pmOS - microkernel OS for RISC-V, x86, and LoongArch | Operating System Development | Page 5

dense carbon Jan 2, 2025, 7:45 PM

#

it's usually so fast as to warrant no asynchronicity

carmine nacelle Jan 2, 2025, 7:45 PM

#

Which then talks to drivers

#

Via IPC, which is asynchronous

#

And stores the cache

#

Idk, it's a bit of a dumb system, I can just put part of it into each driver

carmine nacelle Jan 2, 2025, 7:46 PM

#

carmine nacelle Which then talks to drivers

to traverse the tree

carmine nacelle Jan 2, 2025, 7:47 PM

#

dense carbon it's usually so fast as to warrant no asynchronicity

what if a filesystem starts reading from disk

dense carbon Jan 2, 2025, 7:47 PM

#

carmine nacelle what if a filesystem starts reading from disk

that's the unfortunate, uncached case

carmine nacelle Jan 2, 2025, 7:48 PM

#

which would make my vfs cache server block

#

(weird design)

left dew Jan 2, 2025, 7:50 PM

#

why do you have a vfs cache server?

carmine nacelle Jan 2, 2025, 7:52 PM

#

Because after the filesystem "handle" (I only have IPC ports atm) for a file is resolved from it, the libc then communicates with filesystems directly

left dew Jan 2, 2025, 7:52 PM

#

that is a weird design

carmine nacelle Jan 2, 2025, 7:52 PM

#

Yeah, but what else can I do

#

Kernel has no knowledge of filesystems, it's all IPC for it

left dew Jan 2, 2025, 7:53 PM

#

well, have the vfs server do the talking for you?

#

you = whoever is trying to open/read/write a file

carmine nacelle Jan 2, 2025, 7:53 PM

#

How would that work?

#

It would be a message to VFS then a message to filesystem for everything

left dew Jan 2, 2025, 7:53 PM

#

wdym how? it would just talk to the server responsible for a given file system

#

yeah that's how, i guess

#

managarm has some optimizations for that iirc

#

i think the call that "passthrough lanes"

carmine nacelle Jan 2, 2025, 7:54 PM

#

And I just let processes talk directly to filesystems

left dew Jan 2, 2025, 7:55 PM

#

how does the process know what file system server to talk to

carmine nacelle Jan 2, 2025, 7:55 PM

#

Also, VFS can disappear and make of the opened files would be closed

#

It asks VFS cache

#

It's a bad name

#

Should be something like namespaces server

hexed acorn Jan 2, 2025, 9:51 PM

#

left dew i think the call that "passthrough lanes"

yeah, in managarm the open request gives you an ipc lane you talk to to make file requests like read/write/etc

#

and the ipc on that lane is handled by the fs server responsible for that particular file

carmine nacelle Jan 3, 2025, 3:23 AM

#

I'm kinda thinking about pulling a galaxybrain move and exposing all ACPI devices to devfs (which I don't have yet)

#

Sounds like a nice abstraction abstraction

#

And also maybe adding "namespaces" is a solution to root remounting

#

Like I would have "dev" namespace, which would get mounted to /dev

#

And then various servers will then be able to mount their stuff there

carmine nacelle Jan 3, 2025, 4:05 AM

#

This is so good

#

It can probably just expose AML methods as Unix files

#

Lazily executing it

carmine nacelle Jan 3, 2025, 11:52 AM

#

A bit of a random thought: why not use "banks" of memory for struct Pages on 32 bit platforms? This way I can support "infinite" RAM with PAE

carmine nacelle Jan 3, 2025, 6:54 PM

#

I hate gcc

#

How do I make this go away? nooo

#

what the hell is this? https://gcc.gnu.org/pipermail/gcc-patches/2024-March/647835.html

#

misha@Ryzen-PC:~/gcc-pmos/gcc$ cat ./config/linux-android.opt.urls
; Autogenerated by regenerate-opt-urls.py from gcc/config/linux-android.opt and generated HTML

mandroid
UrlSuffix(gcc/GNU_002fLinux-Options.html#index-mandroid)

tno-android-cc
UrlSuffix(gcc/GNU_002fLinux-Options.html#index-tno-android-cc)

tno-android-ld
UrlSuffix(gcc/GNU_002fLinux-Options.html#index-tno-android-ld)

carmine nacelle Jan 3, 2025, 7:36 PM

#

Maybe I'll just remove pmos.opt

#

fuck it

carmine nacelle Jan 4, 2025, 7:28 PM

#

Anyway, I think I'm gonna disappear for two weeks, until I'm finished with the exams

#

The plan after that is to

Implement AHCI reading
Redo VFS mess (although maybe it's not too bad, but C just makes me feel miserable)
S3 sleep?
Finish implementing signals
Implement USB
Look at GPU stuff

#

(I wanna see if the whole "ACPI and all drivers in userspace" will survive S3)

twilit talon Jan 4, 2025, 7:37 PM

#

thats a good sturdiness test

#

of both kernel & driver design

hoary moat Jan 4, 2025, 7:44 PM

#

twilit talon thats a good sturdiness test

also a good patience test

carmine nacelle Jan 4, 2025, 7:54 PM

#

In theory I just need to save and restore all the state

#

It can't be that hard

#

Just a few extra callbacks

hoary moat Jan 4, 2025, 7:55 PM

#

lmao

#

what about D states?

#

and power resources?

carmine nacelle Jan 4, 2025, 7:55 PM

#

What about them?

hoary moat Jan 4, 2025, 7:55 PM

#

you need them

carmine nacelle Jan 4, 2025, 7:55 PM

#

Idk how they work

hoary moat Jan 4, 2025, 7:56 PM

#

you kinda need them for suspend

hoary moat Jan 4, 2025, 7:59 PM

#

hoary moat you kinda need them for suspend

well you actually only need them for wake

#

so if you have no problem with not being able to wake properly

#

that's fine

hoary moat Jan 4, 2025, 7:59 PM

#

hoary moat so if you have no problem with not being able to wake properly

(if at all)

carmine nacelle Jan 4, 2025, 7:59 PM

#

It would be cool to be support them in general

lunar orchid Jan 7, 2025, 12:44 PM

#

why pmos is removed from active threads

carmine nacelle Jan 7, 2025, 12:44 PM

#

I have exams

lunar orchid Jan 7, 2025, 12:45 PM

#

carmine nacelle I have exams

https://tenor.com/view/yellow-emoji-no-no-emotiguy-no-no-no-gif-gif-9742000569423889376

Tenor

#

osdev more important than exams

untold pollen Jan 7, 2025, 1:01 PM

#

Bump

carmine nacelle Jan 7, 2025, 1:01 PM

#

The build

#

📎 pmOS.iso

#

FDT doesn't do much

#

And AHCI driver is broken for some reason (because of broken fork)

#

My galaxybrain kernel can't save the registers properly so forked processes call exit for some reason immediately after forking

untold pollen Jan 7, 2025, 1:02 PM

#

As long as it loads everything it needs via uefi to do anything I guess we will be fine :P

carmine nacelle Jan 7, 2025, 1:03 PM

#

I've tested uboot half a year ago

#

If it doesn't work now I'll return to it in a week

untold pollen Jan 7, 2025, 1:03 PM

#

Place your bets if it still boots

#

I need to fine two more probes though

#

nvm one more probe

carmine nacelle Jan 7, 2025, 1:06 PM

#

Does this board have ns16550-style controller?

untold pollen Jan 7, 2025, 1:07 PM

#

No

carmine nacelle Jan 7, 2025, 1:08 PM

#

then there will probably be no logs to uart from userspace

#

framebuffer should still work though

untold pollen Jan 7, 2025, 1:18 PM

#

hang on this thing is acting odd

untold pollen Jan 7, 2025, 1:34 PM

#

reee

untold pollen Jan 7, 2025, 1:49 PM

#

Not sure why but this thing isnt working at all

carmine nacelle Jan 7, 2025, 1:49 PM

#

the board?

untold pollen Jan 7, 2025, 1:50 PM

#

yeah

#

will try with og image really quick iguess

carmine nacelle Jan 7, 2025, 1:53 PM

#

Is it not working with pmOS or not working at all?

lunar orchid Jan 7, 2025, 1:57 PM

#

carmine nacelle Is it not working with pmOS or not working at all?

why you are not in my friends

carmine nacelle Jan 7, 2025, 1:57 PM

#

?

lunar orchid Jan 7, 2025, 1:57 PM

#

😡

carmine nacelle Jan 7, 2025, 1:57 PM

#

idk

untold pollen Jan 7, 2025, 2:01 PM

#

hmm seems like at all

#

not sure whats going on there

untold pollen Jan 7, 2025, 2:19 PM

#

I’ll try again later not sure what’s going on

noble mason Jan 9, 2025, 1:49 PM

#

carmine nacelle

for what arch is it?

#

rv or x86?

carmine nacelle Jan 9, 2025, 1:49 PM

#

risc-v

noble mason Jan 9, 2025, 1:49 PM

#

virt machine will work?

carmine nacelle Jan 9, 2025, 1:49 PM

#

yes

#

it's the only thing i've tested it on

noble mason Jan 9, 2025, 1:56 PM

#

windows ❤️

carmine nacelle Jan 9, 2025, 1:56 PM

#

what type of virus did it detect? lol

noble mason Jan 9, 2025, 1:56 PM

#

#

~~pmos is actually trojan do not download~~

#

fine

#

ill try pmos on linux tomorrow

lunar orchid Jan 9, 2025, 2:52 PM

#

noble mason

windows moment

vale hazel Jan 11, 2025, 8:21 PM

#

lunar orchid osdev more important than exams

nahhh let him chill

#

exams are ass

#

also im giving pmos a star

#

:)

#

@carmine nacelle

carmine nacelle Jan 11, 2025, 8:21 PM

#

lunar orchid https://tenor.com/view/yellow-emoji-no-no-emotiguy-no-no-no-gif-gif-974200056942...

Then I get thrown out the university

vale hazel Jan 11, 2025, 8:22 PM

#

carmine nacelle Then I get thrown out the university

what is ur opinion on me staring the project

#

u deserve way more stars

carmine nacelle Jan 11, 2025, 8:22 PM

#

thanks for starring

#

idk, I'm too stressed

vale hazel Jan 11, 2025, 8:22 PM

#

dont force urself to work on pmos

#

work on it WHEN u want to

#

focus on ur exams

#

university is crazy scary i would be shitting my pants as well

carmine nacelle Jan 11, 2025, 8:23 PM

#

I'm not working on it atm

vale hazel Jan 11, 2025, 8:23 PM

#

good

carmine nacelle Jan 16, 2025, 1:06 AM

#

carmine nacelle The plan after that is to 1. Implement AHCI reading 2. Redo VFS mess (although m...

