OBOS (not vibecoded) | Operating System Development | Page 33

flint idol Jan 1, 2025, 6:27 PM

#

oh really

lean glen Jan 1, 2025, 6:27 PM

#

you can do it per-commit iirc

#

I've done it before

#

yea

#

curl -sL https://github.com/user-or-org/repo/archive/sha1-or-ref.tar.gz | tar xz

flint idol Jan 1, 2025, 6:28 PM

#

lean glen ``` curl -sL https://github.com/user-or-org/repo/archive/sha1-or-ref.tar.gz | ta...

marker

#

well that's useful

thick jolt Jan 1, 2025, 7:02 PM

#

give me ur laptop

#

😡

flint idol Jan 1, 2025, 7:07 PM

#

thick jolt give me ur laptop

what if hte reason your laptop is bricked is because you tested nyaux on it

#

*th

#

**the

thick jolt Jan 1, 2025, 7:16 PM

#

flint idol what if hte reason your laptop is bricked is because you tested nyaux on it

lmao 💀

#

no not rlly

#

it was nobara actually

#

that killed my system

#

idk what it did but motherboard is fried!

flint idol Jan 1, 2025, 8:36 PM

#

anyway, it is now time to separate host and target recipes

flint idol Jan 1, 2025, 8:52 PM

#

flint idol anyway, it is now time to separate host and target recipes

is there really a point though thonk

#

No

#

there isn't

#

time for buildall

#

installall is like the exact same but with an extra step

flint idol Jan 1, 2025, 11:04 PM

#

flint idol time for `buildall`

because the compiler needs to be able to find compiled libraries in the sysroot, buildall will be aliased to installall

flint idol Jan 1, 2025, 11:40 PM

#

and I am renaming those commands to build-all and install-all

thick jolt Jan 2, 2025, 12:09 AM

#

flint idol and I am renaming those commands to `build-all` and `install-all`

i am in ur walls

#

also guess who found their old ah laptop

#

i did!!!

flint idol Jan 2, 2025, 12:09 AM

#

thick jolt i am in ur walls

You got it the wrong way

#

I am in your walls

thick jolt Jan 2, 2025, 12:09 AM

#

nooo

#

wait no this is perfect

#

😈

flint idol Jan 2, 2025, 12:09 AM

#

And I was the reason your laptop broke

thick jolt Jan 2, 2025, 12:09 AM

#

u dont mind me drilling into the walls right

thick jolt Jan 2, 2025, 12:10 AM

#

flint idol And I was the reason your laptop broke

😡

#

wow

#

okay im drilling into the walls

#

#

anyways i found my old laptop

#

so osdev can continue

#

just compile times are gonna suck ass

#

nooo

weary hound Jan 2, 2025, 12:13 AM

#

i have the same cpu in my current machine lol

thick jolt Jan 2, 2025, 12:17 AM

#

weary hound i have the same cpu in my current machine lol

😭

flint idol Jan 2, 2025, 12:44 AM

#

thick jolt just compile times are gonna suck ass

still probably better than my cpu

#

"Intel(R) Core(TM) i5-4590T (4) @ 3.00 GHz"

empty kernel Jan 2, 2025, 12:46 AM

#

definitely better than what I used to have to deal with all the time on my laptop: "Intel Pentium Silver N5000 (4) @ 1.10 GHz" (boost of 2.70 GHz)

flint idol Jan 2, 2025, 2:07 AM

#

now that I implemented this

#

it's actually really smart

#

if I want to multithread this

#

which I do

#

so with that, I should have the build-all command implemented

#

it tries to make the most of multithreading

#

basically, in build_pkg_and_dependents, if it can, it will try to start a new thread to build a dependent of that package

#

and then at the end of that function, it joins the thread

#

which I actually don't know if I need to do that

#

or even should join the new thread

flint idol Jan 2, 2025, 2:16 AM

#

flint idol so with that, I should have the build-all command implemented

it's very lightly tested
the dependency graph literally looks like this:
hello depends bash

flint idol Jan 2, 2025, 2:17 AM

#

flint idol or even should join the new thread

I use semaphores to synchronize thread creation

#

#

yeah maybe I should just pthread_detach the thread after creating it

#

and not join it after looping through the dependants

#

if (!(--data->missing_dep_count))
{
    int val = 0;
    sem_getvalue(&awake_threads, &val);
    if (val)
    {
        if (sem_trywait(&awake_threads) != 0)
            goto no_thread;
        pthread_t thr_id = {};
        // We can now start a thread.
        int ec = pthread_create(&thr_id,
                                NULL,
                                build_thread, data);
        if (ec)
        {
            perror("pthread_create");
            fprintf(stderr, "Ignoring error.\n");
            goto no_thread;
        }
        pthread_detach(thr_id);
        continue;
    }
    no_thread:
    res = build_pkg_and_dependants(data, curl_hnd);
    if (!res)
        break;
    continue;
}```

weary hound Jan 2, 2025, 2:24 AM

#

why spawn threads?

flint idol Jan 2, 2025, 2:24 AM

#

parallelism

weary hound Jan 2, 2025, 2:24 AM

#

all you do is run through a graph and fork off to run other commands

lean glen Jan 2, 2025, 2:24 AM

#

I'd just use gnu parallel maybe

flint idol Jan 2, 2025, 2:24 AM

#

gnu 👎

#

maybe I'll look into that soon

weary hound Jan 2, 2025, 2:25 AM

#

i don't understand what you want to parallelize here though

flint idol Jan 2, 2025, 2:25 AM

#

I am parallelizing the building of packages

weary hound Jan 2, 2025, 2:25 AM

#

and why does that need threads?

flint idol Jan 2, 2025, 2:25 AM

#

so I can build the package

#

in parallel

weary hound Jan 2, 2025, 2:26 AM

#

well for individual packages the parallelism is handled by the build system

#

e.g. make/ninja with -jN etc

flint idol Jan 2, 2025, 2:27 AM

#

pitust said otherwise

lean glen Jan 2, 2025, 2:27 AM

#

./configure tho

weary hound Jan 2, 2025, 2:27 AM

#

and if you want to build multiple packages in parallel at once you don't need threads

flint idol Jan 2, 2025, 2:28 AM

#

oh yeah I get what ur saying now

lean glen Jan 2, 2025, 2:28 AM

#

use fork

flint idol Jan 2, 2025, 2:29 AM

#

flint idol oh yeah I get what ur saying now

nvm

weary hound Jan 2, 2025, 2:30 AM

#

weary hound and if you want to build multiple packages in parallel at once you don't need th...

you can do something like this more or less ```
job_queue = []
... add jobs to queue ...

free_slots = N
in_flight = {}
while not job_queue.empty():
while free_slots > 0 and not job_queue.empty():
job = job_queue.pop_back()
pid = fork_and_exec(job)
in_flight[pid] = job
free_slots--
while free_slots != N:
pid = wait()
... add dependents of in_flight[pid] ...
in_flight.remove(pid)
free_slots++

#

and ig some logic to add dependents once a job is complete

flint idol Jan 2, 2025, 2:30 AM

#

weary hound you can do something like this more or less ``` job_queue = [] ... add jobs to q...

marker

#

yeah I will do that soon™️

#

but now I really want to get back to work on OBOS

weary hound Jan 2, 2025, 2:31 AM

#

but imo it's better to build packages one by one

#

and tell the package's build system to parallelize instead

#

you can parallelize configure or whatever but the build itself doesn't make sense imo

lean glen Jan 2, 2025, 2:36 AM

#

yeah just configure

#

tho idk how the output would look

#

would be hard to debug

weary hound Jan 2, 2025, 2:37 AM

#

although even then configure doesn't take that long either

lean glen Jan 2, 2025, 2:37 AM

#

it does tho

#

sucks

flint idol Jan 2, 2025, 2:37 AM

#

PSA: autotools sucks

weary hound Jan 2, 2025, 2:37 AM

#

i mean usually if it takes long then the package is large and takes even longer to build

#

so it's not a substantial amount of time compared to the whole build

lean glen Jan 2, 2025, 2:38 AM

#

idk sometimes builds take faster

#

unironically

weary hound Jan 2, 2025, 2:39 AM

#

sure true yeah

#

also, another problem with not doing -jN or similar, if the current job is a dependency of every other job after it you're only using 1/N of your system to build it

flint idol Jan 2, 2025, 2:40 AM

#

weary hound also, another problem with not doing `-jN` or similar, if the current job is a d...

yeah I had thought of that

weary hound Jan 2, 2025, 2:41 AM

#

so it takes N times as long

#

consider a gcc cross compiler, where on my laptop with 6 cores it takes about minutes to build, but if i built it with -j1 it would be more like 30 minutes instead

flint idol Jan 2, 2025, 2:41 AM

#

for those packages I will probably do -jN

weary hound Jan 2, 2025, 2:41 AM

#

and i can't really build any packages alongside gcc (ig i can build some other tools but most don't take as long as gcc)

flint idol Jan 2, 2025, 2:41 AM

#

because it's not like the rest of the cpus are doing anything

flint idol Jan 2, 2025, 2:42 AM

#

weary hound also, another problem with not doing `-jN` or similar, if the current job is a d...

https://tenor.com/view/meme-gif-25244209

Tenor

haughty abyss Jan 2, 2025, 7:11 AM

#

for all the autocrap you can use a make jobserver

flint idol Jan 2, 2025, 2:49 PM

#

I now have recipes for x86_64-obos-gcc and binutils for obos-strap

flint idol Jan 2, 2025, 3:10 PM

#

I have also added a recipe for mlibc

#

and these recipes almost work

#

except gcc can't find any headers

#

#

so I guess I have to make mlibc spit it's headers and stuff into sysroot/usr/

#

and I will also be adding something else to obos-strap to make clean work better

#

specifically, I will add clean-commands to recipes

#

but for now I can just pass --wipe to meson

#

or not..

