alr have fun :)
#Nyaux
surreal path
thorn bramble
bit 0 of the command port tells you if there's anything to read
surreal path
yea i know
thorn bramble
you should also flush the keyboard buffer when you initialize the keyboard
surreal path
yea i know
ill just the bytes keyboard sends me onto some buffer (reading until there is no more data left), do the state machine until the buffer is no more and then send the eoi
surreal path
i'm just saying
makes sense
thorn bramble
ill just the bytes keyboard sends me onto some buffer (reading until there is no...
just move your current irq handling to a separate function that takes in a byte and replace the irq handler with while (inb(cmdPort) & 0x1) handleByte(inb(dataPort));
surreal path
just move your current irq handling to a separate function that takes in a byte ...
yea thats easier
thorn bramble
nothing ever happens
thorn bramble
this shit is not gonna do anything until next year
surreal path
nahhhh
thorn bramble
it's not like i've been doing it for couple years atp
surreal path
thats completely wrong lol
the amount of contributions you've done torwards other projects + ur own shit as well
anyways brb i go brush teeth and drink wata
like a british
thorn bramble
sounds good
silver yarrow
the ps2 buffer is more than one byte?
surreal path
the ps2 buffer is more than one byte?
apparently yes
silver yarrow
wtf, what type of cement block have I been living under
nah it's def one byte, the one in the controller
surreal path
nah it's def one byte, the one in the controller
not the keyboard tho it seems
thorn bramble
well then maybe it's not a multi-byte buffer
but you should still read until the controller says there's nothing to read anymore
surreal path
just move your current irq handling to a separate function that takes in a byte ...
attempting changes now
ok @thorn bramble weird thing, i did the thing u suggested but it just belives theres an infinite number of the byte you pressed? my while loop is correct and i do update scancode so this is weird as shit
surreal path
lol
thorn bramble
wait am i
surreal path
i dont know
thorn bramble
why did u write it so weird
surreal path
how so
thorn bramble
oh yeah
its like u did not read waht i told u
while (inb(cmdPort) & 0x1) handleByte(inb(dataPort));
read again
carefully
thorn bramble
exactly
thorn bramble
just read
surreal path
ill just read the status reg for shit
thorn bramble
oh yeah status register
surreal path
lol
surreal path
pretty much i htink
thorn bramble
no it is
surreal path
yea
thorn bramble
idk
surreal path
yea may be cooked here
still deadlocking with ze mem lock
not epic
i wonder why im not getting any interrupts in general
i feel like kmalloc here is being called after interrupts are disabled in some part of the code
therefore the lock never gets released, nyaux gets stuck and deadlock time
surreal path
idk
okay i think i may know why, for some reason interrupts are disabled on the mmap syscall
however enabling them explcitly page faults
rip 0x0 my worst enemy
ebon needle
what did i fall asleep watching
thorn bramble
most normal page fault
surreal path
real
not only is demend paging fucked
but if i just dont demend page some weird shit happens
2 more bugs
i literarly found EVEN more bugs
i literarly have 1000000 bugs bro
surreal path
i literarly found EVEN more bugs
and by found i mean encountered, idk where the issue is NOR do i know how to fix it
i want to cryyy in a cornerr
wtf...
and sometimes if i rerun the vm i dont get the page fault??/
wtf
:c
im afraid im gonna need some help with this
i am too scared to ask for anyones help with this cause i have fuckin done that too much
kinda stuck ig
maiden ridge
Eat a 5 star
tender gorge
i am too scared to ask for anyones help with this cause i have fuckin done that ...
Is okie 🥺
tender gorge
i literarly have 1000000 bugs bro
My code has 99999999999999999 bugs 
silver yarrow
my code has bugs²
broken depot
My code has :{:|:&};: bugs
edgy pilot
My code has :{:|:&};: bugs
I'm feeling like crashing my phone
surreal path
maybe turning on ubsan might hlep
idk
im out of ideas
okay i turned on ubsan
its catching a few things related to pvclock
ima fix and see what changes
okay how is that out of bounds??
thats lit on the osdev wiki
and its deadlocking on uacpi with ubsan on?
changing it from 1 << 31 to 1 << 30 fixes it???
the deadlock for uacpi
and it still thinks theres pvclock
wtf
surreal path
changing it from 1 << 31 to 1 << 30 fixes it???
okay casting 1 << 31 and ecx into a u64 solves that
we still having the same issue and ubsan caught jackshit
@brisk zenith i know i ask help from you a lot but im like really really stuck with this, it would be helpful if you could take a look, apologies for the ping
broken depot
changing it from 1 << 31 to 1 << 30 fixes it???