I'm done with exams so I guess it's time to do that

#

(and I think I have a month of holidays)

hoary moat Jan 16, 2025, 1:28 AM

#

letsgo

carmine nacelle Jan 16, 2025, 5:44 PM

#

Ok I guess I'm semi-back
I caught a flu (?)

#

Anyways, I need to think how to implement partition probing and stuff like that

#

And if I want it to be another server

#

Which would also expose itself as a filesystem

#

so the filesystems can then just do open("/dev/sda") or whatever

#

Thonk

#

(because wtf not)

#

And the network subsystem is gonna be in Go

carmine nacelle Jan 17, 2025, 11:35 AM

#

carmine nacelle Ok I guess I'm semi-back I caught a flu (?)

I'm feeling much better today, so maybe I'll do something

carmine nacelle Jan 17, 2025, 8:34 PM

#

Did I just catch the framebuffer bug in QEMU?

#

wtf

#

kernel just decided it doesn't want to work anymore

#

i'm done

#

working with kvm but not with tsc is something new

carmine nacelle Jan 17, 2025, 8:54 PM

#

somehow -bios ovmf-x86/OVMF.fd breaks it...

#

I think my timers are broken...

carmine nacelle Jan 17, 2025, 11:52 PM

#

average C++

#

And fun templates

#

Thonk timers are still broken...

#

sus

#

(I think I don't EOI spurious interrupts)

#

I don't get it

Dispensing the spurious-interrupt vector does not affect the ISR, so the handler for this vector should return without an EOI.

vale hazel Jan 18, 2025, 1:02 AM

#

carmine nacelle <:sus:824242714219642890>

one shot

#

one kill

#

go to sleep

#

please

#

dont be like me

carmine nacelle Jan 18, 2025, 1:02 AM

#

I usually go to sleep at 3 am

scarlet salmon Jan 18, 2025, 2:12 AM

#

vale hazel go to sleep

You should

carmine nacelle Jan 18, 2025, 2:41 AM

#

I can't sleep iwannadie

scarlet salmon Jan 18, 2025, 2:59 AM

#

I was talking more to nyaux, but you should as well

carmine nacelle Jan 18, 2025, 10:54 AM

#

Debugging continues: it looks like I get spurious int as soon as I enter userspace, even if I kill IOAPIC code

#

fuck kvm

twilit talon Jan 18, 2025, 10:57 AM

#

wait whats happening, didnt u have working userspace

carmine nacelle Jan 18, 2025, 11:12 AM

#

My interrupts just casually stop working

#

like userspace works, but I loose timers and external ints

#

I think that's also the reason why it was breaking on some physical PCs

carmine nacelle Jan 18, 2025, 11:14 AM

#

carmine nacelle like userspace works, but I loose timers and external ints

like without timer it can't even poll stuff (which I do periodically for all devices galaxybrain )

#

thonk

#

Wtf is this code

#

(for reference)

#

I'm pointing LAPIC timer to spurious int vector, and wondering why I'm getting that interrupt

#

great

#

Nice 2 years old code

#

What's a better question is why did it only break now

#

I think the whole timer/clock code has to go...

#

I'm too fed up with it

carmine nacelle Jan 18, 2025, 11:42 AM

#

When did lld get this good?

#

This is fun...

carmine nacelle Jan 18, 2025, 12:49 PM

#

(average c++ skill issue)

carmine nacelle Jan 18, 2025, 5:52 PM

#

I guess AHCI is cooking...

carmine nacelle Jan 18, 2025, 6:14 PM

#

(now I just need to think about IPC)

toxic torrent Jan 18, 2025, 6:23 PM

#

carmine nacelle When did lld get this good?

lld 20 is neat

carmine nacelle Jan 18, 2025, 6:30 PM

#

I'm using 19.1.0-rc3

#

but I guess I could rebase onto that

carmine nacelle Jan 19, 2025, 11:13 AM

#

Ok, now that I've slept on it, I think I'll implement sending memory objects

#

And just do zero-copy disk io

#

no read() on block devices

carmine nacelle Jan 19, 2025, 11:47 AM

#

(now that I think of it, it should not even be very difficult to implement)

#

no new syscalls

carmine nacelle Jan 19, 2025, 8:17 PM

#

and somehow, I think the permissions system is building itself

carmine nacelle Jan 20, 2025, 2:57 PM

#

I think I know how to handle multithreaded filesystems

#

(explanation: currently, my ports are tied to threads, which kinda makes some stuff difficult)

#

I've been thinking about breaking that up, but didn't come up with anything I liked

#

But, why not just return sets of ports

#

And let userspace decide what it wants to do with them

#

With some scheduler changes, and considering that every process has a worker thread anyway, it can also get notifications from scheduler, and keep track of which core thread is bound to

#

So in other words, all threads can know which CPU they are on, without making syscalls

#

And with some caching memes, my libc can just decide where to send messages for each request

#

(like how did my stupid IPC just work out?)

carmine nacelle Jan 20, 2025, 6:26 PM

#

Wtf declaring overloaded 'operator()' as 'static' is a C++23 extension [-Wc++23-extensions]

restive junco Jan 20, 2025, 6:33 PM

#

cursed

#

lol

carmine nacelle Jan 20, 2025, 6:35 PM

#

how was it only added in c++23?

toxic torrent Jan 22, 2025, 12:35 AM

#

i was skimming through the code

#

https://gitlab.com/mishakov/pmos/-/blob/main/loader/src/load_elf.c?ref_type=heads#L26

GitLab

loader/src/load_elf.c · main · Mikhail Kovalev / pmOS · GitLab

GitLab.com

#

isn't this magic reversed?

#

ah nvm the 7f is at the start

carmine nacelle Jan 22, 2025, 12:44 AM

#

this is dead code

#

from when I was using multiboot2 to boot bruh

toxic torrent Jan 22, 2025, 12:45 AM

#

chad

carmine nacelle Jan 22, 2025, 12:46 AM

#

this would get loaded by multiboot, load kernel, and continue in ring 3 KEKW

toxic torrent Jan 22, 2025, 12:46 AM

#

bruh

carmine nacelle Jan 22, 2025, 12:49 AM

#

This is what I do now https://gitlab.com/mishakov/pmos/-/blob/main/kernel/generic/processes/tasks.cc?ref_type=heads#L298

#

(now in kernel)

carmine nacelle Jan 23, 2025, 9:14 PM

#

nice c++

left dew Jan 23, 2025, 9:14 PM

#

i feel like this line of code will be a source of at least one bug

#

is begin an iterator?

#

that doesn't make sense nvm

carmine nacelle Jan 23, 2025, 9:20 PM

#

Yeah

#

left dew Jan 23, 2025, 9:20 PM

#

is it some sort of intrusive data structure?

carmine nacelle Jan 23, 2025, 9:20 PM

#

it is

#

my own impl

left dew Jan 23, 2025, 9:20 PM

#

makes perfect sense

#

otherwise that delete &*it would've been very weird

carmine nacelle Jan 23, 2025, 9:23 PM

#

I've just changed std::list-like things of smart pointers (because why not) to intrusive list, removing 2 memory allocations...

#

anyway, I think I'll finally have AHCI disk reading after I implement sending memory objects

#

(memory object is my abstraction for in-kernel buffer of pages)

carmine nacelle Jan 23, 2025, 10:49 PM

#

cool, apparently I don't have syscall for creating memory objects...

eager rose Jan 23, 2025, 10:53 PM

#

do I understand it correctly as that you're doing the microkernel equivalent of memfd_create + sendmsg(SCM_RIGHTS)?

#

also, are your userspace "processes" (or "tasks" or whatever) isolated from each other in memory?

carmine nacelle Jan 23, 2025, 10:55 PM

#

eager rose do I understand it correctly as that you're doing the microkernel equivalent of ...

Basically yes

carmine nacelle Jan 23, 2025, 10:56 PM

#

eager rose also, are your userspace "processes" (or "tasks" or whatever) isolated from each...

Threads are not, processes are

#

"tasks" can share or not page tables

#

so I'm creating a memory buffer, DMAing to it, and sending it to another process

#

which can then move pages to other files

#

and there are no memory copies

eager rose Jan 23, 2025, 10:58 PM

#

iirc Linux has a refcounted struct mm_struct, and clone() and whatever other syscalls there are can "unshare" it by setting some flag (CLONE_MM or something). I guess that would be similar in semantics to what you are doing (except afaik Linux only allows "unsharing" the mm_struct at clone or fork (done implicitly) time if some constraints apply)

carmine nacelle Jan 23, 2025, 10:59 PM

#

I have done very little linux userspace development so idk

carmine nacelle Jan 23, 2025, 10:59 PM

#

eager rose iirc Linux has a refcounted `struct mm_struct`, and `clone()` and whatever other...

I guess

eager rose Jan 23, 2025, 11:09 PM

#

carmine nacelle I have done very little linux userspace development so idk

well linux userspace usually doesn't touch clone() that much itself. and basically never does weird CLONE_MM shit unless it is doing something unusual

#

ok actually it is CLONE_VM

carmine nacelle Jan 23, 2025, 11:10 PM

#

I'm also trying to implement posix api on top of what I have I guess

eager rose Jan 23, 2025, 11:15 PM

#

afaik linux clone3 isn't posix

#

generally linux isn't posix

#

... and that is a good thing

carmine nacelle Jan 23, 2025, 11:15 PM

#

carmine nacelle I'm also trying to implement posix api on top of what I have I guess

so userspace doesn't have to touch my api if it doesn't want to

toxic torrent Jan 24, 2025, 9:41 AM

#

question, wouldn't an infinite loop in one of the servers cause one cpu to be at 100% at all times?

#

or is that intended

carmine nacelle Jan 24, 2025, 10:16 AM

#

My IPC blocks on reads from empty queues

carmine nacelle Jan 24, 2025, 10:16 AM

#

toxic torrent question, wouldn't an infinite loop in one of the servers cause one cpu to be at...

