uACPI - a portable and easy-to-integrate ACPI implementation | Operating System Development | Page 7

fiery turtle Aug 6, 2024, 11:22 PM

#

Didn't know qemu was in such a bad state on ia64

deft canopy Aug 6, 2024, 11:23 PM

#

nothing really supports ia64 emulation

#

there's ski, but it's kinda iffy

#

simics used to support it partially, but was missing key parts

#

qemu-ia64 was part of a summer of code thingy

fiery turtle Aug 6, 2024, 11:23 PM

#

Interesting

#

Why is that?

deft canopy Aug 6, 2024, 11:24 PM

#

and then chickenes made a fork semi-recently to work on it more

deft canopy Aug 6, 2024, 11:24 PM

#

fiery turtle Why is that?

people think it's impossible to properly emulate ia64 (they're stupid imo)

fiery turtle Aug 6, 2024, 11:24 PM

#

How come do they think that? (Im clueless about ia64)

deft canopy Aug 6, 2024, 11:28 PM

#

not really sure, I think it's because it's VLIW

fiery turtle Aug 6, 2024, 11:29 PM

#

Ah

deft canopy Aug 6, 2024, 11:31 PM

#

the main issue is no one has really been motivated enough to implement one that's fully functional

fiery turtle Aug 6, 2024, 11:32 PM

#

Interesting how its less supported than Amiga lol

#

I guess its way more complex

#

And its only for server hardware thats hard to come across

deft canopy Aug 6, 2024, 11:35 PM

#

not quite, it was intended to support consumer use, it just wasn't popular enough
there are systems like the Zx2000 or Zx6000 which could've made great consumer systems, but due to how unpopular and expensive Itanium was they were only really used as workstations
there were also attempts at mobile Itanium chips which could've been great for things like laptops, or low-power systems but those didn't get super far

#

I believe @heady pewter mentioned that there was a cancelled project for a mobile Itanium implementation

fiery turtle Aug 6, 2024, 11:36 PM

#

Oh cool

#

Didn't know about those

deft canopy Aug 6, 2024, 11:36 PM

#

deft canopy not quite, it was intended to support consumer use, it just wasn't popular enoug...

I'm not super sure that early Itanium chips were massively expensive, I would like to note that
the reason I assume they were is how expensive recent chips were, often being thousands at retail price iirc

heady pewter Aug 6, 2024, 11:37 PM

#

yes, there doesn't seem to be anything about it on the Internet

#

I'll have to see if I can track down the original architect and see if he is willing to comment

deft canopy Aug 6, 2024, 11:37 PM

#

fiery turtle Didn't know about those

I think Zx6000s are still sometimes sold on Ebay, and recently I saw a lovely Zx2000, sadly couldn't afford it

deft canopy Aug 6, 2024, 11:37 PM

#

heady pewter I'll have to see if I can track down the original architect and see if he is wil...

If you can I'd love to see what he has to say

heady pewter Aug 6, 2024, 11:37 PM

#

Itaniums were cheap from HP, because Intel had to supply them at cost 😛

fiery turtle Aug 6, 2024, 11:38 PM

#

deft canopy If you can I'd love to see what he has to say

Yeah same

deft canopy Aug 6, 2024, 11:38 PM

#

heady pewter Itaniums were cheap from HP, because Intel had to supply them at cost 😛

wdym? like the Hondo implementation?

heady pewter Aug 6, 2024, 11:40 PM

#

I forget for how long

#

it was part of the partnership

deft canopy Aug 6, 2024, 11:42 PM

#

oh that, did it reduce the cost of Itanium chips?

heady pewter Aug 6, 2024, 11:48 PM

#

for HP 🙂

deft canopy Aug 6, 2024, 11:50 PM

#

Ok that makes sense

sterile egret Aug 7, 2024, 6:02 AM

#

There was a Google Summer of Code project years ago to add IA64 support to qemu, but the person really didn't get all that far. I actually didn't realize how primitive it was until AFTER I managed to apply 13 years of QEMU changes meme I ended up ignoring everything and building a new skeleton, dumping every bit of GSoC code...but that was on top of the original git repo, not qemu's, so I need to apply the changes/additions to qemu master...and I got distracted by other things before I did the latter meme

#

The GSoC thing had like a handful of instructions and the instruction decoder was hacked in from a BSD licensed debugger that I believe was already a dead project in 2011 lmao

#

ski is a good reference, but it's missing a fair bit.

rustic compass Aug 7, 2024, 10:53 AM

#

@fiery turtle I got ownership of the uacpi-rs repository

deft canopy Aug 7, 2024, 11:18 AM

#

sterile egret The GSoC thing had like a handful of instructions and the instruction decoder wa...

