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

[fiery turtle]

so we see how it comes up with that address

[calm latch]

How do I do it?

[fiery turtle]

uacpi_context_set_log_level

[calm latch]

also, you have a fast PC, you can compile it yourself

[fiery turtle]

we will see all io reads and writes

[fiery turtle]

i have wsl so no

wsl is slow af

[calm latch]

I'm also using WSL

[fiery turtle]

oh ok

u probably want the trace log level

debug logs every aml op

trace just logs IO

[calm latch]

This is ok, right?

Do I keep the printf on uacpi_map?

[fiery turtle]

yeah it can be called at any time at all

[fiery turtle]

sure

@fiery ice btw u tested linux shutdown on it recently right

[fiery ice]

ye

via ZorinOS

and it works fine

[fiery turtle]

what even is zorin os

[fiery ice]

linux distro that is similar to windows

it's the only gui one i had on hand

lol

[fiery turtle]

lol

[fiery ice]

it's taking so long to boot now XD

[fiery turtle]

yeah because it traces every io

[calm latch]

Does that PC have UART?

[fiery ice]

no

[calm latch]

Idk why, but flanterm is really slow

[fiery turtle]

bruh

is this why u get bs?

[calm latch]

Idk

[fiery ice]

bruh

you somehow managed to break your OS XD

by adding the log

[fiery turtle]

what happened?

[fiery ice]

keyboard also does not work

[fiery turtle]

pmos is as stable as obos

[calm latch]

This returned 10

Internal error which is an external error probably

[fiery turtle]

yeah

[calm latch]

Maybe it OOMed

[fiery turtle]

if only there was uacpi_status_to_string

[calm latch]

Where did you say I was returning internal error instead of external error?

[fiery turtle]

it was from some screenshot of yours

more interesting is how tf did logs even affect that

[calm latch]

How much ram does that PC have?

[fiery turtle]

you can also try fixing the pci stuff then just retry without debug logs

[calm latch]

I feel like AI is gaslighting me

It's BS

That should be correct

[fiery turtle]

Do u have Gemini or Claude?

[calm latch]

GPT codex 5.2

No

[fiery turtle]

Bruh

How

[calm latch]

This is also correct

[calm latch]

Student license

[fiery ice]

try claude free

[calm latch]

I'm not paying for AI

[hollow scaffold]

its free

[calm latch]

This is fun, I cought the PS/2 bug in a VM

[fiery turtle]

[calm latch]

But I don't care about keyboard now

[hollow scaffold]

talk is cheap

send patches

[calm latch]

To myself?

[hollow scaffold]

or make xhci and hid drivers

[hollow scaffold]

yes

[calm latch]

I've changed the timers API recently, and like I know where to look so it's not interesting to debug

I wanted to study today anyway...

Imma leave codex with it

And go do other stuff

This is with it reading from PCI instead of COM1

[fiery turtle]

Lmao

[calm latch]

Ok, GPT 5.5 is much more reasonable

[north holly]

maybe

but idk why that thing like never boots

[fiery turtle]

Same

Well it used to boot

[calm latch]

I feel like I'm falling into another ACPI rabbit hole

You're supposed to get buses from _CRS

Also, QEMU is trolling?

Or GPT is gaslighting

Or both

So it's the gaslighting

[fiery turtle]

@fiery ice can you test this one as well pls

[calm latch]

I'm fixing the stuff still, but that one might fix it

My PCI code is completely broken

[fiery turtle]

Lol

[calm latch]

Also, I've checked managarm, and it configures PCI bridges??

[fiery ice]

k

give it 5 mins to boot

lol

[fiery turtle]

Lol

[fiery ice]

yeah that debug log is slowing it down A LOT

[fiery turtle]

Maybe should've disabled it for fix check

[fiery ice]

yeah and with the log it's still broken

EC event err 10

oh wait

it now sees the power btn

but it does nothing XD

it just prints shutdown in 3 seconds

but nothing is happening

@calm latch

[calm latch]

yeah

gimme a sec, I'm fixing PCI still

Maybe it will work with that...

[fiery ice]

pls disable that debug log too

lol

[calm latch]

Idk, I feel depressed

Because it's C

[fiery turtle]

Ask chatgpt to fix it

[torpid root]

make no mistakes

[calm latch]

What am I supposed to do instead of UACPI_STATUS_INTERNAL_ERROR?

[fiery turtle]

Well why does it fail

I might add a KERNEL_INTERNAL_ERROR for this tbh

[rustic compass]

also would be nice if you could write down which parts of uacpi_status a function can return

which i plan to do with tons of subenums in uacpi-rs to make error handling easier given rusts strict type system

[fiery turtle]

A lot of helper functions call into aml which can return any status

So thats not really feasible for a lot of them

[rustic compass]

still, then you actually know those calls can return all these error instead of uncertainty, at least from a user side

if its not too much hassle for you

[fiery turtle]

Where do you expect to see them?

Also is that something you can express in rust somehow?

[rustic compass]

i can use subenums that allow me to create a smaller scope from the uacpi_status enum and only contain the elements actually needed and can be converted back into the big enum if necessary

[fiery turtle]

Is that useful?

[rustic compass]

depends on the users error handling tbh, at least for me i dont really like to have to group large amounts of error values into a blanket case in a match statement and just label them as unreachable

[fiery turtle]

We can do that at least for a simple case like the table api

But again that calls into the kernel api which can return anything

[fiery ice]

k, lemme try

[fiery ice]

it worked!

[rustic compass]

at least for the kernel api most error values shouldnt ever be returned as they just dont make sense there

[fiery ice]

pc now shuts off

[calm latch]

Cool

Idk what made it work now, but I've improved everything

[fiery turtle]

That only depends on the sanity of the implementation

There's nothing preventing that

[fiery turtle]

Damn

So it wasn't a uACPI bug, cool

[fiery ice]

seems like no

[calm latch]

yeah, my PCI code was jank

Idk how it even worked before

[fiery turtle]

Lol

[calm latch]

I mean not really, but it wasn't the best

[fiery ice]

and i was about to take inspiration from yours XD

[calm latch]

No, it's mostly the same, but I had a lot of stuff which was commented out, and it was reading from COM1 instead of PCI...

[fiery ice]

wdym reading from COM1?

[calm latch]

0x3f8 instead of 0xcf8

[fiery ice]

💀

[fiery ice]

are you not doing pcie?

[calm latch]

I am

[fiery ice]

then you should not need that port?