#

wait nvm I forgot to run install on mlibc

#

silly me

flint idol Jan 2, 2025, 3:48 PM

#

--- a/support/config.sub
+++ b/support/config.sub
@@ -1731,7 +1731,7 @@ case $os in
             | os9* | macos* | osx* | ios* \
             | mpw* | magic* | mmixware* | mon960* | lnews* \
             | amigaos* | amigados* | msdos* | newsos* | unicos* | aof* \
-            | aos* | aros* | cloudabi* | sortix* | twizzler* \
+            | aos* | aros* | cloudabi* | sortix* | obos* | twizzler* \
             | nindy* | vxsim* | vxworks* | ebmon* | hms* | mvs* \
             | clix* | riscos* | uniplus* | iris* | isc* | rtu* | xenix* \
             | mirbsd* | netbsd* | dicos* | openedition* | ose* \

the patch command isn't enjoying my patch for bash

#

patching file /home/oberrow/Code/obos-pkgs/repos/bash-5.2.37/support/config.sub
Hunk #1 FAILED at 1731.
1 out of 1 hunk FAILED -- saving rejects to file /home/oberrow/Code/obos-pkgs/repos/bash-5.2.37/support/config.sub.rej

#

and I've no idea why

#

patch is part of coreutils right

flint idol Jan 2, 2025, 3:51 PM

#

flint idol `patch` is part of coreutils right

it's something different

flint idol Jan 2, 2025, 3:52 PM

#

flint idol and I've no idea why

the bash I had when I made the patch and the bash I extracted are the exact same bash

white mulch Jan 2, 2025, 4:07 PM

#

flint idol the bash I had when I made the patch and the bash I extracted are the exact same...

and you didn't run autoreconf or anything?

flint idol Jan 2, 2025, 4:07 PM

#

no

white mulch Jan 2, 2025, 4:07 PM

#

weird

flint idol Jan 2, 2025, 4:07 PM

#

I even tried making it on a clean bash

#

making the bash patch

#

ok fixed it

#

somehow

#

/home/oberrow/Code/obos-pkgs/pkgs/lib/gcc/x86_64-obos/14.2.0/../../../../x86_64-obos/bin/ld: /home/oberrow/Code/obos-pkgs/pkgs/lib/gcc/x86_64-obos/14.2.0/crtbegin.o: relocation R_X86_64_32 against hidden symbol `__TMC_END__' can not be used when making a shared object```

#

bruh

torpid wigeon Jan 2, 2025, 4:47 PM

#

fix ur crts

flint idol Jan 2, 2025, 4:48 PM

#

kk

flint idol Jan 2, 2025, 5:33 PM

#

bruh my crts aren't fixing themselves

white mulch Jan 2, 2025, 5:36 PM

#

you enable the pic crts in the config files right and use them in the LINK_SPEC?

flint idol Jan 2, 2025, 5:37 PM

#

idk I just did whatever osdev wiki does

white mulch Jan 2, 2025, 5:37 PM

#

what specifically

flint idol Jan 2, 2025, 5:37 PM

#

but I will check

flint idol Jan 2, 2025, 5:37 PM

#

white mulch what specifically

my os-specific toolchain is basically just this but with obos

OSDev.wiki

OS Specific Toolchain

This tutorial will guide you through creating a toolchain comprising Binutils and GCC that specifically targets your operating system. The instructions below teach Binutils and GCC how to create programs for a hypothetical OS named 'MyOS'.
Until now you have been using a cross-compiler configured to use an existing generic bare target. This is ...

#

and an m68k target

#

ok I never change LINK_SPEC

#

I looked at another patch for gcc and I think I know what I need to change

leaden carbon Jan 2, 2025, 5:55 PM

#

@flint idol I got a question, why do u save IPL into thread

#

afaict the only way you could change thread contexts and come back with a diff IPL is if you yield

flint idol Jan 2, 2025, 5:56 PM

#

I forgor

leaden carbon Jan 2, 2025, 5:56 PM

#

lol fair and relatable

haughty abyss Jan 2, 2025, 5:56 PM

#

leaden carbon afaict the only way you could change thread contexts and come back with a diff I...

assuming you even make this legal

leaden carbon Jan 2, 2025, 5:56 PM

#

Yeah ^

haughty abyss Jan 2, 2025, 5:56 PM

#

which sounds mega scuffed

leaden carbon Jan 2, 2025, 5:56 PM

#

it does

#

but since u can only get preempted on one IPL...

#

and if you make yield just have some proper semantics

flint idol Jan 2, 2025, 5:57 PM

#

I guess back then I assumed IRQL was something I should save

#

I will probablt remove it soon

#

*probably

leaden carbon Jan 2, 2025, 5:57 PM

#

thats fair, and I am not saying there is no reason, I just cant think of one

flint idol Jan 2, 2025, 5:57 PM

#

it is sufficient to not save it in the thread

leaden carbon Jan 2, 2025, 5:58 PM

#

yeah I am not planning on it atleast rn

flint idol Jan 2, 2025, 8:07 PM

#

/home/oberrow/Code/obos-pkgs/repos/bash-5.2.37/examples/loadables/finfo.c:283:14: error: implicit declaration of function 'major' [-Wimplicit-function-declaration]

#

thinkong

#

I get this with my new gcc build

leaden carbon Jan 2, 2025, 8:08 PM

#

lol

flint idol Jan 2, 2025, 8:08 PM

#

except luckily it's in bash

#

in that file:

#

so if major isn't in sysmacros, nor in mkdev

#

that's a bug with my sysdep

#

I think

#

ok I just got bash to build

#

by patching that file to check if major and minor actually exist

#

before using them

#

so now you can probably clone https://github.com/OBOS-dev/obos-pkgs

GitHub

GitHub - OBOS-dev/obos-pkgs: OBOS Package Repository

OBOS Package Repository. Contribute to OBOS-dev/obos-pkgs development by creating an account on GitHub.

#

run ./obos-strap build-all

#

and you get all obos packages conviniently built for you

#

wait no

#

./obos-strap setup-env
./obos-strap install-all```

leaden carbon Jan 2, 2025, 8:33 PM

#

obos-strap is just a binary?

flint idol Jan 2, 2025, 8:34 PM

#

yes

leaden carbon Jan 2, 2025, 8:34 PM

#

sadge

#

cant run on nixos

flint idol Jan 2, 2025, 8:34 PM

#

https://github.com/OBOS-dev/obos-strap

#

source is here

leaden carbon Jan 2, 2025, 8:34 PM

#

fair enough

flint idol Jan 2, 2025, 8:34 PM

#

for obos-strap

flint idol Jan 2, 2025, 8:51 PM

#

fun

#

the compiled bash doesn't run

#

it segfaults

#

on a page that is lazily mapped

#

rather that, or it's a file page

white mulch Jan 2, 2025, 8:59 PM

#

how did it work before then thonk

flint idol Jan 2, 2025, 8:59 PM

#

I mean the build environment did technically change

#

because before crtbegin/crtend weren't working

flint idol Jan 2, 2025, 9:08 PM

#

flint idol rather that, or it's a file page

nvm, the page does not exist

#

there were two syscalls to mmap

#

the range didn't fit in either of them

flint idol Jan 2, 2025, 10:09 PM

#

I think I am passing teh aux values incorrectly

flint idol Jan 2, 2025, 10:30 PM

#

__ensure(Library function fails due to missing sysdep) failededed```
I love it when my functions failededed

flint idol Jan 2, 2025, 10:30 PM

#

flint idol I think I am passing teh aux values incorrectly

anyway I fixed that bug

#

it had something to do with alignment

#

anyway, time make fork

flint idol Jan 3, 2025, 12:10 AM

#

This is how fork will work if I do it userspace:

- Set *child=0
- Fork the current VMM context
- Make a new thread that starts at where ever the "return" statement of the sys_fork sysdep is at
- Start a new process, with the vmm context set to the previously forked one, and with is_fork=true
- Set *child to the process id
- Profit```

white mulch Jan 3, 2025, 12:14 AM

#

flint idol This is how fork will work if I do it userspace: ``` - Set *child=0 - Fork the c...

if you do it like that wouldn't you need a way to create an empty process and then the thread create function would take in a process?

#

that could actually be useful for something else too tbh

thick jolt Jan 3, 2025, 12:14 AM

#

forking ur plate

white mulch Jan 3, 2025, 12:15 AM

#

though idk it might be slightly cursed creating threads in other processes context

#

but ehh

flint idol Jan 3, 2025, 12:15 AM

#

white mulch if you do it like that wouldn't you need a way to create an empty process and th...

my thread create function doesn't work like htat

flint idol Jan 3, 2025, 12:15 AM

#

white mulch though idk it might be slightly cursed creating threads in other processes conte...

iirc the winapi can do that

#

https://learn.microsoft.com/en-us/windows/win32/api/processthreadsapi/nf-processthreadsapi-createremotethreadex

#

yup

white mulch Jan 3, 2025, 12:16 AM

#

how do you know where it should start at tho lol

flint idol Jan 3, 2025, 12:16 AM

#

magic

white mulch Jan 3, 2025, 12:16 AM

#

unless you share the vspace with the process or smth

flint idol Jan 3, 2025, 12:16 AM

#

the thread will start straight in userspace

white mulch Jan 3, 2025, 12:16 AM

#

or well fork ig

flint idol Jan 3, 2025, 12:17 AM

#

at whereever the ret instruction is for sys_fork (not quite, but around there)

main girder Jan 3, 2025, 12:17 AM

#

white mulch though idk it might be slightly cursed creating threads in other processes conte...

Why

flint idol Jan 3, 2025, 12:17 AM

#

you can take an address of a label, right?

#

if not, then my plan would've been foiled

#

nope

#

in GNU C you can actually

#

https://gcc.gnu.org/onlinedocs/gcc/Labels-as-Values.html#:~:text=You can get the address,of that type is valid.

Labels as Values (Using the GNU Compiler Collection (GCC))

main girder Jan 3, 2025, 12:19 AM

#

white mulch how do you know where it should start at tho lol

If ur doing that then you'll usually know for some reason like the entry point of the remote thread is in a library that's always loaded at the same place

flint idol Jan 3, 2025, 12:19 AM

#

although this is kind of scuffed actually, because I don't know where I would need to move the stack pointer

#

if I make the label at the end of the sys_fork function

white mulch Jan 3, 2025, 12:20 AM

#

main girder If ur doing that then you'll usually know for some reason like the entry point o...

ig, it doesn't seem very useful to me in normal use though

#

maybe in some kind of debugger

flint idol Jan 3, 2025, 12:21 AM

#

handle thread_ctx = create_thread_ctx_for_fork(..., &&label);
label:
if (*(child == 0))
    return 0;

#

I could perhaps do that

white mulch Jan 3, 2025, 12:21 AM

#

or just do it in asm

flint idol Jan 3, 2025, 12:21 AM

#

true

#

which is actually what I will be doing because I don't trust the compiler

#

enough

flint idol Jan 3, 2025, 12:58 AM

#

just implemented sys_fork

#

now to test it (here we go)

#

let's see how many times does it panic

#

ENOSYS

#

ah yes

#

I need sys_waitpid

#

and I also need to mangle the symbol in assembly

#

luckily it only panics once

#

I fixed the panic

#

now it's in some weird loop

#

that causes it to hang

#

found a problem

#

through syscall logging

#

turns out I am using Sys_HandleClone wrong

flint idol Jan 3, 2025, 1:47 AM

#

basically, one syscall fails, then another fails, then that chains into a bunch of failed syscalls

