uACPI - a portable and easy-to-integrate ACPI implementation | Operating System Development | Page 23

left orbit Dec 28, 2024, 1:26 AM

#

LULW

gentle peak Dec 28, 2024, 1:26 AM

#

then your cpu doesn't support the same features as mine

left orbit Dec 28, 2024, 1:26 AM

#

How

gentle peak Dec 28, 2024, 1:26 AM

#

it's compiled with -march=native -mtune=native and a hardcoded xcr0 value after all

left orbit Dec 28, 2024, 1:26 AM

#

i mean maybe if u enable avx512 or some shit

fiery turtle Dec 28, 2024, 1:26 AM

#

left orbit How

Maybe u just dont do cpu host

left orbit Dec 28, 2024, 1:26 AM

#

i do

fiery turtle Dec 28, 2024, 1:26 AM

#

Also wsl2 disables some stuff

left orbit Dec 28, 2024, 1:27 AM

#

fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 clflush mmx fxsr sse sse2 ss ht syscall nx pdpe1gb rdtscp lm constant_tsc rep_good nopl xtopology tsc_reliable nonstop_tsc cpuid pni pclmulqdq vmx ssse3 fma cx16 pcid sse4_1 sse4_2 x2apic movbe popcnt tsc_deadline_timer aes xsave avx f16c rdrand hypervisor lahf_lm abm 3dnowprefetch ssbd ibrs ibpb stibp ibrs_enhanced tpr_shadow ept vpid ept_ad fsgsbase tsc_adjust bmi1 avx2 smep bmi2 erms invpcid rdseed adx smap clflushopt clwb sha_ni xsaveopt xsavec xgetbv1 xsaves avx_vnni vnmi umip waitpkg gfni vaes vpclmulqdq rdpid movdiri movdir64b fsrm md_clear serialize flush_l1d arch_capabilities

#

it does avx2

#

sse4_2

fiery turtle Dec 28, 2024, 1:27 AM

#

Hm

#

Thats wsl2 right

#

Not the nested vm

left orbit Dec 28, 2024, 1:28 AM

#

yeah

gentle peak Dec 28, 2024, 1:28 AM

#

xcr0 state enabled: x87, sse, avx, pkru

left orbit Dec 28, 2024, 1:28 AM

#

but i don't imagine any of that disappears when it is nested

#

wtf is a pkru

gentle peak Dec 28, 2024, 1:28 AM

#

idk something my generic xsave code detected my cpu supports

left orbit Dec 28, 2024, 1:28 AM

#

lmao

gentle peak Dec 28, 2024, 1:28 AM

#

so it became the thing i hardcoded

#

here's one with just x87, sse, avx

📎 proxima.iso

fiery turtle Dec 28, 2024, 1:29 AM

#

Private key something

#

Lmao

left orbit Dec 28, 2024, 1:29 AM

#

yeah that works

#

10.4M

fiery turtle Dec 28, 2024, 1:29 AM

#

L cpu

#

Broooo

gentle peak Dec 28, 2024, 1:29 AM

#

hmmm what more can i stub out

fiery turtle Dec 28, 2024, 1:29 AM

#

Can u boot into Linux and run under native kvm?

fiery turtle Dec 28, 2024, 1:30 AM

#

gentle peak hmmm what more can i stub out

Its like stripping car parts lmaoo

gentle peak Dec 28, 2024, 1:30 AM

#

i think i've stubbed out everything i can that still gets called during uacpi init

#

now time to try the uacpi_reset trick

fiery turtle Dec 28, 2024, 1:31 AM

#

Just reset in a loop until the score is satisfactory trl

left orbit Dec 28, 2024, 1:31 AM

#

10.79M

#

dangerously close to 11M

gentle peak Dec 28, 2024, 1:31 AM

#

that definitely did something but it's supposed to warm up 10 times before doing the final init

#

@left orbit can you try this one?

📎 proxima.iso

fiery turtle Dec 28, 2024, 1:32 AM

#

gentle peak that definitely did something but it's supposed to warm up 10 times before doing...

Why is there only 2 runs

gentle peak Dec 28, 2024, 1:32 AM

#

thats what im trying to figure out

left orbit Dec 28, 2024, 1:32 AM

#

12M lol

#

fiery turtle Dec 28, 2024, 1:32 AM

#

Inb4 uACPI bug again

#

New record lmao

gentle peak Dec 28, 2024, 1:33 AM

#

ah its both a uacpi bug and a proxima bug

left orbit Dec 28, 2024, 1:33 AM

#

fiery turtle Dec 28, 2024, 1:33 AM

#

Whats the bug?

north holly Dec 28, 2024, 1:33 AM

#

left orbit

linux or prxima?

left orbit Dec 28, 2024, 1:33 AM

#

proxima

#

time to take this iso onto bare metal

north holly Dec 28, 2024, 1:33 AM

#

flush

fiery turtle Dec 28, 2024, 1:33 AM

#

This is avx in kernel

left orbit Dec 28, 2024, 1:33 AM

#

or do you want me to wait for a new one?

gentle peak Dec 28, 2024, 1:34 AM

#

uacpi bug: uacpi_state_reset doesn't uninstall the sci interrupt handler (i set a breakpoint in uacpi_kernel_uninstall_interrupt_handler and it never got triggered)
proxima bug: pic_install_vector only works the first time it's called

#

or well

#

alloc_irq_vectors to be exact

fiery turtle Dec 28, 2024, 1:34 AM

#

gentle peak uacpi bug: uacpi_state_reset doesn't uninstall the sci interrupt handler (i set ...

Ill investigate tmrw

gentle peak Dec 28, 2024, 1:35 AM

#

yeah now its just this (running in debug mode + tcg that's why the numbers are low)

fiery turtle Dec 28, 2024, 1:35 AM

#

Still hangs?

gentle peak Dec 28, 2024, 1:36 AM

#

see the last message

fiery turtle Dec 28, 2024, 1:36 AM

#

Ah

#

U can stub this for the purposes of testing tbh

gentle peak Dec 28, 2024, 1:36 AM

#

yeah i'm adding a manual uninstall in the warmup code

winter orbit Dec 28, 2024, 1:36 AM

#

it is

fiery turtle Dec 28, 2024, 1:36 AM

#

Nice

gentle peak Dec 28, 2024, 1:37 AM

#

seems after the first time it has minimal impact

#

📎 proxima.iso

left orbit Dec 28, 2024, 1:37 AM

#

lmao thanks for reassuring

dusky glade Dec 28, 2024, 1:37 AM

#

fiery turtle Nice

Nice

fiery turtle Dec 28, 2024, 1:37 AM

#

gentle peak

Dont boot this bare metal 💀

left orbit Dec 28, 2024, 1:37 AM

#

record!!

gentle peak Dec 28, 2024, 1:37 AM

#

oh no absolutely not

fiery turtle Dec 28, 2024, 1:38 AM

#

left orbit record!!

13M with LTO + O3 + AVX2 + SSO

gentle peak Dec 28, 2024, 1:38 AM

#

this is incredibly hacky

fiery turtle Dec 28, 2024, 1:38 AM

#

Lmao

dusky glade Dec 28, 2024, 1:39 AM

#

gentle peak this is incredibly hacky

no 1, whats a interrupt model, do i also need to run uacpI_set_interrupt_model

#

no 2

#

i never ran finalize gpe init

#

do i nee dto run that

gentle peak Dec 28, 2024, 1:39 AM

#

dusky glade no 1, whats a interrupt model, do i also need to run uacpI_set_interrupt_model

interrupt model tells aml what irq controller youre using

#

legacy, ioapic, etc

dusky glade Dec 28, 2024, 1:39 AM

#

oh okay

left orbit Dec 28, 2024, 1:39 AM

#

fiery turtle Dont boot this bare metal 💀

Why not

gentle peak Dec 28, 2024, 1:39 AM

#

i dont know if its necessary but i do it

left orbit Dec 28, 2024, 1:39 AM

#

what could go wrong.

#

TrollDespair

dusky glade Dec 28, 2024, 1:39 AM

#

gentle peak i dont know if its necessary but i do it

will i run finalize gpe init

fiery turtle Dec 28, 2024, 1:39 AM

#

gentle peak i dont know if its necessary but i do it

It is, it changes the _PRT output

fiery turtle Dec 28, 2024, 1:40 AM

#

dusky glade will i run finalize gpe init

Its needed to enable aml event handlers

dusky glade Dec 28, 2024, 1:40 AM

#

fiery turtle Its needed to enable aml event handlers

i will run it

fiery turtle Dec 28, 2024, 1:40 AM

#

Monkuous laptop has them, but its an old mechanism

#

They're also needed for suspend etc

dusky glade Dec 28, 2024, 1:41 AM

#

gentle peak this is incredibly hacky

can i see ur imports

gentle peak Dec 28, 2024, 1:41 AM

#

fiery turtle 13M with LTO + O3 + AVX2 + SSO

dusky glade Dec 28, 2024, 1:41 AM

#

i dont have uacpi_finalize_gpe in scope

#

nooo

fiery turtle Dec 28, 2024, 1:41 AM

#

gentle peak

Oh that was no sso...

gentle peak Dec 28, 2024, 1:41 AM

#

this is all that + SSO

📎 proxima.iso

dusky glade Dec 28, 2024, 1:42 AM

#

dusky glade can i see ur imports

?

gentle peak Dec 28, 2024, 1:42 AM

#

dusky glade i dont have uacpi_finalize_gpe in scope

provided by "uacpi/event.h"

dusky glade Dec 28, 2024, 1:42 AM

#

thanks

fiery turtle Dec 28, 2024, 1:42 AM

#

Bro can't git grep

dusky glade Dec 28, 2024, 1:42 AM

#

there we go

left orbit Dec 28, 2024, 1:42 AM

#

bare metal proxima

dusky glade Dec 28, 2024, 1:43 AM

#

fiery turtle Bro can't git grep

bro doesnt know im lazy

#

bro doesnt know im in his walls

left orbit Dec 28, 2024, 1:43 AM

#

lots of tables

#

ALASKA

fiery turtle Dec 28, 2024, 1:43 AM

#

left orbit bare metal proxima

Solid

gentle peak Dec 28, 2024, 1:43 AM

#

left orbit bare metal proxima

bare metal uacpi comparisons are a bit bogus due to complicated aml

#

but yeah solid

north holly Dec 28, 2024, 1:43 AM

#

dusky glade bro doesnt know im in his walls

you got it wrong, infy is in YOUR walls

gentle peak Dec 28, 2024, 1:43 AM

#

i wonder what it's like on my dev system

dusky glade Dec 28, 2024, 1:43 AM

#

north holly you got it wrong, infy is in YOUR walls

nooo

fiery turtle Dec 28, 2024, 1:43 AM

#

12 blobs and 30k ops

left orbit Dec 28, 2024, 1:43 AM

#

i wanna run the final image u sent.

gentle peak Dec 28, 2024, 1:43 AM

#

is that the mutex stub + half-leaking allocator + AVX2 in kernel ISO or is it the normal one

fiery turtle Dec 28, 2024, 1:43 AM

#

Probably a better bench than qemus blob

#

But alas

left orbit Dec 28, 2024, 1:44 AM

#

left orbit i wanna run the final image u sent.