[calm latch]

But you need AML to do it, and that need PCI

So you've got to use the 0xcf8 if you have no ECAM

[fiery ice]

i support fallback but afaik it is using only ecam

[fiery turtle]

Why do u need aml for ecam

[calm latch]

Not ecam, but some other method

[fiery ice]

you can get it from MADT

or some static table

[fiery turtle]

Wdym

Like the very first aml PCI access you can do with ecam no?

[calm latch]

_CBA

[calm latch]

yes

[fiery turtle]

Yeah but why do u need it so early

[calm latch]

You don't

but you might not have ecam?

[fiery ice]

uacpi loads pci after you can already access CBA i think

[calm latch]

Plus the spec says that calling it activates some stuff

[fiery ice]

at least, i'm doing it like that

this is called after uacpi_namespace_load

uacpi_namespace_initialize is what does stuff with PCI afaik

[fiery turtle]

Well yeah but the question is why weren't you using ecam from the start on his laptop

It has ecam right

[fiery ice]

ye it def has

my os can see it

[calm latch]

I just didn't have it implemented, idk

[fiery ice]

💀

[fiery turtle]

Ah

[calm latch]

Worked on all my PCs without it

[fiery turtle]

There's eval simple integer if you dont have args

[fiery ice]

oh cool

[fiery turtle]

PCs dont do a lot in pts so ye

[calm latch]

Also, look at Managarm

[fiery ice]

nah dw, i already have pci / pcie

i did not test MSI / MSI-X yet tho

[calm latch]

The configure PCI bridges as well

[fiery ice]

but i'm implementing NVMe now

[calm latch]

I don't use those 💀

[fiery ice]

💀

[steel inlet]

when is uAPM releasing

(thanks android autocorrect)

[pine leaf]

what about uDT

(inb4 it's just rebranded libfdt)

[calm latch]

There's smoldtb

(Not by infy, but still a nice library which I use)

[fiery turtle]

i actually already support apm setup in hyper because someone requested it lol

but obviously not the fucked up runtime stuff it needs

[steel inlet]

lmao

[slim panther]

write ur own

[crude summit]

im still waiting for "uSB"

100% would use

[rustic compass]

@fiery turtle need to know if uacpi 5.0 is realistically happening within like a few weeks, if yes i would wait with bumping the uacpi version uacpi-rs uses until then else i just go to 4.? first

[fiery turtle]

A few weeks yes

[fiery turtle]

Added a few more items into the future ™ todo

im wondering how feasibile it would be to put all uacpi global state into a context and then do

#define uacpi_foo(...) uacpi_foo_ctx(&g_uacpi_ctx, ...)

so that rust users and others can be happier theres no global state

but thats not something im doing any time soon, just some thoughts I have

[fiery turtle]

obv this would be for backward compat, with an option like UACPI_NO_GLOBAL_STATE or smth

[frank canopy]

this would actually kinda slap for prekernel stuff - could pass the global state through to the main kernel from a prekernel thats compiled in barebones mode

[fiery turtle]

well the barebones context would obviously be incompatible with a full context, but I guess its possible to construct it such that it could be boostrapped

[rustic compass]

so that i can just put it into a static on the rust side? good idea

i dont really see the need for a non global state, there is only one uacpi instance ever at once unlike a heap allocator like hardend_malloc where its fucking annoying that its fully stored in the .bss meaning in need to do linker black magic to get more than one instance running at the same time

[fiery turtle]

yeah idk its not super useful

[rustic compass]

if you were to write something like a heap allocator i would really haunt you until you add that but uacpi, nah, but i do think it would be kinda funny having aml racing with itself

[frank canopy]

yeah its really not a huge thing lol

[kind mantle]

eliminating all global state for no reason makes no sense to me

[frank canopy]

also, im kinda curious, when making uacpi where did you start?

[kind mantle]

Also Rust globals aren't illegal, you just need to put them in e.g. mutexes or spinlocks

[frank canopy]

yeah the ability to migrate an externally-provided barebones state interests me a lot more than having state not be global

[kind mantle]

but isn't the barebones state also kinda trivial

Because you're just reading the tables, not doing any AML

[frank canopy]

yes, the idea is more for marshalling from a prekernel than it is for within a single thing

use pre-mapped-and-allocated-copy tables

[kind mantle]

Also if you did this you would also need to migrate your prekernel heap to kernel proper heap

Which seems problematic

[frank canopy]

well uacpi already uses fixed buffer for barebones stuff iirc/for early tables access

so its more about reusing the references to that before it copies into real allocations later

[fiery turtle]

sketching out the basic kernel api, then basic table parsing iirc

then tried learning how aml works by just dumping opcodes as text

then did some crappy interpreter impl, that was thrown away

then did it a few more times when i realized it didnt cover some edge cases with reference propagation

then somehow arrived at the microcode thing which kinda fell into place

[frank canopy]

cool

the microcode thing is brilliant btw, i legit never would have thought of it but its so elegant

[fiery turtle]

thanks! theres technically a lot to improve there

some redundant safe deep object copies, some stuff is out-of-line that could be entirely bytecode etc

but fast enough so eh

[frank canopy]

yeah

[fiery turtle]

still waiting for someone with a 300x claude subscription to make a JIT

[frank canopy]

lmao

[fiery turtle]

theres another small vm in uacpi besides the aml thing btw

the resource<->aml resource converter stuff

[frank canopy]

yeah i saw that when trying to debug my resource code lol

[kind mantle]

wait does ACPI define two bytecodes? 💀

[fiery turtle]

nah lol

[frank canopy]

no the resource one is just an implementation thing

[fiery turtle]

it defines very complex resource structures

that are basically unusable in C unless u convert them into something proper

[hallow marten]

I'm sure there are further optimizations before making a JIT lol

Also what's the microcode thing

[fiery turtle]

but then you also need to convert back for _SRS

[frank canopy]

uacpi has microcode for each aml opcode

which is then used to evaluate them

[hallow marten]

So an IR?

[kind mantle]

Microcode isn't really IR

[frank canopy]

no its true microcode - each opcode gets a list of microcode instructions that then get looped over when that opcode is hit

ir would imply it gets translated ahead of time

[hallow marten]

Ah so it's basically a kind of IR that describes how an opcode behaves

Not really IR

DSL or something

[frank canopy]

thats what microcode is

real cpus use the same idea

[fiery turtle]

e.g.

