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

[calm latch]

My kernel is having a giant skill issue returning an integer

[kind mantle]

real->long mode any% speedrun

[calm latch]

I'm dumb

I was returning u16

And 1MB is 20 bits

[north holly]

lol

[fiery turtle]

Return a u20

[kind mantle]

Yes let's return not only an integer type that doesn't exist but also one that doesn't align to bytes

[fiery turtle]

Zig moment

[calm latch]

DSP moment

[gentle peak]

u145

[north holly]

u6

[kind mantle]

PipelineC moment

[calm latch]

This is promising

[north holly]

uACPI

[fiery turtle]

1

[frank canopy]

if its not in a packed struct then zig uses the next representable size and adds the branching etc to implement the right semantics for the specified size so its fine

[calm latch]

I know

[fiery turtle]

Makes sense

[kind mantle]

Me when true()

[kindred beacon]

next do S0ix states

[fiery turtle]

I'll let oberrow figure it out

[north holly]

yeah he can have fun

idk how he managed to do that

[frank canopy]

also good news i updated to uacpi 2.0 on imaginarium and it still compiles

[fiery turtle]

That s cool

[calm latch]

I have a laptop which (only) supports them

[kind mantle]

also good news I updated to uacpi 2.0 on BadgerOS like a month ago and BadgerOS is still useless

[north holly]

what

[calm latch]

So I will be first

[north holly]

uacpi 2.0 didn't exist a month ago

[gentle peak]

uacpi 2.0 was released like a few days ago

[fiery turtle]

Back to the future

[kind mantle]

is funni joke

[frank canopy]

2 days ago yeah

[fiery turtle]

I'm currently running uacpi 467.0

[north holly]

https://github.com/uACPI/uACPI/releases/tag/2.0.0

[kind mantle]

wait 2.0

[north holly]

yes

[kind mantle]

man I really should stop doing stuf at 1am

because my brain melts

anyway sure I'll update to that soon™️

[calm latch]

Wait a second

This is what I wanted

[fiery turtle]

When more badgeros work

[calm latch]

It's %cs:1

[kind mantle]

Probably later this week? I was a bit busy the last couple weeks.

[fiery turtle]

Nice

[kind mantle]

Still staring at AHCI.

[north holly]

x/i $pc

try that instead of disas

[fiery turtle]

Gdb not good at real mode

[frank canopy]

also happy to report that i can fully delete my uacpi_libc.h now that 2.0 is out whee

[fiery turtle]

That's nice

Bochs debugger is much nicer for this

[calm latch]

But I need to fix PCI for uACPI init

And I can't be bothered

Since ... I would need to implement a bunch of stuff because I don't have a shared mutex

[fiery turtle]

F

[north holly]

F

[fiery turtle]

Oberrow should at least send u patched seabios

[north holly]

take it

-bios bios.bin

to qemu

[frank canopy]

what specifically is patched with this?

[frank canopy]

out of curiosity

ah ok cool

[fiery turtle]

It fails to re enable ecam on the chipset

[frank canopy]

mhm

[fiery turtle]

Since it tries to re enable it via ecam

[frank canopy]

lmao

[calm latch]

Through this I can see that I have a skill issue with gdt

[fiery turtle]

Debugging by instruction counting

Also obligatory GDT INIT.... FAIL

[calm latch]

is there no way to see gdtr contents in gdb?

[north holly]

just do monitor info registers

[calm latch]

Ok I'm dumb

forgor to shift ecx by 4

[calm latch]

Also instruction sizes

[left orbit]

uacpi_status uacpi_set_waking_vector(
    uacpi_phys_addr addr32, uacpi_phys_addr addr64
)
{
    struct acpi_facs *facs = g_uacpi_rt_ctx.facs;

    if (facs == UACPI_NULL)
        return UACPI_STATUS_OK;

    // ...
}

why does it return OK if there is no facs?

[calm latch]

And I also confused left and right again

[left orbit]

shouldn't it be like unsupported or something?

[frank canopy]

@fiery turtle how does https://github.com/Khitiara/imaginarium/blob/main/src/krnl/hal/acpi/uacpi/LICENSE look to you?

idk if my wording is any good

the shims.zig file i exclude from the clause is the file of all my kernel-specific exports for the kernel api

[calm latch]

I think I've reached protected mode?

(gdb is really confused)

(which is actually compat mode)

(why am I having so much skill issue with a simple asm trampoline?)

[frank canopy]

nice

[frank canopy]

tbh i was originally planning to make my uacpi zig bindings a separate repo or at least separate folder outside my kernel repo but the callbacks/kernel api proved more annoying than i wanted to deal with to make feel right

i think ive got some ideas though for making it a bit cleaner

the simplest is to just say let the kernel do its own uacpi_kernel shims and only provide type and api bindings and not anything for making the shims prettier but idk

[fiery turtle]

Stole acpica interface iirc

[fiery turtle]

Lgtm by im not a lawyer ofc

[fiery turtle]

Nice

[fiery turtle]

Yeah dunno

[calm latch]

Is there a way to discover legacy PCI io port with uacpi?

Like so I don't have to hardcode 0xCF8

[fiery turtle]

PCI._CRS

It will have decodable IO ports

[fiery turtle]

this is from your pc

[vast kestrel]

Is there any spec for what crs to expect

[fiery turtle]

dont think so, but it's a range of busses covered by this bus, range of io and memory it's able to handle for BARs, and resources that can be used for talking to it

[vast kestrel]

It annoys me so much it's basically a guessing game with the crs

[fiery turtle]

yeah thats kinda annoying

[calm latch]

PCI FIRMWARE SPECIFICATION

Btw

The address that a processor uses to access a device is not necessarily the same as the address stored
in the device’s BAR. The translation (_TRA, _TTP, and _TRS) information is not available to the
operating system until after the operating system brings up the ACPI interpreter. The operating
system must wait until after the ACPI interpreter is up to determine the address at the processor
side associated with the BAR registers configured by firmware

(for people who refuse to support ACPI)

What does this mean?

Before re-enabling a resource, the operating system must reprogram the BAR register using a value that falls within the range reported in the _CRS descriptor of the parent Host Bridge

(what if I implemented PCI hotplug in pmOS? )

I have several PCs with thunderbolt/USB 4...

[calm latch]

Which table is this?

I think I'll redo my PCI enumeration memes...

[fiery turtle]

dsdt lol

[fiery turtle]

which part?