If it's a bug, it would

toxic torrent Jan 24, 2025, 10:17 AM

#

i'm looking at processd atm

carmine nacelle Jan 24, 2025, 10:19 AM

#

It does get_message or whatever (I don't remember), and it should block

toxic torrent Jan 24, 2025, 10:38 AM

#

@carmine nacelle https://gitlab.com/mishakov/pmos/-/blob/main/devicesd/generic/pci/pci.c?ref_type=heads#L324

GitLab

devicesd/generic/pci/pci.c · main · Mikhail Kovalev / pmOS · GitLab

GitLab.com

carmine nacelle Jan 24, 2025, 10:42 AM

#

what's up with it?

restive junco Jan 24, 2025, 11:13 AM

#

it's missing a "n"

carmine nacelle Jan 24, 2025, 11:13 AM

#

bruh

toxic torrent Jan 24, 2025, 12:05 PM

#

carmine nacelle what's up with it?

typo

hoary moat Jan 24, 2025, 5:07 PM

#

@carmine nacelle https://github.com/ThatMishakov/pmOS/blob/main/devicesd/generic/acpi/acpi.c#L257
you should be disabling interrupts here

GitHub

pmOS/devicesd/generic/acpi/acpi.c at main · ThatMishakov/pmOS

Small operating system for x86 and RISC-V. Contribute to ThatMishakov/pmOS development by creating an account on GitHub.

carmine nacelle Jan 24, 2025, 7:02 PM

#

How do I disable them?

hoary moat Jan 24, 2025, 7:40 PM

#

cli

twilit talon Jan 24, 2025, 7:41 PM

#

hoary moat cli

This is userspace

hoary moat Jan 24, 2025, 7:41 PM

#

Set lapic's priority level register to 0xf0

#

Also

#

#

So idk

carmine nacelle Jan 25, 2025, 12:47 PM

#

Ok, I think I'll finally implement sending memory objects

#

I need to decide how to do that...

carmine nacelle Jan 25, 2025, 3:04 PM

#

Did I just discover why it would break sometimes?

carmine nacelle Jan 25, 2025, 3:31 PM

#

I think it's trolling me

#

(wtf)

#

it has to be something dumb

#

Variable just magically becomes 3...

carmine nacelle Jan 25, 2025, 4:22 PM

#

bruh c++ moment

carmine nacelle Jan 25, 2025, 4:39 PM

#

ok, surprisingly sending memory objects was very easy (ignoring the debugging/discovery of unrelated issue)

carmine nacelle Jan 25, 2025, 5:33 PM

#

Time to figure out how to build a disk image with make

carmine nacelle Jan 25, 2025, 6:23 PM

#

Bruh did AHCI driver just work?

#

(238 is 0xEE, which is GPT)

carmine nacelle Jan 25, 2025, 9:25 PM

#

bruh

#

the first entry looks weird...

carmine nacelle Jan 25, 2025, 10:21 PM

#

(I'm gonna stop polluting uACPI thread)

#

This looks credible...

#

I think it's working, now I just need to figure out how to mount filesystems

#

Like how do I do fstab or whatever

#

Cool, the dev branch is juicy...

#

What if I just merge it?

carmine nacelle Jan 25, 2025, 10:40 PM

#

carmine nacelle Like how do I do fstab or whatever

(I think I urgently need execlp...)

#

and to fix fork on ia32 and risc-v

#

which means that I need to fix vfs (even though it isn't really broken...)

#

(by which I mean that I need to write a good hash map header)

#

and "improve the code quality"

#

but I want to do USB...

#

Maybe pmOS will finally have shell in 2026...

carmine nacelle Jan 25, 2025, 10:43 PM

#

carmine nacelle Like how do I do fstab or whatever

I think it's very reasonable to just put that in blockd

#

which will talk to devicesd

#

which will talk to vfsd

#

which will talk to bootstrapd

#

which will talk to filesystems drivers

#

but maybe I want fat32 driver be written in Rust...

torpid vector Jan 26, 2025, 11:16 AM

#

o/

pearl narwhal Jan 26, 2025, 11:22 AM

#

are your filesystem drivers separate servers or are they libraries attached to vfsd or blockd or whatever

carmine nacelle Jan 26, 2025, 11:24 AM

#

I haven't decided

#

They will probably be separate servers

carmine nacelle Jan 26, 2025, 6:28 PM

#

error: unknown type name '__uint128_t' bruh

hoary moat Jan 26, 2025, 7:49 PM

#

ah yes that doesn't exist on i686

#

or the m68k

#

kinda sad, I just coped by using uint64_t instead

#

in the one place I used 128-bit integers

carmine nacelle Jan 26, 2025, 7:51 PM

#

I'm using it to calculate timer frequencies

#

(stolen from managram)

eager rose Jan 26, 2025, 10:27 PM

#

iirc I do something cursed by measuring the frequency, then first multiplying by like a million or something, then dividing by the frequency, followed by multiplying by 1000 again

#

though idr if this was for TSC or HPET

carmine nacelle Jan 27, 2025, 12:49 AM

#

ld.lld: error: undefined symbol: __udivdi3 nooo

toxic torrent Jan 27, 2025, 12:49 AM

#

nooo

carmine nacelle Jan 27, 2025, 1:33 AM

#

I think I'll finally have to add memory region stuff to my PMM

toxic torrent Jan 27, 2025, 1:42 AM

#

please don't remind me

#

i need to do the same

carmine nacelle Jan 27, 2025, 1:43 AM

#

Since ia32 wants cr3 in < 4GB

#

It should be doable...

#

I just don't know how to deal with it

#

Like do I always allocate from > 4GB and then go to < 4GB if that fails?

toxic torrent Jan 27, 2025, 1:44 AM

#

hm

#

wait, does that mean on x86 you cannot physically have more than 1048576 processes ever?

carmine nacelle Jan 27, 2025, 1:46 AM

#

?

carmine nacelle Jan 27, 2025, 1:46 AM

#

carmine nacelle Since ia32 wants cr3 in < 4GB

in ia32

carmine nacelle Jan 27, 2025, 1:47 AM

#

toxic torrent wait, does that mean on x86 you cannot physically have more than 1048576 process...

you can probably recycle page tables..

toxic torrent Jan 27, 2025, 1:48 AM

#

carmine nacelle in ia32

i might have my terminology mixed up

#

isn't ia32 = x86?

carmine nacelle Jan 27, 2025, 1:48 AM

#

32 bit x86

toxic torrent Jan 27, 2025, 1:48 AM

#

yea

#

to be fair you'd surely run out of ram before reaching 1M processes

carmine nacelle Jan 27, 2025, 1:48 AM

#

Idk

#

Yeah

#

I'll probably limit kernel to 16GB of RAM or something anyway

#

because of struct page

carmine nacelle Jan 27, 2025, 4:43 PM

#

Cool, the paging is almost working

#

But it craps itself while doing printf...

carmine nacelle Jan 27, 2025, 5:06 PM

#

This looks reasonable...

#

Silly issue fixed...

carmine nacelle Jan 27, 2025, 5:31 PM

#

toxic torrent to be fair you'd surely run out of ram before reaching 1M processes

PDPT is 32 bytes, and not required to be page aligned...

toxic torrent Jan 27, 2025, 5:33 PM

#

carmine nacelle PDPT is 32 bytes, and not required to be page aligned...

what

#

how is this 32 bytes

carmine nacelle Jan 27, 2025, 5:33 PM

#

4 entries of 8 bytes?

toxic torrent Jan 27, 2025, 5:35 PM

#

ah i didn't read the middle block

carmine nacelle Jan 27, 2025, 5:39 PM

#

I'm too dumb for this

Software can transition between 32-bit paging and PAE paging by changing the value of CR4.PAE with MOV to CR4.

#

Wouldn't it pagefault?

#

If PAE paging would be in use following an execution of MOV to CR0 or MOV to CR4 (see Section 4.1.1) and the
instruction is modifying any of CR0.CD, CR0.NW, CR0.PG, CR4.PAE, CR4.PGE, CR4.PSE, or CR4.SMEP; then the
PDPTEs are loaded from the address in CR3.

#

AMD manual says this

Implicit Invalidations. The following operations cause the entire TLB to be invalidated, including
global pages:
• Modifying the CR0.PG bit (paging enable).
• Modifying the CR4.PAE bit (physical-address extensions), the CR4.PSE bit (page-size
extensions), or the CR4.PGE bit (page-global enable).

carmine nacelle Jan 29, 2025, 2:48 PM

#

(the porting to ia32 continues...)

#

error: undefined symbol: __udivdi3 nooo

carmine nacelle Jan 29, 2025, 3:10 PM

#

Cool, it looks like libclang_rt.builtins-i386 is sane and doesn't emit xmm for that

carmine nacelle Jan 29, 2025, 10:10 PM

#

Thonk

#

c0102e9c: c7 44 24 4c 00 00 00 00          movl    $0x0, 0x4c(%esp)
;         long_string_str long_string = {0, 0};
c0102ea4: 8d 74 24 50                      leal    0x50(%esp), %esi
c0102ea8: c7 44 24 50 00 00 00 00          movl    $0x0, 0x50(%esp)
c0102eb0: c7 44 24 54 00 00 00 00          movl    $0x0, 0x54(%esp)
; void *operator new[](size_t size) { return malloc(size); }
c0102eb8: b9 0a 00 00 00                   movl    $0xa, %ecx
c0102ebd: e8 7e 85 00 00                   calll    0xc010b440 <malloc>
;             long_string.ptr  = new char[length + 1];
c0102ec2: 89 44 24 54                      movl    %eax, 0x54(%esp)
;     serial_logger.printf("%s %i\n", bootstrap.c_str(), bootstrap.length());
c0102ec6: 6a 00                            pushl    $0x0
c0102ec8: 56                               pushl    %esi
c0102ec9: 68 1c e3 11 c0                   pushl    $0xc011e31c             # imm = 0xC011E31C
c0102ece: 68 44 17 12 c0                   pushl    $0xc0121744             # imm = 0xC0121744
c0102ed3: e8 c8 82 00 00                   calll    0xc010b1a0 <_ZN6Logger6printfEPKcz>

#

I'm smelling UB

#

Or stupid code

carmine nacelle Jan 29, 2025, 11:21 PM

#

(for some reason, the kernel is having a skill issue with printf...)

eager rose Jan 29, 2025, 11:26 PM

#

wild guess: since afaik you're porting to 32-bit, it might be having skill issues with int64_t and uint64_t values. assuming you don't mean something like page faults occuring in vsnprintf.

carmine nacelle Jan 29, 2025, 11:28 PM

#

I've had issues with that in allocators because of __builtin_clzl/__builtin_clzll

#

But it doesn't look like it

#

It loads 0 from vtable of a global class

#

for some reason...

#

and calls a null pointer

#

Is it the same dumb problem that I've had about a month ago?

#

#

I think it's stack overflow bruh

hoary moat Jan 29, 2025, 11:51 PM

#

bruh

carmine nacelle Jan 29, 2025, 11:53 PM

#

bruh

#

Fun, it's almost working...

#

Now I just need to write all the boring assembly to handle interrupts...