Lmao

fiery turtle Aug 7, 2024, 12:13 PM

#

rustic compass <@353593948523986945> I got ownership of the uacpi-rs repository

Daamn

#

Thats big

fiery turtle Aug 11, 2024, 8:55 PM

#

uacpi_state_reset vs uacpi_tear_down?

#

acpica does acpi_terminate which is horrible imo, its ambiguous whether it's related to power states

#

for reference, this is what it does:

fiery turtle Aug 11, 2024, 9:36 PM

#

also can't believe i didnt have this check before

torpid root Aug 11, 2024, 9:58 PM

#

fiery turtle `uacpi_state_reset` vs `uacpi_tear_down`?

uacpi_state_reset

north holly Aug 11, 2024, 11:30 PM

#

fiery turtle `uacpi_state_reset` vs `uacpi_tear_down`?

uacpi_implode_all

kind mantle Aug 11, 2024, 11:31 PM

#

fiery turtle `uacpi_state_reset` vs `uacpi_tear_down`?

uacpi_exterminate

#

https://tenor.com/bq9dw.gif

Tenor

left orbit Aug 12, 2024, 7:29 AM

#

fiery turtle also can't believe i didnt have this check before

you forgot the second argument

hollow elm Aug 12, 2024, 7:36 AM

#

fiery turtle also can't believe i didnt have this check before

adding __attribute__((format(printf, 1, 2))) to uacpi_warn would've issued a warning in this case :^)

fiery turtle Aug 12, 2024, 8:02 AM

#

left orbit you forgot the second argument

yes yes i noticed right after i sent it meme

fiery turtle Aug 12, 2024, 8:03 AM

#

hollow elm adding `__attribute__((format(printf, 1, 2)))` to `uacpi_warn` would've issued a...

been a thing for a while

fiery turtle Aug 17, 2024, 7:21 PM

#

https://github.com/aarch64-laptops/build/tree/master/misc

GitHub

build/misc at master · aarch64-laptops/build

Build an Linux OS based image. Contribute to aarch64-laptops/build development by creating an account on GitHub.

#

found a new collection of AML dumps

vast kestrel Aug 17, 2024, 9:18 PM

#

This feels like pokemon

wind fiber Aug 17, 2024, 9:26 PM

#

Gotta catch them all

vast kestrel Aug 17, 2024, 9:27 PM

#

All the broken blobs

fiery turtle Aug 17, 2024, 10:02 PM

#

vast kestrel This feels like pokemon

because it is

#

I think this should work for non-recursive wipe of the entire namespace

void uacpi_namespace_deinitialize(void)
{
    uacpi_namespace_node *current, *next = UACPI_NULL;
    uacpi_u32 depth = 1;

    current = uacpi_namespace_root();

    while (depth) {
        next = next == UACPI_NULL ? current->child : next->next;

        /*
         * The previous value of 'next' was the last child of this subtree,
         * we can now remove the entire scope of 'current->child'
         */
        if (next == UACPI_NULL) {
            depth--;

            // Wipe the subtree
            while (current->child != UACPI_NULL)
                uacpi_node_uninstall(current->child);

            // Reset the pointers back as if this iteration never happened
            next = current;
            current = current->parent;

            continue;
        }

        /*
         * We have more nodes to process, proceed to the next one, either the
         * child of the 'next' node, if one exists, or its peer
         */
        if (next->child) {
            depth++;
            current = next;
            next = UACPI_NULL;
        }

        // This node has no children, move on to its peer
    }
}

fiery turtle Aug 18, 2024, 11:46 AM

#

okay im back from my minecraft hiatus

#

first off now the initialization is atomic, meaning you can't end up with partially initialized uACPI due to some error

#

as a bonus i also added public api to reset api back to initial state

north holly Aug 18, 2024, 11:47 AM

#

is there finally a lock

#

like is it thread-safe

fiery turtle Aug 18, 2024, 11:48 AM

#

#

this much is still to do in terms of locks

#

only OSI and tables are threads safe right now

#

okay uhh some large tests are now failing due to memory leaks

#

this happens because i now reset uACPI after every run

#

this probably uncovered a preexisting bug where i would leak a reference or something

#

but never encountered it because nodes would always still be alive on program exit

fiery turtle Aug 18, 2024, 12:30 PM

#

okay so as far as i can see its related to how deep object copy is handled, which is only used by NameOp...