UACPI_OP(                                                        \
    DivideOp, 0x78,                                              \
    UACPI_OP_PROPERTY_TERM_ARG,                                  \
    {                                                            \
        UACPI_PARSE_OP_OPERAND,                                  \
        UACPI_PARSE_OP_OPERAND,                                  \
        UACPI_PARSE_OP_TARGET,                                   \
        UACPI_PARSE_OP_TARGET,                                   \
        UACPI_PARSE_OP_OBJECT_ALLOC_TYPED, UACPI_OBJECT_INTEGER, \
        UACPI_PARSE_OP_OBJECT_ALLOC_TYPED, UACPI_OBJECT_INTEGER, \
        UACPI_PARSE_OP_INVOKE_HANDLER,                           \
        UACPI_PARSE_OP_STORE_TO_TARGET, 3,                       \
        UACPI_PARSE_OP_OBJECT_COPY_TO_PREV,                      \
        UACPI_PARSE_OP_STORE_TO_TARGET_INDIRECT, 2, 4,           \
    }                                                            \
)                                                                \

[kind mantle]

It really isn't comparable to IR. Just think of it like smaller function calls you would make in a bigger function, but for instructions.

[hallow marten]

Yes I understand what it is

[kind mantle]

Using the term IR implies you're doing preprocessing on it

[hallow marten]

Not necessarily

But microcode is a better name

[fiery turtle]

if u have uacpi in your kernel u can make it trace every uOP being executed if u set the debug log level

but its very noisy

[frank canopy]

zig might actually be good for stuff like this tbh

because you can inline else in a switch plus inline for to remove runtime iteration and force-unroll the microcode iteration loop

[fiery turtle]

the convenient part about this is that since AML is recursive with this microcode I can just "suspend" the current OP if it requires another op before resuming

and then when I resume it, it needs 0 context since it just continues at the instruction pointer where it was suspended

[kind mantle]

presumably some kind of soft stack?

[fiery turtle]

yeah

it has an array of pending ops

with some inline storage

[fiery turtle]

yeah probably

when aml interpreter in zig

[hallow marten]

we did talk about it

[frank canopy]

this is litearlly why i got curious about how you got started

is the itch to make that got too strong

[fiery turtle]

i mean u can try to just start and see how it goes

[frank canopy]

yeah thats the plan

[fiery turtle]

run some basic Return (0) etc

then u can start building the test suite further

[frank canopy]

yeah

[fiery turtle]

i always tell people that they can resue the uacpi python test runner which can accept any other runner binary

[frank canopy]

there is another reason beyond curious i want to do this tho - zig has error return tracing and for uacpi i lose that when going through kernel api because it doesnt marshal through the C stuff

so its a pita to debug if a uacpi kernel api is returning an error code to try and find it

[fiery turtle]

yeah im sure u could do some interesting things in newer languages

[hallow marten]

I have too many side projects already but if you end up deciding to work on it for real I'd gladly help

[flat badge]

a JIT would just be slower

It's not that aml does a lot of compute

By the time the jit has compiled the aml, the interpreter is already done

[fiery turtle]

yeah true

it would only make sense for callbacks that are called all the time at runtime

like even java only jits hot paths

[fiery turtle]

Added a shit ton of tests for the new table api, it should cover everything I think

Hopefully 5.0 this weekend

[steel inlet]

me still on 3.2.0:

[fiery turtle]

💀

[pallid lava]

hmm, qemu does not expose Device() objects for the slots in any extra added root complex

[fiery turtle]

Slots never have a device

Only roots

Or pci express root ports or hotpluggable ports

[pallid lava]

then qemu does not do that

it does for any other device in the primary root bridge

[fiery turtle]

Yes because its hotpouggable

And you cant unplug a bridge

[pallid lava]

i mean

say you add a root port in root bridge 1 and 2, and add devices in those root ports

in the main root bridge, there will be a device() for those root ports and devices

in the other one there is not

and moreover acpi hotplug and hotunplug does not work for the secondary root bridge

it requires handling the native pcie way

qemu-system-x86_64 -M q35 -m 512M -enable-kvm -device pcie-root-port,id=rp1,chassis=1,slot=1,x-do-not-expose-native-hotplug-cap=on -device nvme,serial=nvme1,id=nvme1,bus=rp1 -device pxb-pcie,bus_nr=128,id=pcie.1 -device pcie-root-port,id=rp2,chassis=2,slot=1,x-do-not-expose-native-hotplug-cap=on,bus=pcie.1 -device nvme,serial=nvme2,id=nvme2,bus=rp2 -cdrom Core-15.0.iso ```

if you dump the dsdt you will see that the primary root bridge and descendants will be described by Device() objects and will have _ADR objects, which is how you can link a pci device to its acpi object

for the other root bridge, this does not happen, and since i explicitely disable native hotunplug, you will see that if you do device_del nvme1, it will work because acpi hotunplug is supported for the primary root bridge, but because of a byproduct of acpi nodes not existing for the devices in the secondary root complex, device_del nvme2 will fail with an error message

[rustic compass]

Same version as uacpi-rs

[frank canopy]

zuacpi is using a few commits back on the for-5.0 branch

[fiery turtle]

yeah

thats just limitations of the qemu acpi hotplug

[pallid lava]

hm, now that i think, it kinda makes sense for devices in secondary root complexes to not support acpi style stuff

since all acpi interactions and memes come from the south bridge

[flat badge]

PCIe hotplug is more versatile in general

I think ACPI hotplug exists mostly to handle edge cases where the PCIe bus does not support native hotplug properly

Linux prefers native hotplug as well

[rustic compass]

@fiery turtle
do other tables beside the fadt get cleaned up or is that the only table with special handling?

[fiery turtle]

Its the only one

[fiery turtle]

great, either github ci is fucked again, or intel is blocking github actions from downloading stuff

[fiery turtle]

uACPI 5.0 branch (for-5.0) is basically done
I'd really appreciate testing

Changelog:

• uacpi_kernel_map is now expected to return UACPI_MAP_FAILED on error, and 0x0 is treated as a valid return value to allow hosts that use a direct map
• added UACPI_{START,END}_OF_LOG_MSG that is applied to every single uACPI log message. That means you can add a prefix, or remove a newline at the end (by default, UACPI_END_OF_LOG_MSG is set to \n)
• Added a whole bunch of table API
  • find_by_signature_at (allows specifying the offset where to start looking)
  • find_nth_by_signature (allows specifying the Nth instance of a table to look for)
  • uacpi_table_{ref,unref}_by_index
  • uacpi_table_count
  • uacpi_table_get_by_index
  • uacpi_for_each_table & uacpi_table_info_get_by_index these are for more sophisticated use cases as they return all the internal information available about a table, including its origin, refcount, flags, etc
• UACPI_NATIVE_MMIO, a new option that allows a kernel to define its own MMIO accessors instead of the naive uACPI volatile helpers
• uacpi_enter_sleep_state_simple a simple shortcut helper
• A whole bunch of other shit im too lazy to list atm

[slim panther]

W

i should finally write a proper uacpi crate for zinnia instead of using raw bindgen

[flat badge]

there is also uacpi-rs, right?