the one that reinitializes uacpi 10 times meme

dusky glade Dec 28, 2024, 1:44 AM

#

left orbit bare metal proxima

can u bench nyaux

#

trl

gentle peak Dec 28, 2024, 1:44 AM

#

fiery turtle Probably a better bench than qemus blob

yeah not really comparable

dusky glade Dec 28, 2024, 1:44 AM

#

i actually added something that i think will make nyaux faster tho

#

hold on lemme pull out ze iso

#

this will beat proxima

📎 NyauxKC-x86_64.iso

#

if im wrong hang me to the floorboards

fiery turtle Dec 28, 2024, 1:45 AM

#

Inb4 100/s

gentle peak Dec 28, 2024, 1:45 AM

#

@fiery turtle do you thinnk i should rip out the 10+ inits before testing this on bare metal or do i just go for it

north holly Dec 28, 2024, 1:45 AM

#

rip out the 10+ inits

fiery turtle Dec 28, 2024, 1:45 AM

#

Worst case it reboots

#

Nothing a power cycle won't solve

north holly Dec 28, 2024, 1:46 AM

#

fiery turtle Worst case it reboots

but the question is, what happens after the reboot

left orbit Dec 28, 2024, 1:46 AM

#

fiery turtle Nothing a power cycle won't solve

why'd u guys tell me not to test it 😭

north holly Dec 28, 2024, 1:46 AM

#

fiery turtle Nothing a power cycle won't solve

what about writing to proprietary bios registers that happen to have a checksum, and the system doesn't boot unless the checksum is what it is expects

fiery turtle Dec 28, 2024, 1:46 AM

#

If its a pc I wouldnt worry tooooo much

north holly Dec 28, 2024, 1:47 AM

#

most laptops' blobs are delusion right?

fiery turtle Dec 28, 2024, 1:47 AM

#

left orbit why'd u guys tell me not to test it 😭

I mean u can lol

left orbit Dec 28, 2024, 1:47 AM

#

nah i ain't rebooting now

fiery turtle Dec 28, 2024, 1:47 AM

#

Its just precautions

left orbit Dec 28, 2024, 1:47 AM

#

too lazy

fiery turtle Dec 28, 2024, 1:47 AM

#

north holly most laptops' blobs are delusion right?

Laptops have more controllers that do complex power management

dusky glade Dec 28, 2024, 1:47 AM

#

bench nyaux or begone ur c-c-c-car!

north holly Dec 28, 2024, 1:48 AM

#

when will I be able to run nyaux in my car

dusky glade Dec 28, 2024, 1:48 AM

#

soon ™️

gentle peak Dec 28, 2024, 1:48 AM

#

I'll bench nyaux once I've tested all 3 variants of proxima bare metal

dusky glade Dec 28, 2024, 1:48 AM

#

troll2

dusky glade Dec 28, 2024, 1:48 AM

#

gentle peak I'll bench nyaux once I've tested all 3 variants of proxima bare metal

lets goooo

#

monkuous W

median crest Dec 28, 2024, 1:48 AM

#

my new astral lock is almost done.... soon:tm;

dusky glade Dec 28, 2024, 1:48 AM

#

im going to steal it

#

™️

fiery turtle Dec 28, 2024, 1:48 AM

#

gentle peak I'll bench nyaux once I've tested all 3 variants of proxima bare metal

Would be funny if it just has sleep calls in there

median crest Dec 28, 2024, 1:48 AM

#

finishing release and will just add a quick lock try later

fiery turtle Dec 28, 2024, 1:48 AM

#

Which is possible

fiery turtle Dec 28, 2024, 1:49 AM

#

median crest finishing release and will just add a quick lock try later

Lets goo

median crest Dec 28, 2024, 1:49 AM

#

release is super annoying

left orbit Dec 28, 2024, 1:49 AM

#

cool i can't even check how fast it went

dusky glade Dec 28, 2024, 1:49 AM

#

fiery turtle Would be funny if it just has sleep calls in there

no new nyaux version has properierty technology no one has ever thought about before

dusky glade Dec 28, 2024, 1:49 AM

#

left orbit cool i can't even check how fast it went

trl

left orbit Dec 28, 2024, 1:49 AM

#

serial works

#

uacpi: successfully loaded 1 AML blob, 1705 ops in 101ms (avg 16838/s)

#

how the fufck do u manage that

dusky glade Dec 28, 2024, 1:49 AM

#

hax

fiery turtle Dec 28, 2024, 1:49 AM

#

trl

dusky glade Dec 28, 2024, 1:50 AM

#

trl

median crest Dec 28, 2024, 1:51 AM

#

left orbit how the fufck do u manage that

honestly its impressive

dusky glade Dec 28, 2024, 1:51 AM

#

told u i improved nyaux

#

properity technology

#

properietory

fiery turtle Dec 28, 2024, 1:51 AM

#

How many orders of magnitude is that from proximas score

dusky glade Dec 28, 2024, 1:51 AM

#

20c

#

20x*

left orbit Dec 28, 2024, 1:51 AM

#

about 8

fiery turtle Dec 28, 2024, 1:51 AM

#

Not bad

dusky glade Dec 28, 2024, 1:51 AM

#

times

#

faster

left orbit Dec 28, 2024, 1:52 AM

#

revolutionary

#

BIG

dusky glade Dec 28, 2024, 1:52 AM

#

!!!!

left orbit Dec 28, 2024, 1:52 AM

#

nyaux supremacy

dusky glade Dec 28, 2024, 1:52 AM

#

frfr!!!!

fiery turtle Dec 28, 2024, 1:52 AM

#

Indeed

gentle peak Dec 28, 2024, 1:52 AM

#

normal (iso straight from GitHub actions): 10097 ops, 4,737,793/s
crazy (everything except the 10 warmup inits): 6,373,181/s
crazy² (everything): 6,313,331/s

fiery turtle Dec 28, 2024, 1:52 AM

#

gentle peak normal (iso straight from GitHub actions): 10097 ops, 4,737,793/s crazy (everyth...

Bare metal?

north holly Dec 28, 2024, 1:52 AM

#

infy can you update me score

gentle peak Dec 28, 2024, 1:52 AM

#

yes

fiery turtle Dec 28, 2024, 1:53 AM

#

Damn

left orbit Dec 28, 2024, 1:53 AM

#

aml fears this man

fiery turtle Dec 28, 2024, 1:53 AM

#

I wonder if executing it this fast can reveal bugs with missing waits etc

north holly Dec 28, 2024, 1:53 AM

#

gentle peak yes

what does proxima even use uacpi for
flexing rights?

left orbit Dec 28, 2024, 1:53 AM

#

yes

#

doesn't everybody use uacpi for that?

gentle peak Dec 28, 2024, 1:54 AM

#

left orbit uacpi: successfully loaded 1 AML blob, 1705 ops in 101ms (avg 16838/s)

wtf how, even when running with a completely unoptimized profiler that records every single function call i got 60k/s

gentle peak Dec 28, 2024, 1:54 AM

#

north holly what does proxima even use uacpi for flexing rights?

future proofing

#

and right now flexing rights

left orbit Dec 28, 2024, 1:54 AM

#

either that or runtime allocator tests :^)

north holly Dec 28, 2024, 1:54 AM

#

left orbit doesn't everybody use uacpi for that?

uh I use it for power management, allocator tests, and device discovery

fiery turtle Dec 28, 2024, 1:54 AM

#

gentle peak wtf how, even when running with a completely unoptimized profiler that records e...

It used to be 120/s

dusky glade Dec 28, 2024, 1:54 AM

#

im afraid ive been lying to you all...

#

this is the real nyaux score

north holly Dec 28, 2024, 1:54 AM

#

🤯

dusky glade Dec 28, 2024, 1:54 AM

#

🤯

fiery turtle Dec 28, 2024, 1:55 AM

#

Doesnt count

#

Pci unimplemeted

dusky glade Dec 28, 2024, 1:55 AM

#

nooo

median crest Dec 28, 2024, 1:55 AM

#

trolled

fiery turtle Dec 28, 2024, 1:55 AM

#

You'll get em next time

gentle peak Dec 28, 2024, 1:55 AM

#

on my cpu with kvm: uacpi: successfully loaded 1 AML blob, 1705 ops in 73ms (avg 23342/s)

#

thats nyaux btw

north holly Dec 28, 2024, 1:55 AM

#

📎 obos.iso

fiery turtle Dec 28, 2024, 1:55 AM

#

Thanks for clarifying KEKW

north holly Dec 28, 2024, 1:55 AM

#

test obos 🥺

dusky glade Dec 28, 2024, 1:55 AM

#

gentle peak on my cpu with kvm: `uacpi: successfully loaded 1 AML blob, 1705 ops in 73ms (av...

nooo nooo

fiery turtle Dec 28, 2024, 1:55 AM

#

north holly infy can you update me score

Update your pinned comment

gentle peak Dec 28, 2024, 1:56 AM

#

north holly test obos 🥺

369756/s

fiery turtle Dec 28, 2024, 1:56 AM

#

north holly test obos 🥺

On real hw

left orbit Dec 28, 2024, 1:56 AM

#

dusky glade im afraid ive been lying to you all...

wtf literally how ,,,

dusky glade Dec 28, 2024, 1:56 AM

#

i wonder if there is a way to profile nyaux

median crest Dec 28, 2024, 1:56 AM

#

obos faster than astral before I got my shit together

north holly Dec 28, 2024, 1:56 AM

#

gentle peak 369756/s

and it crashes loading /init right?

dusky glade Dec 28, 2024, 1:56 AM

#

dusky glade i wonder if there is a way to profile nyaux

to see where its spending most of its time being so slow

dusky glade Dec 28, 2024, 1:56 AM

#

left orbit wtf literally how ,,,

its just

median crest Dec 28, 2024, 1:56 AM

#

north holly and it crashes loading `/init` right?

omg init on / this is just like the hit operating system astral

dusky glade Dec 28, 2024, 1:56 AM

#

a difference in ze skill

#

trl

left orbit Dec 28, 2024, 1:56 AM

#

dusky glade to see where its spending most of its time being so slow

pinned message

dusky glade Dec 28, 2024, 1:56 AM

#

oohhh

north holly Dec 28, 2024, 1:57 AM

#

median crest omg init on / this is just like the hit operating system astral

omg!!!!11!!!!11!!!1

dusky glade Dec 28, 2024, 1:57 AM

#

im gonna throw the profiler he has in lib

gentle peak Dec 28, 2024, 1:57 AM

#

north holly and it crashes loading `/init` right?

nevermind i used an older downloaded iso instead of obos(1).iso, it's actually 216154/s

north holly Dec 28, 2024, 1:57 AM

#

it fluctuate

#

try again

gentle peak Dec 28, 2024, 1:57 AM

#

no this is consistent

north holly Dec 28, 2024, 1:57 AM

#

bru

fiery turtle Dec 28, 2024, 1:57 AM

#

north holly it fluctuate

Cope

gentle peak Dec 28, 2024, 1:57 AM

#

i'll try it on bare metal

fiery turtle Dec 28, 2024, 1:58 AM

#

Inb4 blows up

north holly Dec 28, 2024, 1:58 AM

#

I think if I undo my last commit

#

it should speed up

gentle peak Dec 28, 2024, 1:58 AM

#

fiery turtle Inb4 blows up

id sure hope not

fiery turtle Dec 28, 2024, 1:58 AM

#

What did u do there

north holly Dec 28, 2024, 1:58 AM

#

allocator changes

gentle peak Dec 28, 2024, 1:58 AM

#

but really if 11 uacpi inits don't brick my machine what will

fiery turtle Dec 28, 2024, 1:58 AM

#

True

gentle peak Dec 28, 2024, 1:59 AM

#

fiery turtle Inb4 blows up

triple fault before any output was visible, close enough

fiery turtle Dec 28, 2024, 1:59 AM

#

Is this also amd?