#

until it finally hangs

flint idol Jan 3, 2025, 2:32 AM

#

wtf

#

I have managed to corrupt the irq handlers array

#

or wait no

#

I have managed to corrupt something

#

yeah the irq handlers array

#

it looks like a stack overflow to me

flint idol Jan 3, 2025, 3:16 AM

#

I think obos has fallen to this nooo

#

#

actually it only uses mlibc

#

and isn't unix like

flint idol Jan 3, 2025, 3:45 AM

#

yeah I think my fork implementation is fucked

#

https://tenor.com/view/ratatouille-meme-memes-reads-letter-gif-20713760

Tenor

flint idol Jan 3, 2025, 4:09 AM

#

flint idol it looks like a stack overflow to me

it is because gs_base is zero

#

well I fixed that

#

so the new process segfaults first thing

#

but that is certainly better than the new process causes the system to triple fault

flint idol Jan 3, 2025, 4:21 AM

#

flint idol so the new process segfaults first thing

but bash hangs after that

#

or the libc

#

only god knows

#

the thread had not syscalled once, it just SIGSEGVed

limber wave Jan 3, 2025, 5:52 AM

#

flint idol and isn't unix like

u need nyaux like

short mortar Jan 3, 2025, 10:38 AM

#

flint idol yeah I think my `fork` implementation is fucked

Fork is just a fucked thing in itself

vale nymph Jan 3, 2025, 10:58 AM

#

flint idol

Should add uacpi

weak kestrel Jan 3, 2025, 11:04 AM

#

flint idol

me

torpid wigeon Jan 3, 2025, 12:21 PM

#

flint idol and isn't unix like

malding

flint idol Jan 3, 2025, 12:43 PM

#

I have found a problem with my Mm_Fork function

#

I don't ever ref any physical page after having marked it as CoW

flint idol Jan 3, 2025, 1:17 PM

#

flint idol the thread had not `syscall`ed once, it just `SIGSEGV`ed

I have fixed the SEGV

#

but now it hangs

flint idol Jan 3, 2025, 1:23 PM

#

flint idol I have fixed the SEGV

nvm

#

fixed the segfault

#

now I need to figure out what syscall 52 is doing

#

Sys_SigProcMask is being ran in a loop

#

wtf

#

it's with the same parameters each time

#

until it calls Sys_SigAction

#

then Sys_ProcessGetPID

#

then Sys_ThreadOpen

#

which, in turn, returned invalid handle

#

the function was passed a PID

#

pid_t sys_gettid()
{
    return (pid_t)syscall1(Sys_ThreadGetTid, HANDLE_CURRENT);
}
pid_t sys_getpid()
{
    return (pid_t)syscall1(Sys_ProcessGetPID, HANDLE_CURRENT);
}
pid_t sys_getppid()
{
    return (pid_t)syscall1(Sys_ProcessGetPPID, HANDLE_CURRENT);
}

perhaps it is a bug with one of these functions

#

specifically sys_getpid seems sus

#

I'll make it return sys_gettid and see what that does

white mulch Jan 3, 2025, 1:43 PM

#

what uses Sys_ThreadOpen?

flint idol Jan 3, 2025, 1:44 PM

#

I believe it is sys_kill

white mulch Jan 3, 2025, 1:45 PM

#

that takes in a pid

flint idol Jan 3, 2025, 1:45 PM

#

so the getpid value or gettid value

white mulch Jan 3, 2025, 1:45 PM

#

getpid

flint idol Jan 3, 2025, 1:45 PM

#

ok

white mulch Jan 3, 2025, 1:45 PM

#

it takes in a process id and sends the signal to whatever thread that can accept it

flint idol Jan 3, 2025, 1:45 PM

#

handle Sys_ProcessOpen(uint64_t pid)
{
    OBOS_UNUSED(pid);
    // Unimplemented.
    return HANDLE_INVALID;
}```
bruh

#

KEKW

flint idol Jan 3, 2025, 1:46 PM

#

white mulch it takes in a process id and sends the signal to whatever thread that can accept...

ok

#

also idk why but bash gets stuck in an infinite sys_sigprocmask loop

#

until it triple faults

white mulch Jan 3, 2025, 1:48 PM

#

triple faults?

flint idol Jan 3, 2025, 1:48 PM

#

yes

#

it triple faults the kernel after a bit

torpid wigeon Jan 3, 2025, 2:11 PM

#

flint idol ```c handle Sys_ProcessOpen(uint64_t pid) { OBOS_UNUSED(pid); // Unimple...

fent

flint idol Jan 3, 2025, 2:27 PM

#

fent?

weak kestrel Jan 3, 2025, 2:42 PM

#

fent.

short mortar Jan 3, 2025, 3:22 PM

#

fent...

flint idol Jan 3, 2025, 3:37 PM

#

wtf is happening

#

the kernel

#

is mapping memory

#

in the user part of the address space

flint idol Jan 3, 2025, 4:15 PM

#

damn the person who designed fork

#

thank you for coming to my ted talk

#

bro what the actual fuck

#

WHY IS THE KERNEL ALLOCATING

#

KENREL MEMORY

#

BELOW 0xffff'8000'0000'0000

#

something has driven my kernel to total insanity

leaden carbon Jan 3, 2025, 4:25 PM

#

AHAHAHAH

#

it is evolving

#

it is becoming userspace

torpid wigeon Jan 3, 2025, 4:27 PM

#

sentinece

vale nymph Jan 3, 2025, 4:57 PM

#

Obos ascension

flint idol Jan 3, 2025, 4:57 PM

#

I think it was because of a slight change in how PIDs were made

#

yup

#

I made the kernel's struct process be pid zero

#

before it was pid one

#

[ LOG ] User thread 7 SIGSEGV (rip 0, cr2 0)

#

https://cdn.discordapp.com/emojis/816178972672131083.gif?size=64

short mortar Jan 3, 2025, 5:01 PM

#

(install SIGSEGV handler?)

flint idol Jan 3, 2025, 5:01 PM

#

it's in bash

#

I don't want to touch bash source

short mortar Jan 3, 2025, 5:02 PM

#

do it in libc?

#

the default behaviour is to terminate the process

flint idol Jan 3, 2025, 5:03 PM

#

yes

short mortar Jan 3, 2025, 5:03 PM

#

which is done by libc

flint idol Jan 3, 2025, 5:03 PM

#

wait so the kernel wasn't supposed to do it?

short mortar Jan 3, 2025, 5:03 PM

#

it sends a signal

white mulch Jan 3, 2025, 5:03 PM

#

usually the kernel does it

flint idol Jan 3, 2025, 5:03 PM

#

yeah

short mortar Jan 3, 2025, 5:03 PM

#

huh

flint idol Jan 3, 2025, 5:03 PM

#

unless the handler is not SIG_DFL

#

then it propagates the signal to userspace

short mortar Jan 3, 2025, 5:04 PM

#

i thought it was glibc which printed "segmentation fault"

white mulch Jan 3, 2025, 5:04 PM

#

its bash/zsh

#

I think

flint idol Jan 3, 2025, 5:05 PM

#

here I go hang debugging once again

#

nooo

short mortar Jan 3, 2025, 5:05 PM

#

anyway, posix doesn't say where you do it

#

so you can do it there

#

SIG_DFL -> call libc signal handler (?)

flint idol Jan 3, 2025, 5:06 PM

#

idk

white mulch Jan 3, 2025, 5:06 PM

#

but idk what you'd gain from that

short mortar Jan 3, 2025, 5:06 PM

#

you can print stack traces

flint idol Jan 3, 2025, 5:06 PM

#

flint idol idk

I define it as null and have a special handler for that in the code that sets up the interrupt frame for a signal

short mortar Jan 3, 2025, 5:06 PM

#

without it potentially crashing the kernel

flint idol Jan 3, 2025, 5:06 PM

#

short mortar without it potentially crashing the kernel

use gdb?

short mortar Jan 3, 2025, 5:07 PM

#

for what?

flint idol Jan 3, 2025, 5:07 PM

#

stack trace

short mortar Jan 3, 2025, 5:07 PM

#

it becomes very difficult to use once you have 30 threads on 8 cores

flint idol Jan 3, 2025, 5:07 PM

#

bt

flint idol Jan 3, 2025, 5:07 PM

#

short mortar it becomes very difficult to use once you have 30 threads on 8 cores

make a gdb stub

short mortar Jan 3, 2025, 5:08 PM

#

does it change anything?

flint idol Jan 3, 2025, 5:08 PM

#

you can choose specific threads instead of specific cores

short mortar Jan 3, 2025, 5:09 PM

#

idk

short mortar Jan 3, 2025, 5:09 PM

#

short mortar anyway, posix doesn't say where you do it

my kernel doesn't even have signals, so I have userspace thread do it

vale nymph Jan 3, 2025, 5:14 PM

#

white mulch its bash/zsh

Yes

flint idol Jan 3, 2025, 5:22 PM

#

flint idol here I go hang debugging once again

bru

#

#

(that's an address on the stack)

#

oops

#

I forgor to restore rbp

#

--- a/sysdeps/obos/x86_64/fork.S
+++ b/sysdeps/obos/x86_64/fork.S
@@ -113,6 +113,7 @@ fork_shim:
 
     xor rax,rax
 
+    pop rbp
     ret
 
 #if MLIBC_BUILDING_RTLD

theoeretically this is the fix

#

we just love the fourth parameter passed to this function

#

anyway, after having fixed the SEGV in the forked program

[  LOG  ] User thread 6 SIGSEGV (rip 2412500c8, cr2 2412500c8)```

#

https://cdn.discordapp.com/emojis/816178972672131083.gif?size=64

#

before I debug this, I need a wall

#

to bash my head on

short mortar Jan 3, 2025, 7:01 PM

#

unrelated question - which gcc are you using in your toolchain?

flint idol Jan 3, 2025, 7:11 PM

#

short mortar unrelated question - which gcc are you using in your toolchain?

14.2.0

short mortar Jan 3, 2025, 7:12 PM

#

how did you patch it?

flint idol Jan 3, 2025, 7:12 PM

#

Same way as OSDev wiki

short mortar Jan 3, 2025, 7:12 PM

#

it fails with make[1]: *** No rule to make target '../../gcc-pmos/gcc/config/pmos.opt.urls', needed by 's-options'. Stop.

flint idol Jan 3, 2025, 7:12 PM

#

But with same stuff taken from Elysium

#

Try taking obos' patches and modifying them to pmos

short mortar Jan 3, 2025, 7:14 PM

#

I had my patches working on 12

#

But rebasing on 14 doesn't

flint idol Jan 3, 2025, 7:14 PM

#

https://github.com/OBOS-dev/obos-pkgs/tree/master/patches%2Fgcc

GitHub

obos-pkgs/patches/gcc at master · OBOS-dev/obos-pkgs

OBOS Package Repository. Contribute to OBOS-dev/obos-pkgs development by creating an account on GitHub.