[calm latch]

Idk, all of it

[fiery turtle]

if u enable memory decoding on a pci device, its bar must be configured to a range from _CRS

[calm latch]

Hot-Plug Surprise Control

[flat badge]

the BAR must also be configured within the window of its bridge (and that also applies recursively to the bridge window itself)

[fiery turtle]

@gentle peak why did u do this btw?

also when proxima thread

[gentle peak]

don't remember

proxima thread idk

[fiery turtle]

damn

[kind mantle]

pls gib proxima thread

[fiery turtle]

Ikr

[median crest]

taking the heap for a spin

[surreal current]

to make it possible to use the heap from interrupt handlers?

[kind mantle]

Most allocators don't allow it

[surreal current]

thats the normal constraint that causes someone to turn a blocking mutex into a spinlock so im just guessing why they did that

[kind mantle]

Because if e.g. a thread has the spinlock and is interrupted, that interrupt cannot then also take the spinlock.

[surreal current]

well

thats why you disable interrupts

when you hold spinlocks

its cooler to have everything possible be mutex guarded though and use worker threads if you need an interrupt to cause an access to something mutex-guarded

but for instance nonpaged pool in NT is guarded with a spinlock specifically so that interrupt handlers (DPCs technically but same thing for these purposes) can free blocks of memory

[vale isle]

i assume that's ziggy object notation

RON is the best for debug printing

[calm latch]

I've looked at Linux and FreeBSD and still have no idea how they discover 0xCF8/0xCFC

Like do I just assume it exists if there's no MCFG or _CBA?

[left orbit]

zig object notation, but yeah :^)

[frank canopy]

speaking of zon i may look into splitting my zig uacpi bindings out into their own repo either today or tomorrow

wonder if theres a way to turn a subdirectory into a repo in git without losing history

[left orbit]

probably not without altering it at the very least

[frank canopy]

i just want to keep the history of that directory

[left orbit]

unless u are very strict about not touching kernel and zuacpi in the same commit

[frank canopy]

i am not strict about that no lol

[vale isle]

git filter-branch

[left orbit]

yoooo

[frank canopy]

either that or filter-repo, im not sure which and id have to look into the exact command to use but yeah

ive got filter-branch experience too lmao

albeit many years ago

[vale isle]

I have no idea what filter-repo does but sounds cool

probably more of what you're looking for

[frank canopy]

they kinda split up filter-branch after the last time i used it

filter-repo is th other half of that split

oh wait no filter-repo isnt official

im stupid

its a third party addon command

i didnt realise it wasnt official because git officially recommends it lmao

[vale isle]

[Keep vs. remove] Instead of just providing a way for users to easily remove selected paths, also provide flags for users to only keep certain paths. Sure, users could workaround this by specifying to remove all paths other than the ones they want to keep, but the need to specify all paths that ever existed in any version of the repository could sometimes be quite painful. For filter-branch, using pipelines like git ls-files | grep -v ... | xargs -r git rm might be a reasonable workaround but can get unwieldy and isn't as straightforward for users; plus those commands are often operating-system specific (can you spot the GNUism in the snippet I provided?).

yes you want filter-repo

[frank canopy]

yeah filter-branch kinda sucks

filter-repo just didnt exist last time i used filter-branch

[fiery turtle]

Damn

We can put it under the uacpi org

If you wish that is

[frank canopy]

yeah im good for that

thinking MIT license for zuacpi, since thats the same as what uacpi proper does

[fiery turtle]

That would be fitting indeed

[frank canopy]

but yeah ill look into filter-repo-ing it out of imaginarium once im done making cookies

[fiery turtle]

Damn which cookies

[north holly]

I usually don't accept the cookies of websites

but these cookies I might

[frank canopy]

gluten free chocolate chip

[fiery turtle]

Nice

[frank canopy]

(i have celiac so has to be gf)

[calm latch]

@fiery turtle do you have dumps of AML with multiple host bridges?

[fiery turtle]

yes

[calm latch]

thanks

[fiery turtle]

looks like second bus _CRS just returns the bus range

[calm latch]

Yeah

And it's not a PCIe bus

So no MCFG?

But I think I'm just supposed to use IO

[fiery turtle]

this bus is covered by mcfg

[frank canopy]

if its present mcfg will probably cover all host bridges yeah

[calm latch]

But I'm trying to target systems that don't have PCIe

E.g. bochs

[fiery turtle]

so grab the io port resource from the pci bus

0xCF8 is there

[calm latch]

yeah

also

How would uACPI handle hotplug surprise events?

Wouldn't this need to be removing resources?

[fiery turtle]

thats not how pcie hotplug works

also these resources are static, they just describe memory ranges you may use for BARs

PCIe hotunplug works by first notifying you, then you do the unplug work, and then let the system know the device may be removed by the user safely

if done via acpi u call the _EJ0 method

[calm latch]

But what about something like thunderbolt

[fiery turtle]

thats pcie

[calm latch]

but user can just yank the cable out (?)

[fiery turtle]

do u have a dump from that laptop or whatever

[calm latch]

My lenovo laptop has USB 4

[left orbit]

i imagine it works similar to usb?

[calm latch]

And the macbook had firewire and thunderbolt

[left orbit]

except different protocol that's much faster

[fiery turtle]

yeah maybe its just usb, idk

[left orbit]

but there's probably still a controller that manages all of that, no?

[calm latch]

My old laptop with Thunderbolt 3 died so I can't take its dump

[left orbit]

even pci has to go through some other controller that translates usb->pci

(probably the usb4 controller)

[calm latch]

USB 4 is a free version of Thunderbolt 3

[left orbit]

or thunderbolt controller or whatever

so that controller is in charge of actually notifying you of a device disappearing

[calm latch]

I think you can take old thunderbolt 3 devices and they should just work with USB 4

There's also cardbus

(I have a couple of laptops with that)

[frank canopy]

ok i really like filter-repo

this slaps

ok heres a question

should zuacpi handle the relevant c headers or no?

im leaning no

[fiery turtle]

What does that mean?

[torpid root]

zuacpi is a great name lol

[frank canopy]

should it bring in the freestanding C headers to make uacpi compile or should it rely on the consumer to bring in c headers

im leaning rely on the kernel

[left orbit]

yeah same

what if the kernel already has some headers for exactly this purpose

[fiery turtle]

Yeah

[frank canopy]

yeah

[left orbit]

or you could put this behind an option

like, if the kernel does have them it can tell zuacpi to fuck off

[frank canopy]

option is kinda where i was leaning but im not 100% sure overall

[left orbit]

only downside is that they need to be fetched regardless, but oh well

does recursive build.zig.zon stuff work as expected?

if you fetch a zig package, does the build system also go through the fetched dep's zon file?

[fiery turtle]

at least a readme mention would be nice ig

[calm latch]

Btw, it there an equivalent of ACPI_BUS_NUMBER_RANGE in uACPI?