1 will be parsed as an int, so it will be signed 32-bit integer. Shifting it left by 31 changes the sign, which may have triggered the ubsan.
surreal path
1 will be parsed as an `int`, so it will be signed 32-bit integer. Shifting it l...
makes sense
broken depot
Were you to do 1u << 31 you wouldn't have this issue.
surreal path
Were you to do `1u << 31` you wouldn't have this issue.
yea i did that, solved it
broken depot
What's the thing you're debugging right now?
surreal path
What's the thing you're debugging right now?
its a really complicated issue ill ping you to what im experiencing
broken depot
I see. The mmap syscall is broken and crashes when interrupts get enabled, but deadlocks otherwise?
surreal path
yep
broken depot
Your master branch appears to contain an incorrect invokation of sprintf here: https://github.com/rayanmargham/NyauxKC/blob/69ebbb8bf2c0b215529328bd36e76b43b6dffd1c/kernel/src/arch/x86_64/syscalls/syscall.c#L45
Unless you have sprintf defined as something else?
surreal path
Your master branch appears to contain an incorrect invokation of `sprintf` here:...
sprintf is just serial printf
broken depot
I see
supple robin
unfortunate naming
supple robin
💀
surreal path
Anyway do you know where it crashes and potentially the backtrace?
no thats the thing, i hava no idea where it dies and the fact its rip 0 and it dies on instruction fetching a userspace program is something
https://github.com/rayanmargham/NyauxKC also @broken depot you should look at master branch
broken depot
Is it consistent enough you could re-run and debug until it crashes?
surreal path
Is it consistent enough you could re-run and debug until it crashes?
also ive tried to figure out why it dies but i literarly cannot find it
broken depot
RIP becoming 0 can mean one of a couple things:
- Return address corrupted into 0
- Noreturn function does return
- Null function pointer called
I would suspect that last one first
surreal path
if it was a null function pointer with the vfs
that would happen on kernel land
i dont use function pointers anywhere else
as for noreturn function, i usually have panics on functions that arent supposed to return
broken depot
It's userland that goes RIP=0?
surreal path
It's userland that goes RIP=0?
yep
page fault is error 0x14
instruction fetch on userspace
broken depot
ok
surreal path
i do???
ebon needle
what its trying to do
surreal path
that lit cannot happen
surreal path
what its trying to do
read above
ebon needle
i dont finding qemu log
surreal path
what
ebon needle
do you post your qemu log
surreal path
i dont need the qemu log for this as my idt catches the exception??
broken depot
Presumably userspace gets RIP=0 immediately after the syscall drops back to ring 3? Wouldn't that suggest that either something corrupted the syscall return address to 0 or that you restore it wrongly?
I'm looking at your syscalls.asm for x86_64 rn
surreal path
Presumably userspace gets RIP=0 immediately after the syscall drops back to ring...
could be? but it should be restored i believe
broken depot
I cannot immediately see any problem in that file.
surreal path
I cannot immediately see any problem in that file.
yea i lit dont know
broken depot
And this relatively directly calls syscall_mmap
surreal path
pretty much
broken depot
How large is the stack this runs on? Is there any possibility there's an undetected stack overflow?
surreal path
How large is the stack this runs on? Is there any possibility there's an undetec...
the stack is quite big
#define KSTACKSIZE 16384
#define USTACKSIZE 1048576
thats the stack
sizes
broken depot
That won't overflow like this
ebon needle
the stack is quite big
can you say what is the problem, maybe i can help
broken depot
Do you know the mmap parameters that cause the crash?
surreal path
can you say what is the problem, maybe i can help
read above i explain the issue above
ebon needle
ok
surreal path
Do you know the `mmap` parameters that cause the crash?
i believe so
ill check for ya
syscall_mmap(): size 602112 flags 32
Page Fault! CR2 0x0
RIP is 0x0. Error Code 0x14
flags in hex btw
broken depot
Hint nonnull? Or null?
surreal path
ill print hint hold on
ebon needle
okay i think i may know why, for some reason interrupts are disabled on the mmap...
wait you enabling interrupts on syscall???
surreal path
yes?
surreal path
ill print hint hold on
syscall_mmap(): size 4096 flags 32, hint 0x41600000
syscall_mmap(): size 90112 flags 32, hint 0x41601000
syscall_mmap(): size 847872 flags 32, hint 0x41617000
again its kinda inconsistent
but there is hint
surreal path
syscall_mmap(): size 4096 flags 32, hint 0x41600000
syscall_mmap(): size 90112 f...
inconsisent meaning it will either do the page fault or deadlock or work
usually page fault
broken depot
wait you enabling interrupts on syscall???
That should be fine as long as the syscall stack is dedicated to it and either no preemption or it's set up as a proper thread. BadgerOS does it in this last way.
ebon needle
That should be fine as long as the syscall stack is dedicated to it and either n...
i always disabling interrupts in syscalls and in interrupts handlers
flat nymph
changing it from 1 << 31 to 1 << 30 fixes it???