gentle peak Dec 28, 2024, 2:00 AM

#

same machine so yeah

#

5800x

fiery turtle Dec 28, 2024, 2:00 AM

#

Obos doesnt like amd computers

north holly Dec 28, 2024, 2:00 AM

#

yeah idk why

gentle peak Dec 28, 2024, 2:00 AM

#

the only Intel I have is my Haswell laptop

#

I'll try it on there

north holly Dec 28, 2024, 2:00 AM

#

sure

#

all my machines are haswell

fiery turtle Dec 28, 2024, 2:00 AM

#

Haswell gang

north holly Dec 28, 2024, 2:01 AM

#

I genuniely have no idea what's so special about amd that it triples before even getting to gdt init

dusky glade Dec 28, 2024, 2:01 AM

#

yo he good

north holly Dec 28, 2024, 2:01 AM

#

Arch_KernelEntryBootstrap:
    cmp rsi, 0x554c5442
    je .ok ; should triple fault on failure
    xor rax,rax
    jmp rax ; All hope is lost.
.ok:
    push rdi
    push rsi
; Get a hardware-generated random value.
; If both methods are unsupported, it will fallback to using the default value used.
.rdrand:
    mov eax, 1
    xor ecx,ecx
    cpuid
    bt ecx, 30
    jnc .rdseed
    rdrand rax
    jnc .rdrand
    jmp .move
.rdseed:
    mov eax, 7
    xor ecx,ecx
    cpuid
    bt ebx, 18
    jnc .done
    rdseed rax
    jnc .rdseed```

dusky glade Dec 28, 2024, 2:01 AM

#

wha

north holly Dec 28, 2024, 2:01 AM

#

it does some cpuid

north holly Dec 28, 2024, 2:01 AM

#

north holly it does some cpuid

then rdseed/rdrand for teh stack check guard

#

then disables the pic

fiery turtle Dec 28, 2024, 2:01 AM

#

north holly I genuniely have no idea what's so special about amd that it triples before even...

Is it possible the framebuffer is too high and u try to access it

north holly Dec 28, 2024, 2:01 AM

#

sets cr0.wp

#

so I need to wait until the fb gets sober?

fiery turtle Dec 28, 2024, 2:02 AM

#

Yes

gentle peak Dec 28, 2024, 2:02 AM

#

north holly sure

on said laptop it just hangs until eventually getting to the "insert system disk in drive" bios prompt

#

im beginning to think i fucked up the usb

dusky glade Dec 28, 2024, 2:02 AM

#

dusky glade wha

yea i do not know what its complaining

north holly Dec 28, 2024, 2:02 AM

#

gentle peak on said laptop it just hangs until eventually getting to the "insert system disk...

well idk if that's my fault or not

gentle peak Dec 28, 2024, 2:02 AM

#

dusky glade yea i do not know what its complaining

does your compiler support -finstrument-functions?

dusky glade Dec 28, 2024, 2:03 AM

#

let me try and add the flag hold on

north holly Dec 28, 2024, 2:03 AM

#

try uefi ig

gentle peak Dec 28, 2024, 2:03 AM

#

yeah im on uefi

#

ah. that'd probably be why.

fiery turtle Dec 28, 2024, 2:03 AM

#

💀

gentle peak Dec 28, 2024, 2:03 AM

#

it's empty in the iso as well

dusky glade Dec 28, 2024, 2:03 AM

#

even with the flag @gentle peak

north holly Dec 28, 2024, 2:03 AM

#

wtf

gentle peak Dec 28, 2024, 2:03 AM

#

crazy that my main pc triple faults when there's no efi boot image available

gentle peak Dec 28, 2024, 2:03 AM

#

dusky glade even with the flag <@938223886439436320>

one of your headers is fucked ig

fiery turtle Dec 28, 2024, 2:03 AM

#

gentle peak crazy that my main pc triple faults when there's no efi boot image available

Least broken uefi implementation

gentle peak Dec 28, 2024, 2:04 AM

#

idk

dusky glade Dec 28, 2024, 2:04 AM

#

gentle peak one of your headers is fucked ig

no not rlly? i made a seperate file called profiler.h

#

and i only imported this

#

nothing else

gentle peak Dec 28, 2024, 2:04 AM

#

those last 2 things are supposed to be in your task_t struct or equivalent

#

but idk what could be causing the syntax error

dusky glade Dec 28, 2024, 2:05 AM

#

gentle peak those last 2 things are supposed to be in your task_t struct or equivalent

oh im not running this in the scheduler

north holly Dec 28, 2024, 2:05 AM

#

gentle peak ah. that'd probably be why.

then why would it boot on my uefi pc

#

tho

dusky glade Dec 28, 2024, 2:05 AM

#

before kentry

north holly Dec 28, 2024, 2:05 AM

#

and with ovmf

#

cursed iso

gentle peak Dec 28, 2024, 2:05 AM

#

dusky glade oh im not running this in the scheduler

then they can be static vars in prof.c itself but they shouldn't be global and outside the header guard

dusky glade Dec 28, 2024, 2:06 AM

#

gentle peak then they can be static vars in prof.c itself but they shouldn't be global *and ...

still

#

i dont think clang likes any of this

#

even my language server is not happy

gentle peak Dec 28, 2024, 2:06 AM

#

i use clang myself

dusky glade Dec 28, 2024, 2:06 AM

#

its just saying its not allowing the function defination here

gentle peak Dec 28, 2024, 2:07 AM

#

@north holly now it's stuck on Using AGING PRA

north holly Dec 28, 2024, 2:07 AM

#

bruh

#

vmm hang

#

it's after the vmm actually

#

I'm gonna send an iso with more logs wait

#

📎 obos.iso

gentle peak Dec 28, 2024, 2:08 AM

#

@dusky glade apparently there's a } missing at the end of prof_stop. oops

dusky glade Dec 28, 2024, 2:08 AM

#

gentle peak <@787341689333350450> apparently there's a `}` missing at the end of `prof_stop`...

even doing that its still not happy

#

nvm

gentle peak Dec 28, 2024, 2:09 AM

#

and missing ) at the cursor here

#

i really should've implemented prof_start/prof_stop in my ide instead of in the github gist editor

dusky glade Dec 28, 2024, 2:09 AM

#

yea

#

current task is not defined

#

last thing its complaining about

gentle peak Dec 28, 2024, 2:10 AM

#

since you're using global frames/cur_frame just remove the current_task-> parts

dusky glade Dec 28, 2024, 2:10 AM

#

okay

gentle peak Dec 28, 2024, 2:10 AM

#

they're supposed to be task-local since they represent the call stack

dusky glade Dec 28, 2024, 2:10 AM

#

like this?

gentle peak Dec 28, 2024, 2:10 AM

#

yeah

dusky glade Dec 28, 2024, 2:10 AM

#

okay

#

that solved it

#

thanks @gentle peak

#

i will now test the profiler

gentle peak Dec 28, 2024, 2:12 AM

#

@north holly

rn_image_picker_lib_temp_71babd91-790d-4df7-b6fc-ced27a9ab7c7.jpg

north holly Dec 28, 2024, 2:12 AM

#

fuck

#

it hangs in PCI

gentle peak Dec 28, 2024, 2:12 AM

#

dusky glade i will now test the profiler

don't forget the pitfalls listed here https://gist.github.com/monkuous/5752282d03995080e99671ecb9969b3f?permalink_comment_id=5364478#gistcomment-5364478

dusky glade Dec 28, 2024, 2:12 AM

#

uhhh oop

gentle peak Dec 28, 2024, 2:13 AM

#

that's what happens when you put them in the header instead of prof.c

dusky glade Dec 28, 2024, 2:13 AM

#

okay

#

i have an idea

#

called

#

#pragma once

#

trl

gentle peak Dec 28, 2024, 2:13 AM

#

that wouldn't fix it

#

you'd still have one declaration per source file

dusky glade Dec 28, 2024, 2:13 AM

#

nvm ur right

gentle peak Dec 28, 2024, 2:13 AM

#

and i have that in the form of #ifndef+#define+#endif

dusky glade Dec 28, 2024, 2:14 AM

#

uhhh

#

no 1

#

it died

#

no 2

#

wha

#

why

gentle peak Dec 28, 2024, 2:15 AM

#

addr2line time

dusky glade Dec 28, 2024, 2:15 AM

#

i mean i stopped the prof

#

okay hold on

north holly Dec 28, 2024, 2:15 AM

#

gentle peak <@841762704766140486>

can you make sure that the pci device I'm trying to initialize isn't a bridge or smth similar

#

through lspci

gentle peak Dec 28, 2024, 2:16 AM

#

dusky glade i mean i stopped the prof

before prof_start and after prof_stop the only variable accessed should be prof_active which is a static var so

dusky glade Dec 28, 2024, 2:16 AM

#

starting in kentry

dusky glade Dec 28, 2024, 2:16 AM

#

gentle peak before prof_start and after prof_stop the only variable accessed should be `prof...

very beginning

dusky glade Dec 28, 2024, 2:16 AM

#

gentle peak before prof_start and after prof_stop the only variable accessed should be `prof...

it is static

gentle peak Dec 28, 2024, 2:17 AM

#

north holly can you make sure that the pci device I'm trying to initialize isn't a bridge or...

0c:03:30 doesn't exist

north holly Dec 28, 2024, 2:17 AM

#

that's an hid

#

it's on 00:14.00

gentle peak Dec 28, 2024, 2:17 AM

#

oh

north holly Dec 28, 2024, 2:17 AM

#

north holly that's an hid

as in class codes

#

n'stuff

gentle peak Dec 28, 2024, 2:17 AM

#

that's USB controller: Intel Corporation 8 Series/C220 Series Chipset Family USB xHCI (rev 04)

north holly Dec 28, 2024, 2:18 AM

#

gentle peak that's `USB controller: Intel Corporation 8 Series/C220 Series Chipset Family US...

can you show me what resources it has?

dusky glade Dec 28, 2024, 2:18 AM

#

dusky glade starting in kentry

so why is this happening thonk nyaux works fine on real hw even and the scheduler is like really simple so

north holly Dec 28, 2024, 2:18 AM

#

like bars

#

lspci -v

dusky glade Dec 28, 2024, 2:18 AM

#

is it something with ze profiler idk

#

also nothing came out of debugcon

#

nooo

#

oh wait need to call prof show

gentle peak Dec 28, 2024, 2:19 AM

#

north holly `lspci -v`

00:14.0 USB controller: Intel Corporation 8 Series/C220 Series Chipset Family USB xHCI (rev 04) (prog-if 30 [XHCI])
        Subsystem: Toshiba Corporation Device 0001
        Flags: bus master, fast devsel, latency 0, IRQ 28
        Memory at e4620000 (64-bit, non-prefetchable) [size=64K]
        Capabilities: [70] Power Management version 2
        Capabilities: [80] MSI: Enable+ Count=1/8 Maskable- 64bit+
        Kernel driver in use: xhci_hcd

north holly Dec 28, 2024, 2:20 AM

#

interesting

#

the first thing it does

#

is discovering BARs

#

so that's probably where it hagns

gentle peak Dec 28, 2024, 2:21 AM

#

it's bar0 if i'm reading lspci -vv correctly

north holly Dec 28, 2024, 2:21 AM

#

since there should be more logs around the bars of that device

dusky glade Dec 28, 2024, 2:21 AM

#

@gentle peak this is the results of the nyaux profile before it page faults dying in the scheduler, uhhhh the first function is acpi.c:line 9???