#

but i cant come up with a repro atm, will have to dig deeper and actually look at the dsdts

#

only 700 cases of Name to look through in the reporducer blob...

#

okay uhh this fixes all leaks and all tests pass...

north holly Aug 18, 2024, 12:42 PM

#

has the oberrow curse gotten to the reaches of the original thing that cursed obos (uacpi)

fiery turtle Aug 18, 2024, 12:42 PM

#

nah its a bug thats always been there, just never showed itself because the global namespace was never freed before

#

basically a reference leak somewhere

#

anyway i need to figure out wtf is going on and for that i need a way to reproduce the leak

#

also i lied the bug didnt go anywhere, only the symptoms changed

#

aka the source of the leak changed

#

so it wasnt the deep copy thats at fault here

#

could it be the alias i wonder

#

Nope, the reporducer blob only has aliases to processors and the leaked object is a buffer

#

❯ runner/build-linux-64bits/test-runner bin/large-tests/tablet_microsoft_surface_surface_laptop_3_4e426ab8062d/dsdt.dat --log-level warning
[uACPI][ERROR] unable to lookup named object 'EPCS' within (or above) scope '\_SB_.EPC_._STA': not found
[uACPI][ERROR] aborting method invocation due to previous error: AML referenced an undefined object
[uACPI][ERROR]     #0 in \_SB_.EPC_._STA()
[uACPI][ERROR] failed to disable fixed event 0

=================================================================
==11399==ERROR: LeakSanitizer: detected memory leaks

Direct leak of 32 byte(s) in 1 object(s) allocated from:

#

okay found a blob with exactly one leak

#

this will hopefully help track it down

#

i have invented an absolutely genius object leak track down

#

technique

#

#

#

now i know exactly which node is responsible for the leak

#

because of the leak size

#

...aaand its a field connection

#

actual genius debug strat

#

and sure enough that was missing there

#

the only reason why i never saw this leak before was because no blob ever creates a temporary field unit with a connection

#

@north holly curse solved (also do u like my debug strat?)

north holly Aug 18, 2024, 1:17 PM

#

yes

fiery turtle Aug 18, 2024, 1:17 PM

#

thanks

#

also i kinda hate that this is unrefed recursively

#

but the connection must be a buffer object so

#

it cant recurse further

#

oh yeah its typechecked

#

so its not going to recurse

#

thats two more things off the checklist

rustic compass Aug 18, 2024, 1:28 PM

#

whats the next todo you plan to do?

fiery turtle Aug 18, 2024, 1:28 PM

#

the thing we talked about, aka the last box in the checklist

fiery turtle Aug 18, 2024, 2:42 PM

#

Now that i can cleanly reset global state i can literally run the same test twice and they all pass!

#

not a single leak or variable i forgot to reinitialize

wind fiber Aug 18, 2024, 6:18 PM

#

Nice work

fiery turtle Aug 18, 2024, 6:21 PM

#

Thanks!

fiery turtle Aug 19, 2024, 9:08 PM

#

wake up babe new uACPI kernel api just dropped

#

this time its optional and just convenience stuff

#

i dont like the name of this define

#

because its only part of what it does

#

but idk of a better idea

jaunty fox Aug 19, 2024, 9:12 PM

#

fiery turtle this time its optional and just convenience stuff

what does it?

fiery turtle Aug 19, 2024, 9:13 PM

#

jaunty fox what does it?

e.g. in managarm https://github.com/managarm/managarm/blob/67656190ccb7846f3e04be2f550126ad0bbdb520/kernel/thor/system/acpi/madt.cpp#L336 this wouldnt have to be done manually anymore

GitHub

managarm/kernel/thor/system/acpi/madt.cpp at 67656190ccb7846f3e04be...