[frank canopy]

yes. also theres lazy deps in build zig zon too

that only get fetched if used

idk the details of how to use lazy deps tho

[fiery turtle]

hm?

UACPI_RANGE_BUS

if u mean the resource type

[calm latch]

I think so

[fiery turtle]

just call get_current_resources and find out

[frank canopy]

alright @fiery turtle https://github.com/Khitiara/zuacpi

[fiery turtle]

damn thats really cool

[frank canopy]

im happy to transfer over to the uacpi org and update the readme to match that if you want to do that now but otherwise i dont have the energy to do any more with this today

[fiery turtle]

yeah that would be great

how do we do that?

[frank canopy]

i need perms to create repos in the org at least long enough to do the transfer apparently

[fiery turtle]

i guess i need to add you to the org for that?

yeah

[fiery turtle]

first official uacpi org member

[frank canopy]

https://github.com/uACPI/zuacpi 🎉

[fiery turtle]

ill add a mention/link to the bindings to the uacpi repo readme

thanks for your work, very cool

[rustic compass]

@fiery turtle add me too

[fiery turtle]

sure

done

[frank canopy]

as noted in the zuacpi readme its wip so theres functions missing but ill be swapping over imaginarium to use the zuacpi repo as a submodule when i next work on it and most of what youd need for simple stuff (no suspend or shutdown yet really but device/namespace stuff is there) is probably already bound in zuacpi

[rustic compass]

"You don’t have the permission to create public repositories on uACPI"

[fiery turtle]

huh

[fiery turtle]

accept the invite first

[frank canopy]

did you accept the invite from your email or wherever it showed up?

oh beat me to it

[rustic compass]

[fiery turtle]

[rustic compass]

https://github.com/uACPI/uacpi-rs

[fiery turtle]

[rustic compass]

only took a few months

[fiery turtle]

lol

[rustic compass]

*to move the repository

[fiery turtle]

how close are they in general?

(to being done)

[rustic compass]

well types are mostly done like 95% looking at the edge cases and handler stuff

[fiery turtle]

nice

[rustic compass]

rest is relatively easy

[fiery turtle]

thats cool

[hollow elm]

rename to ruacpi

[fiery turtle]

@left orbit do u want the org invite as well since you basically did uacpi-linux yourself?

[rustic compass]

"rest is relatively easy" famous last words

[fiery turtle]

and also monkuous for uacpi userspace work

[left orbit]

i won't say no to that :^)

but i don't promise i will bring anything of value in lol

[fiery turtle]

thats okay lol

whats your github again

[frank canopy]

for zuacpi:

• eval only for integers atm
• missing shutdown/power binding stuff
• resources arent finished yet or tested
• has shims for putting through an allocator for uacpi to use and implements alloc/free and logging kernel apis
• basically i can fully init uacpi complete with namespace enumeration and get enough to start working on basic drivers but not much more yet and the rest of the shims the kernel has to provide

[left orbit]

@48cf

[fiery turtle]

🤔

[left orbit]

lmao

[frank canopy]

i was able to receive power button events and enumerate all devices with these zuacpi bits but didnt bother to actually do shut down with them

lol

[left orbit]

that is most definitely not me on the picture

[fiery turtle]

czapek face reveal

[left orbit]

[rustic compass]

#polls message

[fiery turtle]

lol

[frank canopy]

oh and i put my table types into zuacpi so you can have HPET, MADT, FADT, and MCFG tables typed

[fiery turtle]

sounds good as a start

[fiery turtle]

awesome

[frank canopy]

yeah its got the basics at least

[left orbit]

shouldn't the table types be pulled from uacpi itself?

[frank canopy]

zuacpi is very much me adding bindings as i need them in imaginarium lol

[frank canopy]

the types are the same so it doesnt matter

im not using cImport

[fiery turtle]

zig bindings for structs

[frank canopy]

so its an extern struct either way

[rustic compass]

thanks for creating the zig bindings, i will probably never use them but their mere existance forces me to work on the rust bindings to be earlier finished than yours

[frank canopy]

you joke but @cImport can do that lol

[left orbit]

@cImport is amazing

[fiery turtle]

lol

[left orbit]

truly one of the best features of zig

[frank canopy]

also zig translate-c

[fiery turtle]

yeah thats a neat feature

[frank canopy]

i started the bindings by running translate-c on some of the first headers i needed and then refining that lol

[fiery turtle]

thats probably a good way to do it

[frank canopy]

yeah it was a good staring point for sure

it didnt recognize the status as an enum though which was pain

its def not a perfect tool

but thats why im making proper bindings anyway

[fiery turtle]

what did it think it was?

[frank canopy]

it set uacpi_status to be an alias for c_int and then made a bunch of consts for the numeric values

[fiery turtle]

ah

[left orbit]

probably because in C the enum is really just a weird integer type

[frank canopy]

yeah

[fiery turtle]

i could really use : type for enums in a lot of places

but its super recent C

i didnt want to force that

[slim panther]

yea probably don't

that's a c23 thing iirc

[mortal yoke]

gnu supported it as an extension, though idk about msvc

[left orbit]

msvc does it too iirc

[fiery turtle]

TCC doesnt tho

[rustic compass]

good thing in rust you can "tag" enum values with integers and tell the enum to behave like a integer

[left orbit]

actually msvc doesn't

interesting

[fiery turtle]

bruh

[left orbit]

me when <source>(1): error C2059: syntax error: ':'

[rustic compass]

and then use transmute on that, what could go wrong

[fiery turtle]

embrace extinguish etc went wrong

[frank canopy]

zig has enums and tagged unions separately so it works out here too where it treats it as an int on the abi level

[left orbit]

mfw you just commited at least 5 war crimes by writing 3 lines of rust code

[frank canopy]

sadly zig tagged unions arent c abi compatible because theres no standard for where to put the tag

but i can make an extern union for the things that go like that

[fiery turtle]

it sucks there isnt an abi for automatic optimized struct layout

[frank canopy]

and do some actual nonsense too https://github.com/uACPI/zuacpi/blob/main/src/uacpi/resources.zig#L151-L187

i love comptime

[left orbit]

i love zig war crimes

[fiery turtle]

bruh

[left orbit]

might not be the most readable code ever

but it just works

[rustic compass]

then you havent seen my transmute_copy on generic types to read/write them unaligned as byte arrays

[left orbit]

and you learn to embrace it

[frank canopy]

i think that code is actually broken

[left orbit]

lmao

[frank canopy]

zig comptime war crimes are fun

ill fix it when i actually try calling the resources functions lol

but yeah the idea is i can take the extern union and automatically generate at comptime the tagged union

and then use inline else to convert the two without having to serialize the field names to the binary