#

north holly Dec 28, 2024, 2:21 AM

#

then it hangs reading bar0 thinkong

gentle peak Dec 28, 2024, 2:22 AM

#

dusky glade <@938223886439436320> this is the results of the nyaux profile before it page fa...

i made a python script to format this nicely but i accidentally deleted it lmao

#

anyway that's sorted by ns/call

dusky glade Dec 28, 2024, 2:22 AM

#

gentle peak i made a python script to format this nicely but i accidentally deleted it lmao

bru

#

why is uacpi taking so much ns

#

acpi.c line 9??

#

bro

gentle peak Dec 28, 2024, 2:22 AM

#

#!/usr/bin/env python3
import os
import re
import subprocess

with open('dump.txt', 'r') as f:
    text = f.read()

def translate(matchobj):
    addr = matchobj.group(0)
    out = subprocess.check_output(['addr2line', '-f', '-e', 'hydrogen/kernel/hydrogen', addr], stderr=subprocess.DEVNULL).decode().strip().split('\n')
    func = out[0]
    path = out[1].split(':')[0]
    file = os.path.basename(path)
    line = out[1].split(':')[1]
    return f'{func} ({file}:{line} {addr})'

text = re.sub(r'(0x[0-9a-f]*)', translate, text)
print(text)

gentle peak Dec 28, 2024, 2:22 AM

#

dusky glade why is uacpi taking so much ns

it's the call including all subcalls

dusky glade Dec 28, 2024, 2:22 AM

#

ohhh

gentle peak Dec 28, 2024, 2:23 AM

#

gentle peak ```python #!/usr/bin/env python3 import os import re import subprocess with ope...

@dusky glade this will take the dump at dump.txt, addr2line everything using hydrogen/kernel/hydrogen, and print it to stdout

dusky glade Dec 28, 2024, 2:23 AM

#

gentle peak <@787341689333350450> this will take the dump at dump.txt, addr2line everything ...

yea ik

gentle peak Dec 28, 2024, 2:23 AM

#

alright just making sure you didn't miss it

north holly Dec 28, 2024, 2:24 AM

#

I suspect it hangs reading the size

#

of the BAR

dusky glade Dec 28, 2024, 2:24 AM

#

more readable

gentle peak Dec 28, 2024, 2:25 AM

#

here's some alternative impls for pred in prof.c btw ```c
attribute((no_instrument_function)) static bool pred(record_t *a, record_t *b) {
return a->total < b->total; // total time, high to low
}

attribute((no_instrument_function)) static bool pred(record_t *a, record_t *b) {
return a->calls < b->calls; // number of calls, high to low
}```

dusky glade Dec 28, 2024, 2:25 AM

#

slab allocator is slow?

#

it seems?

gentle peak Dec 28, 2024, 2:25 AM

#

the time unit there is tsc ticks

#

you may want to add -cpu tsc-frequency=1000000000 to turn it into nanoseconds

dusky glade Dec 28, 2024, 2:26 AM

#

kinda slow too

#

init hpet is SUPER slow lol

#

anyways we know that nyaux isnt a complete dumpster fire

#

its just

#

slab allocator is REALLY ass

#

like

#

REALLY

#

REALLY REALLY

dusky glade Dec 28, 2024, 2:27 AM

#

dusky glade slab allocator is slow?

nvm it seems vmm is slower

#

well i was kinda expecting it to be my allocators, not that i have a reason why it would be

#

but

#

it prob cause my code is ass or smt

#

idfk

gentle peak Dec 28, 2024, 2:28 AM

#

i personally find sorting by total time spent in function more useful, because it also indicates how necessary it is to optimize a given function

dusky glade Dec 28, 2024, 2:28 AM

#

this shouldnt be THAT slow tho

north holly Dec 28, 2024, 2:28 AM

#

gentle peak i personally find sorting by total time spent in function more useful, because i...

can you try this

📎 obos.iso

dusky glade Dec 28, 2024, 2:29 AM

#

void* slab_alloc(cache* mod)
{
    if (mod->slabs == NULL || mod->size == 0)
    {
        return NULL;
    }
    slab* cur = (slab*)mod->slabs;
    assert(cur);
    while (true)
    {
        if (cur->freelist == NULL)
        {
            if (cur->next != NULL)
            {
                cur = (slab*)cur->next;

                continue;
            }
            break;
        }
        assert(cur != NULL);
        void* guy = cur->freelist;
        cur->freelist = ((pnode*)cur->freelist)->next;
        memset(guy, 0, mod->size);
        return guy;
    };
    assert(cur);
    cur->next = (struct slab*)init_slab(mod->size);
    cur = (slab*)cur->next;
    assert(cur);
    void* guy = cur->freelist;
    cur->freelist = ((pnode*)cur->freelist)->next;
    memset(guy, 0, mod->size);
    return guy;
}

#

like

#

its really just this code

north holly Dec 28, 2024, 2:29 AM

#

while(true)

dusky glade Dec 28, 2024, 2:29 AM

#

what u have a better idea

#

meme

north holly Dec 28, 2024, 2:29 AM

#

simple != fast

dusky glade Dec 28, 2024, 2:29 AM

#

north holly simple != fast

okay, then how would this be faster

gentle peak Dec 28, 2024, 2:29 AM

#

north holly can you try this

that works

north holly Dec 28, 2024, 2:29 AM

#

wut

gentle peak Dec 28, 2024, 2:29 AM

#

acpi init benchmark scrolled away though

#

it got to the panic while trying to load /init

north holly Dec 28, 2024, 2:30 AM

#

all I did was add more logs thinkong

north holly Dec 28, 2024, 2:30 AM

#

gentle peak acpi init benchmark scrolled away though

I'll disable the extra logs

#

and send you an iso

dusky glade Dec 28, 2024, 2:30 AM

#

dusky glade ```C void* slab_alloc(cache* mod) { if (mod->slabs == NULL || mod->size == 0...

like is there some way to optimize this?

gentle peak Dec 28, 2024, 2:30 AM

#

north holly and send you an iso

bet it's gonna hang at pci init again

north holly Dec 28, 2024, 2:31 AM

#

only one way to find out

#

📎 obos.iso

#

maybe it was just a one-time thing, sometimes obos randomly hangs in stuff where it usually doesn't

#

which is usually solved by a reboot

gentle peak Dec 28, 2024, 2:32 AM

#

it's now having at [ DEBUG ] updating bar 0

#

hanging*

north holly Dec 28, 2024, 2:32 AM

#

well I'll be damned

#

could this be timing-related?

gentle peak Dec 28, 2024, 2:32 AM

#

possibly

#

you should add in all the previous messages except replace \n with \r

#

stays on a single line while still doing all the printing

north holly Dec 28, 2024, 2:33 AM

#

before I do that I want to test something
maybe I get an irq in the middle of the bar being in a weird state, causing a hang

gentle peak Dec 28, 2024, 2:33 AM

#

i doubt that'd be this consistent

dusky glade Dec 28, 2024, 2:34 AM

#

i got slab alloc to be a tiny bit faster by removing not needed if statement

#

statements

#

151 instead of 171

#

now

gentle peak Dec 28, 2024, 2:34 AM

#

dusky glade ```C void* slab_alloc(cache* mod) { if (mod->slabs == NULL || mod->size == 0...

i can't really tell you how fast or slow this is most likely going to be just based on the code

#

what you really have to think about is what the common path is and how much work it's doing on that path

north holly Dec 28, 2024, 2:34 AM

#

gentle peak what you really have to think about is what the common path is and how much work...

📎 obos.iso

gentle peak Dec 28, 2024, 2:35 AM

#

north holly

worked, but again, uacpi benchmark scrolled away

dusky glade Dec 28, 2024, 2:35 AM

#

it seems the -finstrument flag is causing nyaux to die

north holly Dec 28, 2024, 2:36 AM

#

gentle peak worked, but again, uacpi benchmark scrolled away

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

gentle peak Dec 28, 2024, 2:36 AM

#

dusky glade it seems the -finstrument flag is causing nyaux to die

all -finstrument-functions does is insert a function call at the start of each function and another at the end

dusky glade Dec 28, 2024, 2:36 AM

#

yea maybe its messing something up with the scheduler

#

and something to be fair

north holly Dec 28, 2024, 2:36 AM

#

📎 obos.iso

#

try this

dusky glade Dec 28, 2024, 2:37 AM

#

nyaux is a TINY bit faster on tcg

#

trying kvm nopw

gentle peak Dec 28, 2024, 2:37 AM

#

north holly

ok that one worked

#

78964/s

north holly Dec 28, 2024, 2:37 AM

#

yeah the only changes I added there was masking irqs while getting the bar size

dusky glade Dec 28, 2024, 2:38 AM

#

nyaux running on kvm

gentle peak Dec 28, 2024, 2:38 AM

#

north holly yeah the only changes I added there was masking irqs while getting the bar size

weird that that's a consistent issue

north holly Dec 28, 2024, 2:38 AM

#

north holly yeah the only changes I added there was masking irqs while getting the bar size

and the logs never happened, as you could see

#

maybe it's because of an LAPIC timer IRQ

#

or hpet irq

#

which happen pretty consistently in my kernel

dusky glade Dec 28, 2024, 2:38 AM

#

dusky glade nyaux running on kvm

nvm it made no difference

north holly Dec 28, 2024, 2:39 AM

#

so yeah obos boots on intel cpus, but hates AMD cpus

#

how interesting

gentle peak Dec 28, 2024, 2:39 AM

#

even more interesting that it works on AMD in KVM

north holly Dec 28, 2024, 2:39 AM

#

yeah

gentle peak Dec 28, 2024, 2:39 AM

#

because that functions almost exactly the same as AMD bare metal

north holly Dec 28, 2024, 2:40 AM

#

the only thing I could think of doing is digging through the AMD SDM

#

for instructions behaving differently

dusky glade Dec 28, 2024, 2:40 AM

#

dusky glade nyaux is a TINY bit faster on tcg

removed some not needed memset calls

dusky glade Dec 28, 2024, 2:42 AM

#

dusky glade removed some not needed memset calls

removed even more not needed memset calls

gentle peak Dec 28, 2024, 2:42 AM

#

dusky glade ```C void* slab_alloc(cache* mod) { if (mod->slabs == NULL || mod->size == 0...