#

(and I also want to support PAE next)

#

(which implies adding zoned to my pmm)

#

But after thinking about it, it should be quite easy with my design

#

And I also need to figure out SMP bringup...

carmine nacelle Jan 30, 2025, 12:03 AM

#

carmine nacelle (and I also want to support PAE next)

Because I want it to support 16GB of RAM with 32 bits

toxic torrent Jan 30, 2025, 12:07 AM

#

carmine nacelle <:bruh:761542241381908491>

that's enough

carmine nacelle Jan 30, 2025, 12:14 AM

#

I'm allocating 8KB arrays on stack

#

For TLB shootdowns

left dew Jan 30, 2025, 12:41 AM

#

that 4KiB stack must be having lots of fun

carmine nacelle Jan 30, 2025, 3:52 PM

#

I kinda have userspace (now I just need PF handler and memes meme )

carmine nacelle Jan 30, 2025, 4:51 PM

#

thonk apparently hardware task switching is not that bad...

left dew Jan 30, 2025, 4:53 PM

#

printing registers in decimal is diabolical

toxic torrent Jan 30, 2025, 5:02 PM

#

chad

little smelt Jan 30, 2025, 5:03 PM

#

left dew printing registers in decimal is diabolical

completetly diabolical

carmine nacelle Jan 30, 2025, 5:12 PM

#

Anyway, maybe I will have it fully working by the end of today?

#

(the same thing I said on Sunday)

carmine nacelle Jan 30, 2025, 6:17 PM

#

New least overcomplicated save_registers function trl ```x86asm
.type save_registers, @function
save_registers:
pushl %eax
movw $0x1b, %ax
cmpw 8(%esp), %ax
jne 1f

movw $0x38, %ax
movl %ax, %gs
movl %gs:16, %eax
popl (%eax)
movl %ebx, 4(%eax)
movl %ecx, 8(%eax)
movl %edx, 12(%eax)
movl %esi, 16(%eax)
movl %edi, 20(%eax)
movl %ebp, 24(%eax)

movl 8(%esp), %ecx # %eip
movl 16(%esp), %edx # eflags
movl %ecx, 32(%eax)
movl %edx, 36(%eax)
movl 20(%esp), %ecx # esp
movl 4(%esp), %edx # error code
movl %ecx, 28(%esp)
xorl %ecx, %ecx
movl %ecx, 52(%eax)
xorl %eax, %eax
ret

1:
movw $0x70, %ax
cmpw 8(%esp), %ax
jne 1f

movl %gs:16, %eax
popl (%eax)
movl %ebx, 4(%eax)
movl %ecx, 8(%eax)
movl %edx, 12(%eax)
movl %esi, 16(%eax)
movl %edi, 20(%eax)
movl %ebp, 24(%eax)

popl %ecx
popl %edx
addl $4, %esp
popl 32(%eax)
addl $4, %esp
popl 36(%eax)
movl %esp, 28(%eax)
movl %gs:4, %esp
pushl %ecx
movl $4, %ecx
movl %ecx, 52(%eax)
xorl %eax, %eax
ret

1:
xchgl %ecx, 4(%esp)
xchgl %ecx, (%esp)
pushl %edx
pushl %ebx
pushl %ebp
pushl %esi
pushl %edi

movl %esp, %eax
movl 28(%esp), %edx
jmp *%ecx

eager rose Jan 30, 2025, 10:16 PM

#

honestly I should just ask some osdevers in here to hand-"optimize" some assembly routines... then have it as a CTF reversing challenge

left dew Jan 30, 2025, 10:52 PM

#

wtf is j *%ecx

hexed acorn Jan 30, 2025, 10:59 PM

#

ig that was intended to be jmp?

left dew Jan 30, 2025, 11:14 PM

#

definitely

#

i'm just saying "wtf is this" meme

#

because it's definitely not a valid mnemonic

carmine nacelle Jan 30, 2025, 11:40 PM

#

hexed acorn ig that was intended to be jmp?

yeah it was supposed to be jmp

#

It kinda worked after an hour of debugging meme

#

Now SSE gives me #UD instead of #NM for some reason sadmeme

#

even though cr0 = 0xe000001b

#

(which is CR0.MP and CR0.TS set and CR0.EM cleared)

carmine nacelle Jan 30, 2025, 11:43 PM

#

carmine nacelle yeah it was supposed to be jmp

since it's j on risc-v

carmine nacelle Jan 30, 2025, 11:54 PM

#

carmine nacelle Now SSE gives me #UD instead of #NM for some reason <:sadmeme:659848285510565907...

Nvm, it does

#

QEMU log was just showing 6 instead of 7 for some reason...

carmine nacelle Jan 31, 2025, 12:30 AM

#

Cool, now it works in bochs but breaks in qemu...

#

And in bochs it's returning ECONNREFUSED

#

because why wouldn't it I guess

carmine nacelle Jan 31, 2025, 12:56 AM

#

Bruh clang's -O3 -flto assembly is weird...

#

Cool, it's trying to load userspace stuff (the errors are expected since it's 64 bit binaries because I can't be bothered to fix the build scripts...)

#

And it #GPs in SSE instruction in QEMU for no reason

#

(I think)

#

(I don't have GP handler...)

carmine nacelle Jan 31, 2025, 1:10 AM

#

carmine nacelle And it #GPs in SSE instruction in QEMU for no reason

(I'm blaming getCR4() function)

#

Yep

c0105766: 0f 20 e0                         movl    %cr4, %eax
;     cr4 |= (0x01 << 10); // CR4.OSXMMEXCPT (enable #XF exception)
c0105769: 89 c1                            movl    %eax, %ecx
c010576b: 81 c9 00 06 00 00                orl    $0x600, %ecx            # imm = 0x600
; inline void setCR4(unsigned long val) { asm("movl %0, %%cr4" : : "r"(val)); }
c0105771: 0f 22 e1                         movl    %ecx, %cr4
;     if (xsave_supported) {
c0105774: 80 3d f8 4e 13 c0 01             cmpb    $0x1, -0x3fecb108
c010577b: 75 14                            jne    0xc0105791 <kmain+0x3611>
;         setCR4(getCR4() | CR4_OSXSAVE);
c010577d: 0d 00 00 04 00                   orl    $0x40000, %eax          # imm = 0x40000
; inline void setCR4(unsigned long val) { asm("movl %0, %%cr4" : : "r"(val)); }
c0105782: 0f 22 e0                         movl    %eax, %cr4

#

void enable_sse()
{
    auto cr0 = getCR0();
    cr0 &= ~(0x01UL << 2); // CR0.EM
    cr0 |= (0x01 << 1);    // CR0.MP
    cr0 |= (0x01 << 3);    // CR0.TS (Lazy task switching)
    setCR0(cr0);

    auto cr4 = getCR4();
    cr4 |= (0x01 << 9);  // CR4.OSFXSR (enable FXSAVE/FXRSTOR)
    cr4 |= (0x01 << 10); // CR4.OSXMMEXCPT (enable #XF exception)
    setCR4(cr4);

    if (xsave_supported) {
        setCR4(getCR4() | CR4_OSXSAVE);
        set_xcr(0, xcr0);
    }
}

#

(that was it, qemu also works now)

carmine nacelle Jan 31, 2025, 3:24 PM

#

Userspace kinda works more, but framebuffer driver is crashing...

twilit talon Jan 31, 2025, 3:26 PM

#

Letsgo

carmine nacelle Jan 31, 2025, 3:29 PM

#

(something's passing nullptr to flanterm_write 💀)

#

(if I can read the assembly)

#

Or not

#

which makes no sense

#

I think my registers saving/restoring is broken

carmine nacelle Jan 31, 2025, 3:51 PM

#

I can't get bochs to boot nooo

#

Could hyper be pagefaulting?

#

#

It boots fine in QEMU

#

@twilit talon ?

twilit talon Jan 31, 2025, 4:57 PM

#

Doubt

#

It worked last time

carmine nacelle Jan 31, 2025, 9:28 PM

#

This is pagefaulting with kernel address..?

#

Nope

#

I'm dumb

#

uint64_t's are haunting me

carmine nacelle Jan 31, 2025, 9:35 PM

#

carmine nacelle This is pagefaulting with kernel address..?

It's crashing in strlen...

carmine nacelle Jan 31, 2025, 10:44 PM

#

trl

#

a.result is int64_t, aka long long

#

(uint64_ts are really haunting me)

carmine nacelle Jan 31, 2025, 11:00 PM

#

Thonk

#

(I think I know what's killing the framebuffer/hanging the system)

#

it's i8042

#

(I don't have timers...)

twilit talon Jan 31, 2025, 11:08 PM

#

why are u doing bochs btw

carmine nacelle Jan 31, 2025, 11:09 PM

#

?

#

it's qemu

#

idk bochs debugger is convenient to me

#

when there are a lot of context switches and gdb can't keep up

carmine nacelle Jan 31, 2025, 11:10 PM

#

carmine nacelle <:Thonk:1271231807945441402>

It just freezes on inb

#

qemu that is

twilit talon Jan 31, 2025, 11:16 PM

#

ah

carmine nacelle Jan 31, 2025, 11:20 PM

#

carmine nacelle It just freezes on inb

bruh i'm dumb it was tripple faulting

carmine nacelle Feb 1, 2025, 12:15 AM

#

More stuff is working...

#

(I really need to fix timers)

carmine nacelle Feb 1, 2025, 12:16 AM

#

carmine nacelle More stuff is working...

and this pagefault is very concerning

#

(it looks like something has destroyed the stack after the syscall)

hoary moat Feb 1, 2025, 12:30 AM

#

trl

carmine nacelle Feb 1, 2025, 1:06 AM

#

carmine nacelle (I really need to fix timers)

I've asked ChatGPT and it did it for me lol

#

And it even seems to work

#

bruh

#

Anyway, the only things that's broken is ACPI (should be trivial to reimplement) and that weird pagefault

#

(and maybe fork)

#

(since it was broken for no reason with 64 bit kernel)

#

and then I want to implement zones in pmm

#

(which will be also beneficial for 64 bit memes)

#

and PAE because I like to suffer

#

(though I don't expect to)

#

oh, and SMP

#

I need to investigate how to load the trampoline

#

I will probably add an under 1MB zone to PMM

#

Which I'll also eventually use for ISA DMA trl

#

And try to make it relocatable

carmine nacelle Feb 1, 2025, 10:07 AM

#

Fun, I'm not getting lapic interrupts for some reason...

#

(definitely a qemu bug)

#

Thonk

#

I didn't know IF was respected by ring 3

#

Now we're talking...

carmine nacelle Feb 1, 2025, 11:58 AM

#

bruh it's pagefaulting in movl %gs:0x0, %edi

#

somehow everything is working

#

(except interrupts trl )

#

but that's the kernel skill issue

carmine nacelle Feb 1, 2025, 1:20 PM

#

The kernel is trolling me

#

It #GPs on keybord interrupt then pagefaults on ACPI interrupt

#

thonk

#

And it pagefaults inside malloc nooo

#

thonk that pagefault might be normal though

#

I wonder if that #GP is unrelated to interrupts thonk

#

It's sus that it's in framebuffer's driver

carmine nacelle Feb 1, 2025, 3:46 PM

#

carmine nacelle The kernel is trolling me

through the magic of printf, I've deduced that that pagefault was indeed normal, but now I get gpf bruh

carmine nacelle Feb 1, 2025, 4:47 PM

#

This is kinda weird... thonk

#

(Or maybe not, if I have a memory leak, for example...)