#

Those are my gcc patches if you want to compare

short mortar Jan 3, 2025, 7:17 PM

#

I'm not seeying anything that's much different

#

and related to it

#

hmm

#

I forgot about this

flint idol Jan 4, 2025, 5:23 PM

#

bump

#

I'm gonna rewrite Mm_ForkContext

flint idol Jan 4, 2025, 6:18 PM

#

one thing I am noticing

#

is after a fork

#

read/writes from read only regions break

#

in both the parent and child

#

mostly page faults

short mortar Jan 4, 2025, 6:29 PM

#

How are you doing your CoW?

flint idol Jan 4, 2025, 6:30 PM

#

if (curr->cow && curr->un.cow_type == COW_SYMMETRIC)
{
    info.prot.rw = false;
    MmS_SetPageMapping(toFork->pt, &info, info.phys, false);
}
MmS_SetPageMapping(into->pt, &info, info.phys, false);

#

in short, basically just that

#

for each page

#

if (!curr->cow)
{
    curr->cow = true;
    curr->un.cow_type = COW_SYMMETRIC;
}```

flint idol Jan 4, 2025, 6:30 PM

#

flint idol ```c if (!curr->cow) { curr->cow = true; curr->un.cow_type = COW_SYMMETR...

this is before

short mortar Jan 4, 2025, 6:31 PM

#

How do you invalidate TLB?

flint idol Jan 4, 2025, 6:31 PM

#

invlpg

short mortar Jan 4, 2025, 6:31 PM

#

but what if process has several threads

flint idol Jan 4, 2025, 6:32 PM

#

this is on a uniprocessor build

flint idol Jan 4, 2025, 6:32 PM

#

short mortar but what if process has several threads

it would send an IPI

#

telling other CPUs to invlpg this page

short mortar Jan 4, 2025, 6:32 PM

#

because afaik the fork is the way that it is because it was very easy to implement in unix

#

from what I've been taught, you just swap the whole process onto disk

#

and then fork

flint idol Jan 4, 2025, 6:33 PM

#

the idea behind fork is that you have like 1 byte of memory but need like 50 processes

flint idol Jan 4, 2025, 6:33 PM

#

short mortar from what I've been taught, you just swap the whole process onto disk

*taught trl

#

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

Tenor

short mortar Jan 4, 2025, 6:34 PM

#

flint idol ```c if (curr->cow && curr->un.cow_type == COW_SYMMETRIC) { info.prot.rw = f...

but your pages don't have refcount?

flint idol Jan 4, 2025, 6:34 PM

#

they do

#

                page what = {.phys=info.phys};
                page* phys = (info.prot.is_swap_phys) ? nullptr : RB_FIND(phys_page_tree, &Mm_PhysicalPages, &what);
                if (phys)
                {
                    MmH_RefPage(phys);
                    if (!(phys->pagedCount++))
                    {
                        Core_SpinlockRelease(&toFork->lock, oldIrql);
                        Mm_HandlePageFault(toFork, addr, PF_EC_RW|((uint32_t)info.prot.present<<PF_EC_PRESENT)|PF_EC_UM);
                        oldIrql = Core_SpinlockAcquire(&toFork->lock);
                        MmS_QueryPageInfo(toFork->pt, addr, nullptr, &info.phys);
                        // At this point, paged count is two.
                    }
                }```

#

physical page(s) are referenced before mapping them

#

if they're paged out, then they're paged back in

#

I have an idea

#

that wasn't the problem

#

but it was a problem

#

found a bug with signals

#

frame->rflags &= ~ALLOWED_FLAGS;```

#

in sigreturn

#

I was clearing the allowed flags

#

although for some dumb reason everything be faulting on a null pointer access

#

and for some stupid reason, bash sends SIGSEGV to itself after receiving a SIGSEGV

#

[  LOG  ] (thread 6) syscall Sys_ProcessGetPID(0xfe000000, 0x0, 0x0, 0x0, 0x0)
[  LOG  ] (thread 6) syscall Sys_ProcessGetPID returned 0x0000000000000001
[  LOG  ] (thread 6) syscall Sys_ProcessOpen(0x1, 0x0, 0x0, 0x0, 0x0)
[  LOG  ] (thread 6) syscall Sys_ProcessOpen returned 0x000000000400000a
[  LOG  ] (thread 6) syscall Sys_KillProcess(0x400000a, 0xb, 0x0, 0x0, 0x0)
[  LOG  ] (thread 6) syscall Sys_KillProcess returned 0x0000000000000000```

#

sys_kill(sys_getpid(), SIGSEGV);

#

there also seems to be a bug with reading from cloned FDs

#

when I tested fork manually

#

#include <unistd.h>
#include <stdio.h>

int main()
{
    printf("testing fork\n");
    pid_t pid = fork();
    if (pid == 0)
        printf("in child, we are %d\n", getpid());
    else if (pid > 0)
        printf("in parent, child is %d\n", pid);
    else
        printf("error\n");
    return 0;
}```

flint idol Jan 4, 2025, 7:33 PM

#

I have fixed a bug

#

because now it at least prints the message in the parent

#

[  LOG  ] (thread 7) syscall Sys_FdWrite(0x1, 0x9fe015, 0x16, 0x8024d8, 0x0)```

#

apparently the child also tries to printf

flint idol Jan 4, 2025, 7:35 PM

#

flint idol apparently the child also tries to printf

or more accurately, the child tries to write(stdout, ...)

#

wait what

flint idol Jan 4, 2025, 7:36 PM

#

flint idol

the child is the one printing this

#

[  LOG  ] (thread 6) syscall Sys_HandleClose(0x5000003, 0x5000003, 0xffffffff801f1800, 0xb000004, 0x7ad65000)```

#

we just love it when we leak kernel addresses to usermode

flint idol Jan 4, 2025, 8:01 PM

#

found a bug

#

[  LOG  ] (thread 6) syscall Sys_ProcessGetPID(0x4000006, 0x5000003, 0xffffffff801f1800, 0xb000004, 0x7ad61000)
[  LOG  ] (thread 6) syscall Sys_ProcessGetPID returned 0x0000000000000002
[  LOG  ] (thread 6) syscall Sys_HandleClose(0x4000006, 0x5000003, 0x0, 0xb000004, 0x7ad61000)

#

in between these two syscalls

#

there should be a page fault

#

#

because rdi should've been marked as RO

#

by Mm_ForkContext

#

which was called earlier

#

#

ok I believe I fixed it

#

no random PFs either

#

#

holy

#

except bash hangs after

real pecan Jan 4, 2025, 8:10 PM

#

are u talking to obos over serial?

flint idol Jan 4, 2025, 8:10 PM

#

yes

real pecan Jan 4, 2025, 8:10 PM

#

thats pretty cool

flint idol Jan 4, 2025, 8:10 PM

#

flint idol except bash hangs after

maybe it's because I don't send SIGCHLD

flint idol Jan 4, 2025, 8:10 PM

#

real pecan thats pretty cool

yes (it's definitely not because I couldn't be bothered to implement TTYs)

flint idol Jan 4, 2025, 8:11 PM

#

flint idol maybe it's because I don't send SIGCHLD

nor do I implement waitpid

#

OBOS_PAGEABLE_FUNCTION obos_status Core_ProcessTerminate(process* proc, bool forced)
{
    OBOS_UNUSED(proc);
    OBOS_UNUSED(forced);
    return OBOS_STATUS_UNIMPLEMENTED;
}```

#

I suppose it is time to implement this

#

actually

#

I will remove that

#

and replace it with Core_ExitCurrentProcess

real pecan Jan 4, 2025, 8:13 PM

#

flint idol yes (it's definitely not because I couldn't be bothered to implement TTYs)

do u poll it or actually have irqs and stuff

flint idol Jan 4, 2025, 8:13 PM

#

IRQs

#

the UART driver is an actual uart driver

#

and I have /dev/COM1

#

which controls IO to the serial port

#

as well as opening a connection through an ioctl

real pecan Jan 4, 2025, 8:14 PM

#

nice

flint idol Jan 4, 2025, 8:14 PM

#

https://github.com/OBOS-dev/obos/tree/master/src/drivers/x86/uart

#

which means if:

you can get obos to boot on real hw
have a serial port
and obos' uart driver doesn't shit itself
you can run obos bash

#

on real hw

flint idol Jan 4, 2025, 8:15 PM

#

real pecan nice

it even uses uacpi to detect a UART

real pecan Jan 4, 2025, 8:16 PM

#

#include <uacpi_libc.h> trl

flint idol Jan 4, 2025, 8:17 PM

#

ah yes

#

that's because I didn't have a strnlen in memmanip.h

#

lol

#

so I just used uacpi_strnlen

real pecan Jan 4, 2025, 8:18 PM

#

flint idol it even uses uacpi to detect a UART

thats pretty cool

flint idol Jan 4, 2025, 8:18 PM

#

iirc, it also detects the ports that exist through resources

real pecan Jan 4, 2025, 8:19 PM

#

this tho

    uacpi_find_devices("PNP0500", match_uart, nullptr);
    uacpi_find_devices("PNP0501", match_uart, nullptr);

🤢

#

there's uacpi_find_devices_at btw which takes in a list

flint idol Jan 4, 2025, 8:19 PM

#

real pecan there's `uacpi_find_devices_at` btw which takes in a list

I probably wrote the driver before that existed

#

if it's something relatively new

real pecan Jan 4, 2025, 8:19 PM

#

very old

flint idol Jan 4, 2025, 8:20 PM

#

weird

#

yeah so I can detect devices using uacpi/pci

#

but I don't necessarily have a way

#

to pass that to the driver

#

so I need to do that until I decide to fix that

real pecan Jan 4, 2025, 8:20 PM

#

    .acpiId.nPnpIds = 2,
    .acpiId.pnpIds = {
        "PNP0500", // Standard PC COM port
        "PNP0501", // 16550A-compatible COM port
    },

So this is not used basically?

flint idol Jan 4, 2025, 8:21 PM

#

real pecan ```c .acpiId.nPnpIds = 2, .acpiId.pnpIds = { "PNP0500", // Stand...

nah it's used to detect if the driver exists

#

*device exists

vale nymph Jan 4, 2025, 8:21 PM

#

real pecan this tho ```c uacpi_find_devices("PNP0500", match_uart, nullptr); uacpi_...

y not do a thing where buses can discover their own devices and register them instead of searching for them (what I plan on doing when I redo my astral device driver stuff)

real pecan Jan 4, 2025, 8:21 PM

#

yup

#

IMO your DriverEntry should do a register_device(&pnp_id_list, com_probe)