refactored it a bit, this does exactly the same but is much cleaner code imo and easier to analyze and reason about and thus easier to optimize ```c
void *slab_alloc(cache *mod) {
if (mod->slabs == NULL || mod->size == 0) return NULL;

slab *cur = (slab *)mod->slabs;
assert(cur != NULL);

while (cur->freelist == NULL) {
    if (cur->next == NULL) {
        cur->next = (struct slab *)init_slab(mod->size);
    }
    
    cur = (slab *)cur->next;
    assert(cur != NULL);
}

void *ptr = cur->freelist;
cur->freelist = ((pnode *)cur->freelist)->next;
memset(ptr, 0, mod->size);
return ptr;

}```

dusky glade Dec 28, 2024, 2:42 AM

#

gentle peak refactored it a bit, this does exactly the same but is much cleaner code imo and...

will try

gentle peak Dec 28, 2024, 2:42 AM

#

north holly the only thing I could think of doing is digging through the AMD SDM

what i usually do when encountering something that causes phys hw to triple fault is to do a "binary search" with cli; hlt loops

north holly Dec 28, 2024, 2:43 AM

#

if you wanna do that we can

gentle peak Dec 28, 2024, 2:43 AM

#

dusky glade will try

it should function and perform the exact same, it's just a lot easier to read, analyze, and optimize

gentle peak Dec 28, 2024, 2:43 AM

#

north holly if you wanna do that we can

don't really have the time rn (it's 4 AM) sorry

north holly Dec 28, 2024, 2:43 AM

#

it's np

dusky glade Dec 28, 2024, 2:43 AM

#

gentle peak it should function and perform the exact same, it's just a lot easier to read, a...

it went slower lol

#

nvm its not consisent

gentle peak Dec 28, 2024, 2:44 AM

#

anyway

#

due to that refactor, i can tell you that the common path should be cur->freelist != NULL

#

so your performance issues are most likely due to something causing that to never happen

#

or almost never anyway

#

is it open source so i can take a look at more of the allocator?

dusky glade Dec 28, 2024, 2:46 AM

#

dusky glade removed even more not needed memset calls

exact same on kvm

dusky glade Dec 28, 2024, 2:46 AM

#

gentle peak is it open source so i can take a look at more of the allocator?

YEPP

#

heres the link

#

https://github.com/rayanmargham/NyauxKC/

GitHub

GitHub - rayanmargham/NyauxKC

Contribute to rayanmargham/NyauxKC development by creating an account on GitHub.

gentle peak Dec 28, 2024, 2:46 AM

#

awesome

north hull Dec 28, 2024, 2:47 AM

#

gentle peak is it open source so i can take a look at more of the allocator?

it is not for the weak minded

dusky glade Dec 28, 2024, 2:47 AM

#

north hull it is not for the weak minded

😭 trl

gentle peak Dec 28, 2024, 2:48 AM

#

i think i'm gonna need a debugger for this because i really can't analyze this code purely by source

north hull Dec 28, 2024, 2:48 AM

#

the low res trollface is fucked

dusky glade Dec 28, 2024, 2:48 AM

#

for love of god do not look at ioapic.c for love of god do not look at ioapic.c

median crest Dec 28, 2024, 2:49 AM

#

bruh using my new locks barely changed anything at all wtf is this

north hull Dec 28, 2024, 2:49 AM

#

DONT LOOK AT SCHED.C

median crest Dec 28, 2024, 2:49 AM

#

it goes all over the place

gentle peak Dec 28, 2024, 2:50 AM

#

why does git submodule update --init --recursive not even work

dusky glade Dec 28, 2024, 2:50 AM

#

median crest it goes all over the place

mathew looks at nyaux ioapic.c challenge

dusky glade Dec 28, 2024, 2:51 AM

#

gentle peak why does `git submodule update --init --recursive` not even work

i do not know

#

i do have a .gitsubmodules

#

dont know why it doesnt work tho

north hull Dec 28, 2024, 2:51 AM

#

I didnt even do that

#

i just cloned it and it worked

median crest Dec 28, 2024, 2:52 AM

#

can anyone try this I'm coping that its my computer being bad

📎 astral.iso

#

qemu-system-x86_64 -cdrom astral.iso -M q35 -enable-kvm -cpu host,migratable=off

gentle peak Dec 28, 2024, 2:53 AM

#

right well the sources are apparently here without updating the submodules so i'm just gonna ignore that

north hull Dec 28, 2024, 2:53 AM

#

I can throw it on my server

#

sec

dusky glade Dec 28, 2024, 2:53 AM

#

gentle peak right well the sources are apparently here without updating the submodules so i'...

yea lol

#

if u have any questions ping me

gentle peak Dec 28, 2024, 2:53 AM

#

median crest can anyone try this I'm coping that its my computer being bad

median crest Dec 28, 2024, 2:54 AM

#

bruh it made it even worse

dusky glade Dec 28, 2024, 2:54 AM

#

oops

#

trl

zenith crescent Dec 28, 2024, 2:54 AM

#

median crest can anyone try this I'm coping that its my computer being bad

median crest Dec 28, 2024, 2:54 AM

#

kms

gentle peak Dec 28, 2024, 2:54 AM

#

is the compile_commands.json supposed to be fucked

dusky glade Dec 28, 2024, 2:54 AM

#

gentle peak is the compile_commands.json supposed to be fucked

uhhh

#

run bear -- make

#

it usually works after that

#

its generated by bear

gentle peak Dec 28, 2024, 2:55 AM

#

+ clone_repo_commit https://github.com/osdev0/cc-runtime.git cc-runtime 9c281448fe40498a6341d74f55c9af4f642c0b1d
+ test -d cc-runtime/.git
+ test -d cc-runtime
+ set +x
error: 'cc-runtime' is not a Git repository
make: *** [GNUmakefile:161: kernel-deps] Error 1``` fun

#

i just removed the dir

#

works now

lilac knoll Dec 28, 2024, 2:56 AM

#

this whole thread is really manifesting goodhart's law

north hull Dec 28, 2024, 2:56 AM

#

@median crest

#

does ur kernel do serial logs by any chance

#

cuz my server is headless

#

lol

median crest Dec 28, 2024, 2:57 AM

#

I can do e9

#

hold on

north hull Dec 28, 2024, 2:57 AM

#

❤️

north hull Dec 28, 2024, 3:00 AM

#

lilac knoll this whole thread is really manifesting goodhart's law

I dont see an issue with turning off every feature but the ones required by uacpi to measure the success of my kernel

gentle peak Dec 28, 2024, 3:01 AM

#

wow it was surprisingly painless to get a vscode debug configuration working

north hull Dec 28, 2024, 3:01 AM

#

ikr

#

I did it once

#

but then I realized how unergonomic it actually is

median crest Dec 28, 2024, 3:02 AM

#

@north hull

📎 astral.iso

#

kinda funny how my new lock is much better than the old one but did not even affect speed

#

still keeping it though, having a word sized lock is great

gentle peak Dec 28, 2024, 3:04 AM

#

nyaux with -O0 how fast will it be

north hull Dec 28, 2024, 3:04 AM

#

how do I get the serial logs to showup again

median crest Dec 28, 2024, 3:04 AM

#

-debugcon stdio

north hull Dec 28, 2024, 3:04 AM

#

ah

#

i tried -serial stdio

zenith crescent Dec 28, 2024, 3:05 AM

#

e9 isn't serial

north hull Dec 28, 2024, 3:05 AM

#

yeah yeah

north hull Dec 28, 2024, 3:05 AM

#

median crest <@226721688203558913>

dusky glade Dec 28, 2024, 3:06 AM

#

gentle peak nyaux with -O0 how fast will it be

oh no

median crest Dec 28, 2024, 3:06 AM

#

north hull

interesting

north hull Dec 28, 2024, 3:06 AM

#

unfortunate

median crest Dec 28, 2024, 3:06 AM

#

I need to get a serial driver one dya

zenith crescent Dec 28, 2024, 3:06 AM

#

median crest interesting

at least it is consistent

north hull Dec 28, 2024, 3:07 AM

#

here is the entire log @median crest

📎 message.txt

gentle peak Dec 28, 2024, 3:07 AM

#

dusky glade oh no

~1k/s

#

i think make run is tcg too so

median crest Dec 28, 2024, 3:07 AM

#

north hull here is the entire log <@215823874615541761>

that assert fail is expected, it couldn't find nvme0n1

north hull Dec 28, 2024, 3:07 AM

#

ah

#

wonder why

dusky glade Dec 28, 2024, 3:08 AM

#

gentle peak i think make run is tcg too so

it is

median crest Dec 28, 2024, 3:08 AM

#

cuz there is no nvme set up lol

north hull Dec 28, 2024, 3:08 AM

#

right

#

but why does it require it

median crest Dec 28, 2024, 3:08 AM

#

its in the boot options

gentle peak Dec 28, 2024, 3:08 AM

#

dusky glade it is

with kvm on -O0 its 20k

median crest Dec 28, 2024, 3:08 AM

#

theres also an initrd option but I only shippsed the bare minimum there to make it small

#

bare minimum being not even enough to boot

north hull Dec 28, 2024, 3:09 AM

#

oh uacpi was loaded

#

im dumb

north hull Dec 28, 2024, 3:14 AM

#

north hull

this was with a AMD Ryzen 9 7900 (24) @ 5.482GHz

gentle peak Dec 28, 2024, 3:19 AM

#

god okay this is worse than i thought

#

i fixed the main bug causing the slab to be slow

#

(the bug was that each slab only had at most one object on its freelist)

#

but that exposed a bug where too-small slabs were getting picked for allocations, which wasn't caught before since due to the aforementioned bug each allocation was effectively 4k

#

so i fixed that

#

and now there's ANOTHER bug

#

which i haven't investigated yet

dusky glade Dec 28, 2024, 3:20 AM

#

gentle peak (the bug was that each slab only had at most one object on its freelist)

WAIT WHT???

