ReactOS's Modern Hardware Woes

[bright gale]

so it is mapped somewhere

[buoyant thorn]

pfn? page tables? they are mapped. But is it hhdm?

[bright gale]

i mean this function looks like a typical phys_to_virt

[buoyant thorn]

I looked at what it does briefly and it's like some macro would do. It gets pfn number from the phys addr, then it gets pointer to the corresponding pfn entry and from there it gets pte address. Bingo. After you have this, you can get vaddr of the requested phys addr. because of so called recursive pagging. Have no idea why it's called that, bc it's recursive only in regard of mapping the page tables themselves.

[bright gale]

im confused

whats a pfn entry

[buoyant thorn]

An entry in the pfn database. Each such holds info on a ram page. Among which pte address for the mapping currently using this ram page.

[bright gale]

what if there's multiple address spaces mapping it

[buoyant thorn]

It must be for the current context I'd say. There are prototype ptes as well for shared memory.

[bright gale]

hm

so if its not mapped anywhere that would return null?

[buoyant thorn]

I guess. There's such a check.

[bright gale]

i see

[hard flower]

I got jumpscared by this

But yeah kind of

I figured it out when I working on another PR and confirmed with Windows

[hard flower]

The recursive paging part kind of tripped me

[pseudo surge]

i love walking ntdll exports and extracting the syscall numbers and doing direct syscalls

[bright gale]

Wdym kind of?

[hard flower]

How PTE addresses get convered to the VA equivalent

zaval explained it

its kinda like that

What makes it page
chapter 15

[stable topaz]

Told you that was a good book

[hard flower]

it is lol

[stable topaz]

And yet

Our Mm is still broken

Depsite half of us reading that

[hard flower]

#DownWithARM3

[bright gale]

does this mean u have to know its mapped beforehand and u have to know it wont get unmapped from under your feet somehow?

[hard flower]

It first checks if there is an entry in the PfnDatabase

And if it is then it's mapped

Otherwise nope

[bright gale]

but pfndatabase contains every physical page

[hard flower]

Yes

[bright gale]

so like, literally all physical memory is always mapped?

[hard flower]

I think so

[bright gale]

then its literally hhdm lol

just kinda indirect because stuff isnt mapped at a fixed offset

[hard flower]

Yeah

[normal reef]

how does that work on 32 bit then

[bright gale]

@thin temple btw did u see that? Nt also has an HHDM now

[scarlet nova]

good for them tbh

[kindred cliff]

real

[pseudo surge]

i have a project for a custom windows runtime but i haven't pushed in ages

mandatory self plug; https://github.com/notbonzo-com/win-rt

[kindred cliff]

oh nice

i have this nice file: https://github.com/RandomcodeDev/chifir-engine/blob/main/base%2Floader_win32.cpp

wait is that the f00f bug in your bio

[hard flower]

They do something similar for 32 bit as well

Source: I tried the test on x86 build as well and it works

[buoyant thorn]

It doesn't. pfn is not a range where all ram mapped to, pfn is an array (sparse) each entry of which describes a page. It's not hhdm at all.

[grim briar]

ig it was meant as describes every physical page?

well idk

[bright gale]

Yes but the page description contains a pointer to the mapping

So for all intents and purposes it's the exact same thing just with an indirection

[autumn pond]

windows didnt have it on 32bit

[buoyant thorn]

Then page tables are also hhdm. Pfn is not for the mapping role finally, it's for state dynamics maintenance and ram allocation (working sets management). The mere fact you can derive a vaddr for a given physaddr using a pfn entry doesn't make it hhdm. Hhdm is just a range in virtual address space where all ram is mapped. whereas pfn is not like that, it's metadata about a lot of mm stuff put in a structure organized in a quite complex way. This drastic difference is because these 2 were meant to serve absolutely different things.

[bright gale]

Sorry I don't understand the difference

[buoyant thorn]

Are page tables hhdm too?

[bright gale]

The idea of HHDM is u have a way to access any physical memory without mapping it explicitly

The fact that its easy to calculate that place doesnt mean its fundamentally needed

It can be indirect like nt

[autumn pond]

where does the term HHDM come from?

i mean who was the first to call it so

[buoyant thorn]

The idea (or rather purpose) of pfn is totally different. It's not for "accessing any physical memory". It's for keeping the state of every page. Either it's free, standby, modified or in use and if in use then by what. Telling that it the same as hhdm is like telling that page tables are hhdm, page tables then would be even closer. Accessing PFN you don't access "any physical memory". Nor even you can do that. You wouldn't be able to access a page if it weren't mapped. Because there is no any hhdm and pfn doesn't do any hhdm trickery. Maybe julysky could convince.

[bright gale]

Mint invented it for limine, but the idea is super old

[bright gale]

I understand what struct page is, its irrelevant here

What's relevant is given a physical address u can access its contents without mapping it

[buoyant thorn]

No you can't

[bright gale]

Thats what that function does tho

[buoyant thorn]

No

[bright gale]

#1325160121474486372 message

Literally the guy who sent the pr

[buoyant thorn]

If the page hasn't been mapped you wouldn't be able to access it.

so like, literally all physical memory is always mapped?

The answer to this question is no

[bright gale]

Well then you know better than a ros dev

[buoyant thorn]

You didn't understand him correctly.

[bright gale]

Ill wait for their input on this

[buoyant thorn]

While you are waiting, can you point to the line in the function, where you think, it accesses content of the page with a given physical address "without mapping it"?

[bright gale]

No lol I'm just repeating what I've been told and what I've briefly seen in the tests, other than that I dont really care about arguing about it I just wanna know the truth

[pseudo surge]

that's a monolithic beast

[buoyant thorn]

I see. I tried to explain, but I guess the authority of julysky's level is needed here. But I think he doesn't like ros, so I am not sure if tagging him here would be a good idea.

[pseudo surge]

yes

[kindred cliff]

about half of it is just importing functions, only a couple interesting bits

little messy but it works really well. it's backed by some assembly files to get the names right and stuff

[pseudo surge]

i also tried making my own ldr, however API sets and SxS finished that attempt

[kindred cliff]

oh yeah the modern stuff is tricky

[autumn pond]

lol

[kindred cliff]

i compiled the windows xp kernel with clang, and i had to make a tool that checked all the circular imports between the debugger, kernel, and hal to make sure none were missing

[thin temple]

I've known that for years

Not news

They use one on 64 bit to accelerate arbitrary page frame accesses (no need to set up a temporary mapping) especially during times where the other mechanisms wouldn't work because they're at elevated irql and can't take page faults and so on

They still use the recursive mapping to access the page tables and they still use the file mappings to access page cache data

Also that has nothing to do with the what makes it page thing that was posted literally at all

I believe you guys are confused

No

There is an "hhdm" but this function has absolutely nothing to do with that, it isn't even adjacent

[thin temple]

No it doesn't

[buoyant thorn]

Well. I tried to explain that.