Maybe we should make a joint effort to make it usable

[kind mantle]

#1287451030241542205

there is

[flat badge]

yeah i'm aware, it's just not in a state that can be used in production

[rustic compass]

I'm slowly picking up speed with writing it

And barebones mode is finished minus missing tables and uacpi 5.0

The current head of 1.0-dev is buildable, but untested

I plan to finish fadt and madt today

Any wishes for tables?

[rustic compass]

This greatly simplifies my table api

[fiery turtle]

The debug output tables are useful

[rustic compass]

are they a thing on x86 hardware or only on arm and co?

[rustic compass]

DBGP and DBPG2?

[flat badge]

to use it in Managarm, we'd need full AML support, not just tables though

[fiery turtle]

On all arches

[fiery turtle]

Yeah

[flat badge]

I wonder if we could speed it up by letting AI generate a baseline (+ human review ofc)

[fiery turtle]

Yeah that sounds doable

[rustic compass]

Me when I get replaced by AI

[flat badge]

lol

I think this depends on how well uACPIs ownership model maps onto rust, i haven't look into this

if it needs a lot of design work, then AI is probably not the right tool

[rustic compass]

Its mostly pretty good but I haven't looked at other parts that much

[slim panther]

I don't want to use it

nothing against the crate, I'm sure it's good

but i want full control over how uacpi is built

[flat badge]

what does that buy you?

[slim panther]

i can build with different cflags for example

[rustic compass]

You can fork the repo, that's what I will do to when I run the interpreter in userspace

[slim panther]

i could

[full nimbus]

Userspace ACPI based

[rustic compass]

Avx10 + apx go brrr

[slim panther]

W

[full nimbus]

Leaderboard maxxing

[slim panther]

i can't wait for my fred laptop

it arrives in the next two weeks

[full nimbus]

Insert 30 requests for you to test mine and others kernels on it

[slim panther]

i mean sure

[rustic compass]

I plan to buy a cheap zen6 cpu + itx server board for automated testing of my kernels and anyone who asks but that's at least half a year in the future

[full nimbus]

Holy basssdeeeedddd

You can probably do it in VMs for FRED atleast

Since KVM and QEMU supports it if the host does

Also iirc @north hull was working on similar for testing things

[rustic compass]

But but the poor ipmi interface 🥺

[full nimbus]

It’s probably easier to manage and more secure if you open it up more than wux did

Plus you can give every person like 4 cores

And then split the machine up etc

[rustic compass]

I think the server board has a socketed bios flash

[full nimbus]

Inb4 AMD has the 7th vulnerability with microcode uploading though 💀

[rustic compass]

Bold of you to assume it gets that many cores to beginn with with those ram prices 😭

[full nimbus]

Linux has to have a list of valid hashes to upload for certain amd CPUs

I was just so tired of looking at that BS I just skipped supporting that

Esp since I’m not integrating sha256 into my kernel

[slim panther]

ipmi my beloved

[north hull]

well yes but not zen6 lol

random x86 sbc instead

but completely unvirtualized 🙂

[fiery turtle]

Holy

[north hull]

I just need to get some sort of capturecard I can VNC

well maybe two of them, some cheap bs

I kinda wanna VNC both the integrated gpu and the shitty gpu I have on there

lol

[full nimbus]

16650 UART + VNC would be fire

[north hull]

uart is alr done

[full nimbus]

yeah but support both at the same time

[north hull]

well ofc

[full nimbus]

make tabs to switch between them

[north hull]

well I wanna do some sort of api I can integrate into CI

[full nimbus]

(also offtopic?)

[north hull]

the website will just be ontop of that for memes

yeah we can move this

lmao

[fiery ice]

@fiery turtle if i use barebones mode, how can i change this?

and this too

[fiery turtle]

inl()

[fiery ice]

inl?

[fiery turtle]

X86 port io

[fiery ice]

and to map it?

[fiery turtle]

Im not sure what you're asking

[fiery ice]

the uacpi_map_gas function

[fiery turtle]

Im saying you will have to read the counter yourself, you dont need to map ports

[fiery ice]

ah

[fiery turtle]

Anyway, back to 5.0, if someone could test on real hw especially with custom MMIO helpers that would be really nice

[full nimbus]

If you have a test kernel I can use and insert my MMIO helpers into I would but my kernel rewrite isn’t in a proper state

[fiery turtle]

Yeah same lol, I hope to use mine for testing uacpi releases when its more mature

[full nimbus]

I do have a fuck ton of laptops at my house to test on though

[pallid lava]

didnt uacpi os exist?

[full nimbus]

So if you got some ISOs will throw them all onto them

[fiery turtle]

Tomato started it but it was never finished

[fiery turtle]

Lol nice

[pallid lava]

but, isn't it usable to some degree?

[fiery turtle]

Not sure

[full nimbus]

Wait actually

I think my old C++ kernel attempt

That I did with a bunch of others

And ended up dropping out of

Had uACPI working to some degree

Atleast getting a score displayed?

That should be somewhat minimal to cram some bullshit into

[pallid lava]

i would test 5.0 but im very busy rn

next week for sure ill test it

[fiery turtle]

[full nimbus]

fuck

@lethal trench do you remember who got uACPI working at all on there?

When my new phone gets here and shit il make a temp fork of my kernel for this instead

It should have enough memory management shit to make it atleast work enough to test?

It just won’t be able to do threading or interrupt shit

[fiery turtle]

I guess just shutting down is a fine test

[full nimbus]

That should be doable

🤞

As long as nothing has an EC

Oh that would just be power button too so should be fine

[calm latch]

I can rebase this evening

Or like not rebase, but switch to a different commit

Idk you get the ide

[full nimbus]

My kernels goal rn is to get to a very stable base for shit before I do more stuff. So it could be repurposed for a uACPI OS thing easily once I get to the happy base state later

Minimal state being able to load userspace hello world and exit etc and having very solid codebase and structure for everything

[calm latch]

My OS usually works

(keyword usually)

[full nimbus]

Real (for my old kernel atleast, my new one is much better)

[calm latch]

What about uacpi_kernel_mmio_read64/write64 on 32 bit platforms?

Do I implement it as 2 32 bit reads?

I think clang compiles it into that??

[fiery turtle]

Yeah

[lethal trench]

What????