#

TF??/

#

explain the bug

gentle peak Dec 28, 2024, 3:21 AM

#

right so in init_slab you build a new freelist right

#

but you were doing new->next = cur; cur = new; and then not ultimately setting hdr->freelist to cur at the end

#

so effectively the entire freelist was just the first object

dusky glade Dec 28, 2024, 3:22 AM

#

gentle peak but you were doing `new->next = cur; cur = new;` and then not ultimately setting...

oh my god.

gentle peak Dec 28, 2024, 3:22 AM

#

i fixed it in a slightly different way so that the ascending address order is preserved

left orbit Dec 28, 2024, 3:22 AM

#

is that nyaux?

gentle peak Dec 28, 2024, 3:22 AM

#

yes

left orbit Dec 28, 2024, 3:22 AM

#

damn monkuous turned into a nyaux developer

gentle peak Dec 28, 2024, 3:22 AM

#

now i'm getting a bunch of USELESS SLAB messages which are (i assume) ruining perf

left orbit Dec 28, 2024, 3:22 AM

#

that means nyaux is the new kernel we should all be working on

gentle peak Dec 28, 2024, 3:23 AM

#

@dusky glade with that + the selecting slab bug fixed + removing USELESS SLAB, i get 146061/s on -O0 + KVM

#

on -O2 this is now uacpi: successfully loaded 1 AML blob, 1705 ops in 3ms (avg 505754/s)

dusky glade Dec 28, 2024, 3:24 AM

#

oh my god.

left orbit Dec 28, 2024, 3:24 AM

#

holy fuck what

#

how is nyaux doing THIS WELL

#

successfully loaded 1 AML blob, 1705 ops in 3ms (avg 443800/s)

#

how am i slower 😭😭

hollow elm Dec 28, 2024, 3:25 AM

#

LMFAO

gentle peak Dec 28, 2024, 3:26 AM

#

here's the patch ```patch
diff --git a/kernel/src/mem/pmm.c b/kernel/src/mem/pmm.c
index 838b4e5..a69ea75 100644
--- a/kernel/src/mem/pmm.c
+++ b/kernel/src/mem/pmm.c
@@ -119,7 +119,8 @@ slab* init_slab(uint64_t size)
for (uint64_t i = 1; i < obj_amount; i++)
{
pnode* new = (pnode*)((uint64_t)first + (i * size));

```
   new->next = (struct pnode*)cur;
```

```
   new->next = NULL;
```

   cur->next = (struct pnode*)new;
   cur = new;

};
return hdr;
@@ -158,7 +159,7 @@ void* slab_alloc(cache* mod)
}

   cur = (slab*)cur->next;

```
   // assert(cur != NULL);
```

```
   assert(cur != NULL);
```
}

void* ptr = cur->freelist;
@@ -173,7 +174,7 @@ void* slaballocate(uint64_t amount)
for (int i = 0; i != 7; i++)
{
cache* c = &caches[i];

```
   assert(c->size != 0);
```

```
   if (c->size < amount) continue;
   return slab_alloc(c);
```
}
return NULL;
@@ -257,6 +258,6 @@ void slabfree(void* addr)
}
if (howmanynodes == guy->obj_ammount)
{

```
   kprintf("USELESS SLAB\n");
```

```
   //kprintf("USELESS SLAB\n");
```
}
}

zenith crescent Dec 28, 2024, 3:26 AM

#

left orbit how am i slower 😭😭

different CPU

left orbit Dec 28, 2024, 3:26 AM

#

successfully loaded 1 AML blob, 1705 ops in 2ms (avg 809178/s)

#

there we go.

dusky glade Dec 28, 2024, 3:26 AM

#

gentle peak and now there's ANOTHER bug

the other "bug" was just useless slab being printed right?

gentle peak Dec 28, 2024, 3:26 AM

#

yeah

left orbit Dec 28, 2024, 3:26 AM

#

zenith crescent different CPU

i have a faster cpu strictly speaking

zenith crescent Dec 28, 2024, 3:26 AM

#

oh

#

I didn't read properly

left orbit Dec 28, 2024, 3:27 AM

#

but like there is a lot of factors at play

gentle peak Dec 28, 2024, 3:27 AM

#

i figured that was some debug message you put in as a pseudo assert

dusky glade Dec 28, 2024, 3:27 AM

#

gentle peak yeah

THIS ONE FUCKING BUG

gentle peak Dec 28, 2024, 3:27 AM

#

turns out it was more like a todo

dusky glade Dec 28, 2024, 3:27 AM

#

SLOWED NYAUX INTO ASS

dusky glade Dec 28, 2024, 3:27 AM

#

gentle peak turns out it was more like a todo

yea it was

#

thank you @gentle peak

#

actually.

#

2 lines of code

#

slowed down nyaux like crazy

#

your a very smart developer.

left orbit Dec 28, 2024, 3:28 AM

#

if those kinds of bugs are "hiding" in nyaux

#

literally hiding in plain sight

#

i wonder what other kinds of bugs there are

dusky glade Dec 28, 2024, 3:28 AM

#

lol

left orbit Dec 28, 2024, 3:28 AM

#

no like im not even being mean at this point

#

sorry if it sounds like it

dusky glade Dec 28, 2024, 3:28 AM

#

no u arent dw

#

even im surprised

gentle peak Dec 28, 2024, 3:29 AM

#

fun fact

#

this is actually the exact same bug i had in one of the first iterations of proxima

#

and fixing it made my perf go from 1.1M to 4.5M

lilac knoll Dec 28, 2024, 3:31 AM

#

so this is basically an allocator benchmark?

#

did someone try with a bump allocator already

gentle peak Dec 28, 2024, 3:32 AM

#

slower than proxima due to all allocations being in cold memory

lilac knoll Dec 28, 2024, 3:32 AM

#

damn

#

that's very surprising

hollow elm Dec 28, 2024, 3:32 AM

#

lilac knoll so this is basically an allocator benchmark?

allocator + mutex/spinlock + memcpy/memset meme

gentle peak Dec 28, 2024, 3:33 AM

#

@dusky glade is the profiler code in the git repo? or did you take it out before pushing

#

ah nvm found it

left orbit Dec 28, 2024, 3:33 AM

#

successfully loaded 1 AML blob, 1705 ops in 1ms (avg 956441/s)

#

compiling out all the asserts

#

because for some reason i didn't do that in !debug meme

lilac knoll Dec 28, 2024, 3:34 AM

#

gentle peak slower than proxima due to all allocations being in cold memory

how much memory is allocated in total?

north holly Dec 28, 2024, 3:35 AM

#

if this profiler was able to fix nyaux

gentle peak Dec 28, 2024, 3:35 AM

#

lilac knoll how much memory is allocated in total?

don't know

north holly Dec 28, 2024, 3:35 AM

#

surely it can fix me

lilac knoll Dec 28, 2024, 3:35 AM

#

like roughly, megabytes?

north holly Dec 28, 2024, 3:35 AM

#

iirc only a few hundred kb

left orbit Dec 28, 2024, 3:36 AM

#

1835534

#

bytes

#

so about 2mb for the qemu q35 blob

lilac knoll Dec 28, 2024, 3:36 AM

#

so really a bump allocator is slower nooo

even prefetching the entire memory blob ahead of time?

left orbit Dec 28, 2024, 3:36 AM

#

or, well, all memory needed to parse the tables and dsdt

dusky glade Dec 28, 2024, 3:37 AM

#

on tcg

#

oh no

#

on kvm

#

its triple faulting

#

nooo

#

oh fuck

gentle peak Dec 28, 2024, 3:37 AM

#

lilac knoll like roughly, megabytes?

if i make my heap never give memory back to the pmm then the pmm reports 848K used at end of kernel init

#

so

dusky glade Dec 28, 2024, 3:38 AM

#

@gentle peak any idea?

gentle peak Dec 28, 2024, 3:38 AM

#

depends

#

what changes did you make

dusky glade Dec 28, 2024, 3:38 AM

#

i just applied the patch

#

is all

#

it just triple faults

#

on kvm

#

like

#

here

#

consistently

#

???

gentle peak Dec 28, 2024, 3:39 AM

#

like, with git apply?

dusky glade Dec 28, 2024, 3:39 AM

#

yea

gentle peak Dec 28, 2024, 3:39 AM

#

then you have the exact same source as me and i'm not getting the bug

dusky glade Dec 28, 2024, 3:39 AM

#

gentle peak then you have the exact same source as me and i'm not getting the bug

did u try running kvm?

gentle peak Dec 28, 2024, 3:39 AM

#

what command are you using to run?

#

this is my command

lilac knoll Dec 28, 2024, 3:39 AM

#

gentle peak if i make my heap never give memory back to the pmm then the pmm reports 848K us...

that's still really surprising bump would be slower no? i mean the prefetcher should be going brr, and that can fit in cache still

dusky glade Dec 28, 2024, 3:40 AM

#

gentle peak Dec 28, 2024, 3:40 AM

#

lilac knoll that's still really surprising bump would be slower no? i mean the prefetcher sh...

idk that's what the experimental data told us

#

bump certainly can't be slower for the allocating itself so the speed diff must be due to cache

dusky glade Dec 28, 2024, 3:40 AM

#

i dont know whats going on

#

running make clean

#

yep

gentle peak Dec 28, 2024, 3:41 AM

#

@dusky glade only happens with -smp 20

dusky glade Dec 28, 2024, 3:41 AM

#

triple faults

gentle peak Dec 28, 2024, 3:41 AM

#

if it's with -smp 1 it's fine

dusky glade Dec 28, 2024, 3:41 AM

#

gentle peak <@787341689333350450> only happens with `-smp 20`

yea what.

gentle peak Dec 28, 2024, 3:41 AM

#

so it's probably a race condition

dusky glade Dec 28, 2024, 3:41 AM

#

tf...

gentle peak Dec 28, 2024, 3:41 AM

#

i noticed that your allocator has no locking? or at least none that i could see

dusky glade Dec 28, 2024, 3:41 AM

#

it does have

#

in kmalloc.c

#

#

just a global memory lock

lilac knoll Dec 28, 2024, 3:42 AM

#

and your bump allocator was literally, get a block and increment pointer each time?

I would also try to populate the tlb since tlb refills will be quite slow inside kvm

#

and a bit of prefetching to warm things up

#

that's quite intruiging though, almost makes me want to whip out the performance counters

gentle peak Dec 28, 2024, 3:43 AM

#

lilac knoll and your bump allocator was literally, get a block and increment pointer each ti...

yep

#

as for prefetching & stuff

#

we tried double init using uacpi_state_reset in between

#

did noticeably improve performance

#

but obv not useful in real world

lilac knoll Dec 28, 2024, 3:43 AM

#

improve better than the slab?

gentle peak Dec 28, 2024, 3:44 AM

#

i didn't try that together with bump

#

only together with slab

lilac knoll Dec 28, 2024, 3:44 AM

#

it is worth checking the performance counters in general though, might reveal useful info

dusky glade Dec 28, 2024, 3:45 AM

#

this is weird

#

maybe its the same bug that the scheduler to die when u run the profiler

gentle peak Dec 28, 2024, 3:45 AM

#

@dusky glade your create_region gets called from other cpus and bypasses the global memory lock

dusky glade Dec 28, 2024, 3:45 AM

#

i dont know

gentle peak Dec 28, 2024, 3:45 AM

#

at least i think it bypasses the lock

dusky glade Dec 28, 2024, 3:46 AM

#

gentle peak <@787341689333350450> your `create_region` gets called from other cpus and bypas...

to call create region

#

u need to call kvm_region_calloc

#

alloc*

#

to call that

#

u need to go through the mem lock

gentle peak Dec 28, 2024, 3:46 AM

#

ah ok

#

oh for fucks sake kvm why aren't you giving me the rip of the cpu that triple faulted

#

wait @dusky glade that 500k/s measurement was with UBSan enabled, with it disabled it's 1,459,252/s

dusky glade Dec 28, 2024, 3:52 AM

#

gentle peak wait <@787341689333350450> that 500k/s measurement was with UBSan enabled, with ...

damn

#

1 million

#

@left orbit

#

trl

lilac knoll Dec 28, 2024, 3:53 AM

#

stonks

dusky glade Dec 28, 2024, 3:53 AM

#

but we still have a bug

#

nooo

#

cringe triple fault

gentle peak Dec 28, 2024, 3:53 AM

#

ok i managed to reproduce the triple fault with TCG

#

so i can now actually inspect state

dusky glade Dec 28, 2024, 3:54 AM

#

how did u reproduce it

gentle peak Dec 28, 2024, 3:54 AM

#

-smp 255

dusky glade Dec 28, 2024, 3:54 AM

#

oh i see

gentle peak Dec 28, 2024, 3:54 AM

#

ran it a few times

left orbit Dec 28, 2024, 3:55 AM

#

gentle peak `-smp 255`

lmfao

left orbit Dec 28, 2024, 3:55 AM

#

dusky glade 1 million

ill beat u eventually

#

don’t u worry

#

but now time for eep

#

eepy

dusky glade Dec 28, 2024, 3:55 AM

#

get eepy

gentle peak Dec 28, 2024, 3:55 AM

#

dusky glade oh i see

https://hst.sh/desecidamo.apache here's the exception state

#

gonna go to sleep now as well

dusky glade Dec 28, 2024, 3:55 AM

#

general protection fault

#

i see

#

in flanterm @gentle peak ??