Pragmatic microkernel-based OS with fully asynchronous I/O - managarm/managarm

#

and wouldnt have to be exported anywhere

#

as uACPI would call it on its own

#

basically if your glue needs any initialization or state, this will allow for automatic initialization of that

#

aka just init hook

#

added some description on top

#

or maybe i should reword that to "glue code that needs global constructable state in order to work"

left orbit Aug 20, 2024, 1:55 PM

#

fiery turtle but idk of a better idea

weak symbols

fiery turtle Aug 20, 2024, 2:28 PM

#

Not a thing on msvc

#

But yes more elegant

north holly Aug 20, 2024, 8:26 PM

#

@fiery turtle is it possible for uacpi to use after free while being accessed in a multithreaded context?

#

I'm asking because I get a use after free in uACPI in types.c:1065 (uacpi_unwrap_internal_reference)

#

and the call stack goes down to uacpi_find_devices

#

while another CPU was running uacpi_get_pci_routing_table

#

yeah it's something to do with thread safety

#

Day one of waiting for uACPI thread safety

#

(I plan on saying that every day in this thread until uACPI is thread safe)

fiery turtle Aug 20, 2024, 9:04 PM

#

It is possible yes

fiery turtle Aug 20, 2024, 9:05 PM

#

north holly Day one of waiting for uACPI thread safety

Thats my number one priority

#

Aka just wrapping uacpi up for a 1.0 release

#

Rn you can just wrap any uacpi invocation that might race under one mutex

north holly Aug 20, 2024, 9:07 PM

#

ok

fiery turtle Aug 20, 2024, 9:08 PM

#

north holly while another CPU was running `uacpi_get_pci_routing_table`

I can't immediately think how that would race with anything, but possible probably

north holly Aug 21, 2024, 12:41 PM

#

Day two of waiting for uacpi thread safety

fiery turtle Aug 21, 2024, 12:44 PM

#

Lmao

tired crater Aug 21, 2024, 2:09 PM

#

he REALLY REALLY wants it

north holly Aug 21, 2024, 2:26 PM

#

north holly (I plan on saying that every day in this thread until uACPI is thread safe)

.

#

(when it finally is, I will start saying "Day n of uACPI thread safety) /j

north holly Aug 22, 2024, 6:42 PM

#

day three of waiting for uACPI thread safety

fiery turtle Aug 22, 2024, 6:44 PM

#

Rookie numbers, I've been waiting for a few months

vast kestrel Aug 22, 2024, 6:59 PM

#

Add a global lock

#

And later you add an optimization release that magically improves the multi-threaf performance

fiery turtle Aug 22, 2024, 7:07 PM

#

yeah for 1.0 i think ill do a GIL

#

but other subsystems will have granular locks

deft canopy Aug 22, 2024, 8:47 PM

#

vast kestrel Add a global lock

i think we should have UNfine locking everywhere
the OS should completely lock whenever a core enters supervisor mode
!!!!!!!!!!!!!!!!!!!!!!!!!!!

fiery turtle Aug 22, 2024, 8:49 PM

#

a global kernel lock is relatively common for older kernels espeically when a lot of cores werent common

deft canopy Aug 22, 2024, 8:50 PM

#

ya, but it's a bad idea on SMP systems, which tend to be the most common in modern desktop/server systems

fiery turtle Aug 22, 2024, 8:50 PM

#

yeah

vast kestrel Aug 22, 2024, 8:54 PM

#

Didn't freebsd have one til recently

#

Or still maybe

fiery turtle Aug 22, 2024, 8:54 PM

#

maybe

jaunty fox Aug 22, 2024, 8:57 PM

#

vast kestrel Didn't freebsd have one til recently

all the BSDs (and Linux too) had one within recent memory

#

that's how they started to smp'ify

#

freebsd's one has had limited use for long time though

vast kestrel Aug 22, 2024, 8:59 PM

#

I think the Linux one was removed in like Linux 2.6.30 or something

#

I remember freebsd still has their gil but not for everything

jaunty fox Aug 22, 2024, 9:00 PM

#

https://github.com/search?q=repo%3Afreebsd%2Ffreebsd-src+path%3Asys+"mtx_lock(%26Giant)%3B"&type=code these appear to be the remaining uses of the giant lock in freebsd