(uint32_t)1 << 31?
surreal path
thats why i enable it lol
because it SEEMS like some thread x is holding the mem lock and then thread y calls the fuckin memmap syscall and it deadlocks on that
so i need preemption on syscalls
broken depot
Certain code can't safely run with interrupts disabled actually. That's the case for a lot of internal things in my kernel because it's made to be preempted basically everywhere.
ebon needle
i want preemption to happen during a syscall
what if other process use syscall when other process already in syscall on same cpu
surreal path
Certain code can't safely run with interrupts disabled actually. That's the case...
ik
ebon needle
what if other process use syscall when other process already in syscall on same ...
do you changing stacks ?
surreal path
what if other process use syscall when other process already in syscall on same ...
should run fine?
surreal path
do you changing stacks ?
they are different stacks
surreal path
(uint32_t)1 << 31?
i realized
broken depot
what if other process use syscall when other process already in syscall on same ...
Just use a separate stack. I always pair a kernel stack to a userland thread for this reason.
Anyway, back on topic, hint is nonnull
surreal path
yep
broken depot
I'm curious as to why this slabfree is here: https://github.com/rayanmargham/NyauxKC/blob/master/kernel/src/mem/vmm.c#L228
surreal path
I'm curious as to why this `slabfree` is here: https://github.com/rayanmargham/N...
the vmm region structure is allocated with the slab allocator
ebon needle
they are different stacks
i mean do you have per process syscall stack ?
broken depot
-# that would make your asnwer "no"
broken depot
If you are able, maybe you should print the stuff that gets used by syscall for return address when running mmap?
surreal path
If you are able, maybe you should print the stuff that gets used by syscall for ...
wdym by this
broken depot
At enter and also just before returning.
broken depot
wdym by this
Print what the syscall return address is
broken depot
Both when just entering syscall_mmap and just before returning.
surreal path
i dont wanna do
id rather not cause that means i need to fuckin do some weird ass shit
im too braindead
broken depot
Well, rcx gets stored at an offset of from the value of [gs:8], which would be your kernel syscall stack top.
So that would be *(uint64_t*)(stack_top_here - 3*8)
surreal path
ill try that
surreal path
ill see hold on
broken depot
That would be the address I just told him to read
surreal path
^
broken depot
Literally what I was figuring out the pointer to.
surreal path
Literally what I was figuring out the pointer to.
looks fine
i print on entry and exit of syscall
ebon needle
i print on entry and exit of syscall
then syscall is not cooked
broken depot
Error code still 0x14?
surreal path
yep
broken depot
So IRQ 14 (decimal), error code 20 (decimal)
surreal path
yes
broken depot
SDM agrees that this is user instruction fetch
surreal path
yep ik
broken depot
I'm not super familiar with x86 so I wanted to check
surreal path
I'm not super familiar with x86 so I wanted to check
makes sense
surreal path
maybe user stack is overcooked ???
wha
broken depot
Check that your user stack is correct. If that stack pointer got moved somewhere that made the return address 0, you could see this fault.
surreal path
i can print that on mmap ig
@broken depot yea okay
something cooks the user stack ptr
edgy pilot
bro how does your kernel even run 💀
surreal path
wdym
broken depot
That is the same at enter and exit, which implies it's actually something else that causes the thing at the last mmap
surreal path
That is the same at enter and exit, which implies it's actually something else t...
yea
surreal path
bro how does your kernel even run 💀
??
surreal path
bro is not explaining
broken depot
Because how can this be a valid user stack pointer? I think that something is going wrong in a context switch somewhere.
edgy pilot
bro is not explaining
bugs, bugs everywhere
surreal path
bugs, bugs everywhere
true
broken depot
Be that syscall enter/exit or thread switch, this cannot be correct.
surreal path
yea
broken depot
Perhaps check you disable interrupts before you get here: https://github.com/rayanmargham/NyauxKC/blob/master/kernel/src/arch/x86_64/syscalls/syscalls.asm#L93
surreal path
i can attempt to disable interrupts there
surreal path
Perhaps check you disable interrupts before you get here: https://github.com/ray...
doesnt solve it
same issue
broken depot
Does this thread yield or get preempted between these two mmap calls?
surreal path
Does this thread yield or get preempted between these two mmap calls?
id assume so
broken depot
I think now would be about time to get the debugger to figure out what corrupts the stack pointer.
surreal path
fun
brisk zenith
<@938223886439436320> i know i ask help from you a lot but im like really really...
Can't right now sorry, busy playing through deltarune
surreal path
Can't right now sorry, busy playing through deltarune
BASED???