But they didn't want to believe me.

[thin temple]

For a private page, the PTE address field of the PFN structure points back to the address within the recursive page table that currently tracks the page (NOTE that the recursive page table is different for each process but resides at the same virtual base address in each of their address spaces)

Because recursive page tables are "unrolled" virtually into a linear array, you index them by virtual page number to get the address of the corresponding PTE that maps that virtual page

This means you can easily un-index it with simple arithmetic to recover the virtual address from the pte address

That's what they're doing here

This function lets you figure out where a private page is located in a process's address space given its physical address, not within an hhdm

This function would not work for a shared page (such as a file cache page or page of shared anonymous memory) because the PTE address for those is a "prototype PTE" which does not reside in process page tables

And it definitely is not looking the page frame up in an HHDM

For that they just use a simple macro that adds an offset

[thin temple]

I'm unsure what this would be used for but this is what it does @hard flower

[hard flower]

Wait

So it can tell if a page is located in a process' address space

Does that also include the pages allocated for setting up the address space for the process of not?

[thin temple]

Based on the simplicity of the function I'm guessing the caller is meant to already know it's a private page or is prepared to deal with the nonsense result if it's not

[thin temple]

Don't know what this means

[bright gale]

Ohh

Do u know the api family for hhdm?

[thin temple]

I don't think there is one I think it's just used internally for what I said

[hard flower]

Hold on wait

[thin temple]

One example of usage is when a page frame is reclaimed off the standby list they need to write the value of OriginalPte back into the pte pointed to by PteAddress, which is in some arbitrary process's page tables, and they can't use KeAttachProcess to do this because they're holding spinlocks at raised irql, so they use the hhdm

This works because the page table is guaranteed to exist and be wired in memory for as long as there's a resident standby page with a reference for it

[hard flower]

So wait

I can use this to determine if a page is being used for the page table for the running process?

[thin temple]

Wdym

Use what to determine that

Ah yeah page tables are private pages and their PteAddress points to the upper level table entry, so the resulting virtual address would be the page table's location within the recursive page table range

And you could use that to determine that yes

[hard flower]

that solves another problem I have! Nice

[bright gale]

So tldr its just the virtual address of the PTE mapping it?

[thin temple]

No it's the virtual address that the page frame is mapped to in the process

The virtual address of the pte mapping it is stored in the PteAddress field

[bright gale]

Ah ok

So like one step after that

[bright gale]

How is this not racey btw

[thin temple]

Because there are invariants stopping it from being racey

For example if they do this for a page table then the page table presumably has some reference stopping it from being paged out

They could also hold some spinlock

That's not the province of this function to worry about

[bright gale]

Oh

[buoyant thorn]

Why I have a feeling, that Kreuzer will have words to say about you touching that mm part? 😁

[bright gale]

I mean as long as they implemented it correctly

[thin temple]

Also what's with that nerd in the comments to the PR

🤓 "is it allowed?"

Nobody tell that guy that the vast majority of the reactos kernel was written in this exact fashion (understanding and manually decompiling a disassembly w/ help of debug symbols)

[autumn pond]

its sad that their project to write new mm wasnt completed

when they hired a guy to redo the memory manager

[stable topaz]

He’s retarded ignore him

[stable topaz]

We all have to start somewhere

[stable topaz]

The rules are you just can’t copy out of IDA

If it’s functionally the same because you looked at how it works cool fine

Ida Is only so useful to understand what you’re doing

[thin temple]

Insert Alex Ionescu pushlock.c with "Set a to b." type comments

[stable topaz]

I think everything Alex wrote should be thrown out

fucking bullshit file

Rewrote a bunch of code from him through all this because he didn’t understand what he was doing

[bright gale]

Lol

Is that because he just copied from ida

[stable topaz]

There’s worse example

[thin temple]

It's just a dumb decompilation job with no understanding of how the pushlocks worked

[stable topaz]

Ipi.c was really bad

Parts of Ob

[thin temple]

The comments are only there in an attempt to obfuscate this lack of understanding, seemingly

[thin temple]

This is sus and makes me understand why Axel guy made the infamous accusation

[unique lichen]

He probably followed the existing style. Some time ago (actually a long time ago) I recall there were accusations and proceedings regarding copying some code in ReactOS, but as far as I know it didn't lead to anything.

I can't remember which version of pushlock he did... He must have taken the early version before it became unfair?

[stable topaz]

Alex created the style that led to the accusation

All of the accused code is Alex

But MS also loved Alex and we were audited blah blah

[autumn pond]

its remarkable how different people express different opinions about alex

apart from the comments the reactos version of push locks looks good, but im no expert. have you guys noticed any technical problems with it ?

[thin temple]

No because it's an exact functional replica of how it was in windows in the early '00s

It doesn't have priority inheritance, there's that

[stable topaz]

My opinion is that ReactOS will never become a half functioning OS limping on every bs stupid thing windows does

And Alex is the personification of the opposite opinion

I want reactos to be compatible with windows drivers and apps.

That means we have to replicate a lot of the NT architecture,

That doesn’t mean everything

For instance the arbiter system from NT is one of the stupid over dramatizations of how handing IRQs I’ve ever seen

I understand it better than anyone on the team at this point, I’m not reimplementing it

[thin temple]

I don't think the memory manager needs to be identical to NT's internally

[stable topaz]

true, and ARM3 is a poor attempt at making something closer but sacrificing a good design

The previous Mm was buggy but a little better

Had the foundation to be something usable

ARM3 is alex’s design remember

We tried to have a contract to fix the Mm but it was all spent undoing Alex bs

[thin temple]

I would do something a little like what I'm planning on which is to make it a lot like UVM (including how the fork support works) but with NT-like page lists and working set management

[pseudo surge]

Is that the IRQL_NOT_LESS_OR_EQUAL? I kept getting it when writing a driver and couldn't understand how I was triggering it

[hard flower]

It's not complete yet

But yeah I know how to figure this out now

[hard flower]

Oh he did

I wanted to use MI_TRACE_PFNS and fixed compilation on amd64

And he told me to avoid it since it doesn't play well on SMP

[hard flower]

Oh thank god
I thought I was screwed since I've decompiled Windows 11 for the MmMapIoSpace check as well

[stable topaz]

Don’t don’t make it a decomp’ed fucntion

Just understand what you’re writing

understanding what windows does with Ida isn’t a sin

TLDR just don’t be Alex

[hard flower]

Yeah got it

[autumn pond]

because of fork?