[calm latch]

This is not UB, right?

inline uint64_t mmio_readq(uint64_t *ptr)
{
    uint32_t low, high;
    uint32_t * __attribute__((__may_alias__)) ptr32 = (uint32_t *)ptr;
    asm volatile ("movl %2, %k0 \n\t"
                  "movl %3, %k1"
                  : "=r"(low), "=r"(high)
                  : "o"(*ptr32), "o"(*(ptr32 + 1)) : "memory");
    return ((uint64_t)high << 32) | low;
}
inline void mmio_writeq(uint64_t *ptr, uint64_t data)
{
    uint32_t low = data & 0xFFFFFFFF;
    uint32_t high = data >> 32;
    uint32_t * __attribute__((__may_alias__)) ptr32 = (uint32_t *)ptr;
    asm volatile("movl %k0, %2 \n\t"
                 "movl %k1, %3"
                 :: "ri"(low), "ri"(high), "o"(*ptr32), "o"(*(ptr32 + 1)) : "memory");
}

[slim panther]

what could possibly do 64-bit mmio on a 32-bit machine?

[frank canopy]

it wouldnt make sense but why trust firmware to be normal anyway

better safe than not

[fiery turtle]

I mean also it might be just reading 64 bits from ram

But it probably wouldn't be QwordAccess

Usually its AnyAcc

[calm latch]

i686 build on 64 bit hardware?

[slim panther]

mentally ill if you do that

https://cdn.discordapp.com/attachments/819519506572771388/1424183808302780628/WeDoALittlePosting-ny5nqrc4jnsf1.gif

[calm latch]

Or 32 bit userspace with 64 bit kernel

Anyway, if someone wants to try it

(https://gitlab.com/mishakov/pmos/-/commits/dev?ref_type=heads)

The old version is normal, right?

[calm latch]

cc (and also @fiery turtle)

[fiery turtle]

yeah i havent bumped the actual integer yet

[fiery turtle]

does it shut down on power button press?

[calm latch]

In qemu

I haven't tried it on physical hardware yet

It seems to work on the MacBook as well

[fiery turtle]

thats great

[fiery turtle]

ACPICA has been such a shitshow, some random guy came in with AI bug reports, opened a shit ton of PRs, and the maintainer merged them with barely any review, theres no tests nor CI run required to merge a PR

it caused regressions all over the code base

the compiler is fucked, the disassmebler is fucked

what a joke

i spent maybe 3 hours total fixing shit for uacpi tests that broke due to that

[fiery ice]

bruh

[fiery turtle]

I opened a PR to fix one of the regressions a month ago

0 fucks given

the backlog of prs fixing regressions is growing but the maintainer is gone it seems

[fiery ice]

let's hope this forces linux and other kernels away from it

[full nimbus]

uACPI Linux upstream?

[fiery turtle]

i hope he didnt send that garbage to the linux upstream

theres a reason why they use a fork it seems

because intel doesnt give a shit about the quality of that project

thank god ubuntu 26 ships acpica thats one release behind

[full nimbus]

If ACPICA keeps getting worse

Will things be forced to use uACPI

[fiery turtle]

the only reason it keeps working at all is because a ton of people use it so they are just forced to work as QA

seriously like they didnt even bother slapping a fucking github workflow that would start tests that they have

[fiery turtle]

well no because uacpi doesnt have a compiler/disassembler

so it will have to use that crap

[fiery ice]

well, we now have a new feature for 6.0

lol

[full nimbus]

Well at least for the Linux kernel to not have it regress so bad if they ever have to pull from downstream

[hallow marten]

one day i will vibecode one trust

[fiery turtle]

im seriously considering it

especially since asl syntax is really simple

a token is basically almost always a direct match to an opcode

[hallow marten]

I think there's stuff that's a bit tricky though no?

[fiery turtle]

ive never written a compiler so idk what would be tricky

[hallow marten]

it's not a compiler

barely

[fiery turtle]

yeah its more of a transpiler tbh

since output can be converted back to input without any loss

[full nimbus]

What would be needed for Linux upstreaming anyways?

I know there is some basic support for it on an out of tree repo

[fiery turtle]

compatibility API

basically go through acpica headers, and implement all of those functions using uacpi api

[hallow marten]

should i make one in rust

[fiery turtle]

yes

[full nimbus]

Couldn't a clanker do that with relative ease?

[fiery ice]

yes

[fiery turtle]

probably

[hallow marten]

ok so what is stopping us from claiming uACPI superiority

[full nimbus]

I mean it can't hurt to test it rn

[fiery turtle]

it's a shit ton of tokens

theres a lot of api and context

[fiery ice]

huuh

i have unlimited

if you wanna try

[full nimbus]

If infy was able to get this merged and working that would be fucking huge for both infy in life and the hobby as a whole

And it definitely is capable

[fiery turtle]

i agree but its going to be really tough

[fiery ice]

lemme go grab the 2 repo urls

[full nimbus]

It's worth it though imo

[hallow marten]

no rate limit?

[fiery ice]

nope

[hallow marten]

what

[fiery turtle]

the only possibility is to merge it as an opt-in config, then wait as it gains more trust overtime from cutting edge repos

[hallow marten]

which plan is that

all plans have a rate limit

unless you pay like 39583983948$ of extra usage

[fiery turtle]

infinite bank account

[full nimbus]

CachyOS would probably be open to enabling it

[fiery ice]

perplexity pro

via their web ui

unlimited requests

[hallow marten]

bro

thast like the worst vibecoding

[fiery ice]

and can go and read repos

and it's very good at that

it has direct github access btw

[fiery turtle]

perhaps

[fiery ice]

via the connector

[hallow marten]

i think if u get a 20$/month codex plan u could probably do it in a few sessions

[fiery turtle]

but you have to explain to the clanker that you want to implement compatibility api on top of uacpi so that a kernel can be built against either one with 0 code changes

[fiery ice]

ok, it printed stuff

@fiery turtle wanna try?

it's just 2 files

[fiery turtle]

wdym by 2 files?

[hallow marten]

what

[fiery ice]

1 header and 1 c file

for compat between acpica and uacpi

[fiery turtle]

why would there even be a header

[hallow marten]

it's already done?