DELTARUNE IS FUCKING BASED
yooo monkuous have fun
brisk zenith
Thanks
brisk zenith
Accidentally lost my save some time over the last few years so I have to play through the whole game again
surreal path
that makes sense lol
brisk zenith
youll love chapter 3 and 4
I mean it's a Toby fox game I'm sure I'll love them
surreal path
you will
surreal path
I SAW THAT STAR INFY, DELTARUNE PLAYER TOO?
kind root
not my thing
surreal path
fair
kind root
i like games with graphics 
surreal path
@broken depot
so something happens and the rsp is swapped with syscall_stack_ptr_tmp
thats fucking weird as shit
oh
@broken depot I FIXED THE PAGE FAULT ISSUE!!!!
nvm
kind root
@surreal path btw u know u can debug from vscode directly right
surreal path
<@787341689333350450> btw u know u can debug from vscode directly right
ik
but im too lazy to set that shit up
kind root
a made a script to autogenerate vscode configs for that for my kernel
works pretty well
surreal path
where can i find it
broken depot
<@322418405095505920> I FIXED THE PAGE FAULT ISSUE!!!!
kind root
im gonna implement idt and gdt in my kernel now, should I add gdt.. ok and idt.. ok @surreal path ?
surreal path
<:letsgo:1282155513156468747>
bro did not see the nvm
surreal path
im gonna implement idt and gdt in my kernel now, should I add gdt.. ok and idt.....
yes 
broken depot
broken depot
im gonna implement idt and gdt in my kernel now, should I add gdt.. ok and idt.....
Yes
surreal path
slient gdt init...ok
edgy pilot
slient gdt
daring juniper
Write the string "gdt init... ok" to a random memory address
kind root
maybe i should detect avx512 support and then that would enable verbose gdt
elder shoal
im gonna implement idt and gdt in my kernel now, should I add gdt.. ok and idt.....
and you dont ask me, the biggest gdt ok idt ok connoisseur... I know the real ones
surreal path
okay i feel like its in relation to syscall stack tmp and syscall user sp
but doing what i should do breaks mlibc (which means my kernel is broken)
i dont know what im supposed to do in the actual scheduler itself with syscall stack tmp and syscall user sp
do i just leave them
idk
okay
i think i have a general idea whats happening
because i enable interrupts for the syscalls, at the start of the syscall i throw the user rsp into syscall_stack_ptr_tmp and load kernel_stack_ptr from gs then before sysret i load from syscall_stack_ptr_tmp into rsp
now what is probs happening is that on preemption i save the context, but that saves the user context not wherever the syscall is, sched yield works because it manually builds its own stackframe. so basically when preemption happens i go to another thread and whenever i switch back to the thread i load the USER stack and rip, now the user thread assumes the syscall went well and just dies
how to fix this idrk
surreal path
because i enable interrupts for the syscalls, at the start of the syscall i thro...
@brisk zenith im assuming this is kinda correct?
brisk zenith
idk enough about your kernel design to say
surreal path
right
brisk zenith
if that is indeed what's happening then it'd definitely cause issues
brisk zenith
idk
surreal path
fair
brisk zenith
sorry for the short responses im busy fighting jevil
surreal path
sorry for the short responses im busy fighting jevil
spam the act pirroute if your trying to beat him peacefully
its op
and dont apologise lol glad ur having fun
surreal path
spam the act pirroute if your trying to beat him peacefully
if not, just defend for rude buster and spam rude buster
brisk zenith
this is where im at rn, just finished the giant fullscreen attack
not looking too good item wise
brisk zenith
no this is on linux with proton
surreal path
this is where im at rn, just finished the giant fullscreen attack
okay so defend, rude buster and heal prayer susie
surreal path
yea good
brisk zenith
well then
surreal path
you made a program to keep track of this?
brisk zenith
yeah my inventory is full of them
surreal path
okay
use them sparringly
during the fight
you wanna take advantage of ralseis heal prayer
if your like <20hp then yea def heal with an item
surreal path
because i enable interrupts for the syscalls, at the start of the syscall i thro...
wait no this shouldnt happen???? because its happening via an interrupt??? wtf
idk anymore
time to cry in a cornerrr
kind root
time to cry in a cornerrr
u mean lock in
surreal path
u mean lock in
no
kind root
yes
kind root
try to imagine what u would do if u were monkuous
surreal path
im not monkuous i cant imagine what goes in their head lol
kind root
try really hard
surreal path
all i can think of is printing the rip and rsp of a frame every preempation
but thats not because i imagined i am monkuous
i lit just came up with that rn
kind root
good start
surreal path
all i can think of is printing the rip and rsp of a frame every preempation
ill also print syscall tmp ptr
in mmap
rip and rsp looks sane
kind root
try different opt levels as well maybe?
broken depot
try different opt levels as well maybe?