#

then com_probe is called with the namespace node that matched

flint idol Jan 4, 2025, 8:22 PM

#

I don't quite enjoy the idea of loading a driver solely for the purpose of probing a device that it wants

real pecan Jan 4, 2025, 8:22 PM

#

usually userspace decides what to probe

#

or its statically linked

#

so

flint idol Jan 4, 2025, 8:23 PM

#

anyway, I need to add a execve sysdep

white mulch Jan 4, 2025, 8:23 PM

#

you could have a separate manifest thing along with the driver too

#

that contains the devices that it supports (basically like windows ini)

flint idol Jan 4, 2025, 8:24 PM

#

I do that, but in the "driver header" which is embeded into the driver ELF

flint idol Jan 4, 2025, 8:46 PM

#

idk why but telloff on FD 0

#

always causes a PF

vale nymph Jan 4, 2025, 8:48 PM

#

astral ass naming convention

flint idol Jan 4, 2025, 8:48 PM

#

oh no

#

I was just making it not be obos naming convention

#

the REAL function name is Sys_FdTellOff

#

typedef struct process
{
    // threads waiting for our death
    struct waitable_header waiting_threads;```

flint idol Jan 4, 2025, 9:36 PM

#

so for some reason

#

I think handling signals is causing all to go to shit

#

[ DEBUG ] (pid 1) Handling page fault at 0x0...
[  LOG  ] User thread 6 SIGSEGV (rip 0, cr2 0)
[ DEBUG ] (pid 1) Handling page fault at 0x0...
[  LOG  ] User thread 6 SIGSEGV (rip 400002, cr2 0)```

flint idol Jan 4, 2025, 10:02 PM

#

bruh

#

SIGMAX = 64

#

fixed

#

it was a bug with how I was 'dispatching' signals

#

and it turns out SIGCHLD is masked by bash

#

so that isn't the problem

#

I guess I need to implement sys_waitid

flint idol Jan 4, 2025, 10:18 PM

#

flint idol I guess I need to implement `sys_waitid`

I basically have everything except for three things:

I need to signal waiters on the process* when a process exits (not really that hard)
I need to make process handles waitable in the syscall Sys_WaitForObjects
I need some way to get a process handle for any child process (for the value -1 passed to waitid)

#

the first thing I just did

#

and the second thing I also just did

#

all I need is to do the third thing

#

which I will do in 5 minutes

#

ah yes I also need a syscall to get a process' exit status

flint idol Jan 4, 2025, 10:21 PM

#

flint idol the first thing I just did

I think I also need to make it so that this happens on signals

#

I need to signal a process object when:

it terminates
it receives SIGCONT or SIGSTOP

#

then additionally, if it terminates because of a fatal signal, I need to encode that in the exit status

#

given how much stuff this does, maybe I should just do it in the kernel

#

anyway I will be back

flint idol Jan 4, 2025, 11:03 PM

#

wait what's the difference between sys_waitid and sys_waitpid

#

ok I think I figured that out

thick jolt Jan 4, 2025, 11:18 PM

#

good luck

#

obos doom port when

#

trl

flint idol Jan 4, 2025, 11:24 PM

#

thick jolt obos doom port when

soon

flint idol Jan 5, 2025, 12:53 AM

#

holy shit

#

I can almost do shit

#

#

nvm I think it only works the first time

#

the 2nd time bash gets stuck in a waitpid loop

#

fixed the waitpid loop

#

but instead it just hangs

weak kestrel Jan 5, 2025, 1:04 AM

#

noice
2025 really is the OBOS year

flint idol Jan 5, 2025, 1:04 AM

#

fr

#

I love writing stuff like exit process:

    // Disown all children.
    for (process* child = proc->children.head; child; )
    {```

#

[  LOG  ] User thread 6 SIGSEGV (rip 4563db, cr2 0)```

#

nooo

#

I'm also able to get it to hang

real pecan Jan 5, 2025, 1:18 AM

#

flint idol

Damn literal year of obos

flint idol Jan 5, 2025, 1:19 AM

#

ok, it is able to not hang

#

on the 2nd wait pid

#

flint idol Jan 5, 2025, 1:19 AM

#

flint idol on the 2nd wait pid

or well, any others

#

the problem was it was picking a dead process to wait on if pid == -1

#

which would cause a hang

#

I kinda wanna see if I can run bash in bash

#

it complains about tcsetattr

#

then dies

flint idol Jan 5, 2025, 1:22 AM

#

flint idol then dies

well, it doesn't run the 2nd bash

#

how do I get my bash to do that

#

iirc there was an environment variable I had to set

#

but idr it

#

found it

#

it's the PS1 environment variable

flint idol Jan 5, 2025, 1:35 AM

#

flint idol

something I've noticed is if I do it twice

#

bash just segfaults

#

before I port coreutils

#

I want to stub sys_tcsetattr

#

and sys_tcgetattr

#

and I also want to make it so bash recognizes that a child segfaulted and prints the nice "segmentation fault" message

#

memory corruption at it's finest

#

[  LOG  ] (thread 7) syscall Sys_FdAlloc(0x0, 0x0, 0x0, 0x0, 0x0)
[  LOG  ] (thread 7) syscall Sys_FdAlloc returned 0x000000000000000c
[  LOG  ] (thread 7) syscall Sys_FdOpen(0xc, 0xbfefc0, 0x1, 0x0, 0x0)
[  LOG  ] (thread 7) syscall Sys_FdOpen returned 0x0000000000000000
2
[  LOG  ] (thread 7) syscall Sys_FdTellOff(0xc, 0x0, 0x0, 0x0, 0x0)
[  LOG  ] (thread 7) syscall Sys_FdTellOff returned 0x0000000000000000
[  LOG  ] (thread 7) syscall Sys_FdSeek(0xc, 0x0, 0x2, 0x0, 0x0)
[  LOG  ] (thread 7) syscall Sys_FdSeek returned 0x0000000000000000
[  LOG  ] (thread 7) syscall Sys_FdTellOff(0xc, 0x0, 0x0, 0x0, 0x0)
[  LOG  ] (thread 7) syscall Sys_FdTellOff returned 0x00000000003f18f7
[  LOG  ] (thread 7) syscall Sys_FdTellOff(0xc, 0x0, 0x0, 0x0, 0x0)
[  LOG  ] (thread 7) syscall Sys_FdTellOff returned 0x00000000003f18f7
[ DEBUG ] (pid 2) Handling page fault at 0x0...
[  LOG  ] User thread 7 SIGSEGV (rip 41213307, cr2 0)

#

also for some reason trying to start any child process that actually exists causes this to happen

#

I think it is a bug with sys_seek

#

while that is not the case

#

there was still a bug in sys_seek

#

[  LOG  ] (thread 8) syscall Sys_FdAlloc(0x0, 0x0, 0x0, 0x0, 0x0)
[  LOG  ] (thread 8) syscall Sys_FdAlloc returned 0x000000000000000d
[  LOG  ] (thread 8) syscall Sys_FdOpen(0xd, 0xbfef80, 0x1, 0x0, 0x0)
[  LOG  ] (thread 8) syscall Sys_FdOpen returned 0x0000000000000000
[  LOG  ] (thread 8) syscall Sys_FdSeek(0xd, 0x0, 0x2, 0x0, 0x0)
[  LOG  ] (thread 8) syscall Sys_FdSeek returned 0x0000000000000000
[  LOG  ] (thread 8) syscall Sys_FdTellOff(0xd, 0x0, 0x0, 0x0, 0x0)
[  LOG  ] (thread 8) syscall Sys_FdTellOff returned 0x00000000003f18f7
[  LOG  ] (thread 8) syscall Sys_FdSeek(0xd, 0x0, 0x0, 0x0, 0x0)
[  LOG  ] (thread 8) syscall Sys_FdSeek returned 0x0000000000000000
[  LOG  ] (thread 8) syscall Sys_FdTellOff(0xd, 0x0, 0x0, 0x0, 0x0)
[  LOG  ] (thread 8) syscall Sys_FdTellOff returned 0x0000000000000000
[ DEBUG ] (pid 3) Handling page fault at 0x0...
[  LOG  ] User thread 8 SIGSEGV (rip 41212df4, cr2 0)```

#

although now it PFs somewhere else

#

I think my uart driver is broken lul

torpid wigeon Jan 5, 2025, 2:33 AM

#

flint idol I think my uart driver is broken <:lul:575445162906615818>

oh no

#

it's not broken

#

you're about to find out a funny

flint idol Jan 5, 2025, 2:34 AM

#

lapfedmoment

#

what

torpid wigeon Jan 5, 2025, 2:34 AM

#

https://tenor.com/view/pou-sad-rain-shower-depressed-gif-670686772125037083

Tenor

#

does this happen by chance when you do a backspace

flint idol Jan 5, 2025, 2:35 AM

#

torpid wigeon https://tenor.com/view/pou-sad-rain-shower-depressed-gif-670686772125037083

what does that mean

#

like bash dies?

torpid wigeon Jan 5, 2025, 2:35 AM

#

torpid wigeon does this happen by chance when you do a backspace

.

flint idol Jan 5, 2025, 2:36 AM

#

wait

#

when I do a backspace

#

that's when it breaks

weary hound Jan 5, 2025, 2:36 AM

#

trl

torpid wigeon Jan 5, 2025, 2:36 AM

#

trl

flint idol Jan 5, 2025, 2:36 AM

#

what are you guys hiding from me

torpid wigeon Jan 5, 2025, 2:36 AM

#

reading in a backspace does not delete the character

flint idol Jan 5, 2025, 2:36 AM

#

yes

#

but this is in nc, which iirc, does it's own buffering

#

unless it doesn't, then I'll be damned

torpid wigeon Jan 5, 2025, 2:38 AM

#

changing /bin/bash to /bin
becomes /bin/bash\b\b\b\b\b

#

trolled

#

check out this funny for example

#

flint idol Jan 5, 2025, 2:40 AM

#

I'm pretty sure netcat deals with backspaces

#

because it holds the write until enter is pressed

weary hound Jan 5, 2025, 2:40 AM

#

netcat is probably using cooked mode so yeah

flint idol Jan 5, 2025, 2:41 AM

#

guess my serial driver is buggy then

#

when I talk about it, it goes awau

#

*away

#

damn uart driver

flint idol Jan 5, 2025, 2:49 AM

#

flint idol ``` [ LOG ] (thread 8) syscall Sys_FdAlloc(0x0, 0x0, 0x0, 0x0, 0x0) [ LOG ] ...