fiery turtle Aug 22, 2024, 9:01 PM

#

lmao

#

Giant

vast kestrel Aug 22, 2024, 9:01 PM

#

Til the concept is called a giant lock lol

#

https://en.m.wikipedia.org/wiki/Giant_lock

jaunty fox Aug 22, 2024, 9:02 PM

#

kernel module loading/unloading, some keyboard drivers, a few random drivers, smb (the network fs?), some terminal stuff

fiery turtle Aug 22, 2024, 9:02 PM

#

some stuff in acpi

#

but not too bad honestly

vast kestrel Aug 22, 2024, 9:02 PM

#

Yeah

fiery turtle Aug 22, 2024, 9:03 PM

#

god forbid my keyboard LED event races with module loading

#

this is known to kill people

vast kestrel Aug 22, 2024, 9:03 PM

#

Exactly

jaunty fox Aug 22, 2024, 9:04 PM

#

vast kestrel https://en.m.wikipedia.org/wiki/Giant_lock

mach had one called the unix funnel

#

later the bsd funnel

fiery turtle Aug 22, 2024, 9:04 PM

#

like the cake

jaunty fox Aug 22, 2024, 9:07 PM

#

https://github.com/evolver56k/Darwin-0.3/blob/745305afd074437bb660bcf02146a2754eeef5ae/kernel-7/kern/parallel.c#L45 mach's funnel

#

unix_master() is much cooler than mtx_lock(&Giant)

torpid ferry Aug 22, 2024, 9:32 PM

#

vast kestrel Didn't freebsd have one til recently

openbsd still does it, mostly

#

iirc they got rid of Giant entirely a while back

#

now they just have like 3 massive locks

jaunty fox Aug 22, 2024, 9:38 PM

#

openbsd is moving very slowly to wreck their giant locking (which they call the kernel lock or mp lock)

#

a lot of the netbsd inheritance was already equipped with fairly fine-grained locking (like UVM) but they've trashed it and are reconstructing it gradually, e.g. in https://github.com/openbsd/src/commit/19dcab733a71ecac0688127acdc039e90a471eba

north holly Aug 23, 2024, 2:16 PM

#

day four of waiting for uACPI thread safety

fiery turtle Aug 23, 2024, 2:34 PM

#

Same

kindred beacon Aug 23, 2024, 3:14 PM

#

Be the change you want to see ultrameme

fiery turtle Aug 23, 2024, 3:26 PM

#

Lmao

deft canopy Aug 23, 2024, 6:04 PM

#

kindred beacon Be the change you want to see <:ultrameme:729026898666979428>

maybe I shoudl take a shit ton of weed and submit a PR with my own thread safety

north holly Aug 24, 2024, 10:17 AM

#

Day 5 of waiting for uacpi thread safety

north holly Aug 24, 2024, 11:29 PM

#

well seems like uACPI has caused lai to be completely unmaintained

#

fiery turtle Aug 24, 2024, 11:32 PM

#

that last patch there is a fix for a bug i found in lai lol

deft canopy Aug 24, 2024, 11:43 PM

#

north holly

Oops

hollow elm Aug 24, 2024, 11:47 PM

#

north holly well seems like uACPI has caused lai to be completely unmaintained

it wasn't really being maintained before that either :^)

hasty plinth Aug 25, 2024, 8:41 AM

#

https://github.com/UltraOS/uACPI/blob/master/include/uacpi/uacpi.h#L57C66-L57C71 found a spelling mistake (https://github.com/UltraOS/uACPI/blob/master/include/uacpi/utilities.h#L98C17-L98C26 this too)

fiery turtle Aug 25, 2024, 9:15 AM

#

Thanks, will fix

north holly Aug 25, 2024, 3:02 PM

#

day 6 of waiting for uacpi thread safety

jaunty fox Aug 25, 2024, 3:14 PM

#

i should take a look at my storport shim again some day

#

https://learn.microsoft.com/en-us/windows-hardware/drivers/ddi/storport/nf-storport-storportinvokeacpimethod this would be fun to implement

fiery turtle Aug 25, 2024, 4:17 PM

#

jaunty fox https://learn.microsoft.com/en-us/windows-hardware/drivers/ddi/storport/nf-storp...

yeah sounds interesting

dusky glade Aug 25, 2024, 7:41 PM

#

hey does uacpi have rust bindings?