Higher opt levels can make debugger info worse and are also likelier to make optimizations that would break with ub
kind root
Higher opt levels can make debugger info worse and are also likelier to make opt...
maybe hes already doing o2 or smth
surreal path
im already using o2
broken depot
maybe hes already doing o2 or smth
All I know is that I go to -O0 anytime I get a heisenbug
kind root
now try that on linux
broken depot
now try that on linux <:trl:829316641191886888>
Pain
-O0 Linux kernel sounds slow as mollasses
kind root
nah it just wont compile
broken depot
That is very surprising
kind root
why wouldnt they??/
idk, tech debt
broken depot
Are they relying on different behaviours of I suppose specifically GCC on -O2
kind root
Are they relying on different behaviours of I suppose specifically GCC on -O2
hard to say because it doesnt even compile because stuff doesnt get optimized out
but it probably wont boot
99%
broken depot
Ahhhh
kind root
surreal path
the addr are always the same but sometimes ptrtmp doesnt print
broken depot
Not optimized out -> uses some functions that are compiler built-ins?
kind root
Not optimized out -> uses some functions that are compiler built-ins?
see some examples in the so question
kind root
the addr are always the same but sometimes ptrtmp doesnt print
big ub
surreal path
ill turn on O0
kind root
wait wtf were u using
broken depot
the addr are always the same but sometimes ptrtmp doesnt print
That implies the MMAP isn't always anon? Otherwise UB.
surreal path
wait wtf were u using
O2
kind root
bruh
surreal path
That implies the MMAP isn't always anon? Otherwise UB.
no i handle that, if it isnt anon i enosys that shit
kind root
bro is debugging with O2 
gray hill
but even with O0 some values still get optimized out
kind root
Og
surreal path
if u ignore the preemption taking shit it seems to be that syscall user tmp gets changed on the same thread
by something
gray hill
i use O0 and -g
surreal path
if u ignore the preemption taking shit it seems to be that syscall user tmp gets...
bptrtmp 0xfcf10 t 3
and then
ptrtmp 0x100a48 t 3
broken depot
if u ignore the preemption taking shit it seems to be that syscall user tmp gets...
Set a watch point at the address of syscall user tmp
kind root
-ggdb3 for most debug info
surreal path
-ggdb3 for most debug info
what does that do
surreal path
Set a watch point at the address of syscall user tmp
will do
gray hill
-ggdb3 for most debug info
never seen this flag
kind root
what does that do
adds debug info
gray hill
works with clang too?
broken depot
-ggdb3 for most debug info
I knew of -ggdb but there's even higher tiers of it?
kind root
yeah lol
surreal path
it made it a hw watchpoint?
kind root
works with clang too?
should work ig
surreal path
adds debug info
like what
surreal path
it made it a hw watchpoint?
this aint kvm
broken depot
it made it a hw watchpoint?
GDB thinks it's hardware because QEMU reports it as such
gray hill
i also finally managed to start debugging with CLion
broken depot
like what
More accurate tracking of register <-> variable mappings for the most part AFAIK
gray hill
and it's gdb integration is SO GOOD
kind root
i also finally managed to start debugging with CLion
how did u set that up
with qemu
i found some old extension but it didnt work
gray hill
how did u set that up
no extension needed
don't mind the mappings
they are not needed for me
they are a leftover from windows
i had a ton of trouble trying to use it from wsl tho
surreal path
Set a watch point at the address of syscall user tmp
its not an address its a value so i need to watch the cpu
gray hill
but native linux works very well
kind root
oh so its native linux
kind root
and u have to run qemu manually right
gray hill
yes
kind root
i have a more automated thng with vscode
kind root
how?
kind root
ig, sucks that this only works on naitve linux
gray hill
using before launch
surreal path
if u ignore the preemption taking shit it seems to be that syscall user tmp gets...
nope its inconsisent
sometimes that doesnt happen
kind root
couldnt make it work on wsl
gray hill
ig, sucks that this only works on naitve linux
it theoretically could work on wsl too
maybe it was bc i had gef installed
kind root
strange because it should be the exact same speed
gray hill
and basically instant with cmake config and compilation
gray hill
strange because it should be the exact same speed
it is def way faster
surreal path
and it still page faults for the same reason
still on O0
gray hill
can run the whole os till scheduler start in less than 1s
frigid cliff
I usually recognize people by pfp and I was extremely confused who this random person monopolizing the nyaux thread and then I realized 
gray hill
- i got at least double the speed of uacpi loading
surreal path
I usually recognize people by pfp and I was extremely confused who this random p...
whos monopolizing
huh
me confusion
surreal path
crazy
frigid cliff
whos monopolizing
I thought you were a random person taking over your own thread
surreal path
lmao
gray hill
oh XDD
frigid cliff
I didn’t recognize you because you changed you pfp 😭
gray hill
yes
surreal path
we lock in
gray hill
agree
frigid cliff
Anyway
kind root
imagine using the nyaux thread to debug nyaux
surreal path
😱
frigid cliff
imagine using the nyaux thread to debug nyaux
gray hill
anyway, good resources on ipc?
frigid cliff
I’m a little slow sometimes
kind root
fast memcpy
gray hill
?
surreal path
WTF
kind root
= good ipc
gray hill
yeah i have a "fast" memcpy
kind root
yeah i have a "fast" memcpy
even for user copies?