[thin temple]

well yeah thats one reason

NT's does implement COW fork

but in a stupid way

that could be avoided by just designing fork into it better

[autumn pond]

is the dumb way something architecturally unfixable or maybe its been revised to better support wsl1 ?

[autumn pond]

i should check how it works in reactos

[stable topaz]

It doesn’t

[bright gale]

How complex is ros memory manager?

[autumn pond]

damn

actually i wonder how fork works in nt

[hard flower]

You can make your way around it after a few hours

I was initially looking at the ROS memory manager because of something else

But I think I mostly got the high level idea

[unique lichen]

In MM it's a sequence of phases and a PTE state machine.

The memory manager initializes a new address space for the child process. It can attach into the parent's context and walk the parent's VAD tree. This tree tracks which virtual addresses have been reserved.

For each inheritable VAD it allocates a special inheritance descriptor and charges quota. For each created descriptor duplicates the VAD into the child's.

Then for each PTE decides the type; for example, for fork PTE would just increment its reference count I believe.

It also needs to attach into the child's context to add working set elements. Then it links each descriptor into the parent's clone tree.

I was going to say that conceptually it's not difficult, but it's actually quite racey. And it's probably at least 6 levels of synchronization.

The kernel clones the handle table as well.

It's probably been rewritten for NT 10, but the core is still the same. Probably, at least he improved the creation of cloning chains to improve parallelism.

It's not so surprising that ReactOS might not support this, I think today the need isn't that great in a Windows-like ... Maybe for debugging.

[hard flower]

I think ROS might need it considering there are plans for the POSIX subsystem on it

[unique lichen]

PosixSpawn also can be supported. I think you could make a flag to create in frozen state and attach.

[stable topaz]

Reactos can now burn CDs

And use CD DRM in latest nightly

[stable topaz]

Congrats on getting #Canelled

https://reactos.org/forum/viewtopic.php?t=24594

I'm kidding, we replied with a massive comment that is a NOT SO tldr of "What ReactOS is allowed to do"

So if you're interesting in the subject, yeah lol

[bright gale]

exactly what i wanted in 2025

but thats kinda cool anyway, i remembed CDs being kinda complex

[stable topaz]

it's just ATA passthrough

not too bad

fun though!!

I actually like that standard a lot

you can do a lot of cool things with it

[bright gale]

ATAPI or whatever?

[stable topaz]

Actually passthrough exists in a lot of things

like USB cd drives

for example

usbstor.sys itself is a SCSI driver

It's just one of those things that "just works" on windows that you don't think about for a hobbiest OS

[hard flower]

I mean what I did can literally be done using windbg as well

🥶

[stable topaz]

bingo

that's what i do

you did nothing wrong

afaic

[hard flower]

God I'm reminded of that weird drm

[stable topaz]

[hard flower]

The function call graph was some anime girl

💀

And it used tls callbacks

Oh yeah does ROS do tls callbacks?

[stable topaz]

yes

[stable topaz]

if any luck first PR that allows the new ACPI driver to be USABLE will open before july 20th

[bright gale]

Daamn

[stable topaz]

I ended up writing a Ryzen GPIO driver out of annoyance

[bright gale]

The one required for the power button?

[surreal void]

Wasn't it also some ACPI namespace or something?

[stable topaz]

few more fixes have been thrown into latest nightly, hopefully the new ATA + AHCI combo driver will kick into high gear

because if we try to merge this ACPI work without it funnny things will happen

[autumn pond]

@thin temple sorry for pinging again, but does all that describe what you found problematic fork in nt mm vs uvm ?

[thin temple]

its more complicated than the uvm way and slower

[stable topaz]

I took off Friday and Thursday

So I can get this going

But also it’s my birthday on Saturday so

What I want for my birthday is functioning ACPI

[bright gale]

lmfao

great goal what can i say

[bright gale]

https://www.youtube.com/watch?v=RnXdazuukN8

thought id post it here as well

[rotund ibex]

Pog

[marsh folio]

cloned managarm but not nyaux 😔

[bright gale]

wdym

[tender ferry]

this says managram

no nayux

[severe quartz]

[bright gale]

lol

[rotund ibex]

doesnt that say managarm?

[tender ferry]

yes

it's a joke

people often misspell managarm as "managram"

which is also why i spelt nyaux "nayux" on purpose

[buoyant thorn]

nanogram. but, of course, mammogram is the best.

[stable topaz]

Getting ready for first server motherboard uACPI boot with reactos

[stable topaz]

UACPI did work

our VBE driver did not lmao

[hard flower]

@stable topaz happy birthday!!

[hard flower]

Idk if it's 19th over there for you
It is here

So h

[stable topaz]

4 hours till for me so close enough

Thank you

[gloomy burrow]

Happy Birthday

[bright gale]

Lol

[rotund ibex]

Happy Birthday!

[kindred cliff]

happy birthday

[marsh folio]

yea thats the issue. when nyaux folder

[marsh folio]

ok vro

[marsh folio]

HAPPY B DAY!!!

[fervent mica]

Aaaaa happy birthday hope I'm not too late

[sleek badger]

happy late birthday!

[stable topaz]

A reverse sleeper so the mb is ofc older but still uACPI handles it well

[bright gale]

what's a reverse sleeper lol

[hard flower]

Looks great outside
Outdated hardware inside

Basically the opposite of a sleeper build

Where you have an old case with brand new hardware inside

[livid marsh]

324gb of ram?

[stable topaz]

3.24

[bright gale]

ah lol

[livid marsh]

ahh

[bright gale]

is it running your new driver?

[stable topaz]

yeah it’s using uACPI, and the new pci.sys

[bright gale]

nice

how close is it to being PR'able?

[stable topaz]

Pretty close I only really have 2 regressions I need to solve

[bright gale]

thats pretty cool

[buoyant thorn]

Is this setup noisy?