#

huh

#

and now its suddenly a segment related issue

#

crazy

#

hollow elm Dec 28, 2024, 3:57 AM

#

the address is not likely to match between builds

dusky glade Dec 28, 2024, 3:57 AM

#

probs not

#

bros some UB

north holly Dec 28, 2024, 3:57 AM

#

gentle peak https://hst.sh/desecidamo.apache here's the exception state

ur profiler thing broke obos

#

it causes all code to get corrupted, it seems

gentle peak Dec 28, 2024, 3:58 AM

#

dusky glade probs not

this is the call stack

dusky glade Dec 28, 2024, 3:58 AM

#

gentle peak this is the call stack

yea thats so fucking weird

#

wtf

north holly Dec 28, 2024, 3:59 AM

#

north holly it causes all code to get corrupted, it seems

wait I think I stack overflowed

#

yeah I did

dusky glade Dec 28, 2024, 3:59 AM

#

i had a similar issue in which there was something wrong with the far return

#

and it also faulted on that instruction

#

that is solved now

gentle peak Dec 28, 2024, 3:59 AM

#

#GP(selector) If the source selector points to a segment that is not a TSS or to one for a task that is already busy.

#

the two gdt entries at that selector:

0x00908ba5802c0067
0x00000000ffff8000

dusky glade Dec 28, 2024, 4:00 AM

#

first one looks so fucking weird

#

tf

#

memory corruption?

#

yea looking at the bit sitself

#

first one is corrupted

#

i think

#

nvm second one looks weirder

gentle peak Dec 28, 2024, 4:03 AM

#

keep in mind this is a tss descriptor

#

2nd one is top 32 bits of base zero extended to 64 bits

#

base 0xffff800000a5802c
limit 0x67
attr 0b1001
pres 1
dpl 0
type 0xb

#

yeah so the tss is busy

dusky glade Dec 28, 2024, 4:04 AM

#

that is not at all correct

#

at all

gentle peak Dec 28, 2024, 4:04 AM

#

everything except the type is correct for a tss

dusky glade Dec 28, 2024, 4:05 AM

#

im really suspecting memory corruption

#

as i do set the correct bits in gdt_init

gentle peak Dec 28, 2024, 4:05 AM

#

yeah no this is just a race condition

dusky glade Dec 28, 2024, 4:05 AM

#

wdym

gentle peak Dec 28, 2024, 4:05 AM

#

you're setting the correct bits in gdt_init yes, but you're setting them in a global shared variable without locking said variable

#

multiple cpus are running gdt_init at the same time and setting the same memory

dusky glade Dec 28, 2024, 4:05 AM

#

hold on

#

ur right

#

i will put a lock and tes tit

#

no it still triple faults

north holly Dec 28, 2024, 4:07 AM

#

gentle peak you're setting the correct bits in gdt_init yes, but you're setting them in a gl...

you should probably go to sleep

dusky glade Dec 28, 2024, 4:07 AM

#

yea

#

to be fair u need to go to sleep

gentle peak Dec 28, 2024, 4:07 AM

#

dusky glade no it still triple faults

here's the patch i used which successfully fixes it ```patch
diff --git a/kernel/src/arch/x86_64/gdt/gdt.c b/kernel/src/arch/x86_64/gdt/gdt.c
index 218233d..09759db 100644
--- a/kernel/src/arch/x86_64/gdt/gdt.c
+++ b/kernel/src/arch/x86_64/gdt/gdt.c
@@ -3,6 +3,7 @@
#include <stdint.h>

#include "sched/sched.h"
+#include "utils/basic.h"

typedef struct
{
@@ -10,11 +11,13 @@ typedef struct
uint64_t offset;
} attribute((packed)) GDTR;
uint64_t gdt[11];
+static spinlock_t gdt_lock;
extern void gdt_flush(void*);
GDTR gptr = {};

void init_gdt()
{

spinlock_lock(&gdt_lock);
gdt[0] = 0x0;
gdt[1] = 0x00009a000000ffff;
gdt[2] = 0x000093000000ffff;
@@ -44,6 +47,7 @@ void init_gdt()
gptr.size = sizeof(gdt) - 1;

gdt_flush(&gptr);
spinlock_unlock(&gdt_lock);
}
void change_rsp0(uint64_t stackaddr)
{

gentle peak Dec 28, 2024, 4:07 AM

#

north holly you should probably go to sleep

you're right

#

good night

dusky glade Dec 28, 2024, 4:07 AM

#

gentle peak here's the patch i used which successfully fixes it ```patch diff --git a/kernel...

i just added it

north holly Dec 28, 2024, 4:08 AM

#

gentle peak good night

gn

dusky glade Dec 28, 2024, 4:08 AM

#

wait

#

nvm

#

forgot to run make

#

goodnight monkuous have a goodnights rest

#

thanks for the stupid race condition u caught as well!

median crest Dec 28, 2024, 4:11 AM

#

oh wtf booting with 64 mbs of memory gives a huge boost to performance what

#

I go from 900k to 1.4mil-1.5mil consistently

#

on my crap laptop

dusky glade Dec 28, 2024, 4:12 AM

#

median crest I go from 900k to 1.4mil-1.5mil consistently

nyaux has 1 million ops now

#

😎

median crest Dec 28, 2024, 4:12 AM

#

neat

dusky glade Dec 28, 2024, 4:12 AM

#

no like actually

#

without ubsan

#

it runs 1m ops

#

ish

#

500k with ubsan

#

because of one slab allocator bug lol

#

@fiery turtle update my score when u wake up

median crest Dec 28, 2024, 4:15 AM

#

seems like running with over 2 gbs of memory slows it down

#

wonder why

#

if I run with 2.9gb it gets the consistent ~1.4mil and with 3 gb it gets 1 mil wtf??

zenith crescent Dec 28, 2024, 4:54 AM

#

dusky glade nyaux has 1 million ops now

You haven't implemented PCI though have you?

fiery turtle Dec 28, 2024, 11:15 AM

#

Bruh wtf have I missed

gentle peak Dec 28, 2024, 11:16 AM

#

nyaux is 1.5M/s if you change 3 lines of code and disable ubsan

fiery turtle Dec 28, 2024, 11:16 AM

#

Nyaux single handedly fixed by monkuous

fiery turtle Dec 28, 2024, 11:16 AM

#

gentle peak nyaux is 1.5M/s if you change 3 lines of code and disable ubsan

💀 thats actually crazy

left orbit Dec 28, 2024, 11:17 AM

#

slab was broken basically

fiery turtle Dec 28, 2024, 11:17 AM

#

Oberrow missed his chance at getting obos decursed as well

#

Now its last place

gentle peak Dec 28, 2024, 11:17 AM

#

every time a slab was allocated only the first object was put onto its freelist so effectively every allocation went to the pmm

fiery turtle Dec 28, 2024, 11:17 AM

#

Ah

left orbit Dec 28, 2024, 11:18 AM

#

also @\infy if you're still curious about zig's ability, you can take ghostty (the zig terminal emulator) and actually effortlessly cross compile it on linux for windows, which i find kinda cool that it just produces a working windows binary with no dependencies other than zig, everything including the C/C++ dependencies is built by zig and statically linked into the final binary

gentle peak Dec 28, 2024, 11:19 AM

#

oh btw @dusky glade i figured out why the profiler caused a page fault: when creating a task your scheduler sets its rsp to the top of the stack, but the sysv abi expects the return address to be stored at rsp, so when it tried to read the return address it page faulted

fiery turtle Dec 28, 2024, 11:19 AM

#

Oh yeah that's pretty nice, I know people who do osses in zig literally compile elf on windows

gentle peak Dec 28, 2024, 11:19 AM

#

fixed by just subtracting 8 from the top of the stack when creating a task

left orbit Dec 28, 2024, 11:20 AM

#

fiery turtle Oh yeah that's pretty nice, I know people who do osses in zig literally compile ...

yeah that is also possible, if you want to do that for whatever reason

fiery turtle Dec 28, 2024, 11:20 AM

#

Bruh inb4 nyaux beats proxima trl

left orbit Dec 28, 2024, 11:20 AM

#

xorriso for windows is another beast

fiery turtle Dec 28, 2024, 11:20 AM

#

Xorriso is kind of insane yeah

left orbit Dec 28, 2024, 11:20 AM

#

i don't think i've seen a working version

#

but it's out there supposedly

gentle peak Dec 28, 2024, 11:20 AM

#

i've used a working xorriso for windows multiple times

fiery turtle Dec 28, 2024, 11:21 AM

#

I want to try it out for sure, you've shilled it enough LULW

gentle peak Dec 28, 2024, 11:21 AM

#

i think this is the place i downloaded it from usually? https://github.com/PeyTy/xorriso-exe-for-windows

GitHub

GitHub - PeyTy/xorriso-exe-for-windows: Latest xorriso.exe 1.5.2 pr...

Latest xorriso.exe 1.5.2 pre-build for Windows to make iso disk images to get CD to boot on EFI system - PeyTy/xorriso-exe-for-windows

fiery turtle Dec 28, 2024, 11:21 AM

#

Surprisingly its from a guy with fake os

gentle peak Dec 28, 2024, 11:21 AM

#

seems like it's just a cygwin build though

left orbit Dec 28, 2024, 11:22 AM

#

fiery turtle I want to try it out for sure, you've shilled it enough <:LULW:54342055302188237...

yeah i just wanted to note my little experiment, i think ghostty is a nice benchmark for the compiler and showcases it's in a much better state than 1-1.5 years ago where it crashed if you called a function in a funny way lol

fiery turtle Dec 28, 2024, 11:22 AM

#

Yeah

torpid root Dec 28, 2024, 11:27 AM

#

damn I gotta pump my numbers up

#

step 1: figure out how

#

step 2: give up

#

that concludes the work, goodbye

left orbit Dec 28, 2024, 11:28 AM

#

i wonder what the numbers will look like if infy decides to work on a kernel of his own

fiery turtle Dec 28, 2024, 11:31 AM

#

Im shit at algorithms sooo

#

At the bottom with obos

north hull Dec 28, 2024, 11:32 AM

#

I will soon join obos at the bottom

torpid root Dec 28, 2024, 11:59 AM

#

my kernel refuses to get above 1.5mil

#

on a slowish cpu of course

fiery turtle Dec 28, 2024, 12:16 PM

#

north hull I will soon join obos at the bottom

more people at the bottom, easier to look cool with my potential future kernel next to obos so im happy LULW

left orbit Dec 28, 2024, 12:17 PM

#

can't wait

#

(i need a good project to contribute to)

fiery turtle Dec 28, 2024, 12:17 PM

#

well proxima

#

or astral

#

or managarm

#

or obos lol

left orbit Dec 28, 2024, 12:18 PM

#

lol obos

fiery turtle Dec 28, 2024, 12:18 PM

#

obos has suspend which is cool

left orbit Dec 28, 2024, 12:18 PM

#

astral and obos nah because of code quality, my "ocd" makes it impossible to work with code like that