[loud ice]

you can macro and ifdef around it

lol

[left orbit]

ugly

[fiery turtle]

u wont get warnings for type punning etc

which is the entire point sort of

[left orbit]

wait for pitust to come up with a way to achieve this with the most cursed set of macros ever

[fiery turtle]

literally any problem in C can be solved with insane macros

[frank canopy]

lol

[loud ice]

i mean enum E UACPI_IS(base) could work?

you can also just newtype uacpi_status to a struct

at the cost of breaking api

[rustic compass]

you mean the ones that broke msvc and clang on windows for me

[fiery turtle]

lol yeah

they broke, but only some versions of msvc

which is the cool part

[rustic compass]

If we can decide on what our real hardware benchmark machine should be, I have found a aspect that should be considered. Newer Intel cpus tell you the tsc frequency, using such a machine removes discrepancies from inaccurate tsc calibration

[fiery turtle]

true ig, although newer cpus allow you to have tsc scaling in VMs so u can set any frequency you want

also kvm reports it as well

[rustic compass]

But it would make the test more complex if you also have to compensate for tsc calibration or the hypervisor needs to trap cpuid calls and emulate those leafs or similar

[fiery turtle]

true

[calm latch]

Is this normal that bochs returns resources->entries[0].irq.num_irqs == 0 for all IRQs in uacpi_get_pci_routing_table?

[fiery turtle]

show code

[calm latch]

I think I've fixed it

(I was reading from 0xd00 instead of 0x3fc)

Though now, suspiciously, all interrupts point to GSI 9 and GSI 11??

[fiery turtle]

hm, how is that related to rsources?

[fiery turtle]

yeah thats where bochs routes pci irqs

[calm latch]

It was reading from PCI ports to get the interrupt numbers

[fiery turtle]

ah

[calm latch]

I guess I have uACPI working in Bochs now?

(I just need to reimplement ECAM)

[fiery turtle]

thats nice

[surreal current]

this was a literal trillion yrs ago but wrk pushlocks are literally identical to the reactos implementation - that isnt to say he literally looked at the wrk itself, that was just an observation i made at some point while looking at reactos pushlocks to figure out how to implement them myself

i got suspicious because of what you mentioned about the comments not understanding what the pushlocks were actually doing

[winter orbit]

i see

yeah i dont blame the suspicion

[surreal current]

but that could be just as much an effect of 1:1 decompilation without really getting what it was doing

[fiery turtle]

Btw how's reactos new acpi coming along?

[surreal current]

as opposed to literally copying code or something

[winter orbit]

maybe

[winter orbit]

ive been focusing on getting ready for CLT ill continue post march (28~ specifically)

[fiery turtle]

ah

whos gonna be first, haiku or ros

[winter orbit]

Idk probably haiku

this is what im losing my mind over atm

last thing i need to do before clt

[fiery turtle]

seems to work?

[winter orbit]

it's using a file from windows 2003 still

win32k.sys

[fiery turtle]

ah

[winter orbit]

im trying to fix that yeah

like this is all ive done for like.. the past 6-7 days

[fiery turtle]

damn, good luck

[calm latch]

How do I wake up bochs from S3?

[north holly]

try pressing a key in the window

bochs window

[fiery turtle]

i've studied proxima sources for a bit

pretty smart stuff

well the pre-split proxima

[strong heath]

how do powerbutton events work with uacpi

how does it notify the OS

[fiery turtle]

there is like 4 ways a power button can be routed

the simplest one aka the fixed event is just a callback from interrupt context

if u add support for embedded controllers then you just install a notify handler to a power button device

Notify() is how AML calls into a kernel

for the fixed event there's a uacpi_install_fixed_event_handler

the third way is a GPIO device, this is applicable to modern AMDs, works similarly to an EC

the fourth way is a generic event device, also similar idea, but needs a separate os driver as well

only applicable to embedded

and qemu on aarch64 i think

[strong heath]

okay so what do I do lol

[fiery turtle]

uacpi_install_fixed_event_handler

this will cover most hardware

well most desktops

and older laptops

[strong heath]

and for covering all hardware I gotta do that plus GPIO device? whats a GPIO device

[fiery turtle]

let's say this covers 60% of hardware

EC driver covers 30%

9% is GPIO

1% is GED

thats how id put it

[strong heath]

okay life sucks is the takeaway

[fiery turtle]

ec and gpio drivers are both like 200 loc of C

for a crappy version that will mostly work

if u want it fancy and async then maybe 1k loc

e.g. pmos supports all 4

well no it doesnt support the GED

but thats mostly for arm crap

https://www.qemu.org/docs/master/specs/acpi_hw_reduced_hotplug.html

[fiery turtle]

@calm latch has a spec for AMD GPIO

[calm latch]

It's in the AMD manual whatever

[fiery turtle]

really?

[fiery turtle]

ah

[calm latch]

Linux source for it was not too insane

[fiery turtle]

and this was on your amd laptop or pc?

[gentle peak]

thanks

[calm latch]

on my laptop

[strong heath]

ill give a look to what pmos does, the 60% setup should be easy

[fiery turtle]

ah ok it was on qookies laptop as well

maybe it doesnt concern modern amd desktops

[calm latch]

the primary sources were manual and linux

[calm latch]

All my desktops have normal fixed events (I think)

Not even EC

[fiery turtle]

thats nice

[calm latch]

Btw, how does power button work on RISC-V in QEMU?

[fiery turtle]

see the GED link I sent above

for EC-routed power buttons i have a short summary on the osdev wiki

[fiery turtle]

exact same idea as gpio but simpler

[strong heath]

is qemu of those machines that do the fixed event handler thing

or not

[fiery turtle]

yeah of course

[hollow elm]

qemu uses PNP0C0C yeah

[strong heath]

so it uses the notify thing

thats the notify thing

[fiery turtle]

nah

i dont think it even has a PNP0C0C

even if it does, its still routed as a fixed event

[hollow elm]

oh my bad, i thought PNP0C0C was for the fixed event

i misremembered

[fiery turtle]

it's the device u hook up with a notify handler usually

[strong heath]

well I can tell you my fixed event aint firing

now I gotta figure out why

[fiery turtle]

issue of the skill

[keen shard]

I’ve ported all of the sysdeps for uacpi @fiery turtle

Now I have to actually implement it

Should I init uacpi at the beginning of my kernels main function?

[strong heath]

how does this thing work in the background, does it just call the handler from an ISR?

could it be the irq not firing?

[fiery turtle]

yes and yes

is your install_irq_handler implemented properly?