[stable topaz]

No?

[buoyant thorn]

Why the ? You aren't sure or what? I asked because I haven't used these modern glassy cases with a dozen fans. Everything is on laptops. My last PC (lost with the home) had way less fans and they were smaller, but of good quality. Still the noise drove me crazy. It was not even from fans, but from air flowing through the small holes in the back side of the case. And here I see a dozen of pretty big fans, that's why I asked. Genuinely interested. So it's quiet? With all these fans spinning it doesn't do even that waterfall sound?

[bright gale]

If you look at the pic only one fan is connected out of 3 on the front

[buoyant thorn]

Well, at least 3 of the visible then. In addition to this on the right, CPU's one and on the back look spinning too. Plus, there should be something on GPU and maybe with SSD.

[stable topaz]

I’m confused why you’d think it wouldn’t be quiet

That’s why the “?”

I don’t think I’ve seen a PC case that was genuinely audible in years

[buoyant thorn]

I explained why. It's probably we have different noise sensitivity. On the scale 0 to 10, how would you evaluate its noise level. 0 being no any noise, 10 unbearable, 5 - after 30 mins getting uncomfortable.

[rotund ibex]

thats just a Waker

[bright gale]

https://www.youtube.com/watch?v=vh-icbJuQgg

@stable topaz are u guys feeling scared now

[stable topaz]

lol no

But I’m glad they’re making progress :

[bright gale]

lol

[stable topaz]

[bright gale]

nvm it doesnt have much

[stable topaz]

Minesweeper doesn’t need much

[bright gale]

is it really just this?

[stable topaz]

Ehh

Doubt

[bright gale]

lol

fake video

i mean they have a literal firefox icon on the desktop

[thin temple]

this person is a relatively notorious hack fraud

at some point they ran a "rival" osdev discord ("rival" is in quotes because nobody here really noticed they existed) with the ExectOS guy who plagiarized minoca and misrepresented "his" work on hackernews

[bright gale]

ah lmao

[hard flower]

I mean Polaris has a recipe for openrw but that doesn't mean it runs GTA 3

I've tried it before it fails

[marsh folio]

average ibuypower pc

[hard flower]

I buy problems

[scarlet nova]

btw, i forgot to ask earlier today, but what's the lore with Bob Pony? @stable topaz

[stable topaz]

I just thought what he did was kinda weird

“Let me grab someone’s exe, change it and reload it on my site”

Ignoring the Patreon stuff for a moment

Sounds like a good way to exploit your readers

Then implying it was directly from him..

Just net weird

[scarlet nova]

interesting, thanks

that is indeed a weird thing to do

and then he deleted the tweet

[stable topaz]

Indeed so it’s a non issue now I guess

[hard flower]

He is a dumbass

[stable topaz]

Lot's of insane progress lately

we're looking pretty good for interoperability

[bright gale]

Nice

[rotund ibex]

W

[stable topaz]

Doing a small PC upgrade to Imrpove reactos performance

I hope it’ll be enough for minesweeper

[rotund ibex]

How many millions of fps will you get 👀

[stable topaz]

30 because reactos can’t render more than that without crashing

Requires a gigabyte of RAM per frame

[rotund ibex]

should have gotten 128gb then

[stable topaz]

damn you’re right

[rotund ibex]

ReactOS support networking, right?

then you can probably just Download more ram

[stable topaz]

Oh, you’re right

Perfect

[kindred cliff]

put swapfile.sys on google drive and call it a day

[rotund ibex]

Uploading (*and* downloadinh ig) ram 🗣️ 🔥

[stable topaz]

this isn't really a software channel

but ill dump a few successes just because they're prep for what im about to do

[kindred cliff]

holy shit

[kindred cliff]

i can try digging up a screenshot of my game engine on reactos

[queen perch]

real

[stable topaz]

ok yall get one failure

[kindred cliff]

my game engine uses various warcrimes to function on basically every version of windows after 2000

on windows 2000 it crashes before the loader finishes

i think it's my fault for using a modern linker and not one from 25 years ago

somehow all of my render backends broke independently of each other

gonna add reactos support to the program i'm making now though

[stable topaz]

bool reactos goes hard

[sweet matrix]

😮

[sweet matrix]

I can compile apps for Windows NT 3.1 with GCC 13.1.0 and Mingw-w64 13.0.0

(though I did have to compile them custom and patch some things out)

you dont need a 25 yo linker to target old windows versions

[kindred cliff]

i don't trust mingw though, it never works for what i need

[sweet matrix]

what do you need it for

[kindred cliff]

i dunno. things i can make work fine with clang or msvc seem to be much more annoying with gcc

i think the complete lack of msvc compatibility just makes it a lot harder to write code for windows in the ways i find pleasant

not being able to use the official windows headers is a pain in the ass

[sweet matrix]

Or is it

[kindred cliff]

no it is

maybe if you don't care it's not, but it always causes me problems

[sweet matrix]

I have found the mingw replacements to work fine for me, what problems have you noticed

[kindred cliff]

phnt, a set of headers that makes working with the native api a lot nicer, completely breaks with mingw

i also remember having issues with d3d12, but that might've been unrelated

[sweet matrix]

I see

[sweet matrix]

i guess phnt was made to work with msvc primarily and that's why it doesnt work

[kindred cliff]

it was made to work with the official windows headers, because they're the most reliable

[sweet matrix]

anyway, isn't it possible to use the latest MSVC version with a different UCRT

the UCRT is open source to my knowledge

[kindred cliff]

what does that have to do with anything?

[sweet matrix]

You said, "my fault for using a modern linker and not one from 25 years ago"

[kindred cliff]

i wasn't using the crt

[sweet matrix]

oh i see

[kindred cliff]

the executable's headers caused a problem in the loader

i'm not sure why even lld didn't work, it should've

[stable topaz]

https://github.com/reactos/reactos/pull/8396

and so

:3

step 2

this allows reactos WDF to reach feature parity with windows 10 15xx

[kindred cliff]

woah

[bright gale]

damn\

[stable topaz]

technically we can go up to later windows too but its done like this for simplicity

although after some more progress ill add a flag to build with windows 22xx compatibility

[stable topaz]

https://github.com/reactos/reactos/pull/8359

support for modern windows 10 audio codecs

[kindred cliff]

woah wtf

[stable topaz]

nice

[kindred cliff]

if i was gonna write a new subsystem, i should just study csrss, right?