I have found out the page fault is in sys_execve

flint idol Jan 5, 2025, 3:20 AM

#

flint idol I have found out the page fault is in `sys_execve`

it was in sys_seek

#

but for some reason

#

it still doesn't execve properly

#

it just does random shit

#

until it exits

#

I get EFAULT after execve

#

so it's a bug in my execve syscall

#

[  LOG  ] (thread 7) syscall Sys_VirtualMemoryAlloc(0xfe000000, 0x0, 0x1017, 0x802dc0, 0x802db4)
[  LOG  ] (thread 7) syscall Sys_VirtualMemoryAlloc returned 0x00000000003a4000
[  LOG  ] (thread 7) syscall Sys_ExecVE(0x3a4000, 0x1017, 0x802e60, 0x410c7b40, 0x0)
[  LOG  ] (thread 7) syscall Sys_ExecVE returned 0x000000000000001f```

#

0x000000000000001f is OBOS_STATUS_PAGE_FAULT

#

char** kargv = allocate_user_vector_as_kernel(ctx, argv, &argc, &status);
if (obos_is_error(status))
    return status;```

#

it reports a PF here

#

after making a change, I get EINVAL

#

uhh

#

why is the argv passed to execve scuffed

#

and by that I mean

#

argv[1] == 1

#

'

#

wth

#

#

that's the string it passes me

#

in argv[0]

#

int sys_execve(const char *path, char *const argv[], char *const envp[])
{
    int ec = 0, fd = 0;
    ec = sys_open(path, O_RDONLY, 0, &fd);
    if (ec)
        return ec;
    off_t off = 0;
    sys_seek(fd, 0, SEEK_END, &off);
    sys_seek(fd, 0, SEEK_SET, 0);
    void* buf = nullptr;
    ec = sys_vm_map(nullptr, off, PROT_READ|PROT_WRITE, 0, fd, 0, &buf);
    if (ec)
        return ec;
    obos_status st = (obos_status)syscall4(Sys_ExecVE, buf, off, argv, envp);
    switch (st)
    {
        case OBOS_STATUS_UNIMPLEMENTED: return ENOSYS;
        case OBOS_STATUS_INVALID_ARGUMENT: return EINVAL;
        case OBOS_STATUS_PAGE_FAULT: return EFAULT;
        case OBOS_STATUS_INVALID_FILE: return ENOEXEC;
        case OBOS_STATUS_NOT_FOUND: return ENOENT;
        default: __builtin_unreachable();
    }
}

#

my execve sysdep doesn't modify argv

#

execl("/bin/shutdown", "");```

#

that's my invocation of exec

#

some cursed shi is happenin bru

#

perhaps it is a problem with fork

#

yeah it turns out to be a problem with fork

#

because with that same call before calling fork

#

it doesn't return EINVAL

flint idol Jan 5, 2025, 12:52 PM

#

I mean not really

#

I have some kernel functions for posix compatibility

#

and my mlibc sysdeps are basically a kernel->posix translation layer

short mortar Jan 5, 2025, 1:19 PM

#

How many kernels using mlibc actually implement POSIX as native API? Thonk

torpid wigeon Jan 5, 2025, 1:20 PM

#

short mortar How many kernels using mlibc actually implement POSIX as native API? <:Thonk:127...

i do

celest acorn Jan 5, 2025, 1:21 PM

#

what do you mean

#

POSIX does not specify anything relating syscalls, only what is exposed in the libc

#

there is no native API for the kernel

flint idol Jan 5, 2025, 1:22 PM

#

was that directed toward me or mishakov?

celest acorn Jan 5, 2025, 1:22 PM

#

mishakov

real pecan Jan 5, 2025, 1:23 PM

#

celest acorn what do you mean

meaning they generally map posix to syscalls 1:1

short mortar Jan 5, 2025, 1:26 PM

#

real pecan meaning they generally map posix to syscalls 1:1

This

#

Like POSIX functions as syscall api

celest acorn Jan 5, 2025, 1:27 PM

#

its common

#

if you are making a POSIX kernel theres little advantage for not doing POSIX-like syscalls

short mortar Jan 5, 2025, 1:34 PM

#

But there are at least 2 microkernels using it + a bunch of other operating systems which are also not Unix clones

flint idol Jan 5, 2025, 1:35 PM

#

then they translate a sysdep to whatever their kernel needs to do

#

obos_status st = (obos_status)syscall4(Sys_ExecVE, buf, off, argv, envp);```

#

I wonder if it's valid to pass arrays like that

#

this is in:

int mlibc::sys_execve(const char *path, char *const argv[], char *const envp[])```

#

it probably is correct, as execve works if I don't fork

#

so it's a problem with fork

celest acorn Jan 5, 2025, 1:38 PM

#

i am personally skeptical of the approach managarm takes

#

and others

#

and I have voiced it for years, I dont really understand why you would have your own API with its own incompatibilities and rough edges for nothing to use it and have to use separate servers for translating to POSIX, instead of using POSIX directly and shaving layers

#

POSIX as a spec allows a lot of extensions on top of it, when I want to do something in Ironclad that is not in POSIX I just add my extensions, not make parallel layers

timid elk Jan 5, 2025, 1:51 PM

#

i mean for my OS I don't want POSIX to be the native interface and I'll only implement what I need for ports

#

so the managarm approach does make sense to me but one has to be careful

short mortar Jan 5, 2025, 1:52 PM

#

I feel like making POSIX native does impose certain restrictions or whatnot on your design

timid elk Jan 5, 2025, 1:52 PM

#

exactly

short mortar Jan 5, 2025, 1:53 PM

#

like in case of microkernels certain interfaces make no sense

timid elk Jan 5, 2025, 1:53 PM

#

I'm fine having a broken or lacking implementation of POSIX to allow for more freedom for my design

short mortar Jan 5, 2025, 1:54 PM

#

but you can certainly emulate them for programs that want it

timid elk Jan 5, 2025, 1:54 PM

#

and if somebody is complaining about that,
a) wtf are you seriously targeting my system
b) use the native interface instead

flint idol Jan 5, 2025, 1:55 PM

#

hmm this is weird, when I use execve instead of execl, nothing breaks

flint idol Jan 5, 2025, 1:55 PM

#

flint idol ```c execl("/bin/shutdown", "");```

as seen here

#

maybe my usage of execl is just wrong

timid elk Jan 5, 2025, 1:57 PM

#

flint idol hmm this is weird, when I use `execve` instead of `execl`, nothing breaks

maybe you just implemented it so well that nothing wants to break

short mortar Jan 5, 2025, 1:58 PM

#

maybe it's has an extra bug which cancels the first one out trl

flint idol Jan 5, 2025, 2:02 PM

#

[  LOG  ] User thread 7 SIGSEGV
[ DEBUG ] Exitting process 2 after receiving signal 11

#

nooo

#

that's a GPF probably

#

otherwise it would've printed cr2

#

if it were a PF

#

[  LOG  ] (thread 7) syscall Sys_ExecVE(0x3a4000, 0x1017, 0xbfefa0, 0xe7fe00, 0x0)```

#

last syscall before disaster struck

#

qemu-system-x86_64: hw/core/cpu-sysemu.c:76: cpu_asidx_from_attrs: Assertion `ret < cpu->num_ases && ret >= 0' failed.```

#

bruh

#

CHAT

#

execve works in bash now

#

flint idol Jan 5, 2025, 2:24 PM

#

time to port coreutils

#

configure: error: could not determine how to read list of mounted file systems```

#

checking for fs_stat_dev... no
checking for fs_info.h... no
checking for BEOS mounted file system support functions... no
configure: error: could not determine how to read list of mounted file systems```

#

https://cdn.discordapp.com/emojis/816178972672131083.gif?size=64

#

@vale nymph

#

help

#

it seems like coreutils needs statvfs

#

vale nymph Jan 5, 2025, 2:33 PM

#

look at the vinix or ironclad sysdeps, they have some file for that

#

since thats linux stuff but they build with the linux option disabled

flint idol Jan 5, 2025, 2:33 PM

#

thanks mathewnd

vale nymph Jan 5, 2025, 2:34 PM

#

you are welcome struct thread* oberrow;

flint idol Jan 5, 2025, 2:57 PM

#

I got it to configure

#

In file included from src/stat.c:19:
./lib/config.h:4831:20: error: expression in static assertion is not an integer
 4831 | #   define alignof _Alignof
      |                    ^~~~~~~~
src/stat.c:879:24: note: in expansion of macro 'alignof'
  879 |         static_assert (alignof (STRUCT_STATVFS) % alignof (fsid_word) == 0);
      |                        ^~~~~~~
In file included from ./lib/stddef.h:80,
                 from ./lib/stdio.h:77,
                 from src/stat.c:31:
src/stat.c:880:24: error: invalid use of undefined type 'struct statfs'
  880 |         static_assert (offsetof (STRUCT_STATVFS, f_fsid) % alignof (fsid_word)
      |                        ^~~~~~~~
src/stat.c:880:24: error: expression in static assertion is not an integer
src/stat.c:882:40: error: invalid use of undefined type 'const struct statfs'
  882 |         static_assert (sizeof statfsbuf->f_fsid % alignof (fsid_word) == 0);
      |                                        ^~
src/stat.c:882:24: error: expression in static assertion is not an integer
  882 |         static_assert (sizeof statfsbuf->f_fsid % alignof (fsid_word) == 0);
      |                        ^~~~~~
src/stat.c:883:54: error: invalid use of undefined type 'const struct statfs'
  883 |         fsid_word const *p = (fsid_word *) &statfsbuf->f_fsid;
      |                                                      ^~
src/stat.c:888:37: error: invalid use of undefined type 'const struct statfs'
  888 |         int words = sizeof statfsbuf->f_fsid / sizeof *p;
      |                                     ^~
src/stat.c:910:54: error: passing argument 1 of 'human_fstype' from incompatible pointer type [-Wincompatible-pointer-types]
  910 |       out_string (pformat, prefix_len, human_fstype (statfsbuf));

#

but alas

#

I do have statvfs in abi-bits

#

and it is installed through install_headers

#

same thing with statfs.h

#

oberrow@AcerAIO:~/Code$ x86_64-obos-gcc test.c
test.c: In function 'main':
test.c:5:12: error: variable 'fs' has initializer but incomplete type
    5 |     struct statfs fs = {};
      |            ^~~~~~
test.c:5:19: error: storage size of 'fs' isn't known
    5 |     struct statfs fs = {};

#

with my test program the same happens

#

statvfs works

#

in fact, I can't even find the header for statfs

#

#

man page says that

#

tried both headers, neither of them exist

#

ok I think I just need to copy them from the linux option

#

lib/freadseek.c: In function 'freadptrinc':
lib/freadseek.c:69:3: error: #error "Please port gnulib freadseek.c to your platform! Look at the definition of getc, getc_unlocked on your system, then report this to bug-gnulib."
   69 |  #error "Please port gnulib freadseek.c to your platform! Look at the definition of getc, getc_unlocked on your system, then report this to bug-gnulib."