[strong heath]

I am checking if it is

[fiery turtle]

last time i checked it was a no op

[strong heath]

last time you checked it was less full than it is now

:3

[fiery turtle]

lol

[strong heath]

okay it works now

I wasnt unmasking it in my lapics

I pagefault on return but thats details

[fiery turtle]

ah

cool

just dont return and shutdown instantly

[strong heath]

is that a thing

or a joke

[fiery turtle]

nah u shouldnt do that

its a joke ofc

[strong heath]

you got me scared for a second

[fiery turtle]

lol

a lot of kernels using uacpi do it because they dont have softirq or workqueue etc

[strong heath]

does the return address or anything of the sort get passed in the uacpi handler context of the ISR

[fiery turtle]

wdym? uacpi doesnt care about anything like that, it just reads a few regs and stuff

[strong heath]

you know that you have a variable you gotta pass to the interrupt when called

the context thing

[fiery turtle]

its just uacpi state

[strong heath]

could not respecting that have to do with this misterious write to 0x30 that is pagefaulting me

[fiery turtle]

nah

[fiery turtle]

or wait

do u literally discard the context lol

[strong heath]

that does fit the bill in not respecting it

[fiery turtle]

what are u giving that function then

just null?

[strong heath]

lets say, the void

[fiery turtle]

wtf

that wont work ofc

[strong heath]

yeah the page fault to 0x30 could just be you trying to access a structure element in the handle

[fiery turtle]

yeah it just retrieves some stuff from the ctx associated with that irq

[strong heath]

yeah its probably that

how does linux handle power button events, I will need to expose this to userland somehow

does it just send a SIGPWR and userland picks the pieces?

[fiery turtle]

no lol

the power button gets an object in sysfs or wahtever

then u can listen to it as if it was a keyboard

or w/e

[strong heath]

I am going to send a SIGPWR to init

and it picks the pieces

[fiery turtle]

lol

[fiery turtle]

https://github.com/torvalds/linux/blob/master/drivers/acpi/button.c#L591

[hollow elm]

gobless evdev

[keen shard]

Hi Infy sorry to bother but could you answer my question I asked earlier?

[fiery turtle]

ah right

[keen shard]

This one

[fiery turtle]

i mean depends on you

see when other kernels do it

you probably want early table access so u can bootstrap stuff

[keen shard]

Should I have like the idt and stuff initialized

[keen shard]

Okay

[fiery turtle]

yes

[keen shard]

Okay

[fiery turtle]

when you call uacpi_initialize your scheduler must be online

[keen shard]

Okay so I could go memory, gdt, idt, tmpfs, sched uacpi?

[kind mantle]

You don't need FS

[keen shard]

Oh okay

[left orbit]

you need to initialize it after scheduling primitives are available but before you enumerate any buses

[keen shard]

Oh okay

Got you

[fiery turtle]

uacpi_kernel_mkdir

[left orbit]

it will also need memory management

and pci/port io access

and an allocator

so make sure all of that is available befoe calling uacpi_initialize and friends

[keen shard]

Okay sounds good

[keen shard]

Okay done

I’m going to spend hours on gdb and get back to you guys

[keen shard]

Well now I’m having a great time

[strong heath]

okay, after making an abomination, Ironclad now handles poweroff just fine passing it to userland as a signal, which then the init chooses to poweroff with

ill clean up the thing a bit and I am done

[fiery turtle]

If it works it works

[north holly]

when uacpi JavaScript bindings?

[calm latch]

Cool, it's <200 LoC in Linux...

So it's like EC?

(from linux)

static irqreturn_t acpi_ged_irq_handler(int irq, void *data)
{
    struct acpi_ged_event *event = data;
    acpi_status acpi_ret;

    acpi_ret = acpi_execute_simple_method(event->handle, NULL, event->gsi);
    if (ACPI_FAILURE(acpi_ret))
        dev_err_once(event->dev, "IRQ method execution failed\n");

    return IRQ_HANDLED;
}

Maybe I'll write a driver for it

[calm latch]

Btw, how does uACPI handle device hotplug?

[fiery turtle]

it doesnt

hotplug just calls Notify

it's up to you to handle it

[flat badge]

ACPI based hotplug is not that useful

Implementing PCIe based hotplug probably makes more sense

[fiery turtle]

I dont really have any data on that, but qemu recently switch to acpi hotplug by default because SHPC is racey and sucks

[flat badge]

shpc is not pcie though

[fiery turtle]

Iirc pcie has similar problems

Could be wrong

[flat badge]

PCIe hotplug works fine

[fiery turtle]

Probably

[kind mantle]

TIL ACPI has a hotplug mechanism?

[fiery turtle]

Acpi is best for virtual environments

[fiery turtle]

Yea

[flat badge]

And you can actually use it on real hw on most modern boards

[fiery turtle]

Acpi hotplug is nice because it doesnt have any sort of timeouts

You have to ack it or its denied

[kind mantle]

What's it used for?

[gentle peak]

devices can be marked as "not inserted" and that status can change (when it does you get a notify)

[fiery turtle]

Anything, pci devices, ram sticks, cpus

[kind mantle]

Hotplugging... RAM.

And CPU.

[gentle peak]

yes?

[flat badge]

• on virtual machines ;D

[gentle peak]

both cpu and ram hotplug has been a thing for a while

[calm latch]

You get a notify, what do you do with it?

[fiery turtle]

How does cpu hotplug work on severs

[fiery turtle]

Re read status

[kind mantle]

I guess I'm thinking too much of real HW?

[flat badge]

No

[calm latch]

But devices will always be there?

[fiery turtle]

In the namespace yes

[gentle peak]

insertion: scan the device and its children for devices whose _STA indicates they've been newly inserted (if you encounter enumerable buses like pci i'm pretty sure you're supposed to re-enumerate those too)
removal: same thing but for newly removed

[kind mantle]

Damn, I want to support CPU and RAM hotplug on BadgerOS now

[fiery turtle]

Its trivial on qemu

[gentle peak]

any devices that need to end up in the namespace will be there from the beginning

[fiery turtle]

For ram just read _CRS when its inserted

[flat badge]

PCIe hotplug works on any thunderbolt controller though and also on many ports (they may not expose a physical button for hot plug / hot remove but there are expansion cards that do)

[kind mantle]

BadgerOS doesn't even need any big modifications to support that funnily enough.

[fiery turtle]

NT had supported acpi hotplug since forever so idk if korona is right about it being vm only

For both cpus and ram

[flat badge]

I'm not saying that acpi hotplug in general is VM only

[fiery turtle]

Well and pci ofc

[kind mantle]

Well NT VMs exist too, it's not like windows only gets run on real hardware

[flat badge]

But hot plugging ram or cpus is VM only

[fiery turtle]

Why? Cant u insert ram sticks at runtime lol

[flat badge]

Iirc there is ACPI hotplug for pci devices that also works on real hw

[fiery turtle]

On server hw

[flat badge]

No

[fiery turtle]

Why?