#

Or an interrupt storm

#

Now all userspace just pagefaults when I press the power button

eager rose Feb 1, 2025, 6:59 PM

#

carmine nacelle (Or maybe not, if I have a memory leak, for example...)

why would it page fault and not just panic(OOM)?

carmine nacelle Feb 1, 2025, 7:03 PM

#

Idk

#

Maybe it can't allocate memory

#

(I'm in a long trip and have drained my laptop's battery so I can't debug untill I get home nooo )

hoary moat Feb 1, 2025, 7:05 PM

#

use your brain to transfer energy from your brain to your laptop

#

to charge it

carmine nacelle Feb 1, 2025, 7:07 PM

#

I have a charger but it's 33W so it charges by 1% every 5 minutes or so

#

(maybe faster, but still very slow)

carmine nacelle Feb 1, 2025, 8:42 PM

#

Bruh windows's trolling me

#

(cpu stuck at 530 MHz)

left dew Feb 1, 2025, 9:20 PM

#

i mean it's not utilized so of course it's clocked lower than usual

#

that's how laptops work

carmine nacelle Feb 1, 2025, 9:25 PM

#

No

#

It's not underutilized

left dew Feb 1, 2025, 9:40 PM

#

oh it looks like it is

#

16% lol

#

mine does that too if i don't do anything, but that's probably becuase i dont have some ultra performance thing enabled or whatever

carmine nacelle Feb 1, 2025, 9:44 PM

#

It idles at 1 GHz usually

#

But it gets stuck at some very low frequency

carmine nacelle Feb 1, 2025, 9:47 PM

#

carmine nacelle This is kinda weird... <:thonk:575445129863888896>

Anyway, I was calling call instead of jmp in interrupt handlers, which was shifting stack, and passing stack address instead of the interrupt number, which was confusing the kernel and causing interrupt storm

#

though the PS/2 controller still spams interrupts for some reason

#

(broken timer infrastructure?)

#

(I poll PS/2 devices every few seconds to check that they were not disconnected or died)

#

And uACPI has successfully shut down QEMU!

#

Yep, timers are broken, but spamming is not because of that

#

(and that's also why uACPI was bugged (probably))

#

(or maybe it's actually not)

#

Thonk I think it's PS/2 mouse...

#

Yep, timers just like to die from time to time...

carmine nacelle Feb 1, 2025, 10:21 PM

#

I think it's time to move IOAPIC to the kernel...

#

Also I have a fucked thought

#

What if I make the kernel preemptive by timers

#

But just not reschedule inside them

#

Though I don't know how to notify the lower priority levels

#

Also, I think I know why the RISC-V port is calling exit(0) on fork meme

carmine nacelle Feb 1, 2025, 11:24 PM

#

I'm having skill issue passing arguments from asm to C

left dew Feb 1, 2025, 11:30 PM

#

push arguments in reverse order, don't forget to pop them at the end

#

also something something stack alignment

#

32 bit is weird

eager rose Feb 1, 2025, 11:37 PM

#

carmine nacelle I think it's time to move IOAPIC to the kernel...

how in the world have you even managed to do IOAPIC in userspace??

carmine nacelle Feb 1, 2025, 11:42 PM

#

As usual?

carmine nacelle Feb 1, 2025, 11:43 PM

#

left dew push arguments in reverse order, don't forget to pop them at the end

It was taking 2 uint64_ts and I've been pushing 2 longs instead + I've had some other skill issues with using addl instead of subl for alignment

carmine nacelle Feb 1, 2025, 11:44 PM

#

eager rose how in the world have you even managed to do IOAPIC in userspace??

Just mapped them into the devices server

#

(it also does interrupts allocation at the moment)

#

(and had hpet driver)

#

though I just use lapic now

eager rose Feb 1, 2025, 11:44 PM

#

goddamn

#

cursed

carmine nacelle Feb 1, 2025, 11:45 PM

#

why

eager rose Feb 1, 2025, 11:46 PM

#

idk. it's just an unusual way of doing things

#

your microkernel is like a scheduler, ipc, and a trimmed-down memory manager

carmine nacelle Feb 1, 2025, 11:48 PM

#

it's a fat memory manager

carmine nacelle Feb 1, 2025, 11:49 PM

#

eager rose your microkernel is like a scheduler, ipc, and a trimmed-down memory manager

but that's like the whole point?

eager rose Feb 1, 2025, 11:49 PM

#

yeah

#

but like

#

allocating interrupt lines in userspace, sure. I can buy that.

but actually writing to APIC registers and stuff, that's just... a bit far, maybe

carmine nacelle Feb 1, 2025, 11:58 PM

#

LAPIC is in kernel

#

though it kinda creates problems

#

since I don't know what to do with interrupts masking now

#

I guess kernel could ask userspace, but that seems weird

#

What if all userspace dies and kernel gets stormed by interrupts

#

(also, I've put plic into kernel with risc-v memes)

carmine nacelle Feb 2, 2025, 12:20 AM

#

bruh I like how f'ed it's getting

#

kernel is becoming nonpreemptive with a little * that it actually is

#

(I just need to change amd64 and risc-v ports nooo )

toxic torrent Feb 2, 2025, 12:24 AM

#

lol

carmine nacelle Feb 2, 2025, 12:25 AM

#

carmine nacelle kernel is becoming nonpreemptive with a little * that it actually is

eventually

#

anyway, that was how I wanted to do copy-to-user anyway

#

(by allowing the kernel to pagefault in itself)

carmine nacelle Feb 2, 2025, 12:26 AM

#

carmine nacelle (by allowing the kernel to pagefault in itself)

it's working in i686 port now (except the actual copy-to-user impl)

carmine nacelle Feb 2, 2025, 12:26 AM

#

carmine nacelle This is kinda weird... <:thonk:575445129863888896>

I have a feeling that fork's CoW is returning garbage instead of copying pages for some reason

carmine nacelle Feb 2, 2025, 12:28 AM

#

carmine nacelle it's working in i686 port now (except the actual copy-to-user impl)

But i've fixed a long-standing issue (caused by myself, but that's besides the point) where the newly allocated top level pages were not shared across the processes

#

Now I just let the kernel only map the new top level ptes in the global/idle page table, and copy it to the other ones, on demand, in pagefaults anywhere inside the kernel, if they were not copied during the task creation

#

also I don't like how flanterm has stopped #GPing

#

I guess I'll have to wait for another few months for it to surface again

carmine nacelle Feb 2, 2025, 12:39 AM

#

carmine nacelle I have a feeling that fork's CoW is returning garbage instead of copying pages f...

The kernel is being casually weird

#

It's deducing that the page is only mapped once after fork

#

which sounds fishy...

#

Yep the kernel is being weird

#

or my printf is broken

#

somehow the pages teleport from one page table to the other one

#

bruh

#

magic

#

I think I'm going to sleep risc

#

(I slept for 5 hours today, and I think I'm too energetic (considering that I've taken 2 (or maybe 3) cups of coffee, when I usually don't drink any))

#

(what the fuck is this code)

#

(32 bit addr mask is ~0xfff)

#

cool, kernel decided to pagefault, why not

carmine nacelle Feb 2, 2025, 12:56 AM

#

carmine nacelle cool, kernel decided to pagefault, why not

(while taking a spinlock)

hoary moat Feb 2, 2025, 12:57 AM

#

oh yeah that was me

carmine nacelle Feb 2, 2025, 12:57 AM

#

ok I'm hoping the oberrow curse goes away in the morning

#

my paging code was the thing that was never crashing

#

but uint64_ts are haunting me now

hoary moat Feb 2, 2025, 1:00 AM

#

carmine nacelle ok I'm hoping the oberrow curse goes away in the morning

oberrow curse never goes away

#

sorry you have to find out this way

carmine nacelle Feb 2, 2025, 1:02 AM

#

cleansing negative energy from pmOS kernel

hoary moat Feb 2, 2025, 1:02 AM

#

I've heard if you smoke crack before starting osdev, bugs go away (this is a joke, don't smoke crack)

carmine nacelle Feb 2, 2025, 1:05 AM

#

I'll just change the vscode theme to light so they burn

carmine nacelle Feb 2, 2025, 1:07 AM

#

carmine nacelle cool, kernel decided to pagefault, why not

(I think it's just stack overflowing)

#

(I really need stack guard pages)

#

(especially since I have a good vmm which could actually trivially let me do that)

carmine nacelle Feb 2, 2025, 1:09 AM

#

carmine nacelle cool, kernel decided to pagefault, why not

This printf stuff is sus

#

Why is is copying all of them if it was only told to copy 1 page? thonk

#

(there is no such thing as too many printfs)

#

more wtf code

#

Cool it seems to be working

#

(the fork that is)

#

Now it just fails to map the memory object when reading the AHCI disk (probably uint64_t's striking again)

carmine nacelle Feb 2, 2025, 1:25 AM

#

carmine nacelle Now it just fails to map the memory object when reading the AHCI disk (probably ...

No, it's size_t instead of uint64_t

#

iwannadie

#

Cool, I guess AHCI also works now

#

Also, somehow the i686 port feels very fast

#

(I think it's because of the temp mappings)

#

I wonder how it would perform if I redo the read from user function...

#

bruh

carmine nacelle Feb 2, 2025, 10:13 AM

#

thonk I wanted to do ifunc memes, but I still need to figure out ELF relocations...

#

So I guess I'll do the fast copy from user memes

carmine nacelle Feb 2, 2025, 7:12 PM

#

I'm having weird issue with clang

#

Which pops arguments from stack

#

(it feels like clang is using stdcall instead of cdecl for some reason, or I'm stupid and can't write assembly)

#

anyway, I've changed addl $24, %esp to leal -16(%ebp), %esp since I don't trust clang (aka I don't know my tooling) and it seems to be working...

#

Though I am getting a suspicious ```
/home/misha/pmos/kernel/arch/i686/paging/user_access.cc:5:28: warning: 'user_access_page_fault' has C-linkage specified, but returns incomplete type 'ReturnStr<bool>' which could be incompatible with C [-Wreturn-type-c-linkage]
5 | extern "C" ReturnStr<bool> user_access_page_fault(unsigned access, const char *faulting_addr,
| ^
1 warning generated.

carmine nacelle Feb 2, 2025, 8:00 PM

#

and it's tripple faulting with kvm nooo

carmine nacelle Feb 2, 2025, 11:07 PM

#

carmine nacelle New least overcomplicated save_registers function <:trl:829316641191886888> ```x...