[fiery turtle]

the header is already provided by acpica

[fiery ice]

it's still printing 💀

this file is LONG

oh god

[fiery turtle]

what prompt did u even give

[hallow marten]

how is perplexity even profitable if they give out unlimited usage wtf

[fiery ice]

https://github.com/acpica/acpica
https://github.com/uACPI/uACPI

Can you give me the implementation of a compat layer for the Acpica API for uacpi?

the literal prompt

the model was 5.4 thinking

[fiery ice]

they make money from Computer and enterprise plans

it is still going BRUH

HOW LONG IS IT

[fiery turtle]

we should've worded the prompt better but ok

[hallow marten]

ah perplexity relies on other models

[fiery ice]

the header

[hallow marten]

didnt know you could use claude and gpt

[fiery ice]

the c file

[fiery ice]

with the 20 bucks a month plan (9 as a student) you get 5.4 and sonnet 4.6

[fiery turtle]

/**
 * AcpiInitializeTables — map to uacpi_setup_early_table_access().
 * InitialTableArray / InitialTableCount are ignored; uACPI manages its own
 * internal table buffer.
 */
ACPI_STATUS AcpiInitializeTables(void *InitialTableArray,
                                 UINT32 InitialTableCount,
                                 BOOLEAN AllowResize)
{
    (void)InitialTableArray;
    (void)InitialTableCount;
    (void)AllowResize;
    return AE_OK;
}

yeah thats...

[hallow marten]

if i were paying id rather pay for gpt plus

[fiery turtle]

like i said its going to be rough

[hallow marten]

or claude if they stop lobotomizing it

[fiery ice]

summary here

[fiery ice]

claude is expensive AF

even more now with 4.8

[hallow marten]

it doesnt cost more

[fiery ice]

and it's worse than 5.5

[hallow marten]

the plan is hte same price

you just have less usage

I think chatgpt is a better deal atm tho

[fiery ice]

yes

[hallow marten]

in terms of usage

[fiery ice]

and power too

[hallow marten]

yeah 5.5 is really good

[fiery ice]

5.4 doesn't joke either

and 5.5 is locked in the max plan for perplexity

so 200 bucks a month

no way they offer this too 💀

how much are they losing on this

[hallow marten]

they probably do have a limit

you just havent hit it yet

[fiery ice]

tho i saw that on anthropic models you do have limits

on openai nope

[fiery turtle]

"only supports a single descending callback"

/**
 * Depth-first iterate the namespace starting at the first child of 'parent'.
 *
 * 'descending_callback' is invoked the first time a node is visited when
 * walking down. 'ascending_callback' is invoked the second time a node is
 * visited after we reach the leaf node without children and start walking up.
 * Either of the callbacks may be NULL, but not both at the same time.
 *
 * Only nodes matching 'type_mask' are passed to the callbacks.
 *
 * 'max_depth' is used to limit the maximum reachable depth from 'parent',
 * where 1 is only direct children of 'parent', 2 is children of first-level
 * children etc. Use UACPI_MAX_DEPTH_ANY or -1 to specify infinite depth.
 */
uacpi_status uacpi_namespace_for_each_child(
    uacpi_namespace_node *parent, uacpi_iteration_callback descending_callback,
    uacpi_iteration_callback ascending_callback,
    uacpi_object_type_bits type_mask, uacpi_u32 max_depth, void *user
);

this is hopeless tbh

[fiery ice]

and i spammed a ton of stufff

[hallow marten]

we need someone with more tokens to do it

@ korona

[fiery ice]

if i had computer i could really let it loose in agent mode

[hallow marten]

tf you mean if i had a computer lol

[fiery ice]

tho there i have token limits

[full nimbus]

Doesn't mint have that 100 dollar Claude subscription

[hallow marten]

i think korona has 200$ from work

[fiery ice]

no, i mean computer credits for perplexity

[hallow marten]

is it that hopeless you think or not

i still have my gpt plus free trial

[fiery turtle]

its probably possible with a very detailed description and opus 4.8 highest thinking mode

but thats a shit ton of tokens

[hallow marten]

why very detailed description

[fiery ice]

try to gimme one, i'll re-prompt again

[fiery turtle]

to explain how it has to structure it, that it has to actually read headers, how it should test, etc etc

[fiery ice]

@fiery turtle

updated prompt

[hallow marten]

I doubt that is good

[fiery turtle]

we dont need any headers at all

the headers are already provided by acpica

[hallow marten]

yeah

[fiery turtle]

the idea is its a drop in replacement

[fiery ice]

ok updated it a sec to tell it to use acpica headers

[hallow marten]

Task: port linux to uACPI (uACPI/ directory) from ACPICA. To achieve this task, write a compatibility layer/shim on top of uACPI that implements the ACPICA interface. This means
  that we only need headers from ACPICA and nothing else. Ensure the kernel can be compiled with any implementation selected (i.e we can compile the kernel with ACPICA or uACPI).
  Ensure to test the implementation using QEMU.

I wrote this

I think this is good

[fiery turtle]

yeah that is decent

[fiery ice]

i'll run it in parallel and see

[hallow marten]

I will /goal it and watch it work for 12 hours

[fiery turtle]

do u have enough tokens for that?

[hallow marten]

/goal waits until rate limit

[fiery turtle]

ah ok

[hallow marten]

so like it'll literally wait for days until tis done

[fiery turtle]

lol

[fiery ice]

ok, let it loose on this one too

[rustic compass]

build a workflow in n8n that creates the code and tests it and feed it back into the model and then put the api key in the model

[fiery ice]

heck no XD

it will hoover up my money very fast via the api

[rustic compass]

but at least its automated

[fiery ice]

💀

[fiery turtle]

no shit

[fiery ice]

it's been reading for a few minutes rn

lol

[hallow marten]

yeah we're cooked

[fiery ice]

well, at least it will work even if my pc is off XD

@hallow marten with your prompt it's already generating

[fiery turtle]

a lot of uacpi api is actually very very close to acpica so this is definitely doable, its just a lot of work

[hallow marten]

i am something of a prompt engineer

[rustic compass]

tbh maybe someone with a sufficiently large gpu or one of those amd ai max boards could host a good open source model and use n8n to let the model figure out and test the bindgen in the background

[fiery ice]

bruh, why did it go and check my os too 💀

[fiery turtle]

yeah you're cooked tbh

[hallow marten]

bro is cooked

he doesnt know what torture he's about to endure

Task: port linux to uACPI (uACPI/ directory) from ACPICA. To achieve this task, write a compatibility layer/shim on top of uACPI that implements the ACPICA interface. This means that we only need headers from ACPICA and nothing else. Ensure the kernel can be compiled with any implementation selected (i.e we can compile the kernel with ACPICA or uACPI). Ensure to test the implementation using QEMU. You can start with a partial implementation, but ideally do NOT half-bake things, this should be production ready. Avoid adding extraneous comments and anything unnecessary, remember this is a shim, so NO new headers should be needed.