north holly Aug 25, 2024, 7:42 PM

#

I think someone made some

#

but it's defintely far too early for you to port it

dusky glade Aug 25, 2024, 7:43 PM

#

yea fair point

fiery turtle Aug 25, 2024, 7:46 PM

#

rustic compass <@353593948523986945> I got ownership of the uacpi-rs repository

yes, see this repo

#

although last commit there is may

rustic compass Aug 25, 2024, 7:52 PM

#

developement is planned but currently halted

dusky glade Aug 25, 2024, 7:53 PM

#

understandable

rustic compass Aug 25, 2024, 7:53 PM

#

i will start once i need it for my kernel, which can be anywhere between next week and next year 😛

fiery turtle Aug 25, 2024, 7:53 PM

#

lol

dusky glade Aug 25, 2024, 7:53 PM

#

😭 😭

fiery turtle Aug 25, 2024, 7:53 PM

#

maybe nyaux gets there before u

dusky glade Aug 25, 2024, 7:54 PM

#

watch me

#

😎

fiery turtle Aug 25, 2024, 7:54 PM

#

will do

deft canopy Aug 25, 2024, 8:25 PM

#

rustic compass developement is planned but currently halted

would you be open to PRs?

rustic compass Aug 25, 2024, 8:27 PM

#

deft canopy would you be open to PRs?

in the future yes, but atm its likely that i will restructure the code from the ground up which will break pr code

#

its in a state in which i wouldnt recommend creating prs

#

but ideas are always welcome

fiery turtle Sep 1, 2024, 10:27 AM

#

Noticed that ACPICA has an oversight where the kernel api initialization hook is only called once during early init, which is not enough because different stages of init need different APIs, e.g. work queues don't become available until later

#

so linux had to invent their own acpi_os_initialize1 to combat this oversight and then call into it manually

#

Thats why i decided to make the notion of uacpi_init_level public, as well as call the kernel init hook for each level when appropriate

#

basically this

#

and this is now public

hasty plinth Sep 1, 2024, 10:36 AM

#

nice, sounds handy

fiery turtle Sep 1, 2024, 10:41 AM

#

Thanks, I know im definitely going to use it in Astral

fiery turtle Sep 1, 2024, 11:25 AM

#

hasty plinth <https://github.com/UltraOS/uACPI/blob/master/include/uacpi/uacpi.h#L57C66-L57C7...

thanks, fixed and gave u credit in the commit description

hasty plinth Sep 1, 2024, 11:27 AM

#

ah not necessary, but thank you anyway 🙏

dusky glade Sep 2, 2024, 9:49 AM

#

rustic compass <@353593948523986945> I got ownership of the uacpi-rs repository

where is that repo

#

i dont care if its not mantained i need to use it like right now

mortal yoke Sep 2, 2024, 9:50 AM

#

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

dusky glade Sep 2, 2024, 9:50 AM

#

thank you

fiery turtle Sep 2, 2024, 9:51 AM

#

i dont think its uhh

#

up to date

#

but with a bit of work it might be usable again

mortal yoke Sep 2, 2024, 9:52 AM

#

and ig all of those sysdep functions might not be needed anymore, didn't you add a lot of default impls to uacpi?

fiery turtle Sep 2, 2024, 9:52 AM

#

i did indeed

#

basically none of them are needed

dusky glade Sep 2, 2024, 9:54 AM

#

@rustic compass is there documentation

#

for the crate

#

nvm i know how to set it up

#

fn install_interrupt_handler(&self, irq: u32, handler: Box<dyn Fn()>)
            -> Result<uacpi::Handle, uacpi::Status> {
        unimplemented!()
    }

box is not defined???

#

???

#

what is this trait

#

box isnt defined

fiery turtle Sep 2, 2024, 10:19 AM

#

Resources:

Documentation on the wiki: https://osdev.wiki/wiki/UACPI
Read the headers!
Rust bindings (incomplete): https://github.com/Shinribo/uacpi-rs

OSDev.wiki

UACPI

uACPI is a portable and easy-to-integrate implementation of the Advanced Configuration and Power Interface (ACPI).
Its main focuses are full NT compatibility, safety, and minimization of stack usage by avoiding recursion entirely.
Extra documentation can be found in the project's GitHub repo.