I think this is working too well, so I'll do the same on x86_64 trl

vale hazel Feb 3, 2025, 9:52 AM

#

pmos playing portal 2 when

carmine nacelle Feb 3, 2025, 10:26 AM

#

When I get a disk driver and a shell

#

(although well, AHCI kinda works (it pagefaults on phys hardware))

carmine nacelle Feb 3, 2025, 12:42 PM

#

carmine nacelle I think this is working too well, so I'll do the same on x86_64 <:trl:8293166411...

(the way I've been handling interrupts before was a tiny bit insane)

#

Now I just save nested interrupt registers on stack...

carmine nacelle Feb 3, 2025, 1:17 PM

#

And I suspect I know why the framebuffer was casually breaking

carmine nacelle Feb 3, 2025, 4:10 PM

#

I'm having a skill issue with GDT nooo

carmine nacelle Feb 3, 2025, 5:56 PM

#

(something is destroying the return frame, gdt works as normal (it seems))

carmine nacelle Feb 3, 2025, 6:59 PM

#

thonk

carmine nacelle Feb 4, 2025, 10:29 AM

#

Ok, I think it's time to add zones to my PMM

#

(Then do PAE and SMP on i686...)

carmine nacelle Feb 4, 2025, 9:44 PM

#

I wonder...

#

if this is what could be breaking the framebuffer...

#

(ok, I urgently need watchdog memes)

carmine nacelle Feb 4, 2025, 11:16 PM

#

I've been thinking about this all day, and I haven't come up with something that I like...

#

I think I'll just create a general, and below 4G zone, and a reserved below 1MB zone for now, and that's it

carmine nacelle Feb 8, 2025, 12:26 AM

#

Oof, PAE is not working well...

carmine nacelle Feb 8, 2025, 1:38 AM

#

bruh, auto is trolling me

#

Ok I have absolutely no clue of what the fuck is happening

#

pmm just pretends that addresses above 4GB don't exist, even though they clearly do

#

then some fuckery happens

#

and then the memory read by uacpi becomes corrupt

#

or maybe it isn't?

#

wtf

#

all I did was enabling PAE

#

How nice

#

(the only nice thing is that it's consistent...)

#

thonk

#

(32 bit ints are really haunting me nooo )

#

This is insanity

#

How does it even get demoted to int from unsigned long long?

#

The kernel has really gone rogue...

#

(that looks like ascii)

carmine nacelle Feb 8, 2025, 2:49 AM

#

Could it be a seabios skill issue?

#

I want to try something...

#

bruh I think hyper just tripple faults, with no image

carmine nacelle Feb 8, 2025, 12:34 PM

#

removing -O3 and lto makes it fail to boot...

little smelt Feb 8, 2025, 3:27 PM

#

carmine nacelle removing -O3 and lto makes it fail to boot...

bro how🙏

toxic torrent Feb 8, 2025, 3:29 PM

#

carmine nacelle removing -O3 and lto makes it fail to boot...

linux reference

little smelt Feb 8, 2025, 3:30 PM

#

not even my OS dared to do that

#

altho when my APIC implementation was wrong the keyboard only worked on -O3

#

then i discovered i did not setup the redirect tables

carmine nacelle Feb 8, 2025, 3:39 PM

#

carmine nacelle removing -O3 and lto makes it fail to boot...

It (-O3) optimized a bunch of stuff away, which was broken because I was missing constinit in pmm, and my pmm is initialized before I call global constrictors

#

Now it crashes in different place

#

(all registers get ASCII in them, and then the kernel pagefaults)

carmine nacelle Feb 8, 2025, 5:13 PM

#

bruh it returns to a bogus address, ebp also points to shit, and I have no idea how to debug that

carmine nacelle Feb 8, 2025, 6:15 PM

#

Idk why, but if feels like I'm using wrong calling convention

carmine nacelle Feb 8, 2025, 7:03 PM

#

carmine nacelle removing -O3 and lto makes it fail to boot...

The issue was ratified, and it almost behaves exactly the same nooo

#

(except that the kernel also pagefaults now)

#

I suspect I know what's going on...

carmine nacelle Feb 8, 2025, 10:21 PM

#

I wonder if this could be an issue with tlb...

hoary moat Feb 8, 2025, 10:42 PM

#

*TLS (?)

carmine nacelle Feb 8, 2025, 10:56 PM

#

No

#

Translation Lookaside Buffer

#

I think it never gets invalidated somehow

#

like flanterm works, and it uses the same functions, so phys mappings have to be correct...

carmine nacelle Feb 9, 2025, 8:58 AM

#

Yep, I've compared the logs, and it's not flushing the tlb properly

carmine nacelle Feb 9, 2025, 10:31 AM

#

It was a dumb mistake in page invalidation loop

#

(the only issue now is that AHCI driver pagefaults...)

carmine nacelle Feb 9, 2025, 10:52 AM

#

ok, PAE kinda seems to be fully working...

#

(I just need to fix PMM ignoring physical memory limit)

twilit talon Feb 9, 2025, 10:52 AM

#

carmine nacelle ok, PAE kinda seems to be fully working...

what was the bug?

carmine nacelle Feb 9, 2025, 10:53 AM

#

it was not invalidating tlb

twilit talon Feb 9, 2025, 10:53 AM

#

bruh

carmine nacelle Feb 9, 2025, 10:53 AM

#

between mmap and munmap

carmine nacelle Feb 9, 2025, 10:54 AM

#

carmine nacelle (I just need to fix PMM ignoring physical memory limit)

I've set it to 16GB bruh

carmine nacelle Feb 9, 2025, 10:55 AM

#

carmine nacelle (I just need to fix PMM ignoring physical memory limit)

It also can allocate memory below 4GB now, on 64 bit arches as well

#

For legacy devices DMA

carmine nacelle Feb 9, 2025, 11:00 AM

#

carmine nacelle (I just need to fix PMM ignoring physical memory limit)

Actually nevermind...

#

(if someone wants to try it bruh )

#

The only thing missing is NX (which is also unsupported on x86_64 port)

#

but works on RISC-V, so it should mostly be working...

#

I mean be easy to implement...

#

bruh

#

What's weird is that PAE is supposedly disabled yet it's still doing that

carmine nacelle Feb 9, 2025, 11:32 AM

#

ok, I think I need a PXE server...

carmine nacelle Feb 9, 2025, 11:43 AM

#

carmine nacelle <:bruh:761542241381908491>

it should have been assert(page->allocated_count <= 14) bruh

#

Cool, ignoring the fact that AHCI driver explodes (probably after seeing 6TB drive), it fully works!

#

thonk how do these work?

hoary moat Feb 9, 2025, 11:51 AM

#

Magic

carmine nacelle Feb 9, 2025, 11:51 AM

#

Ok, since my holidays end this week, I think I'll just fix the x86_64 and RISC-V ports and merge the ia32 port without SMP

#

(though I still need to investigate it not working on the Sony laptop...)

#

(also, the not broken build)

📎 pmos-hyper.img

cyan elk Feb 9, 2025, 12:20 PM

#

carmine nacelle <:thonk:575445129863888896> how do these work?

they likely don't

#

at least not in the way you'd think they would

#

it's probably a PS/2 to USB adapter that exposes them as USB devices to the host, given the description of the item and the fact it literally has an extra spare internal USB port

carmine nacelle Feb 9, 2025, 12:22 PM

#

I'm assuming it's just a USB keyboard

toxic torrent Feb 9, 2025, 12:45 PM

#

carmine nacelle <:thonk:575445129863888896> how do these work?

carmine nacelle Feb 9, 2025, 7:02 PM

#

bruh my AHCI driver is trolling me

#

bruh

hoary moat Feb 9, 2025, 7:14 PM

#

carmine nacelle bruh my AHCI driver is trolling me

So am I

carmine nacelle Feb 9, 2025, 7:14 PM

#

It was just me being dump

#

It was doing what I told it to do

hoary moat Feb 9, 2025, 7:14 PM

#

My code does what it wants to do

#

So at least yours does what you told it to do

carmine nacelle Feb 9, 2025, 7:26 PM

#

it doesn't page fault but still doesn't work

#

nooo

#

Just times out

#

But the disk light is on

#

What am I missing? thonk

#

*the activity light that is

#

Even though identify works fine

#

Do I need some special flag for it to clear CI or something?

carmine nacelle Feb 10, 2025, 2:25 AM

#

Fun, the AHCI driver now actually works well enoguh to read partition tables from 2 disks on physical hardware

#

(kinda feel like magic hearing hard drives noises and it actually reading data)

#

(and also, it was done on 32 bit kernel, with allocator allocating data at about 16GB physical addresses)

#

(so PAE also works, and kernel and userspace don't choke on 64 bit addresses and >4GB RAM)

#

Anyway, I'll probably add changes to 64 bit builds (the zoned pmm stuff changed the initialization somewhat) and merge IA32 support sometime soon ™

#

Then I just need to design the interface for filesystems nooo

carmine nacelle Feb 10, 2025, 2:36 AM

#

carmine nacelle Fun, the AHCI driver now actually works well enoguh to read partition tables fro...

Also this thing (AHCI) works with IOAPIC and level triggered interrupts, so I guess props to uACPI for interrupts routing

#

(and also, this whole thing uses c++ coroutines because why not)

carmine nacelle Feb 10, 2025, 9:35 AM

#

Ok, there are too many things I want to do

pearl narwhal Feb 10, 2025, 1:35 PM

#

carmine nacelle Anyway, I'll probably add changes to 64 bit builds (the zoned pmm stuff changed ...

IA64 when

carmine nacelle Feb 10, 2025, 1:36 PM

#

When either Limine or Hyper decides to support it

#

@twilit talon ?

pearl narwhal Feb 10, 2025, 1:39 PM

#

gift me itanium hardware

#

i will port the bootloader of choice of whoever gifts me itanium hardware

carmine nacelle Feb 10, 2025, 1:41 PM

#

Yeah, that's the issue

#

No hardware

#

Maybe I'll port to aarch64 though...

#

thonk

#

Though I really wanted to finally write a filesystem driver

#

Maybe in Rust trl

#

(I wanted to do it in Ada, but that language is too alien for me)

pearl narwhal Feb 10, 2025, 1:45 PM

#

carmine nacelle Though I really wanted to finally write a filesystem driver

that sounds fun

twilit talon Feb 10, 2025, 2:57 PM

#

carmine nacelle <@353593948523986945> ?

Not sure where to test

carmine nacelle Feb 10, 2025, 4:13 PM

#

bruh

#

Nevermind, it was a printf skill issue bruh

#

Great, the pmm now poops itself on 64 bit platforms for some reason...

#

maybe I should just drop the 64 bit support trl

pearl narwhal Feb 10, 2025, 4:31 PM

#

just drop everything

#

pmOS should only run in your mind

carmine nacelle Feb 10, 2025, 4:59 PM

#

bruh limine tripple faults on my PC (?)

