invalid in my kernel, because of demand paging

#

and demand paging means accessing it the first time means a page fault

#

which is the specific problem

#

now my solution is to temporarily lower the irql before doing memcpying into a buffer

uncut narwhal Jan 27, 2025, 2:28 AM

#

feel like the solution is to have a function to "pin" pages to prevent them being paged out when theyre to be used for io

#

i think windows does that with MDLs?

flint idol Jan 27, 2025, 2:28 AM

#

yes I am going to do that

#

that is on my todo list

#

and so is the rewrite of my IO system

uncut narwhal Jan 27, 2025, 2:29 AM

#

mood tbh

flint idol Jan 27, 2025, 2:29 AM

#

huh what is "redirect for host"

uncut narwhal Jan 27, 2025, 2:34 AM

#

never heard of that tbh

flint idol Jan 27, 2025, 2:35 AM

#

the kernel is starting to die

#

At one point it starts to die because of a bunch of use after frees

#

It seems

#

And because of the frame backlog

#

i.e., there are so many pending frames the kernel starts to vanish

#

From the network

#

Until it finally crashes under the load

#

This time it could withstand a lot more packets, at least

vale nymph Jan 27, 2025, 2:34 PM

#

flint idol the kernel is starting to die

Unplug the life support

flint idol Jan 27, 2025, 2:38 PM

#

Done

#

Rip obos rewrite 5 2024-2025

#

/j

leaden carbon Jan 27, 2025, 2:38 PM

#

gg

#

Era of Nyaux supremacy began 2025-

honest citrus Jan 27, 2025, 8:32 PM

#

flint idol however, I suspect this is not an allocator bug, and instead just plain memory c...

kasan meme

flint idol Jan 27, 2025, 8:35 PM

#

I broke that in a recent obos update

flint idol Jan 28, 2025, 12:12 AM

#

I plan to do DHCP soon

#

Should I write my client in userspace

#

Or kernel space

#

And just have some syscall to do DHCP and configure the interface's tables

#

Oh btw how do I know the subnet when I broadcast DHCPDISCOVER

empty kernel Jan 28, 2025, 5:38 AM

#

I think linux mostly does DHCP in userspace

white mulch Jan 28, 2025, 8:57 AM

#

flint idol Oh btw how do I know the subnet when I broadcast DHCPDISCOVER

why would you need to know it? you only need to set the op/htype/hlen/xid fields of the dhcp header for the discover packet (along with the options to specify the msg type and parameter request list with whatever extra info you want like the subnet mask or dns server)

#OBOS (not vibecoded)