#

fuck u

white mulch Jan 5, 2025, 3:15 PM

#

flint idol fuck u

yeah gnulib is a piece of garbage

#

idk who though it was a good idea

#

you need to define SLOW_BUT_NO_HACKS

sharp pike Jan 5, 2025, 3:19 PM

#

add -DSLOW_BUT_NO_HACKS to the CPPFLAGS variable when you configure coreutils

flint idol Jan 5, 2025, 3:32 PM

#

ok

#

horray

#

I have coreutils for obos

#

./configure --host=x86_64-obos CPPFLAGS=-DSLOW_BUT_NO_HACKS=1```

#

although that's how I added it to CPPFLAGS, idk if that's correct

#

patching file /home/oberrow/Code/obos-pkgs/repos/coreutils-9.5/build-aux/config.sub
Hunk #1 FAILED at 1745.
1 out of 1 hunk FAILED -- saving rejects to file /home/oberrow/Code/obos-pkgs/repos/coreutils-9.5/build-aux/config.sub.rej```

#

bruh

#

it configured

#

after doing some fnuy stuff

#

I did cp patches/bash/config.sub.patch patches/coreutils/config.sub.patch

#

then modified the line info in patches/coreutils/config.sub.patch

white mulch Jan 5, 2025, 3:49 PM

#

you could also patch automake and then copy the patched config.sub file from it to everywhere else

#

or run autoreconf with automake from your patched install in PATH, that should also regenerate it afaik

flint idol Jan 5, 2025, 4:00 PM

#

so for some reason

#

bash dislikes space

#

[ LIBC  ] __cxa_guard_acquire contention```

#

bruh

#

I removed my stubs for tcsetattr and tcgetattr

#

to be greeted with that

#

then a SIGILL

#

before a batch prompt

#

well I'll be damned

#

diff --git a/sysdeps/obos/generic/sysdeps.cpp b/sysdeps/obos/generic/sysdeps.cpp
index 78106564..1965770d 100644
--- a/sysdeps/obos/generic/sysdeps.cpp
+++ b/sysdeps/obos/generic/sysdeps.cpp
@@ -453,7 +453,7 @@ int sys_vm_unmap(void *pointer, size_t size)
     return sys_anon_free(pointer, size);
 }
 
-int sys_tcgetattr(int fd, struct termios *attr)
+/*int sys_tcgetattr(int fd, struct termios *attr)
 {
     infoLogger() << __func__ << " is unimplemented" << frg::endlog;
     return 0;
@@ -462,6 +462,6 @@ int sys_tcsetattr(int, int, const struct termios *attr)
 {
     infoLogger() << __func__ << " is unimplemented" << frg::endlog;
     return 0;
-}
+}*/
 
 } // namespace mlibc

#

these were my only changes

#

fuck this

#

[ LIBC  ] __cxa_guard_acquire contention��
[  LOG  ] User thread 6 SIGILL```

#

--- a/sysdeps/obos/generic/sysdeps.cpp
+++ b/sysdeps/obos/generic/sysdeps.cpp
@@ -453,15 +453,4 @@ int sys_vm_unmap(void *pointer, size_t size)
     return sys_anon_free(pointer, size);
 }
 
-int sys_tcgetattr(int fd, struct termios *attr)
-{
-    infoLogger() << __func__ << " is unimplemented" << frg::endlog;
-    return 0;
-}
-int sys_tcsetattr(int, int, const struct termios *attr)
-{
-    infoLogger() << __func__ << " is unimplemented" << frg::endlog;
-    return 0;
-}
-
 } // namespace mlibc

#

meanwhile the diff

#

mlibc is trolling

#

a hello world program worked

#

fuck bash

flint idol Jan 5, 2025, 9:35 PM

#

Guys I got school tomorrow

#

https://cdn.discordapp.com/emojis/816178972672131083.gif?size=64

#

So likely less osdev

#

And I gtg for the rest of the day probably

#

So yeah

flint idol Jan 5, 2025, 9:35 PM

#

flint idol ``` [ LIBC ] __cxa_guard_acquire contention�� [ LOG ] User thread 6 SIGILL```

If anyone has an idea of why the actual fuck this happens

#

It would be nice if you could suggest something

weak kestrel Jan 6, 2025, 3:20 PM

#

#

/jk

flint idol Jan 6, 2025, 3:24 PM

#

Sure

#

Check the license tho

#

Tis Mit

flint idol Jan 7, 2025, 12:51 AM

#

flint idol ``` [ LIBC ] __cxa_guard_acquire contention�� [ LOG ] User thread 6 SIGILL```

https://cdn.discordapp.com/emojis/816178972672131083.gif?size=64

#

x/i 0x4124211a
0x4124211a:  0f 0b                    ud2

#

nooo

flint idol Jan 7, 2025, 1:30 AM

#

I suspect it is a problem within the kernel, and not because of some dumb thing I added to my mlibc sysdeps

#

I turned on debug mode, to be greeted by:

#

bruh

flint idol Jan 7, 2025, 1:54 AM

#

kernel is trolling

#

now gdb is trolling

#

thonk

flint idol Jan 7, 2025, 2:24 AM

#

wat the shit

#

https://cdn.discordapp.com/emojis/816178972672131083.gif?size=64

flint idol Jan 7, 2025, 2:26 AM

#

flint idol <:thonk:575445129863888896>

I set a breakpoint for before that node was made

#

set a watchpoint

#

after it's creation

#

then, to my dismay

short mortar Jan 7, 2025, 2:32 AM

#

Linked lists skill issue?

flint idol Jan 7, 2025, 2:36 AM

#

my linked lists are good, it's my

#

dare I say

#

||allocator||

#

wtf happened

#

the kernel was stable two days ago

#

stashing my latest changes causes the problem to disappear

#

so it's probably something stoobid I did

#

omg

#

the reason I was getting the mlibc error

#

was because I was somehow mismatching mlibc versions

#

[  LOG  ] (thread 7) syscall Sys_FdSeek(0x1, 0x0, 0x1, 0x0, 0x0)
[  LOG  ] (thread 7) syscall Sys_FdSeek returned 0x0000000000000002

#

in cat

#

0x0000000000000002=OBOS_STATUS_INVALID_ARGUMENT

#

which makes sense, what does seeking stdout even do

#

*makes sense for me

#

bash-5.2# /usr/bin/cat /test2.txt
cat: standard output: Operation not implemented (ENOSYS)

#

[ LIBC  ] hit io_seek() error 22```

#

I get that log from mlibc too

flint idol Jan 7, 2025, 3:49 AM

#

yeah what is a seek on stdout supposed to do

flint idol Jan 7, 2025, 3:54 AM

#

flint idol yeah what is a seek on stdout supposed to do

for some reason it works when I put it in a debugger

#

I suppose I need sys_fstat and sys_stat

#

the former doesn't exist, so just sys_stat

#

while I'm doing that, I will be implementing the sys_opendir and sys_readentries sysdeps

#

fsfd_target fsfdt

#

I wonder what the point of that parameter is...

#

perhaps it is so that the kernel can verify that the stated file is actually what the program expects, since iirc, it's not in struct stat

flint idol Jan 7, 2025, 4:20 AM

#

figured out what that does

flint idol Jan 7, 2025, 4:37 AM

#

todo tomorrow probably: sys_fstatvfs

#

and probably also sys_statvfs

#

idk

#

[ LIBC  ] In function fstat, file ../options/posix/generic/sys-stat.cpp:150
__ensure(Library function fails due to missing sysdep) failed```

#

as cat complains about that

#

[ LIBC  ] In function fcntl, file ../options/posix/generic/fcntl.cpp:24
__ensure(Library function fails due to missing sysdep) failediled```

#

and that too

limber wave Jan 7, 2025, 8:08 AM

#

flint idol I turned on debug mode, to be greeted by:

i don’t know where but i saw such a panic kernel somewhere

flint idol Jan 8, 2025, 1:25 PM

#

Bump

limber wave Jan 8, 2025, 1:45 PM

#

flint idol Bump

oberrow why you blocked me 😭

#

the duck is not so bad that it should be blocked

#

I can't send it to him friend and react him

#

but why idk

flint idol Jan 8, 2025, 2:30 PM

#

obos error codes

#

src/oboskrnl/error.h

#

I do it in mlibc sysdeps

#

Translate obos status to errno

limber wave Jan 8, 2025, 2:36 PM

#

oberrow doesn't love me 😔

flint idol Jan 8, 2025, 2:40 PM

#

limber wave oberrow doesn't love me 😔

I unblock you

limber wave Jan 8, 2025, 2:40 PM

#

flint idol I unblock you

yay

#

but why you block me 🤔

flint idol Jan 8, 2025, 2:41 PM

#

Duck spam was getting annoying

limber wave Jan 8, 2025, 2:41 PM

#

flint idol Duck spam was getting annoying

I no more duck spamming

weak kestrel Jan 8, 2025, 5:40 PM

#

limber wave I no more duck spamming

now it's just spamming

limber wave Jan 8, 2025, 5:49 PM

#

weak kestrel now it's just spamming

where

leaden carbon Jan 9, 2025, 10:28 AM

#

oberrow wtf happened to the formatting in this file

#

https://github.com/OBOS-dev/obos/blob/master/src/oboskrnl/scheduler/schedule.c

#

lol

weak kestrel Jan 9, 2025, 10:34 AM

#

mOSs curse

real pecan Jan 9, 2025, 10:36 AM

#

weak kestrel mOSs curse

lmao

empty kernel Jan 9, 2025, 10:44 AM

#

How's OBOS going?

flint idol Jan 9, 2025, 11:29 AM

#

leaden carbon oberrow wtf happened to the formatting in this file

A mix of git merge and obos code style being terrible probably

#

Yeah I have no idea

#

When I looked at it

#

I will be fixing that

flint idol Jan 9, 2025, 11:30 AM

#

empty kernel How's OBOS going?

School started, less time to work on it

real pecan Jan 9, 2025, 11:32 AM

#

flint idol School started, less time to work on it

write code on the phone

flint idol Jan 9, 2025, 11:36 AM

#

I need to fix the build system

#

First

#

On my phone

empty kernel Jan 9, 2025, 8:14 PM

#

flint idol School started, less time to work on it

Unfortunate

leaden carbon Jan 9, 2025, 9:32 PM

#

https://github.com/OBOS-dev/obos/blob/master/src/oboskrnl/locks/mutex.c#L45-L56 I was looking at this and noticed that what if success is set to false and then on the next test of the while condition you actually end up acquiring the lock, success is still false and u end up with a deadlock I think?