#

Unfun: interrupts are broken for some reason on AMD64 build...

carmine nacelle Feb 10, 2025, 5:03 PM

#

carmine nacelle <:bruh:761542241381908491> limine tripple faults on my PC (?)

@vale hinge

carmine nacelle Feb 10, 2025, 5:06 PM

#

carmine nacelle Unfun: interrupts are broken for some reason on AMD64 build...

And it was refusing to post after reboot

carmine nacelle Feb 10, 2025, 5:07 PM

#

carmine nacelle <@563448614509215755>

(btw, what happened to your avatar?)

#

(maybe it's some massive skill issue...)

#

(it's not booting on 2 PCs)

#

ahci driver also worked on intel macbook

#

letsgo

#

(before I lost interrupts...)

#

Fun statistic: limine passes 107 memory entries on that PC

vale hinge Feb 10, 2025, 5:13 PM

#

carmine nacelle <@563448614509215755>

?

#

what PC?

#

open an issue? send additional information

#

do you think i have a crystal ball or something lol

wide spruce Feb 10, 2025, 5:14 PM

#

what does the PM of PMOS mean

vale hinge Feb 10, 2025, 5:14 PM

#

protected mode

wide spruce Feb 10, 2025, 5:14 PM

#

really?

carmine nacelle Feb 10, 2025, 5:18 PM

#

Mikhail and my friend Pau

#

It was a 2 person project

#

(There was pimOS before with Ivan, but it didn't even get to loading its own page tables bruh )

carmine nacelle Feb 10, 2025, 5:19 PM

#

vale hinge what PC?

If I boot with bios, it doesn't load, idk

#

On 2 PCs

#

Which suggests it's a skill issue on my part

#

So probably false alarm

carmine nacelle Feb 10, 2025, 5:20 PM

#

vale hinge what PC?

But it's the PC which broke it before

#

Which remaps LAPIC

#

but hyper does work

#

and the image does boot in qemu

vale hinge Feb 10, 2025, 5:28 PM

#

carmine nacelle If I boot with bios, it doesn't load, idk

well, is it just your kernel or like, any Limine kernel?

#

does it work booting Linux, does it work booting other Limine kernel such as the template or test kernel bundled with the Limine repo?

carmine nacelle Feb 10, 2025, 5:28 PM

#

limine itself doesn't boot, I see the bios logo and it reboots

#

with bios booting

#

which might suggest that I didn't write it to usb drive properly

vale hinge Feb 10, 2025, 5:29 PM

#

have you made sure you created the image properly as well?

carmine nacelle Feb 10, 2025, 5:29 PM

#

it boots in qemu

vale hinge Feb 10, 2025, 5:29 PM

#

that's not necessarily a guarantee

#

qemu is very lenient

carmine nacelle Feb 10, 2025, 5:31 PM

#

I do limine/limine bios-install $(DISK_IMAGE) and it says

Physical block size of 512 bytes.
Installing to GPT. Logical block size of 512 bytes.
Secondary header at LBA 0x3ffff.
Secondary header valid.
GPT partition NOT specified. Attempting GPT embedding.
New maximum count of partition entries: 44.
Stage 2 to be located at 0x1a00 and 0x7ffd600.
Reminder: Remember to copy the limine-bios.sys file in either
          the root, /boot, /limine, or /boot/limine directories of
          one of the partitions on the device, or boot will fail!
Limine BIOS stages installed successfully!

vale hinge Feb 10, 2025, 5:35 PM

#

on some motherboards CSM won't boot GPT images that aren't ISOHYBRID ISOs

#

unless it is an ISO

#

idk you need to be way more specific

carmine nacelle Feb 10, 2025, 5:38 PM

#

it also fails on a pure bios machine

#

It's a GPT image with 2 partitions

#

fat32 and ext2

vale hinge Feb 10, 2025, 5:39 PM

#

is the Hyper image also GPT?

#

pure BIOS machines may also fail to boot disks that present GPT disk labels

carmine nacelle Feb 10, 2025, 5:40 PM

#

Hyper is MBR

#

https://gitlab.com/mishakov/pmos/-/blob/dev/limine/Makefile?ref_type=heads#L118 this is how I create the image

vale hinge Feb 10, 2025, 5:40 PM

#

ok well, that is one difference, you should make sure those 2 things are the same and reevaluate

carmine nacelle Feb 10, 2025, 5:44 PM

#

carmine nacelle it also fails on a pure bios machine

and it has arch linux in bios + gpt mode installed

#

with grub

untold pollen Feb 10, 2025, 5:47 PM

#

Attempt 2

carmine nacelle Feb 10, 2025, 5:49 PM

#

Does it have EFI?

untold pollen Feb 10, 2025, 5:49 PM

#

Idk we will see

toxic torrent Feb 10, 2025, 5:50 PM

#

it does

little smelt Feb 10, 2025, 5:50 PM

#

carmine nacelle <:bruh:761542241381908491>

just one EfiConventionalPage descriptor?

toxic torrent Feb 10, 2025, 5:50 PM

#

@carmine nacelle @untold pollen check #voice-0

vale hinge Feb 10, 2025, 6:02 PM

#

carmine nacelle and it has arch linux in bios + gpt mode installed

ok well then, i will delete the Limine repository thanks for the feedback

carmine nacelle Feb 10, 2025, 6:02 PM

#

vale hinge ok well then, i will delete the Limine repository thanks for the feedback

how can I debug it?

#

like i would like to see limine working on that PC

vale hinge Feb 10, 2025, 6:03 PM

#

can Hyper do GPT on BIOS? for one test that in that case

#

that would be test #1

carmine nacelle Feb 10, 2025, 6:03 PM

#

Hyper can't

vale hinge Feb 10, 2025, 6:04 PM

#

@twilit talon accurate?

carmine nacelle Feb 10, 2025, 6:04 PM

#

I can test mbr + bios

#

also that PC has COM port and 0xb8 for early debugging if that's needed

vale hinge Feb 10, 2025, 6:04 PM

#

yeah, test #2 would be, as i said above, test Limine with BIOS/MBR

carmine nacelle Feb 10, 2025, 6:04 PM

#

idk

#

and I have limine repo cloned already

vale hinge Feb 10, 2025, 6:05 PM

#

also #3 would be to please open an issue with the machine specific and what you've tested so far, just so this issue does not get lost in time if we don't fix it

twilit talon Feb 10, 2025, 6:13 PM

#

vale hinge <@353593948523986945> accurate?

Yeah, the installer cant embed it in gpt because I was lazy

pearl narwhal Feb 10, 2025, 6:16 PM

#

untold pollen Attempt 2

cool!!

vale hinge Feb 10, 2025, 6:29 PM

#

twilit talon Yeah, the installer cant embed it in gpt because I was lazy

i will remove embedding anyways for 9.x so

#

:^)

#

i thought up a better replacement

twilit talon Feb 10, 2025, 6:35 PM

#

vale hinge i will remove embedding anyways for 9.x so

Why is that?

vale hinge Feb 10, 2025, 6:37 PM

#

well, you're seeing part of it here now, also shrinking and extending images causes issues with it, and so on

twilit talon Feb 10, 2025, 6:39 PM

#

I didnt read the entire thread

#

But yeah its a hack

carmine nacelle Feb 10, 2025, 8:18 PM

#

anyway, latest risc-v pmOS build

📎 pmos.img

carmine nacelle Feb 10, 2025, 8:42 PM

#

i686 port is fun...

carmine nacelle Feb 10, 2025, 11:23 PM

#

Ok, so the first step of learning the language is porting its standard library ultrameme

#

(my latop is suffering from compiler compilations)

#

(also, I wonder how insane it would get, using 3 languages)

twilit talon Feb 10, 2025, 11:34 PM

#

which?

left dew Feb 10, 2025, 11:37 PM

#

Do not port rust

#

i Repeat DO NOT port rust

carmine nacelle Feb 10, 2025, 11:54 PM

#

why not

wide spruce Feb 11, 2025, 12:04 AM

#

its a bit of a pita

carmine nacelle Feb 11, 2025, 12:06 AM

#

Why?

#

rustc already recognizes it

left dew Feb 11, 2025, 12:07 AM

#

lol you're in for a treat

#

porting the std WILL be fun!

carmine nacelle Feb 11, 2025, 12:08 AM

#

I had Go kinda working

#

It can't be that hard

#

Will also improve my libc

left dew Feb 11, 2025, 12:09 AM

#

oh it WILL

#

or will it...

#

idk, all i can say is good luck!

carmine nacelle Feb 11, 2025, 12:09 AM

#

The osdev wiki article is very bad though

left dew Feb 11, 2025, 12:09 AM

#

yeah i wonder why

#

(it was probably written by someone who has no idea what they are doing)

carmine nacelle Feb 11, 2025, 12:11 AM

#

I have no idea what I'm doing

#

somehow it had build stage 0 without my target

left dew Feb 11, 2025, 12:11 AM

#

i recommend you take a peek at managarm

#

they have rust already

#

and from what i know it's a giant pain in the ass

carmine nacelle Feb 11, 2025, 12:16 AM

#

I haven't found it

#

(but looked for like 30 seconds)

#

Ok, I think it's trolling me already

#

I've added my target to STAGE0_MISSING_TARGETS and it's pretending it's not there

left dew Feb 11, 2025, 12:31 AM

#

enjoy :^)

carmine nacelle Feb 11, 2025, 1:18 AM

#

rust must have changed something

twilit talon Feb 11, 2025, 7:47 AM

#

left dew enjoy :^)

is it that bad?

left dew Feb 11, 2025, 9:25 AM

#

twilit talon is it that bad?

apparently

twilit talon Feb 11, 2025, 9:29 AM

#

linux abi compat enjoyers win again

wide spruce Feb 11, 2025, 11:25 AM

#

they already have enough pain

#

let them have this one

carmine nacelle Feb 11, 2025, 11:26 AM

#

wdym

wide spruce Feb 11, 2025, 11:26 AM

#

linux abi compat is pain

carmine nacelle Feb 11, 2025, 11:26 AM

#

linux compat is boring

wide spruce Feb 11, 2025, 11:26 AM

#

that aside

twilit talon Feb 11, 2025, 11:33 AM

#

id rather implement namespaces than port rust LULW

carmine nacelle Feb 11, 2025, 11:35 AM

#

I'm gonna join their discord server

#

I'm sure some people will be thrilled to help...

wide spruce Feb 11, 2025, 11:41 AM

#

no they wont

wide spruce Feb 11, 2025, 11:42 AM

#

twilit talon id rather implement namespaces than port rust <:LULW:543420553021882371>

namespaces are not that hard

toxic torrent Feb 11, 2025, 11:52 AM

#

twilit talon id rather implement namespaces than port rust <:LULW:543420553021882371>

nooo

carmine nacelle Feb 11, 2025, 3:32 PM

#

thonk

#

the build system is dumb

toxic torrent Feb 11, 2025, 3:34 PM

#

bruh

#

what i did once is use mrustc

#

to bootstrap another rustc

carmine nacelle Feb 11, 2025, 3:38 PM

#

I still have no idea of what's happening

#

I've looked at managarm and redox in the rust code

#

And I have the same stuff as they

#

But no, fuck you, unsupported target

#

also, greping rust tree for error occurred yields nothing, but matches in binary

#

which makes me wonder if that could be coming from llvm

pearl narwhal Feb 11, 2025, 3:40 PM

#

left dew porting the std WILL be fun!

std porting is the only hard bit (but extremely hard

toxic torrent Feb 11, 2025, 3:41 PM

#

if you used mlibc this would be basically free meme

pearl narwhal Feb 11, 2025, 3:41 PM

#

but it shouldn't be too much worse than idk go probably and go has its own issuds

pearl narwhal Feb 11, 2025, 3:42 PM

#

carmine nacelle I'm gonna join their discord server

did u?