this is the full prompt if u wanna try it

[fiery ice]

oh wait, my prompt is starting too

the structure it wants

i'll just share the prompt once done XD

so you can copy the files manually

[hallow marten]

this looks stupid still

im just using a local linux copy i have and uacpi and getting it to write code

and I'll send a patch

[fiery ice]

i'm still not seeing any stub, so it is good for now i think

[fiery turtle]

Well the internal header is fine

[fiery ice]

oh and it's making tests too

fuck

[fiery turtle]

The problematic stuff may be when it looks at random internal acpica global variables that Linux uses in some places

[fiery turtle]

Thats not a problem, unload is not used I dont think

[hallow marten]

we should get anthropic to sponsor it

like bun

[fiery turtle]

And the unload opcode has never seen any real use nor is it supported by acpica even

[fiery ice]

https://www.perplexity.ai/search/task-implement-an-acpica-to-ua-xPTMpIBcQNSVFg3Ojj7wWA

[fiery turtle]

First thing I read is already bullshit

[fiery ice]

and here the full chat for this

[fiery turtle]

Uacpi has no teardown api

[hallow marten]

simple enough

[fiery turtle]

Did bro read the headers?

[fiery ice]

yeah the .md was a second prompt

so it may have hallucinated the second one

[hallow marten]

@fiery turtle why dont you just do this

are you like

[fiery ice]

the first one should be accurate tho

[hallow marten]

stupdi??

[hallow marten]

yeah ok it seems to not be completely dumb

[fiery turtle]

Yes

[fiery turtle]

Yeah that's good

[hallow marten]

it hasnt even started and it's at like 25k tokens already 💀

[fiery turtle]

Alr im gonna go to bed and see what your clanker comes up with tomorrow

[hallow marten]

i dont think we're getting this done tonight guys

[fiery ice]

did you check the files that chat generated btw

[fiery turtle]

Yeah definitely not

[fiery ice]

gonna leave perplexity running overnight too

[fiery turtle]

Where?

[fiery ice]

these

[hallow marten]

it's doing kconfig

[fiery turtle]

Yeah looks somewhat ok

[fiery ice]

if you wanna try them

tmr

[fiery turtle]

I mean if it can even get something that compiles

[fiery ice]

well, it's worth a try

[hallow marten]

mine is getting something to compile atm

[fiery ice]

also, that chat comes with a cmake file

[fiery turtle]

Cmake file in Linux?

💀

[hallow marten]

i think that kinda stuff is much better to do with a cli agent tool

[fiery ice]

nah, it's just to build the tests

[hallow marten]

because it can integrate nicely

[fiery ice]

true

[hallow marten]

like mine has written kconfig and makefile for uacpi already

[fiery turtle]

Cool

[fiery ice]

i don't wanna pay for tokens, lol

it's too expensive rn

[hallow marten]

does this make sense

[fiery turtle]

Ofc

[hallow marten]

yeah ok the linux shim looks great

the kernel functions it implemented

[fiery ice]

let's try with yours first then

[hallow marten]

looks all good to me

[fiery turtle]

The shim is the first tiny step

[hallow marten]

this is the only weird thing so far

but yeah its obviously the easiest thing

[fiery turtle]

Wtf is that code alignment

But the code itself is correct

[hallow marten]

this is not in the text editor

it looks fine in a text editor

it actually really looks good so far

might actually work lol

[fiery turtle]

Great

[fiery ice]

With intCast's prompt it gave me this

[hallow marten]

what is that

[fiery ice]

well, gonna go to sleep, cya tmr

[hallow marten]

https://tenor.com/view/fire-writing-writing-fire-his-writing-is-fire-is-this-fire-hold-up-gif-15400153686296022809

I think it's actually cooking

holy shit it worked

it linked

and ran

[slim panther]

wtf are yall doing

[hallow marten]

enslaving a clanker to port linux to uacpi

[slim panther]

huge

[hallow marten]

linux SUCKS

[hallow marten]

this was fake

ok now it works

kinda

theres still a bunch of stubs

but it gets linux through boot and reboot() works

[steel inlet]

wdym

[hallow marten]

it still built with ACPICA or ran the ACPICA version somehow

[steel inlet]

ah

[hallow marten]

but now it does run with uACPI

[steel inlet]

how does it do in the benchmark

[hallow marten]

?

[steel inlet]

this

since this you said was wrong

[hallow marten]

this isnt accurate yet if I understand correctly, also this was TCG

[steel inlet]

ahhh

[hallow marten]

this was right

[steel inlet]

mb discord mobile moment

it went to the wrong message

[hallow marten]

it stubbed the resource api

so

[steel inlet]

ah

[hallow marten]

but I think I'll run out of usage before it can fix it

[full nimbus]

Opencode lets you have a free local agent

It's not that bad

[hallow marten]

this is opus 4.8

[full nimbus]

And you can let it do the final fixes

They give you something called

"big pickle"

[hallow marten]

this is the kinda stuff that needs thinking

i could give it to codex 5.5 xhigh tho

but id rather wait till tomorrow

[full nimbus]

True

This will be sick when it's working

If it's cleaned up upstream hopefully

[hallow marten]

well it does work

but its a bit hacked together

[full nimbus]

Yeah but fully working and put together a bit better etc

[hallow marten]

i did make it write a proper document outlining how to work on it tho so maybe we could polish that

i feel like discount monkuous

he would've done it by hand

😔

[fiery turtle]

Lmfao

*in a few hours

But no this is pretty cool honestly

[full nimbus]

And it gives a good base regardless for us to know what to do or whatever else

Even if we don't use it outright

[fiery turtle]

Yeah

[kind mantle]

uACPI in Linux would be good but like how are you getting that merged

Even if you had the shim

[torpid root]

I don't doubt that infy could get it merged

[full nimbus]

there isnt any reason it shouldnt get merged

its am extra AML engine that can help hardware compat

and it is a technically supeiror one

[calm latch]

Did you only refer to Ultra kernel or any iso?

[full nimbus]

any ISO

[calm latch]

Did you miss this?

[full nimbus]

yes

il test soon™

[torpid root]

<old link>

it's a bit large because the sysroot is included in it

[fiery turtle]

is that on uacpi 5.0?