[stable topaz]

yeah

[kindred cliff]

ok

is there alpc yet or not quite

[stable topaz]

No.

[kindred cliff]

ok

does it actually matter that much which i use, even if i wanted to port it to windows 10?

[stable topaz]

Huh?

what are you asking me loll

[kindred cliff]

like csrss uses alpc on newer versions, right? but it doesn't matter if i use lpc instead?

[stable topaz]

oh uh

no really

[kindred cliff]

sorry about the phrasing the first time my brain is fried

[sweet matrix]

so you're using those .sys files from rectos and they work in w10?

[stable topaz]

yes

and windows 8.1

and windows 8

[sweet matrix]

that is crazy

[stable topaz]

and windows 7

and windows vista

[sweet matrix]

wow

[stable topaz]

:P

and its win10 15xx feature parity

to it's a massive compatibility boost

[stable topaz]

already effects things like this

wich adds modern audio support

[sweet matrix]

well that's one way to make sure your implementation is like windows

[stable topaz]

that's what we usually do

audio, kmdf, is all reactos here

i even did a discord call using these drivers!

[stable topaz]

same here but also trackpad and i2c support

[sweet matrix]

holy crap thats awesome

but wait

build 10586

apparently thats a pre-RTM build

...of the november update

so maybe its okay

[stable topaz]

its not pre-RTM

[sweet matrix]

i guess betawiki is wrong then

[stable topaz]

no there's just also a pre-RTM build

that the same ID

[sweet matrix]

oh okay

[stable topaz]

https://betawiki.net/wiki/Windows_10_build_10586_(th2_release)

Windows 10 build 10586 (th2_release) is the official RTM build for the Windows 10 November Update.

[sweet matrix]

i see i see

[stable topaz]

This matters because what was introduced in win10 even in RTM for hardware support

is MASSIVE

and nothing quite did that big of an overhault

so us having this kmdf support means most drivers (stuff that's not wifi,ethernet/gpu) will load now

since most stuff has been kmdf for 15+ years

[sweet matrix]

and that includes modern audio drivers right

and storage?

[rotund ibex]

Windows Vista mentioned

[sweet matrix]

i guess Windows Vista is the one that introduced KMDF

[stable topaz]

yup

[stable topaz]

here's vista with our KMDF and hdaudiobus

(technically bringing win10 functionality to vista LMAO)

It's hard to convince people how important and huge this is

but yeah

its key

[sweet matrix]

that is actually insane ngl

[stable topaz]

in fact, my system right now is using reactos directx, kmdf, hdaudio, a part of WDDM, and the NetAdapterCx (kmdf module for networking)

and as you can see by the fact im still typing

zero stability issues

i just had to compile with win11 compatibility

all of these are kmdf

So yes while all this work has been taking years

it's paying off

very quickly now

and while ros itself as an OS may be lame for awhile in usermode. Im confident hardware support is going to be pretty decent in 2026

[sweet matrix]

i can't wait

you could finally somewhat daily drive ROS on modern hardware

even if userspace leaves to be desired

i also wanna work on my OS but i need to figure out what next

[stable topaz]

whatever is cool

[rotund ibex]

pog

[stable topaz]

It’s kinda rough but I’ll throw it here too

Here vaguely the map / plan of how I’m splitting up all the work I’m doing

Everything on the chart should be in mainline before February of next year ish~

[bright gale]

nice

[stable topaz]

Software renderedVR anyone

[queen perch]

amazing now I can be a furry in reactos

I know the answer is 9000% no but does any vr headset work on ros

[stable topaz]

no but the trackers seem to do something

So.. eventually?

Biggest issue with headset is GPU

[vapid elk]

Just pretend to be an RTX 4090 but it's secretly just software rendering, so the app thinks you have like a 1Mhz 4090

Definitely causes 0 problems and fixes everything

[stable topaz]

i lied

i just went down a rabbit hole seeing some stuff for my blog post when someone in the discord was talking about VRHeadsets.
Maybe ill add this little note to the post but its already kinda technical
but there's a KMDOD called IndirectKMD

it turns out all RDP Sessions, Virtual Desktop software, OBS, and the occlus rift / HTC all use this to render

The WDDM driver is a KMDOD so easy to use for us,
But all the sensors and usb components are just usbvideo, usbaudio, and KMDF?!

Anyway here's an occlus rift accepting the signal from reactos

and yeds

there's a hook to dispatch the bug check screen over the driver

So you can see your favorite reactos bug checks in VR 🎉 🎉 🎉

LFG BUGCHECK OS

[queen perch]

Amazing

[rotund ibex]

lfg?

[tender ferry]

let's fricking go

[stable topaz]

X64 + SMP + working hd audio

[sweet matrix]

but why are you in 16 color mode

[stable topaz]

a few reasons

But tldr x64 uses a fallback mechanism to reach into VGA,
This mechanism is required to behave a certain way for gpu drivers to work

We can’t fix VBE etc until we extend this mechanism which is very complex

And we can’t just hack it without breaking GPUs

Hence

[stable topaz]

holy shit i was tracing a bug

and it went all the way down to pushlock.c

finally a reason to burn that file to the ground asap

[queen perch]

what was the bug

[stable topaz]

It causes the OS to infinitely stall

[stable topaz]

Whenever a lock is held by pushlock it freezes the entire OS and gets exponentially worse for every CPU added

Ladies and gentlemen and neutrals:
ReactOS

[bright gale]

But it had very useful comments its impossible for it to be buggy

[stable topaz]

sir please understand this is exactly how windows does it! I am good programmer look at my comments - source: trust me bro

[bright gale]

Lmao

[stable topaz]

I created a implementation Windows 8+'s WaitOnAddress APIs

and now that im analyze windows behavior a bit I noticed you can hold a push lock over the address

that's an interesting concept

oh wait that's super useful actually wtf

[queen perch]

turnstiles in ros when

[stable topaz]

maybe

[autumn pond]

interesting

did you find out whats the problem was

[stable topaz]

I’m treating Alex’s code as a resource I can’t use and I’m going to rewrite the entire file from scratch

[autumn pond]

like is it a bug in windows 2k3 as well?