surreal path
its this values before preemption
gray hill
for user i'll use libc one
kind root
i mean copy_from_user
gray hill
with xmm and other improvements
surreal path
it fucking changed
gray hill
i mean copy_from_user
huuh, what
kind root
huuh, what
how do u access user pointers in your kernel
surreal path
my kernel is trolling me
idk why its changing
like the only time i touch it is fork()
gray hill
how do u access user pointers in your kernel
i still have not user mode yet, but each process has it's own context
so just load the context
surreal path
ill print before fork happens cause i didnt get a fork log
surreal path
no idea
kind root
i am trying to get there now
u basically need a memcpy that can return an error if it page faulted during the copy so u can return -EFAULT back to user
gray hill
rn i have a bigger issue
surreal path
fork goes fine???
gray hill
the stupid kvm clock resets itself
gray hill
ghosts hunted your kernel
gray hill
u basically need a memcpy that can return an error if it page faulted during the...
rn tho, i want ipc only for kernel threads
for drivers
mossy belfry
fast memcpy
is fast memcpy the fancy compiler intrinsic memcpy
edgy pilot
is fast memcpy the fancy compiler intrinsic memcpy
no, it's either rep movsb, memcpy optimised for certain sizes and alignments, sse
or why not all of them
mossy belfry
rep vmovdqu16
kind root
is fast memcpy the fancy compiler intrinsic memcpy
compiler intrinsic just calls into your memcpy
mossy belfry
oh... 💥
maiden ridge
and can you ask him specifically this: "can you send an iso of your os with doom included so my buddy can beat episode 1 while using the same computer to heat water for his tea?"
@linuxmaster2.0
@surreal path
maiden ridge
Damn when can he expect a working iso
surreal path
when its fixed
surreal path
i will send when its fixed
maiden ridge
i will send when its fixed
Aight cool 
surreal path
we ball at 5 or smt
little tired
im gonna try to watchpoint the fucking syscall ptr tmp like @mossy belfry said
if that doesnt work
out of ideas again
mossy belfry
good luck 💥
surreal path
thanks
surreal path
yo mb
its just a bit too hot
im just waiting for the temp to cool
its really hot today
mossy belfry
average day in the southern us
surreal path
okay bro
mossy belfry
good luck on the weather
surreal path
execution date today
flat nymph
it's 31 Celsius where I live
gray hill
35 here
silver yarrow
12C
gray hill
wow
daring juniper
22 Centigrade for me
edgy pilot
execution date today
daring juniper
bro uses centigrade
Centigrade sounds cooler to me lol
elder shoal
It was 3°C in the morning today, you can take some cold
surreal path
okay so i tried the watchpoint on the syscall user tmp
var
no luck, gdb catches nothing
guess we gotta wait for monkuous i suppose?
fucking hell bro
surreal path
yea i am lit desperate
fucking nothing
i just want this solved
there has to be SOMETHING im missing
mossy belfry
okay so i tried the watchpoint on the syscall user tmp
do you know where it suddenly changes
surreal path
mmap
mossy belfry
i have an idea, you can break it right before that and manually step step step and print it out until it changes
mossy belfry
did it not show up 💀
surreal path
doesnt help
mossy belfry
so u were able to catch the change tho right
surreal path
im even on O0 as @kind root suggested
surreal path
so u were able to catch the change tho right
it happens on mmap after forking a process and attempting to mmap shit for the dynamic linker
mossy belfry
is monokous able to run your kernel too or do i just have to stare at code lol
i'll try to run it 💥
mood.
ok i think the building is going well 🧀
mossy belfry
Creating config file /etc/mercurial/hgrc.d/hgext.rc with new version
/jinx: line 656: cd: /base_dir/base-files: No such file or directory
make[1]: *** [GNUmakefile:13: nyaux.iso] Error 1
make[1]: Leaving directory '/home/gummi/NyauxKC'
make: *** [GNUmakefile:10: all] Error 2
make: Leaving directory '/home/gummi/NyauxKC'
🙁
uh oh
maybe i will rm the git repo and try again
mossy belfry
it's compiling just fine
rayan your build system takes a little while
is this why the build dir was 20gb or something
very many time
gray hill
LOL WHAT
mossy belfry
idk man i managed to write 3 unit tests, eat dinner, and build nodejs and nyaux still isnt done
"build nodejs" isn't a joke here, gentoo decided to update it today
gray hill
XD
surreal path
its called having a build system
mossy belfry
linux kernel scheduler was probably in a bad mood today, it's ok
surreal path
wtf are u one abt lol
mossy belfry
it didnt feel like running the build system
mossy belfry
thats normal for any build system
what is it actually building tho, is it just making sure that your environment == my environment so we dont have "Works on my machine!" scenarios
hmmm
i shall fix diz
kind root
mood.