[torpid root]

oops

I thought this was just "I got shitton of hardware gimme ur isos" lol

[fiery turtle]

lol

nah I was asking people to test uacpi 5.0 before i merge it

[full nimbus]

it still is but it has to be uACPI 5.0

[torpid root]

lemme see if I can quickly update

[fiery turtle]

bonus points for UACPI_NATIVE_MMIO

or ig we already tested it with pmos, maybe builtin mmio is better to test

[calm latch]

My AMD laptop with GPIO button also seemingly works

[torpid root]

is this not updated yet or did I compile the wrong source

[fiery turtle]

i havent bumped the integer yet yes

[fiery turtle]

nice

[torpid root]

@full nimbus https://safe.ilobilo.com.ge/image-qyRQxA6X.iso

[calm latch]

Also, mine is both the latest table only in kernel and in userspace

[torpid root]

hmmm. am I missing something?

[calm latch]

Though I've noticed there are some errors?

Maybe my stuff is just broken again

_SB_.MHSP

[fiery turtle]

looks like its trying to use EC before u register it

u probably register it a bit too late

[fiery turtle]

wdym?

[calm latch]

My EC driver is broken

I've just noticed

[torpid root]

nothing, I think I just somehow regressed my uacpi score

didn't even notice until now

[fiery turtle]

ah

[calm latch]

This one has working EC

[calm latch]

You said that was fine, but there's an error right after it prints the tables

[fiery turtle]

thats just a warning but yeah

it has a duplicate object in there

u will probably see this on linux also

[torpid root]

sending shootdown to every core on uacpi unmap

I'm a genius

[fiery ice]

GG

linux to uacpi finally?

[calm latch]

I kinda do that as well?

[flat badge]

yeah that's needed unless you have other tricks to avoid it

[calm latch]

I mean if if only a single thread of the server is running, there are no remote shoptdowns

[torpid root]

sure but no other core accesses uacpi during init

[hallow marten]

Ideally you batch them

[torpid root]

I know

[hallow marten]

Or ideally you do them asynchronously

Which I described before but I forgot

[fiery turtle]

Alright release 5.0 is officially live https://github.com/uACPI/uACPI/releases/tag/5.0.0

@lofty dragon FYI as u have used the for-5.0 branch before

[fiery ice]

gg

[hallow marten]

@fiery turtle i think its done

[fiery ice]

gg

[hallow marten]

can I send boot logs for you to check cuz idk if they're right

[fiery turtle]

Which part?

I wonder how it solved resource conversion and stuff like that

Like it looks vaguely correct but I don't remember qemu Linux acpi boot logs by heart

[fiery turtle]

Since u have a kconfig option u can try the acpica boot logs and compare them

[hallow marten]

right

[fiery turtle]

Holy

[hallow marten]

hey how cool is that

[fiery turtle]

Nice

[hallow marten]

is that good

[fiery turtle]

Thats the only way to do I guess

The other way would be a small bytecode

[hallow marten]

why is it so different

mhm this is with ACPICA

there's a bunch of different stuff

[fiery turtle]

Well because those resources are not part of the spec, rather its a "convenient" mapping of them from aml to C

And how to do them is up to the library

For most of them its just assigning one field to the other

Like its the same thing in reality

[fiery turtle]

Yep

[hallow marten]

in uacpi there's no "Added _OSI"

[fiery turtle]

Looks like it maybe stubbed a few things? Or didnt figure out how to use uacpi osi helpers

[fiery turtle]

There's also nothing about _OSC

[hallow marten]

there's no in acpica eihter

it fixed it

I think this is fine

I will send the patches if anyone is interested

[fiery turtle]

You should rebase to 5.0, there's now by index api and a few other things

[fiery turtle]

What does it mean by acquire by path

[hallow marten]

no idea, I'm updating it to 5.0 atm

[fiery turtle]

Can you ask it

[fiery turtle]

Huh, thats an interesting api ig, ill grep the linux source to see if thats actually used by anything

Maybe worth adding

But it sounds a bit insane

Since aml paths are not standardized obviously

Nor are object names

[hallow marten]

it's used like 3 times

[fiery turtle]

Which files?

[hallow marten]

drivers/hwmon/asus-ec-sensors.c
947:    return ACPI_SUCCESS(acpi_acquire_mutex(data->mutex.aml,

drivers/platform/chrome/cros_ec_lpc_mec.c
43:     success = ACPI_SUCCESS(acpi_acquire_mutex(aml_mutex,

include/acpi/acpixf.h
700:                        acpi_acquire_mutex(acpi_handle handle,

[fiery turtle]

Lol I see

[hallow marten]

ok yeah now sysfs works with the new api

[fiery turtle]

Nice

[hallow marten]

very cool

[fiery turtle]

Good work

[hallow marten]

yeah i worked so hard

[fiery turtle]

Try booting it on real hw

[fiery turtle]

Lol

[hallow marten]

i will send a humonguous patch if anyone wants to do so

[fiery turtle]

Maybe an ISO

[fiery ice]

what agent you using btw

[fiery turtle]

As well

[hallow marten]

but then i have to have an initrd

and stuff

[fiery ice]

i can boot an iso rn on my resurrected laptop

the one that pmOS had a problem on

[fiery turtle]

Clanker can make a bootable ISO almost instantly

[hallow marten]

mhm ok

[fiery turtle]

With like busybox

[hallow marten]

opus 4.8

[fiery ice]

oh

[fiery turtle]

Did it consume your entire subscription

[fiery ice]

money hoover 4.8

[hallow marten]

barely

[fiery ice]

💀

[fiery turtle]

Huh

[fiery ice]

how much you paying?

[fiery turtle]

Interesting

[hallow marten]

im not but this is the cheapest plan

[fiery ice]

wait what+

the 20 bucks a month one??

[hallow marten]

yea

[fiery ice]

HOW

i heard it was so bad for limits

[hallow marten]

well im not using it every day lol

it did use like ~50%

for this thing

[fiery ice]

💀

[hallow marten]

if you use it everyday and vibecode a lot I can see how the limit would be low

[fiery ice]

well, give us an iso to test on real hw

[hallow marten]

im trying

@fiery turtle @fiery ice

[fiery ice]

k

booting

[hallow marten]

it's pretty cool lol

[fiery ice]

what booting opting should i use

[hallow marten]

just boot the first entry

[fiery ice]

test shell or verbose debug

k

it's stuck at loading initramfs 💀

how much should it take

[hallow marten]

shouldnt take long

[fiery turtle]

Nice

[fiery ice]

it's loading for like 3 mins

[hallow marten]

ok then it may have crashed somewhere

[fiery turtle]

Maybe it died

[hallow marten]

try the debug logs thing

[fiery turtle]

Do u have earlycon enabled?

[fiery ice]

still nope

i'm gonna try loading it as ramdisk

[hallow marten]

?