[flat badge]

Because the memory controller doesn't support it

[fiery turtle]

Like all of them?

[flat badge]

I'm not aware of any x86 machine that supports ram or cpu hotplug

[kind mantle]

AFAIK there exist some high-end expensive stuff that does actually allow for RAM hot plugging with real HW

Like the IBM Z mainframes for example

[flat badge]

But i know that plenty machines (even intel consumer hw) support hotplug of pci devices

[kind mantle]

That's basically mandatory with the whole USB4 situation and its PCIe tunneling

[flat badge]

Yes, afaik there are mainframes that support hot plug of RAM

But they obviously don't run acpi

[kind mantle]

No

[fiery turtle]

Anyway ram hotplug is fun to play around with on qemu

Same goes for cpus

[kind mantle]

Can QEMU also remove said RAM and CPUs again?

[fiery turtle]

Yeah

You have to ack it tho

[flat badge]

CPU hotplug on VMMs is useful, that's true

[fiery turtle]

By calling _EJ0

[kind mantle]

I mean makes sense

[flat badge]

Does acpi really support ram hotplug btw? How does it work?

[kind mantle]

You wouldn't want a CPU core suddenly poofing out of existence while it's still running

[flat badge]

Why would you ever do this instead of virtio balloon?

[fiery turtle]

The devices exist in the namespace from boot, but _STA returns 0. Then on hotplug u get a notify on that ram stick, re read _STA, then get its ranges via _CRS

[fiery turtle]

Its a different idea

[flat badge]

I'm not asking about hotplug in general, I'm asking: how can ram be represented in the namespace

[fiery turtle]

Just a device with pnp id ram?

Or rather nvdimm or whatever

[flat badge]

Does that exist?

[kind mantle]

*that is marked as not present/inserted/whatever

[fiery turtle]

Of course

[mortal yoke]

I wonder if there are any devices doing ram offlining with acpi

[flat badge]

Which pnp id is it?

[fiery turtle]

PNP0A06

Thats what qemu uses I think

https://github.com/qemu/qemu/blob/b69801dd6b1eb4d107f7c2f643adf0a4e3ec9124/hw/acpi/memory_hotplug.c#L350

Wait no its PNP0C80

Memory module

[fiery turtle]

This one is just a generic resources container

[flat badge]

I see. I didn't know that the namespace can represent memory

[slim panther]

https://www.aliexpress.com/item/1005007468105944.html
i think i just found the most cursed device to test acpi on

[fiery turtle]

Chinese acpi

[slim panther]

🔥 hell yes

[fiery turtle]

Wtf is ubundu

[slim panther]

,Linux

obviously

[fiery turtle]

Maybe its saying that windows 11 supports them

[rustic compass]

How bad can it be?
-famous last words

[pine leaf]

I could test uACPI on a UEFI implementation that's compatible with my phone

[fiery turtle]

like aarch64?

i dont think there's an aarch64 kernel with uacpi

[calm latch]

Managarm?

[fiery turtle]

It doesn't support acpi

On arm

[calm latch]

Ok maybe I'll port pmOS

[jaunty fox]

Keyronex

[pine leaf]

next time I'm doing shenanigans w/ my phone I might give it a try then

[fiery turtle]

Ah

Does it actually support acpi in aarch64, or does it mostly rely on dts

[left orbit]

it does support acpi iirc

[fiery turtle]

i started work on UACPI_BAREBONES_MODE, only need to solve these now

it should compile only with this available

[flat badge]

very nice

[slim panther]

pog

what about a non interactive mode?

[fiery turtle]

whats that?

[slim panther]

everything except sleep/aml stuff

or is that already what this is for

[fiery turtle]

every other subsystem besides tables is useless if you have no AML, so yeah

[slim panther]

👍

[fiery turtle]

but

this one has no allocation or thread safety

so there can be like a TABLES_ONLY_MODE that does feature allocation and mutexes

but nothing else

idk if it's as useful

not sure about a use case for that

[slim panther]

there probably isnt

[flat badge]

for what?

[fiery turtle]

an advanced table-only mode

with mutexes and dynamic allocation

[flat badge]

Ah

So a primitive table mode is definitely useful. For example, we'll use it in Managarm to detect CPUs in eir

= the prekernel

[fiery turtle]

yeah, a primitive one can be used in bootloaders, pre-kernels, microkernels, etc

builtin drivers as well

[flat badge]

That's useful because it allows to pre-allocate some data structures

[fiery turtle]

a fully featured one, idk

[flat badge]

Also, in the future we'll use SRAT for numa support etc

[fiery turtle]

yeah

[flat badge]

An advanced one sounds a bit useless, agree

[fiery turtle]

although that would probably be parsed by managarm itself

[frank canopy]

yeah once we get tables-only mode i plan to make a pre-kernel for imaginarium

[fiery turtle]

added some docs to uacpi_kernel_map

[dense steppe]

nice 👍

[fiery turtle]

yay

i've never written the word ifndef this many times in my life

[gentle peak]

busd: info: successfully loaded 1 AML blob, 1705 ops in 3434ms (avg 496/s)
👍

[fiery turtle]

how 💀

[gentle peak]

conditional breakpoints in uacpi's main loop

[fiery turtle]

are you working on proxima 2.0?

[gentle peak]

yeah

[fiery turtle]

nice

[calm latch]

I see d at the end of bus

[gentle peak]

yeah

it's in userspace

[fiery turtle]

bus daemon?

[gentle peak]

indeed

anyway, if i remove all the debugging stuff i get 73k/s on -O0 TCG

[flat badge]

-O0 TCG? is that with qemu in debugging mode?

[gentle peak]

yes

here's a release build (-O3 LTO KVM)

[flat badge]

that sounds painful

[fiery turtle]

damn lol

proxima 2.0 already faster than everything

[gentle peak]

including proxima 1.0! and it's running in userspace!

[fiery turtle]

how tho

[gentle peak]

this is probably because as of right now the libc isn't capable of threading and thus does not bother doing locking

[calm latch]

I've also just found a skill issue in my mutex, so maybe pmOS will be a bit faster

[left orbit]

wait wtf

8M?

crazy

[fiery turtle]

technically its illegal because no locking

[left orbit]

cant wait for acpi.sys

[gentle peak]

iirc i asked you a while ago if it'd be legal if a single-threaded userspace server (that inherently does not need to care about locking) does not lock

and you said yes

so by that standard this is legal

[fiery turtle]

oh yeah

true ig

[gentle peak]

but it's a bit cheesy

[fiery turtle]

this design doesnt require locking

so its fair

[gentle peak]

also this doesn't use any cache-related cheats

this is the first run, nothing has been paged in yet, allocator cache is empty

[fiery turtle]

whats the libc and allocator used?