#

sadly

fiery turtle Dec 28, 2024, 12:18 PM

#

oh yeah astral code style is uhhh

vagrant hull Dec 28, 2024, 12:18 PM

#

mfw no ironclad

fiery turtle Dec 28, 2024, 12:18 PM

#

hardtocomprehend

left orbit Dec 28, 2024, 12:18 PM

#

it's ada lol

#

but for others sure

vagrant hull Dec 28, 2024, 12:19 PM

#

contribute to cynix or something, there is not much so it's not a good project

fiery turtle Dec 28, 2024, 12:19 PM

#

oh ye

vagrant hull Dec 28, 2024, 12:19 PM

#

it's just a project

fiery turtle Dec 28, 2024, 12:19 PM

#

forgot about it

left orbit Dec 28, 2024, 12:19 PM

#

autohell makes me wanna cry but i'll look

vagrant hull Dec 28, 2024, 12:19 PM

#

it doesn't even use autohell

left orbit Dec 28, 2024, 12:20 PM

#

then what

#

plain makefile?

vagrant hull Dec 28, 2024, 12:20 PM

#

yeah

left orbit Dec 28, 2024, 12:20 PM

#

close enough

#

surprised you didn't do autotools

vagrant hull Dec 28, 2024, 12:20 PM

#

bear it up :p

fiery turtle Dec 28, 2024, 12:20 PM

#

implement enough in cynix so it can be put on the leaderboard

#

iirc it scored very high

vagrant hull Dec 28, 2024, 12:20 PM

#

i don't like autotools

#

it does 4M/s on my system

fiery turtle Dec 28, 2024, 12:20 PM

#

tahts probably with old uacpi too

vagrant hull Dec 28, 2024, 12:20 PM

#

yeah

left orbit Dec 28, 2024, 12:22 PM

#

fiery turtle well proxima

i didn't really look at proxima so it could be a target, and managarm is phd code so i'd rather blow my head off than try to wrap my head around all the async hell :^)

vagrant hull Dec 28, 2024, 12:22 PM

#

the thing with Cynix compared with the other examples there is that despite there not being much to show, admittedly, i have tried to keep the code really clean and there is no technical debt

fiery turtle Dec 28, 2024, 12:22 PM

#

my main problem with managarm is its hard to build

#

u need a sequence of commands, files, edit stuff

mortal yoke Dec 28, 2024, 12:22 PM

#

if I ever get this PE kernel anywhere I am going to try uacpi on it, can also do sse because the call conv saves them iirc (so using it in interrupts is not a problem)

left orbit Dec 28, 2024, 12:23 PM

#

fiery turtle my main problem with managarm is its hard to build

it's actually easy with cbuildrt, and you get to pull tools and packages from ci

fiery turtle Dec 28, 2024, 12:23 PM

#

proxima and cynix are both pretty clean i think

vagrant hull Dec 28, 2024, 12:23 PM

#

whose project is Proxima again, sorry?

#

i forgor

left orbit Dec 28, 2024, 12:23 PM

#

@\monkuous

fiery turtle Dec 28, 2024, 12:23 PM

#

monkuous

vagrant hull Dec 28, 2024, 12:23 PM

#

oh right

#

the fast boi

fiery turtle Dec 28, 2024, 12:23 PM

#

they single handedly fixed nyaux

#

now it gets 1.5M

left orbit Dec 28, 2024, 12:23 PM

#

mortal yoke if I ever get this PE kernel anywhere I am going to try uacpi on it, can also do...

i wanted to do a funny PE kernel too lol

strong heath Dec 28, 2024, 12:24 PM

#

Ironclad is unironically the best to contribute, its reasonably stable, very low technical debt, very self contained, very documented, easy to test, it just has a single problem, and that is knowing Ada

left orbit Dec 28, 2024, 12:24 PM

#

i like how PE does imports/exports

#

compare to ELF

#

it actually has a dedicated thing for that, which is awesome

mortal yoke Dec 28, 2024, 12:24 PM

#

yeah

fiery turtle Dec 28, 2024, 12:24 PM

#

PE is also built with reloc in mind

left orbit Dec 28, 2024, 12:24 PM

#

yup

fiery turtle Dec 28, 2024, 12:24 PM

#

not as an afterthought

left orbit Dec 28, 2024, 12:26 PM

#

my plan was to have a loader that boots with limine and gets the kernel + some modules passed as limine modules, does some simple memory management, loads kernel + module into the memory and passes control to the kernel

#

but i hate C build systems

calm latch Dec 28, 2024, 12:26 PM

#

left orbit yeah i just wanted to note my little experiment, i think ghostty is a nice bench...

Is it difficult to patch to have os specific toolchain?

left orbit Dec 28, 2024, 12:26 PM

#

probably not, as long as you add it to llvm

slim panther Dec 28, 2024, 12:27 PM

#

left orbit astral and obos nah because of code quality, my "ocd" makes it impossible to wor...

like what

vagrant hull Dec 28, 2024, 12:27 PM

#

PE support to Limine would honestly be a welcome addition

fiery turtle Dec 28, 2024, 12:27 PM

#

building llvm 💀

left orbit Dec 28, 2024, 12:27 PM

#

vagrant hull PE support to Limine would honestly be a welcome addition

like loading kernels from a PE instead of an ELF?

vagrant hull Dec 28, 2024, 12:27 PM

#

yeah

left orbit Dec 28, 2024, 12:27 PM

#

i mean i could do that but

#

limine boot protocol doesn't really do PE

vagrant hull Dec 28, 2024, 12:28 PM

#

the protocol itself is literally 100% executable agnostic

left orbit Dec 28, 2024, 12:28 PM

#

is it?

vagrant hull Dec 28, 2024, 12:28 PM

#

so no, that's untrue

#

yes

left orbit Dec 28, 2024, 12:28 PM

#

hmm i thought it was ELF specific lol

#

my bad

calm latch Dec 28, 2024, 12:28 PM

#

left orbit astral and obos nah because of code quality, my "ocd" makes it impossible to wor...

Have you looked at pmOS? ultrameme

slim panther Dec 28, 2024, 12:28 PM

#

doesnt the protocol mandate e.g. ET_DYN

vagrant hull Dec 28, 2024, 12:28 PM

#

never mentions ELF once

#

nope

#

not at all

calm latch Dec 28, 2024, 12:28 PM

#

left orbit probably not, as long as you add it to llvm

I have llvm

left orbit Dec 28, 2024, 12:28 PM

#

it actually doesn't, wow

fiery turtle Dec 28, 2024, 12:28 PM

#

limelfline

vagrant hull Dec 28, 2024, 12:28 PM

#

i took care specifically of never making the protocol in any way ELF specific

slim panther Dec 28, 2024, 12:28 PM

#

cool

#

when a.out support

vagrant hull Dec 28, 2024, 12:29 PM

#

it can

left orbit Dec 28, 2024, 12:29 PM

#

The executable, loaded at or above 0xffffffff80000000, sees all of its segments mapped using write-back (WB) caching at the page tables level.

#

segments

#

very nice

#

PE support is very doable then, i guess

fiery turtle Dec 28, 2024, 12:29 PM

#

clearly this means GDT segments meme

slim panther Dec 28, 2024, 12:29 PM

#

meme

left orbit Dec 28, 2024, 12:29 PM

#

especially since i don't have to do imports/exports

mortal yoke Dec 28, 2024, 12:30 PM

#

left orbit my plan was to have a loader that boots with limine and gets the kernel + some m...

I also basically did that, its slightly cursed though because it has to have a page somewhere in the low virtual memory mapped in both the current page map and the new page map where you switch the pagemap

left orbit Dec 28, 2024, 12:30 PM

#

only load the PE into memory and do relocs

slim panther Dec 28, 2024, 12:30 PM

#

left orbit especially since i don't have to do imports/exports

you still have to do relocs

vagrant hull Dec 28, 2024, 12:30 PM

#

segments in here isn't referring to ELF segments specifically, rather just executable image segments

left orbit Dec 28, 2024, 12:30 PM

#

yeah i get that, i just like the fact that it's format agnostic

vagrant hull Dec 28, 2024, 12:31 PM

#

and there can be only 1 segment, for those formats (or really lack thereof) that don't have segmentation

left orbit Dec 28, 2024, 12:31 PM

#

also technically we could implement a dynamic linker for PE into limine lol

vagrant hull Dec 28, 2024, 12:31 PM

#

eh

#

we don't have that for ELF either

left orbit Dec 28, 2024, 12:31 PM

#

it's piss easy as it's just loading multiple PE files and resolving imports

#

yeah but ELF is a bit more complicated in that regard

#

but true, feature parity would be nice

vagrant hull Dec 28, 2024, 12:32 PM

#

it's not something we couldn't do

slim panther Dec 28, 2024, 12:32 PM

#

left orbit also technically we could implement a dynamic linker for PE into limine lol

why tho

left orbit Dec 28, 2024, 12:32 PM

#

slim panther why tho

because it would be nice, and as i said it's really easy with PE

vagrant hull Dec 28, 2024, 12:32 PM

#

yeah i don't think it's that useful honestly

slim panther Dec 28, 2024, 12:32 PM

#

it requires extensive kernel support

vagrant hull Dec 28, 2024, 12:32 PM

#

i'd worry about plain PE support first

mortal yoke Dec 28, 2024, 12:32 PM

#

and for things like kernel modules the kernel can also do that itself if it wants

left orbit Dec 28, 2024, 12:32 PM

#

slim panther it requires extensive kernel support

not really extensive, it's done by the bootloader :^)

#

so you can just dynamically link ur kernel and drop the modules in /boot

slim panther Dec 28, 2024, 12:33 PM

#

well, how do you make sure it's mapped where the kernel won't complain

left orbit Dec 28, 2024, 12:33 PM

#

relocations have a size

#

but the IAT is just a table of addresses

#

so it won't complain

#

different story for elf maybe, i guess

#

idk how dynamic linking works in the ELF land

slim panther Dec 28, 2024, 12:34 PM

#

it's not that bad

#

i have a dynamic ELF linking impl already

left orbit Dec 28, 2024, 12:34 PM

#

well so does mlibc

#

but it's kinda poop lol

#

well, not mlibc

#

but elf dynamic linking in general

slim panther Dec 28, 2024, 12:34 PM

#

i need your opinion on my coding style KEKW

left orbit Dec 28, 2024, 12:35 PM

#

show

slim panther Dec 28, 2024, 12:36 PM

#

https://github.com/menix-os/menix/blob/main/kernel%2Fsystem%2Fmodule.c#L301-L589

GitHub

menix/kernel/system/module.c at main · menix-os/menix

A minimal and expandable Unix-like kernel. Contribute to menix-os/menix development by creating an account on GitHub.

left orbit Dec 28, 2024, 12:36 PM

#

kernel%2Fsystem%2Fmodule.c

#

wtf is github cooking

slim panther Dec 28, 2024, 12:36 PM

#

banger

left orbit Dec 28, 2024, 12:36 PM

#

looks properly formatted

fiery turtle Dec 28, 2024, 12:36 PM

#

tabs are L

#uACPI - a portable and easy-to-integrate ACPI implementation