[stable topaz]

idk this is reactos not windows 2003 :)

[autumn pond]

yeah its just i heard that its a precise reimplementation of 2k3 pushlocks

with some absurd comments

[stable topaz]

Apperently it's not

since it doesn't work

But honestly who cares its clear the person that wrote it didnt understand what they were writing

No one should be shocked it doesn't work

It's likely some one line breaking it tbh

not going to find out

[autumn pond]

i remember you posted screenshots with working firefox which was super cool, that work hasn't been merged yet ?

[stable topaz]

It's making it's way in

that's over 200+ commits over master at the time

now it's about 120~

[stable topaz]

https://reactos.org/blogs/investigating-wddm/

[bright gale]

nice

[rotund ibex]

Common ROS W

[stable topaz]

@bright gale

New Edition

[bright gale]

Not bad

[rotund ibex]

ReactOS Server™ Teaser?

[stable topaz]

[rotund ibex]

I'll take that as a yes :D

[stable topaz]

A lot of stuff needs to be done before we do any real windows servers features but the Se subsystem of our kernel is ready for it

So eventually ™

[rotund ibex]

https://tenor.com/view/jfk-clone-high-i-like-your-funny-words-magic-man-jack-black-gif-18659433

[bright gale]

numa support in react os

[stable topaz]

That’s easier than anything else windows server specifically does

💀

[bright gale]

lol

[stable topaz]

Someone was like looking at Active Directory

And i am pretty sure if we ever actually tried MS would actually annihilate us

[green wraith]

i think AD is supported by other oses too, no?

[bright gale]

that sounds so boring to implement ngl

[stable topaz]

To some degree

[stable topaz]

Well the worst of things like that or group policies is done

we have a Security subsystem

https://github.com/reactos/reactos/tree/master/ntoskrnl/se

So the kernel itself can do it

But even our winlogon has hacks in there because none of the rest of the OS is ready for taking Se seriously

[bright gale]

makes sense

its probably not a priority to get security stuff in order

[green wraith]

i think you can integrate AD in ROS

don't think MS would care that much

especially bc other oses have support for it already

[stable topaz]

But it’s useful for kernel to have it since the basic idea is there for stuff to extend from it

Like part of what has been taking us so long for ntfs

Is mostly because ntfs actually behaves properly on ROS

[stable topaz]

funny progress bits

[bright gale]

Not bad

This is like 1% of your computers?

[stable topaz]

lol

yeah

maybe less at this point man idk

[rotund ibex]

thread not dead

[stable topaz]

Well my new PCI.sys supports resizable bar and now i actually was able to test it

and the WDF PR is now merged into the nightlies

[stable topaz]

the vga fallback for x64 has been resolved so VBE runs properly

[stable topaz]

ugh the new PCI driver is 85 source files

[green wraith]

oof

so many

[stable topaz]

Okay so

Resource management is improving rapidly
With videoprt

[rotund ibex]

Progrss

[delicate pewter]

Chill

Stop cooking

-# keep cooking btw

[sleek cipher]

Wow

[livid marsh]

based gpu

[stable topaz]

something you haven't seen part 7

[sweet matrix]

Multi-monitor support?

[stable topaz]

yeah

can actually see it usable somewhere thanks to the stuff being fixed for resource managment

[bright gale]

That's cool to see

[stable topaz]

Massive work has been done

GTX 960 running with 3D accel (XP driver 368.xx)

8800 GTS (driver 181.22)

https://youtu.be/VlGrngbYMUM?si=bNnJteyjSDWSR0_I

vista-win11 Vbox WDDM driver on reactos

Vbox 7.1

[stable topaz]

Oh this low 50 FPS cap is just because Vsync

[gloomy burrow]

V cool

[bright gale]

Damn

[livid marsh]

based gpu

[stable topaz]

Starting from left
INTEL GMA 945
NVIDIA 8800 GTS
AMD Radeon HD 7530G

[dusty cobalt]

FUCK I need to start cooking

[stable topaz]

yeah when reactos starts looking good you know its time to lock in

[dusty cobalt]

have you tested any nvidia GPU that uses GSP?

(Turing+)

[tender ferry]

i'm pretty sure it's running actual nvidia drivers so it probably doesn't matter which gpu you use, it'll work anyway

but the drivers are 12 years old so who knows, maybe some stuff changed

[rotund ibex]

Holy shit alive

[stable topaz]

Nothing with 3D accel yet

I’m still perfecting my WDDM stacks basics

[stable topaz]

This working was kinda a rush job

Will be another 6 months until I get WDDM 3D accel on hardware at least

Too many other things to do first

But hey

The next release will have xpdm ( XP AND before) gpu drivers

Which is enough for some pretty cool stuff

[stable topaz]

VFIO passthrough

GTX titan X on reactos

[stable topaz]

GPU acceleration, sound, Ethernet all at once

[stable topaz]

Of course let's be clear

ReactOS IS reactos

so in the sam vein :) people are starting ot use the latest work on super old hardware

[sweet matrix]

so fucking based

[gloomy burrow]

Very cool

Can ReactOS run something like Windows' Device manager?

Or any other system programs

[bright gale]

Cant see why it wouldn't

[vocal needle]

Do you got HL2 working?

[rotund ibex]

time to garger all my friends to find the oldest possible pc where reactos boots

[stable topaz]

very old

Pentium 1 64mb of ram

[sweet matrix]

i almost have an older build

pentium 1 and 16 mb of ram

the only reason i "almost" have it is because i don't have its hard drive caddy so i can't use hard drives with it

(i'll have to hackily solder behind the port or something)

[stable topaz]

thats funny lmao

[sweet matrix]

(or find a caddy or adapter, but those are rare as fuck)

i "didn't spend that much" on it (only about 20 euros or so)

but the fact it has no hard drive whatsoever is kinda bollocks

it has a floppy drive

of who knows what quality because i have neither floppy disks nor a USB floppy drive to write them with

[stable topaz]

https://github.com/reactos/reactos/pull/8662
Next modern audio PR is open

[autumn pond]

why jira issues mentioned in the pr cant be opened

[stable topaz]

I think he just made the link wrong

[stable topaz]

So much work talked about here

but why does it matter? whats this all leading to

This is a ryzen embedded platform booting ReactOS with UEFI + GPT

ReactOS for years has had bugs and assumptions across it's code base on how hardware resource managment work, and inorder to actually make a new acpi driver, ive had tounderstand how this works in Windows Vista (our target) and Win10 (to make sure we dont need to redo work)
And of course reactos now.