[gentle peak]

plibc (proxima libc), which uses a stripped-down version of the allocator in hydrogen

[fiery turtle]

are you planning on using it forever?

[gentle peak]

yes

[fiery turtle]

damn

is this an mlibc competitor

or do u want to keep it proxima only

[gentle peak]

oh it's absolutely not a competitor lmao

i'm not really trying to make it portable

[fiery turtle]

well its fast

[gentle peak]

people are welcome to use it if they wanna go through the effort ig

but all the code assumes it runs on proxima

[fiery turtle]

You could make it more portable

[gentle peak]

okay wtf i switched to uacpi's builtin uacpi_kernel_alloc_zeroed instead of mine that's aware of when something's already zeroed and the performance went up

[fiery turtle]

Btw would you switch to bare bones mode for the kernel instead of own impl if I implemented it?

[gentle peak]

i guess the check's more expensive than actually zeroing

[fiery turtle]

Huh lol

[flat badge]

managarm could probably beat managarm's score by running uacpi in userspace as well

[fiery turtle]

Probably

[gentle peak]

idk, i don't really want the kernel to have any outside deps and my acpi code is tiny so it's really not a maintenance burden

[calm latch]

time for a third uACPI in userspace implementation

[fiery turtle]

Makes sense

[gentle peak]

like i'm pretty sure i have as many lines of table defs as actual logic in the acpi stuff

[flat badge]

nah, it's too much work just for the meme

[left orbit]

@gentle peak will you do suspend/resume?

[gentle peak]

i want to do so eventually but it's a really low priority

[fiery turtle]

Lol

But managarm could be #1

[calm latch]

It's not hard if you have libc (which you do)

[flat badge]

no, proxima would probably still be better optimized for this use case

[fiery turtle]

It's hard in managarm world because a lot of stuff in the kernel uses it

[gentle peak]

i mean there aren't really any uacpi optimizations here

i cobbled together this kernel api impl in a few hours

[fiery turtle]

Is the src for everything available btw?

[gentle peak]

well i haven't pushed busd yet but yeah

[fiery turtle]

Nice

I'll take a look when you push

[gentle peak]

might take a while, i haven't implemented all the primitives necessary for the full kernel api impl and i'm not pushing until that's done

[fiery turtle]

Luckily work stuff should be trivial in userspace

[gentle peak]

yeah especially single-threaded

[fiery turtle]

Would you do it with an event loop?

[gentle peak]

probably

[fiery turtle]

Yeah

[calm latch]

I just create a pthread_create

[fiery turtle]

Id just do an event loop probably

Where u can schedule oneshot work

[gentle peak]

i'll need an event loop anyway for ipc

[calm latch]

How are you planning to do your IPC btw?

[fiery turtle]

Would be cool if you make it with coroutines

So that e.g. uacpi sleep pauses instead of blocking

[calm latch]

Only with c coroutines maybe (?)

makecontext stuff

[gentle peak]

two methods, which one is used depends on traffic type (message size, amount of traffic, etc)

• a unix socket like thing which can bidirectionally send/receive bytes and/or handles
• shared memory for submission & completion queues and futexes to notify either side of a new message

[fiery turtle]

I mean qemu has it so proxima can too probably

As in, most qemu work is done in coroutines

[calm latch]

Afaik stackless C++/Rust coroutines need all the code to be async, which uACPI isn't

[fiery turtle]

Stackless sucks anyway

[calm latch]

but this could be done, yeah

but at this point why not just use threads

[fiery turtle]

Proxima fibers when

[calm latch]

You can use userspace scheduler

[fiery turtle]

Because threads are more expensive

Than fibers

And for stuff like busd u probably dont need threads

[gentle peak]

yeah threads are only for stuff that really needs high perf like the posix server

[flat badge]

= basically managarm's design

[calm latch]

But threads are also convenient and easy to use

[fiery turtle]

So are fibers lol except u dont get race conditions

[gentle peak]

is it? i haven't really looked at managarm's internals at all

[flat badge]

Managarm has shared memory queues for kernel <-> userspace communication

and it has mostly single threaded drivers in userspace

[calm latch]

I mean pmOS as well, everything is single threaded not parallel atm

[gentle peak]

ah

yeah it seems kinda like the most obvious design to go for

[fiery turtle]

Didnt u say u use pthreads

[calm latch]

I did, but not for parallel stuff

Idk how to explain it

[fiery turtle]

Lol

[calm latch]

Like there is a single event loop in everything, and I just create threads when I feel like it

[calm latch]

Ok, maybe it's only half true

My pipes server creates a thread per pipe

But my AHCI driver is using C++ coroutines

[fiery turtle]

Aka u use pthreads because you're lazy to design stuff

[calm latch]

yes

Also because of my IPC skill issue

[gentle peak]

i wanted to see what the score would be like if i did the funny uacpi_state_reset thing, instead i got vmm whack a mole

[fiery turtle]

Like a bug?

[gentle peak]

so many bugs

i should've expected this from something i wrote mostly between 2am and 4am

[fiery turtle]

I thought proxima was bug free

[gentle peak]

i should add more asserts

[calm latch]

My IPC "ports" (message queues) have single receiver and are tied to a thread, and I haven't come up with a good abstraction to make stuff multithreaded

[calm latch]

Asserts are good if you don't confuse them with ensure/panic

[fiery turtle]

Slippery slope

[gentle peak]

the stuff that's used a lot is mostly bug free i'm pretty sure, but busd is the first time i'm creating more than like 10 vmm regions

so there's a lot of very untested code paths i'm hitting

[fiery turtle]

Is it an rb tree?

[gentle peak]

avl tree yeah

[fiery turtle]

Ah

[gentle peak]

it's not in the tree management code though

i'm pretty sure at least

[fiery turtle]

Where does it die?