#

you're likely on the wrong server

toxic torrent Feb 11, 2025, 3:42 PM

#

@carmine nacelle no "error occurred" in llvm

carmine nacelle Feb 11, 2025, 3:42 PM

#

pearl narwhal did u?

yeah they said I joined wrong server

pearl narwhal Feb 11, 2025, 3:42 PM

#

carmine nacelle yeah they said I joined wrong server

the "official" discord server for rust is for developing rustc and nobody uses it

#

most rustc development happens on zulip

#

the unofficial rust server is what you're looking for

#

specifically #os-dev or the rustc-dev thread in #dark-arts

carmine nacelle Feb 11, 2025, 3:44 PM

#

toxic torrent <@350748255929499649> no "error occurred" in llvm

llvm/tools/llvm-split/llvm-split.cpp

#

line 89

pearl narwhal Feb 11, 2025, 3:45 PM

#

I can send an invite if you want

toxic torrent Feb 11, 2025, 3:45 PM

#

carmine nacelle llvm/tools/llvm-split/llvm-split.cpp

oh bruh i was in pager mode on the top of the page

carmine nacelle Feb 11, 2025, 3:45 PM

#

is my llvm patch fucked up?

#

I just use x86_64-pmos normally

#

or whatever

#

without unknown

#

bruh

#

Time to dig through that I guess...

#

no, but llvm should be printing error

#

it's not coming from there...

carmine nacelle Feb 11, 2025, 4:10 PM

#

through scientific printfs, I've deduced that it can't find the right C compiler...

carmine nacelle Feb 11, 2025, 6:13 PM

#

bruh

carmine nacelle Feb 11, 2025, 6:36 PM

#

How am I supposed to modify it?

carmine nacelle Feb 12, 2025, 12:01 AM

#

(for a future reference to myself: RUSTC=/home/misha/rust-pmos/build/host/stage2/bin/rustc cargo run -p gen-target-info)

#

thonk why don't I also add RISC-V and i686 and aarch64, while at it?

carmine nacelle Feb 12, 2025, 1:26 AM

#

I think I finally made it recognize pmOS!

#

It's even trying to build std

#

(I guess it's time for some big time stubbing)

#

(tomorrow)

modest nest Feb 12, 2025, 10:08 AM

#

i think your problem is that cargo doesnt know about your toolchain

#

https://rust-lang.github.io/rustup/concepts/toolchains.html
under "custom-toolchain"

Toolchains - The rustup book

#

@carmine nacelle

carmine nacelle Feb 12, 2025, 10:10 AM

#

modest nest https://rust-lang.github.io/rustup/concepts/toolchains.html under "custom-toolch...

I've read that already and it didn't work

carmine nacelle Feb 12, 2025, 10:10 AM

#

carmine nacelle (for a future reference to myself: `RUSTC=/home/misha/rust-pmos/build/host/stage...

It only started working after I cloned cc-rs and ran this with new toolchain

#

Inside it

#

And pointed std Cargo.lock to use my cc-rs

modest nest Feb 12, 2025, 10:11 AM

#

do you use a crt0 object?

carmine nacelle Feb 12, 2025, 10:11 AM

#

I think so

modest nest Feb 12, 2025, 10:12 AM

#

that might cause the need for cc-rs

carmine nacelle Feb 12, 2025, 10:12 AM

#

(the wiki is a bit ambiguous)

#

I want it to link against my libc anyway (?)

#

(tbh idk how rust works)

modest nest Feb 12, 2025, 10:13 AM

#

afaik you have two options either have a crt0 object that inits the stack and stuff or call the entry point directly but you need to do the setup before that

#

you dont need to afaik

#

you could implement your os api directly in rust std without a libc in between

#

but usually a system already has a libc so putting rust std on top of that is reasonable

carmine nacelle Feb 12, 2025, 10:15 AM

#

My libc creates its own thread and manages signals, open files, processes and stuff like that

#

So I kinda want it loaded

#

(I guess I could reimplement it later fully in Rust)

carmine nacelle Feb 12, 2025, 4:43 PM

#

Do I have to patch unix crate if I declare pmOS as so?

#

*libc

modest nest Feb 12, 2025, 7:04 PM

#

probably?

carmine nacelle Feb 13, 2025, 12:06 AM

#

thonk it's not recognizing my target bruh

#

nvm, I'm just dumb

modest nest Feb 13, 2025, 8:34 AM

#

So it works now?

carmine nacelle Feb 13, 2025, 8:57 AM

#

Yeah, but I'm filling definitions from my libc atm

#

And there's a lot of them

carmine nacelle Feb 13, 2025, 1:37 PM

#

Do I even need crt0 objects if in theory my libc should take care of it for me?

left dew Feb 13, 2025, 1:53 PM

#

they need to be statically linked into every program iirc

carmine nacelle Feb 13, 2025, 2:18 PM

#

I don't have dynamic linking anyway...

carmine nacelle Feb 13, 2025, 5:26 PM

#

bruh, rust std wants random generator...

#

(and a bunch of other stuff I don't have yet)

eager rose Feb 13, 2025, 5:35 PM

#

int
random (void)
{
        return 4; // generated by the roll of a fair dice
}

hexed knot Feb 13, 2025, 5:37 PM

#

eager rose ```c int random (void) { return 4; // generated by the roll of a fair di...

/// Generate cryptographically secure random numbers using a novel
/// high-performance patent-pending algorithm.
pub fn random<const N: usize>() -> [u8; N] {
    // I have been informed that the letter 'A' is equally likely to be chosen as any other.
    [0x41; N]
}

left dew Feb 13, 2025, 5:37 PM

#

pub fn random() usize {
  let mut ret: usize = 4;
  #[cfg(target_arch = "x86_64")]
  {
    unsafe {
      core::arch::asm!(
        "rdrand",
        _ = out(ret),
      );
    }
  }
  ret
}```

#

shrimple

eager rose Feb 13, 2025, 5:38 PM

#

eager rose ```c int random (void) { return 4; // generated by the roll of a fair di...

this is a reference to XKCD and a suggestion to stub it out

hexed knot Feb 13, 2025, 5:39 PM

#

left dew ``` pub fn random() usize { let mut ret: usize = 4; #[cfg(target_arch = "x86...

thats actually insecure

#

what if someone has Google Project Zero microcode

left dew Feb 13, 2025, 5:41 PM

#

real...

hexed knot Feb 13, 2025, 5:41 PM

#

lol

#

i am really curious what the exploit is tbh

#

and also

#

how i can exploit it

left dew Feb 13, 2025, 5:43 PM

#

insecure hash function

#

probably adding some garbage data and causing a hash collision or some shit

eager rose Feb 13, 2025, 5:48 PM

#

hexed knot what if someone has [Google Project Zero microcode](<https://github.com/google/s...

then you're screwed anyways lol

hexed knot Feb 13, 2025, 5:49 PM

#

left dew insecure hash function

more details though lol

hexed knot Feb 13, 2025, 5:49 PM

#

eager rose then you're screwed anyways lol

no

#

that microcode doesnt cause a security issue

#

because it doesnt set CF

carmine nacelle Feb 13, 2025, 6:53 PM

#

But there is no RDRAND on Pentium Pro...

#

trl

modest nest Feb 13, 2025, 9:17 PM

#

left dew they need to be statically linked into every program iirc

not for rust programms, all rust cares about is that its std main is called with a specified setup this can either be done by crt0 or by a sophisticated os lib like a userspace scheduler

#

or just in asm in your code in the std

carmine nacelle Feb 13, 2025, 9:30 PM

#

I decided not to rant in porting

#

But

#

Very nice of Rust

#

https://pubs.opengroup.org/onlinepubs/9799919799/

The <sys/types.h> header shall define at least the following types:
gid_t
Used for group IDs.
id_t
Used as a general identifier; can be used to contain at least a pid_t, uid_t, or gid_t.
pid_t
Used for process IDs and process group IDs.
...

All of the types shall be defined as arithmetic types of an appropriate length, with the following exceptions:
...
Additionally:
nlink_t, uid_t, gid_t, and id_t shall be integer types.
blksize_t, pid_t, and ssize_t shall be signed integer types.

#

Very nice

#

My ids are all int64_t

toxic torrent Feb 13, 2025, 9:36 PM

#

carmine nacelle My ids are all int64_t

are you even allowed to do that

carmine nacelle Feb 13, 2025, 9:36 PM

#

toxic torrent are you even allowed to do that

Am I prohibited?

toxic torrent Feb 13, 2025, 9:38 PM

#

looking at the spec, no

carmine nacelle Feb 13, 2025, 9:38 PM

#

toxic torrent are you even allowed to do that

POSIX just says it has to be an integer type of "an appropriate length"

toxic torrent Feb 13, 2025, 9:38 PM

#

weird that rust does that then

#

i guess cast it meme

carmine nacelle Feb 13, 2025, 9:38 PM

#

I can change it to int32_t or whatever, but it's gonna be anoying

#

(I mean the pmOS thing)

#

(Can I complain somewhere? bruh trl )

toxic torrent Feb 13, 2025, 9:39 PM

#

/dev/null

carmine nacelle Feb 13, 2025, 9:40 PM

#

nice

#

Also my getpid can ENOMEM ultrameme

#

But I think that is not allowed

toxic torrent Feb 13, 2025, 9:41 PM

#

carmine nacelle Also my getpid can ENOMEM <:ultrameme:729026898666979428>

how

#

getpid must not fail

carmine nacelle Feb 13, 2025, 9:41 PM

#

#

Yeah

#pmOS - microkernel OS for RISC-V, x86, and LoongArch