For example videoprt (the driver that controls video drivers) had some assumption about how GPU IRQs are handled.
this lead to any attempt at trying ot fix the root cause blowing up the entire Operating system

our uefi bootloader in the nightlies now behaves good,
and many MANY other fixes have happened including the audio bus, new KMDF stack to where the foundation of ros isnt going to fallapart with all my work merged

It's been an uphill battle for sure, but things are looking good for the next step

[stable topaz]

minimal integration with our shoddy PCI driver for _PRT handling

This actually works great!

And means i might be able to get uACPI's logic merged faster

[bright gale]

nice to see

[stable topaz]

i do have a question for you @bright gale

vmware has some nasty logs from uACPI itself

have you tested vmware much?

[bright gale]

yeah it worked fine, i think this is you returning NOT_FOUND from io_map or something?

[stable topaz]

hm

oh i see

thank you

[bright gale]

npnp

[stable topaz]

nice

much better

[bright gale]

nice

was it vmwares aml just probing for present pci devices?

aml usually does that by checking pci reads against 0xFFFF

ur not the first to run into it, i might add support for NOT_FOUND in uacpi in this case i suppose

[stable topaz]

yeah pretty much

[stable topaz]

nice

64bit on realhw

both SATA Drives are detected and working now

but USB is still failing

massive progress though, 64bit finally escaping the VM

[bright gale]

nice!

[stable topaz]

yeah and with the GTX titan working now

that's going to be amusing

[bright gale]

can finally play crisis properly

[stable topaz]

12gb of vram

[bright gale]

big

[stable topaz]

more than 32gb of ram usable now in 64bit as of recent too

and SMPs only remaining issue was this so

crysis "how it was meant to be played"

[bright gale]

exactly

[stable topaz]

oh

I didn’t post my new toy

[bright gale]

it has certainly been through a lot

[stable topaz]

yup but it was free

ASUS G75VW

My new reactos daily driver

[bright gale]

not bad

[stable topaz]

32gb of ram

It has full 3-D acceleration of course

So I’ve been playing through dragon age: origins

The progression this year and the last from the start of this channel is hilarious

[bright gale]

indeed

[stable topaz]

and for the first time

ever

64bit reactos on real hardware (without ramboot)

[stable topaz]

Thank you uACPI

[bright gale]

Damn a whole 800 mhz

[stable topaz]

a wholeeee 800mhz

uefi just boots to usermode without any hacks or weirdness now

which... duh

lol

its uacpi

[bright gale]

nice nice

workqueue deadlock looks scary tho

[stable topaz]

no uefi win32k framebuffer driver

so it just infinitely waits

[bright gale]

Ah

[stable topaz]

now let's see if installing the gpu driver will work

[stable topaz]

optiplex 9020

[stable topaz]

past 8 cpu limit (so just fixing xapic lol....)_

[rotund ibex]

Vista

[stable topaz]

storage

more fixes on this now that uACPI lets me test on UEFI class 3 machines

another new test system

[balmy wind]

ooo hawt

I wonder if I should try to install it on my system. it's similar to your new test machine

[stable topaz]

eventually itll work, lots of stuff to do still though

[bright gale]

cool stuff

[stable topaz]

NVMe and xHCI in the same iso (using uACPI ofc)

[naive hearth]

ReactOS is coming quite far since the last time I’ve seen it a few years ago :o

Btw is there any plans to implement support for a FRED driver /hj

(Basically better IDT on very new hardware and isn’t too bad to implement)

[stable topaz]

I dunno, we're inching to modern hardware little by little though so maybe

[naive hearth]

Mind if I share my driver of it? It does some strange conversion to hook into the make IRQ systems but it’s relatively simple

[stable topaz]

go for it lol

[naive hearth]

Just needs a 64 bit aligned stack and it pushes everything you need there and you have to change a few entries to and from userspace because it forbids swapgs and iretq and sysexit

System calls and everything Goto one entry and interupts Goto the same handles but with seperate ones for ring0/ring3

It’s a very neat little thing they have

https://git.evalyngoemer.com/evalynOS/evalynOS/src/commit/198a270da303cb67569ccd8871b8012513ebd35e/src/kernel/drivers/x86_64/fred/fred.c
https://git.evalyngoemer.com/evalynOS/evalynOS/src/commit/198a270da303cb67569ccd8871b8012513ebd35e/src/kernel/drivers/x86_64/fred/fred.asm

My driver ended up braking after some changes so I’m gonna fix it soon™ and cleanup the conversion logic layer

That’s the commit where it fully works in SIMICS though

[bright gale]

implement it for ROS

[naive hearth]

I very well would but I’m not sure if I’d qualify

[stable topaz]

pretty interesting though

[naive hearth]

I don’t remember any windows code but I know my eyeballs have glanced at it at times

The most I remember are like one or two comments that are curse words

I’d very well help with it though

[stable topaz]

really the rule is you can’t work on anything you’ve seen

[naive hearth]

Well I mean I certainly havent seen any windows FRED implementations

I don’t think they even support it yet

[stable topaz]

or probably most of windows in general

[naive hearth]

[stable topaz]

So most people are fine

They just don’t realize it

[naive hearth]

Hey how hard could it be to try and get this working

[stable topaz]

We barley do APIC right I’m trying to fix it

[naive hearth]

My APIC and x2APIC driver works

I could try and help there lmfao

[naive hearth]

And by works i mean it does the bare minimum but i got time

[naive hearth]

So I can’t work on ||poop.txt||

[stable topaz]

Truly end of the world

[naive hearth]

https://tenor.com/view/bucky-dusted-infinity-war-thanos-snap-gif-14193636

But yeah if I get some free time I’d love to help work on this now that I know I’m actually allowed too and I got some OSdev experience

I was under the impression that if you had ever seen anything about windows internals you would be barred

[stable topaz]

just some scrutiny

[naive hearth]

Hm?

[stable topaz]

conversaions like this

i mean

is what happens

if youve seen anything

the "ok but are you working on what you've seen?"

etc

[balmy wind]

oh cool so just looking at the build scripts doesn't count :)

[stable topaz]

we dont even build remotely like windows

so

[naive hearth]

All I’ve seen is the meme shit tbh 💀 and like maybe some code but I legit could not even tell you one line of source code

[stable topaz]

yeah well

[naive hearth]

Well I could guess and say there is one “int i = 0;” and an “i++;”

[balmy wind]

the closest I've gotten to looking at actual source code is probably the officially supplied PDBs from MS's symbol server 🤣

[stable topaz]

officially supplied PDBs from MS's symbol server 🤣

Which we use to reverse sometimes

so

[balmy wind]

yea exactly

[naive hearth]

Welp guess I’m gonna send patches at some point now

This will be fun especially when taking breaks from other projects

I have the special FRED role for a reason and it’s my job to send patches for it to everyone

It’s like the only thing I got to my name

are yall doing x2apic yet btw?