#

unlikely scenario obviously but

flint idol Jan 9, 2025, 9:36 PM

#

Good catch

#

I have never had it happened

vale nymph Jan 9, 2025, 9:37 PM

#

you need some whitespace in that code goddamn

flint idol Jan 9, 2025, 9:37 PM

#

Bru

empty kernel Jan 10, 2025, 3:54 AM

#

There's a reason I don't look at OBOS code

flint idol Jan 10, 2025, 4:36 AM

#

Bruh

leaden carbon Jan 10, 2025, 5:11 AM

#

lmao

vale nymph Jan 10, 2025, 9:36 AM

#

https://tenor.com/view/aokiji-akainu-aokiji-vs-akainu-one-piece-anime-gif-18756892

Tenor

#

Astral and obos competing for most unreadable code

leaden carbon Jan 10, 2025, 9:37 AM

#

https://tenor.com/view/tony-stark-shrug-tony-stark-shrug-nice-argument-but-i-have-your-home-address-nice-argument-but-gif-23136379

Tenor

#

meanwhile Nyaux

real pecan Jan 10, 2025, 9:41 AM

#

Lmao

limber wave Jan 10, 2025, 9:44 AM

#

vale nymph Astral and obos competing for most unreadable code

nyaux.

torpid wigeon Jan 10, 2025, 10:12 AM

#

vale nymph Astral and obos competing for most unreadable code

real

frigid tide Jan 10, 2025, 2:50 PM

#

limber wave nyaux.

meanwhile floppa os:

#


// Dispatch based on the operation
    if (flopstrcmp(operation, "sin") == 0 && arg_count > 2) {
        result = sin(flopatof(arguments[2]));
    } else if (flopstrcmp(operation, "cos") == 0 && arg_count > 2) {
        result = cos(flopatof(arguments[2]));
    } else if (flopstrcmp(operation, "tan") == 0 && arg_count > 2) {
        result = tan(flopatof(arguments[2]));
    } else if (flopstrcmp(operation, "sqrt") == 0 && arg_count > 2) {
        result = sqrt(flopatof(arguments[2]));
    } else if (flopstrcmp(operation, "exp") == 0 && arg_count > 2) {
        result = exp(flopatof(arguments[2]));
    } else if (flopstrcmp(operation, "log") == 0 && arg_count > 2) {
        result = ln(flopatof(arguments[2]));
    } else if (flopstrcmp(operation, "pow") == 0 && arg_count > 3) {
        result = pow(flopatof(arguments[2]), flopatof(arguments[3]));
    } else if (flopstrcmp(operation, "deg_to_rad") == 0 && arg_count > 2) {
        result = deg_to_rad(flopatof(arguments[2]));
    } else if (flopstrcmp(operation, "rad_to_deg") == 0 && arg_count > 2) {
        result = rad_to_deg(flopatof(arguments[2]));
    } else if (flopstrcmp(operation, "abs") == 0 && arg_count > 2) {
        result = fabs(flopatof(arguments[2]));
    } else {
        echo("Unknown or invalid 'flopmath' operation. Available operations:\n", RED);
        echo("sin, cos, tan, sqrt, exp, log, pow, deg_to_rad, rad_to_deg, abs\n", WHITE);
        return;
    }

#

💀

leaden carbon Jan 10, 2025, 2:59 PM

#

jesus

#

lol

#

you know you could make an array of strings+function pointers and loop over it

#

just an idea

haughty abyss Jan 10, 2025, 3:03 PM

#

vale nymph Astral and obos competing for most unreadable code

they cant win

#

there is always shkwve

#

although i guess i actually use a code formatter because im too lazy to turn off format-on-save

haughty abyss Jan 10, 2025, 3:04 PM

#

haughty abyss there is always shkwve

and i have actual complex logic

#

like a 700 loc NFA generator

#

with 4 comments total

frigid tide Jan 10, 2025, 3:38 PM

#

i do don’t worry this is being deprecated 😭😭

frigid tide Jan 10, 2025, 3:38 PM

#

leaden carbon you know you could make an array of strings+function pointers and loop over it

😭

flint idol Jan 10, 2025, 4:21 PM

#

frigid tide ``` // Dispatch based on the operation if (flopstrcmp(operation, "sin") == ...

wth

#

time for statfs I think

flint idol Jan 10, 2025, 4:29 PM

#

flint idol todo tomorrow probably: `sys_fstatvfs`

nvm this

#

yeah so I don't have anything in my driver interface

#

for sys_statvfs

flint idol Jan 10, 2025, 5:54 PM

#

I am able to get cat to run

#

but not work

#

before it would complain about statvfs

#

now it doesn't, but when I type stuff, it doesn't work as expected (echoes stuff back at me)

#

trying to do a file causes EFAULT somewhere

white mulch Jan 10, 2025, 5:59 PM

#

flint idol for `sys_statvfs`

statvfs can be implemented with statfs afaik

flint idol Jan 10, 2025, 5:59 PM

#

statfs is a linux sysdep

#

and I don't got linux sysdeps

#

I also implemented sys_statvfs

white mulch Jan 10, 2025, 6:00 PM

#

ah yeah

flint idol Jan 10, 2025, 6:01 PM

#

[  LOG  ] (thread 7) syscall Sys_ExecVE(0x3a4000, 0x5a3e7, 0xc7efe0, 0xe7fe00, 0x0)
[  LOG  ] (thread 7) syscall Sys_ExecVE returned 0x000000000000001f```

#

execve returned OBOS_STATUS_PAGE_FAULT

#

thonk

#

nvm, that is the content of test2.txt

#

flint idol Jan 10, 2025, 6:16 PM

#

flint idol execve returned `OBOS_STATUS_PAGE_FAULT`

I fixed that

#

and another bug with execve

white mulch Jan 10, 2025, 6:17 PM

#

this is an unrelated question but do you do lazy ipl in obos? because I'd be interested in knowing how does that work

#

or how you implement it if you do ig

flint idol Jan 10, 2025, 6:18 PM

#

#elif OBOS_LAZY_IRQL
    irql irql_ = OBOS_IRQ_VECTOR_ID_TO_IRQL(frame->vector);
    if (irql_ < CoreS_GetCPULocalPtr()->currentIrql)
    {
        CoreS_SetIRQL(irql_, CoreS_GetIRQL());
        CoreS_DeferIRQ(frame);
        CoreS_SendEOI(frame);
        return;
    }
    if (!CoreS_EnterIRQHandler(frame))
        return;
    CoreS_SendEOI(frame);
    irql oldIrql2 = Core_RaiseIrqlNoThread(irql_);```
This is in the IRQ dispatcher in my IRQ interface

#

if you can't see what that does, it first checks if the IRQL the vector can run at is less than what the current irql should be

#

if it is, then it continues on with it's life

#

otherwise it sets the IRQL in cr8 (CoreS_SetIRQL)

#

then defers that IRQ in an arch-specific way (CoreS_DeferIRQ)

#

sends an EOI, then returns

#

#if !OBOS_LAZY_IRQL
    CoreS_SetIRQL(to, *Core_GetIRQLVar());
#endif

#

in raise irql it does that simple check

#

to see if it should set the IRQL in cr8

flint idol Jan 10, 2025, 6:21 PM

#

flint idol then defers that IRQ in an arch-specific way (CoreS_DeferIRQ)

on x86_64 this does a self IPI on frame->intNumber

white mulch Jan 10, 2025, 6:22 PM

#

what does that ipi do?

flint idol Jan 10, 2025, 6:22 PM

#

it makes it so that when cr8 is lowered < the IRQL that the IRQ can run at, the IRQ gets run again

white mulch Jan 10, 2025, 6:23 PM

#

ah so you basically let the cpu handle re-running the irq after the irql is lowered

flint idol Jan 10, 2025, 6:23 PM

#

yup

white mulch Jan 10, 2025, 6:26 PM

#

that's actually pretty smart, thanks for the explanation

flint idol Jan 10, 2025, 6:26 PM

#

np

frigid tide Jan 10, 2025, 6:30 PM

#

leaden carbon you know you could make an array of strings+function pointers and loop over it

i fixed it

flint idol Jan 10, 2025, 7:29 PM

#

obos userspace doesn't play nicely with SMP (whoops)

#

flint idol Jan 10, 2025, 7:59 PM

#

it's in the UART driver

flint idol Jan 10, 2025, 8:27 PM

#

I believe I fixed it

#

-smp cores=4,threads=1,sockets=1

#

0xffffffff800315d5 <+2069>:       48 8b 04 25 28 01 00 00 mov    rax,QWORD PTR ds:0x128```

#

clang is being mysterious

#

    driver->ftable.get_blk_size(to_stat->desc, &blkSize);

#

generated after this line of code

#

apparently driver is null

#

I think I found it

#

now it hangs for a bit then triple faults

#

anyway

#

I implemented the sys_open_dir and sys_read_entries sysdeps

vale nymph Jan 10, 2025, 8:42 PM

#

year of the obos desktoip

flint idol Jan 10, 2025, 8:42 PM

#

fr

flint idol Jan 10, 2025, 8:57 PM

#

bruh

#

why is it trying to read_entries on stderr

#

[  LOG  ] (thread 10) syscall Sys_OpenDir(0x0, 0x916dbc, 0x0, 0x0, 0x0)
[  LOG  ] (thread 10) syscall Sys_OpenDir returned 0x0000000000000002
[  LOG  ] (thread 10) syscall Sys_ReadEntries(0x2, 0xcff018, 0x800, 0xcff010, 0x0)
[  LOG  ] (thread 10) syscall Sys_ReadEntries returned 0x0000000000000002```

#

interesting

#

oopsie

#

how fun

#

    status = OBOSH_ReadUserString(upath, nullptr, &sz_path);
    if (obos_is_error(status))
    {
        if (statusp)
            memcpy_k_to_usr(statusp, &status, sizeof(obos_status));
        return HANDLE_INVALID;
    }```

#

returned from there in Sys_OpenDir

#

upath is nullptr

#

thonk

#

I think I know what it is

flint idol Jan 10, 2025, 11:18 PM

#

obos has ls

#

now

real pecan Jan 10, 2025, 11:21 PM

#

Nice

#

Now do mesa

flint idol Jan 10, 2025, 11:40 PM

#

this god damn kernel is never fucking done with memory bugs is it

#

I turn off optimizations

#

it fucking hangs

#

IN HTE RB TREE CODE

flint idol Jan 11, 2025, 12:06 AM

#

for future reference: never do cat /usr/share/info/gcc.info

#

unless you have time to burn

#

what should I port next...

#

grep

#

hmm

#

surely it's not too hard

#OBOS (not vibecoded)