[gentle peak]

it dies while trying to initialize a new page for busd's allocator, for some reason it can't find the region it just created

i'm investigating rn

[fiery turtle]

Ah

[redacted] curse

[dense steppe]

things like tree data structures etc. are like the one thing I consider important enough to test at the moment

[fiery turtle]

Yep

[gentle peak]

i probably should do that too

but i really can't be bothered

[fiery turtle]

I wonder why the old proxima was triple faulting on that pc

[gentle peak]

which pc

[fiery turtle]

Oberrows pc

[gentle peak]

ah

i never did figure that out

[fiery turtle]

With an i5 something

Yeah its strange

[gentle peak]

for my own sanity i'm just gonna assume he downloaded one of the march=native images

[fiery turtle]

Lol

[gentle peak]

okay nvm it's definitely the tree code, the region's in the linked list but not in the tree

[fiery turtle]

True true

Classic

[dense steppe]

if you look for vmatree in the commit history of my current kernel, you will see like four or five commits just fixing bugs that I found by testing with randomly-generated test data from a Python script.

and even then, I managed to miss a bug

[fiery turtle]

Yeah lol these self balancing trees are very easy to fuck up

Linux still has code to debug them and verify

[dense steppe]

it wasn't the self-balancing part that I messed up, though

rather

the auxiliary data I keep for fast free space lookup

[fiery turtle]

Eh whatever

Still

[dense steppe]

yeah

[fiery turtle]

Good thing they're easy to cover with tests in userspace

[gentle peak]

what do you do for that? rn i just keep a linked list alongside the tree and do first fit

[dense steppe]

it's not exactly easy to think in terms of invariants everywhere

[fiery turtle]

Yep

[gentle peak]

well that's not good

[mortal yoke]

I haven't actually managed to make a completely working one myself which is why I originally just rewrote the frigg rbtree to be iterative instead of recursive and in c, then later on for the c++ kernel I rewrote my c tree in c++

[dense steppe]

basically, for each node:

• store a prev_gap, the gap between this node and the previous node
• store max_prev_gap, the maximum value of prev_gap in this subtree

[fiery turtle]

Damn, how did you make it iterative

Do u still have the code btw

[gentle peak]

hmm

sounds like a lot of extra edge cases lol

[dense steppe]

it's relatively easy to see how this permits O(log N) find a free area of size >=something

[dense steppe]

yeah. that's mainly why I wrote the testing code

[dense steppe]

however to find an area with >PAGE_SIZE alignment is still O(n^2)

[fiery turtle]

I was pretty amazed at how simple old proxima was yet fast af

[gentle peak]

simple = fast usually

[fiery turtle]

Like pmm and heap allocator were trivial yet fast

[gentle peak]

lines of code in the fast path is in many cases a great estimate for performance

(incl. iterations)

[dense steppe]

what was it that Einstein said now again?

[fiery turtle]

True

[calm latch]

I have my own impl, tested by the kernel's insanity

[mortal yoke]

well it wasn't that hard as frigg mostly had stuff that could be tail call optimized so I basically just replaced them with while loops and setting the node when traversing up

[fiery turtle]

Nice

[flat badge]

that could probably be merged to upstream frigg

[dense steppe]

I tried to implement something like the relatively new Linux maple tree for virtual address space management, based on some blog post or something that described the datastructure.

I won't be going down that road again for a very long time.

[fiery turtle]

Tbh it should have been iterative from the start

[mortal yoke]

and yes I do have it in https://github.com/Qwinci/hzutils/blob/main/include/hz/rb_tree.hpp (and in https://github.com/Qwinci/crescent/blob/main_c/kernel/src/utils/rb_tree.c), there are some other optimizations that could be done like merging the color with the parent but I haven't bothered doing that

[left orbit]

self balancing BSTs are magical

[gentle peak]

they are also a bit of a pain to implement

old proxima just used a plain old non-balancing bst for the vmm

but once you have them they're great yeah

[flat badge]

copyright violation

[fiery turtle]

💀

[dense steppe]

there was one time that, for a competitive programming problem, I implemented a multiplicative segment tree over matrices. (segment tree=poor man'scompetitive programmer's self-balancing BST)

[fiery turtle]

But would be nice if you upstreamed iterative stuff

[mortal yoke]

I did mention it in the commit I think, and I didn't copy paste it but yeah ig

[dense steppe]

Initializing lawsuit... OK

[calm latch]

My vmm uses splay trees

[fiery turtle]

Lol

[flat badge]

lol

don't worry, but it would be nice to add a comment to indicate the origin of the code

[mortal yoke]

I could add a mention about frigg at the top yeah

[dense steppe]

"Initializing <whatever>... OK" is basically osdev brainrot

what did it even originate from?

[gentle peak]

my favorite

[fiery turtle]

Initializing prison sentence... SUCCESS

[slim panther]

initializing gdt... ok

oh you mean who

[dense steppe]

no

[slim panther]

i think lapfed

[left orbit]

every osdev project ever showcased

that did not feature anything more impressive than a shitty ps2 keyboard driver

[dense steppe]

I should print that in my rewrite just for the meme

[slim panther]

ow

[fiery turtle]

basically because beginners print out stupid stuff like that and its funny

[gentle peak]

essential proxima code

[fiery turtle]

how did it even exist without that

[left orbit]

real

petition to add Init GDT... OK to managarm

[dense steppe]

a slight problem for me... I initialize the GDT before any C++ code is even executed 💀

[fiery turtle]

dont forget IDT

[dense steppe]

fake OS

[fiery turtle]

have a ring buffer

that is flushed when logger is initialized

[dense steppe]

just call printk from asm

lol

[fiery turtle]

yes

[dense steppe]

but yeah, I should probably implement a printk ringbuffer

[fiery turtle]

mov rdi, "Initializing GDT..."
call printk

or whatever

[fiery turtle]

lol

[left orbit]

perfect

how will u print

without a valid staack

[fiery turtle]

also inserts sleeps and progress bars

for gdt init

[gentle peak]

limine provides one

[fiery turtle]

to seem like a cool os

[left orbit]

ah

so you just do it

because why not

own stack = cool

:^)

[gentle peak]

because limine's stack is in memory that i want to reclaim

[fiery turtle]

Stage 1 GDT init... starting

[dense steppe]

own stack = I know what I get

[gentle peak]

and the stack used for the entry point stays used for the lifetime of the system because the entry thread becomes the idle thread

[dense steppe]

same with own page tables

etc.

[left orbit]

so i wasn't totally deranged for doing that

good to know

[mortal yoke]

I also do that ```x86asm
.lcomm STACK, 0x8000, 16

.section .text
.globl _start
.def _start
.type 32
.endef
_start:
lea rsp, [rip + STACK + 0x8000 - 40]

mov rax, cr0
// clear EM
and rax, ~(1 << 2)
// set MP
or rax, 1 << 1
mov cr0, rax

// set OSXMMEXCPT and OSFXSR
mov rax, cr4
or rax, (1 << 10 | 1 << 9)
mov cr4, rax

jmp early_start

[calm latch]

Quick question, how do I stop clang from optimizing entered_sleep away?

[left orbit]

asm volatile("" :: "m"(entered_sleep) : "memory");