[stable topaz]

I think someone started redesigning how our APIC code works to support both xAPIC and x2APIC

But i havent touched that yet

[naive hearth]

Because Intel is removing the xapic and won’t support it being used as the default device on systems that still have it

[stable topaz]

yeah

[naive hearth]

It’s pretty much 1:1 to the MMIO minus some things make it easier

[stable topaz]

yeah but there was something windows does that made us have to pause

idk right now

sleepy

[naive hearth]

My kernel does two seperate implementations but I could merge it if I wasn’t partly lazy and trying to make simple to understand code

Would this channel be aptly renamed to ReactOS’s Modern Hardware Wins?

Since it seems a lot of actual good progress has been being made

[stable topaz]

there's a longgggggg way to go stilll

like if you read through the history we recently got up to the GTX 960 and titan x working

GPU accel wise

[naive hearth]

I wonder how close the 6600xt is

I know AMD is better somtimes?

[stable topaz]

well windows had a transition in their driver model for GPUs

there's 2 dramatically different models

mainline ros only supprots one

I started working on the other

[naive hearth]

Damn

[stable topaz]

But once we support that we basically get all Win10/Win11/Win7 GPU drivers easy

[naive hearth]

May be worth making this more clear on a blog post or somthing

[stable topaz]

Like i got display out on my RTX 4060

[naive hearth]

I know a few people who had the same misconception

[stable topaz]

at some point

that's the plan

[naive hearth]

Ahh nice nice

[naive hearth]

:O

ReactOS may be close to being able to run anticheat games and save everyone from windows 11

[stable topaz]

well

there's a lot of stuff there that's happened but it's intentionally left out of this thread

[naive hearth]

:O

[stable topaz]

NT6 apps are lightly toouched on here because we have been pretty quickly adding NT6 APIs but they're still disabled in the nightlies by default

so everything LOOKS like win2k3 stilll, but internally its very different

I mean if you consider the last year of work
ReactOS now supports

• Win10 Audio drivers
• Win10 WDF (ACPI Devices)
• Full XP/2000 GPUs
  etc

so stuff is progressing fast

[naive hearth]

Yall are doing some really insane work here

[stable topaz]

but a lot of things are disabled in nightlies

thanks!

[naive hearth]

I could never get this far myself

[stable topaz]

only takes 30 years

[naive hearth]

Would ROSSL (WSL) be a possibility at some point

[stable topaz]

yes

[naive hearth]

but a lot of smart people too

I haven’t even got bash yet

(Though that’s due to my own footguns lol)

[naive hearth]

Ooooo

[stable topaz]

yeah there's a path there

also we can run the legacy win2000 posix subsystem too

[naive hearth]

Oooo

[stable topaz]

dunno why you'd want that

but you could

[naive hearth]

I was gonna ask about that lmfao

[stable topaz]

we had our own

i not long ago tried to revive

it was a bit cooler

had a funny super power

could compile x11 apps and had an x11 server over win32

[naive hearth]

God damn that’s cool

Also the one thing about FRED is only Intel SIMICS supports it and bochs has a very botched implementation of x86-64 as a whole so it would be a while for QEMU to even support it

So I wouldn’t look at it for a bit unless you wana use Intel SIMICS lmfao

(I did make a guide for it tho if needed)

(I’m just too lazy to use it rn)

But yeah I wish yall luck on this

This is a very cool project

[stable topaz]

thank you!

[balmy wind]

No shot? That's awesome

[stable topaz]

https://x.com/the_darkfire__/status/1610359609923088384?s=46

The_DarkFire_ (@The_DarkFire__)

Fear it

X · Jan 3, 2023, 7:37 PM

yup this is from the NT resource kit and win2000’s posix.exe and psx

[balmy wind]

Your fork, iirc it was the most recent UEFI branch, is meant to build with the latest ROSBE right? I had some assembler issues (one was in acpi.S I believe) with it not accepting h-suffixed immediates

I can get back to you on the exact line and error in a few hours

[stable topaz]

I wouldn’t try to build anything from my fork 💀

It’s such a mess

[balmy wind]

Well, it's the easiest path to trying to run ROS on my hardware :)

[naive hearth]

I wonder if ROS would boot on my desktop with a 5700x and 6600xt and 32gb ram?

I do have other devices though

Like this thing

[balmy wind]

I hope your sata controller isn't buggy, if that's what you're using :D

[stable topaz]

it’s getting pretty close but needs a few more things

[naive hearth]

My desktop has no SATA drives

I may just end up wiping my windows NVMe just for ROS

[stable topaz]

oh no don’t I only got NVMe working ago 15 minutes ago

[naive hearth]

Well I’d partition it into a 128g one for ROS and reuse what’s left for extra game storage ;p

[stable topaz]

ROS will eat your entire drive sometimes

randomly

Well it used to

I have no idea if it still does

[naive hearth]

It will be fineeee. Can’t be worst than windows vibecoding an nvme driver that fully breaks the drive

[stable topaz]

It can get close

[naive hearth]

Atleast it’s made by a human XD

[balmy wind]

Iirc you have hardware pretty similar to mine, but I can guinea pig if you ever want. I do wonder if I need that COM header to serial adapter though lol.
Ryzen 9 5950x, Gigabyte Aorus B550M Pro-P (the one with realtek 2.5g eth), Radeon Rx 6600xt, Nvidia GTX 760, 2x nvme, 1x sata 3. Intel ax210 wifi

[naive hearth]

Hey we all have almost the exact same desktops lmfao

[balmy wind]

It's great hardware ❤️

Intel my beloved (for anything but cpus)

I wanna get a 7900xtx or smth though, when whatever the next gen is drops

[naive hearth]

I like the n100 they make tho

It’s good for immich & jellyfin

[rotund ibex]

All ROS progress is pog

[stable topaz]

basic webcam support has been implemented

[bright gale]

face reveal

[rotund ibex]