@surreal path btw you should write Co-authored-by: username in your commit messages if you want to give attribution
It will even display monkuous pfp and stuff next to the commit
surreal path
ah i see
mossy belfry
is this supposed to do this
surreal path
rm -rf subprojects/uacpi
mossy belfry
same issue
surreal path
<@787341689333350450> btw you should write Co-authored-by: username in your comm...
yea i def should next time they help
can u just
mossy belfry
ok it worked now
mossy belfry
mossy belfry
i did do that
mossy belfry
i did that already and it took 40 minutes to rebuild your project
mossy belfry
i dont really feel like waiting that long again and i already built sysroot
surreal path
u gave up that was ur fault but ok wtv
mossy belfry
your build script is broken, the first time i compiled it i just ran make in sysroot and it created that error
i think you forgot to include uacpi as a submodule or something
mossy belfry
i ran make in sysroot
surreal path
tell me the error
mossy belfry
that was the error
i have a very long list of build output commands one sec
see here it says it took 38 minutes
surreal path
see dms
surreal path
last resort chat
im just gonna ask the osdev forums
and if we get no help there
idk
kind root
im just gonna ask the osdev forums
Letsgo
kind root
Lol idk
kind root
It has some experienced people
mossy belfry
i think i have to open a DE now 😢
it ok tho
12 second boot nice
now how do i make the bug happen
tawdry mirage
im just gonna ask the osdev forums
In total there are 3110 users online :: 10 registered, 0 hidden and 3100 guests (based on users active over the past 5 minutes)
surreal path
run any program that forks
mossy belfry
did it happen
tawdry mirage
ai scrapers are eating the forum rn
surreal path
> In total there are 3110 users online :: 10 registered, 0 hidden and 3100 gues...
what???
surreal path
did it happen
run it again
gray hill
ai scrapers are eating the forum rn
are those the ones that ignore robots.txt?
tawdry mirage
are those the ones that ignore robots.txt?
so basically all of them?
gray hill
the devs of claude are the worst ones
but you can use cloudflare to filter them out i think
mossy belfry
25gb build dir, nice
surreal path
thats fine
gray hill
then idk what im gonna do
?
surreal path
wdym by ?
gray hill
wdym what you gonna do?
surreal path
i mean what i said, i literally dont know how to solve this issue.
i dont know what to do going forward
gray hill
is the forum down?
surreal path
no its just mostly bots now
surreal path
well, try something, like feed your whole project into o3 and hope for the best
what?
gray hill
maybe it can help a bit
gray hill
it is very good at finding issues
surreal path
????
gray hill
well, your problem is probably caused by an issue in your os code no?
surreal path
yes
gray hill
maybe the AI can find it
daring juniper
I find that unlikely lol
gray hill
I find that unlikely lol
well, researchers used it to find some important bugs
so not all hope is lost
gray hill
look it up
daring juniper
🤷♂️
daring juniper
surreal path
just ai models and people saying NOT to use O2-O3 when debugging
gray hill
surreal path
daring juniper
at first when I heard O3 I thought he was talking about the optimization flag XD
gray hill
oh XD
glossy walrus
<https://www.reddit.com/r/linux/comments/1ky83u3/openais_o3_ai_found_a_zeroday_v...
one out of the million made up instances or false positives, congrats
gray hill
well, ofc
glossy walrus
just please stop suggesting LLMs for this
gray hill
but it might help
glossy walrus
don't defer your critical thinking to clankers
glossy walrus
but it might help
it does, for speedrunning atrophying your brain
glossy walrus
it is very good at finding issues
congrats you won the leo prize for shit quote of the day
worldly condor
but it might help
it probably won't
tawdry mirage
no its just mostly bots now
the problem is more that the bots are slowing the forum down a ton
surreal path
oh i see
tawdry mirage
the bots are only reading (since they're guests anyway)
coral dove
22 Centigrade for me
299 Kelvin where im at rn
daring juniper
Eyy!
edgy pilot
don't defer your critical thinking to clankers
I tried to use an llm for programming and I started to just wait for autocompletions instead of actually thinking
I removed it after I realised that
glossy walrus
shit like GH Copilot is good for autocompleting memcpy and appending comments when closing C++ namespaces
for the rest have fun
it can be useful for doing stuff like asking what does linux's function x_y_z() do` tho
mossy belfry
i dont need copilot check out how incredibly useful my autocomplete is, you can find the function you need in only around 2-3 business days
(no idea why this happens)
surreal path
hello
i have just waken up
give like a few hours and we will post onto the forum
wait qwinci why can user tmp change, who changes it???
i am reading old messages as to figure out why some parts of my code do this
maybe this could be our issue?
thorn bramble
give like a few hours and we will post onto the forum
oh god no please
surreal path
oh god no please
why?
thorn bramble
it's just a bunch of boomers
don't
if i were u i'd rather spend an eternity trying to debug it myself than ask the forums
thorn bramble
probably
surreal path
wait qwinci why can user tmp change, who changes it???
i guess we will wait on qwincis response here
thorn bramble
judging by the quality of most posts i've seen
ebon needle
wait qwinci why can user tmp change, who changes it???
maybe memory leak ?
surreal path
maybe memory leak ?
what? no? tf u on about lol
broken depot
maybe memory leak ?
Memory leaks don't lead to memory corruptions
ebon needle
what? no? tf u on about lol
i mean
maybe
you memcpying
and idk how changing this value ?
surreal path
also considering my allocator is pretty battle tested (thats not to say its perfect or wont lead to isssues in the future, heck even managarm had slab corruption issues) i wouldnt blame it
surreal path
and idk how changing this value ?
bro we know something is changing the value
we just dont know why
surreal path
i lit did
gray hill
ah
supple robin
give like a few hours and we will post onto the forum
oh oh
molten grotto
wait qwinci why can user tmp change, who changes it???
I don't exactly remember and can't take a look at the code properly rn as I haven't yet unpacked my pc after moving yesterday (its also the reason why I haven't been that active lately)
thorn bramble
also considering my allocator is pretty battle tested (thats not to say its perf...
but managarm had 10 years to write and test their stuff
and u haven't
surreal path
but managarm had 10 years to write and test their stuff
ik
im not saying its good
im just saying its probs not the allocator for this situation
surreal path
I don't exactly remember and can't take a look at the code properly rn as I have...
ah okay, when are u fully moving in?
surreal path
im just saying its probs not the allocator for this situation
i mean prove me wrong lol, if it is ill eat my words up
daring juniper
I mean I've only asked the forums stuff like, twice before, but it doesn't seem that bad
supple robin
why would they hate people for having fun
they're all old
surreal path
they're all old
surreal path
okay interesting
commenting out these lines in waitpid causes the same fault
i genuinely do not know why
i did this before
yea doing what waitpid does
in mmap
does NOT solve the issue lol
i thought
back to dying
yea i have no choice but to ask the forum
worth a shot
post made
hopefully one smart boomer will respond
ebon needle
extreme measures
surreal path
very much so
ebon needle
very much so
the more extreme measures is rubber duck debugging
surreal path
okay so
surreal path
okay interesting
theory on why this didnt work
id imagine that when mmap got preempted, some other thread made a syscall while interrupts were enabled
and that piece of code which i still do not know why i do doesnt execute cause its only on mmap so
that probs is what happened
i mean plasubile
what ill do is add those 2 lines in every syscall, along with disabling and enabling interrupts beforehand
if that doesnt work well fuck all
im genuinely just throwing shitty idea after shitty idea
just to get something to work lol
idc tho ill never give up with nyaux
even if we dont get help
even if we are stuck for years on this
idfk
cinder plinth
did you log interrupts
surreal path
nyaux is my child
surreal path
ofc
lol
did i just...
solve my page fault issue...
i added cli to the start of the syscall asm
i wanna make sure this is real
i have solved it...
I HAVE SOLVED THE PAGE FAULT ISSUE
yea it still deadlocks on the mem lock BUTTTT
PROGRESSS
@kind root banger right?
surreal path
yea it still deadlocks on the mem lock BUTTTT
if i enable interrupts on every other syscall and do the same shit
it might work
and not deadlock
hope is hope bro
and why it deadlocks now
keyboard interrupt wants to allocate memory for the keyboard packet,
cinder plinth
i added cli to the start of the syscall asm
Well then interrupts aren't enabled during syscalls, wasn't that the point?
surreal path
the userspace thread is holding the mem lock
surreal path
Well then interrupts aren't enabled during syscalls, wasn't that the point?
no, during a syscall function that can be preempted now
the syscall asm cannot
but the syscall function itself can now
cinder plinth
Ah
surreal path
keyboard interrupt wants to allocate memory for the keyboard packet,
^
okay we are closer
i made the memlock disable interrupts then lock then unlock then enable interrupts
seems like forking any program causes some weird ass shit
which is better?
idk
silver yarrow
i made the memlock disable interrupts then lock then unlock then enable interrup...
this is how it should be
surreal path
yea
silver yarrow
otherwise there is a race condition for deadlock
surreal path
pretty much
kind root
i added cli to the start of the syscall asm
yeah just make it so its disabled on entry always
theres a flag for that no?
surreal path
yeah just make it so its disabled on entry always
whats ur opinion on us getting closer to fixing all these issues tho?
surreal path
theres a flag for that no?
i believe so
surreal path
i made the memlock disable interrupts then lock then unlock then enable interrup...
also it seems like sys openat is getting preempted even tho it shouldnt
i think i know why
i dont check if interrupts were enabled in the first place for the mem lock
and just enable them
vfs lookup does some allocation so
yea
lol
makes a lot of ze sense
surreal path
i would fix if VSCODE WASNT A PIECE OF FUCKING SHIT
did arch break it???
it wont launch after i updated it
crystal scarab
kind root
it wont launch after i updated it
linux user
surreal path
linux user
okay bro
surreal path