GitHub

GitHub - Shinribo/uacpi-rs: UACPI bindings for Rust

UACPI bindings for Rust. Contribute to Shinribo/uacpi-rs development by creating an account on GitHub.

mortal yoke Sep 2, 2024, 10:21 AM

#

fiery turtle Resources: - Documentation on the wiki: https://osdev.wiki/wiki/UACPI - Read the...

dusky glade Sep 2, 2024, 10:23 AM

#

thanks

#

i still cannot get Box working

#

ugh

mortal yoke Sep 2, 2024, 10:26 AM

#

dusky glade i still cannot get Box working

what issue are you having with it

#

Box is in alloc

dusky glade Sep 2, 2024, 10:27 AM

#

mortal yoke Box is in alloc

i cannot use Box

#

box is not defined

#

i defined a global allocator

#

but i still cannot use box

mortal yoke Sep 2, 2024, 10:30 AM

#

dusky glade but i still cannot use box

what error does it give? I assume you have use alloc::boxed::Box;

dusky glade Sep 2, 2024, 10:31 AM

#

mortal yoke what error does it give? I assume you have `use alloc::boxed::Box;`

yep

#

that gives me a error

#

no such thing as boxed

#

in core::alloc

mortal yoke Sep 2, 2024, 10:31 AM

#

its not in core

#

its alloc::boxed::Box not core::alloc::boxed::Box

dusky glade Sep 2, 2024, 10:32 AM

#

i still cannot use it

#

no such crate as alloc

#

could not find boxed in alloc

deft canopy Sep 2, 2024, 1:07 PM

#

extern alloc

#

Make sure you're doing extern alloc first

#

@dusky glade

dusky glade Sep 2, 2024, 1:09 PM

#

Yea it works now

deft canopy Sep 2, 2024, 1:43 PM

#

there ya go

fiery turtle Sep 7, 2024, 12:13 PM

#

okay soo.. back to work

#

im looking at how ACPICA implements early table access again

#

and as always its very awkward api

#

if u want early tables u have to do mental gymnastics that could've been done for you implicitly

north holly Sep 7, 2024, 12:15 PM

#

oops I page faulted because uACPI wasn't thread safe and it took a long time to reboot
day n of waiting for uacpi thread safety

jaunty fox Sep 7, 2024, 12:18 PM

#

fiery turtle if u want early tables u have to do mental gymnastics that could've been done fo...

what's it you've got to do?

fiery turtle Sep 7, 2024, 12:20 PM

#

Anyway here's my plan:

make table mappings have refcount again, so introduce uacpi_table_{ref,unref}
introduce uacpi_kernel_get_rsdp as it may now be requested from multiple places
introduce uacpi_setup_early_table_access(initial_buffer, buffer_capacity, may_reallocate) where initial_buffer will be used by uacpi to store initial table descriptors
make uacpi_initialize reallocate the descriptors from initial_buffer into a normal resizable heap array if early table access was requested, otherwise work as it did before

#

uacpi_initialize will also error out if any early tables are still mapped at this point

fiery turtle Sep 7, 2024, 12:21 PM

#

jaunty fox what's it you've got to do?

basically it forces you to call AcpiReallocateRootTable to tell it that your heap is available otherwise the tables are stuck in the fixed buffer

#

which it already assumes heap availability in AcpiInitializeSubsystem anyway

jaunty fox Sep 7, 2024, 12:22 PM

#

one thing i would find a bit dirty on the part of acpica

north holly Sep 7, 2024, 12:23 PM

#

real chads just find the tables themself

jaunty fox Sep 7, 2024, 12:23 PM

#

mapping tables being done with the same callback as mapping stuff that ought to be mapped differently

fiery turtle Sep 7, 2024, 12:24 PM

#

thats an interesting thought, can u elaborate on that a bit?

jaunty fox Sep 7, 2024, 12:25 PM

#

in my case a) i anticipate using the SLIT, SLAT, SLOT, whatever it's called that has NUMA locality information in it ahead of being able to map pages other than those found in the direct map b) if mmio is mapped i want to be able to map it with appropriate cache/shareability attributes, because aarch64 won't forgive me getting that wrong, and i don't want to map tables as if they're MMIO

#

i don't know if uacpi actually does need to map anything MMIO at any time

fiery turtle Sep 7, 2024, 12:26 PM

#

lemme give u some context for these

jaunty fox Sep 7, 2024, 12:26 PM

#

i'm just thinking of the delayed bootup i had when uacpi_kernel_map mapped cacheable

north holly Sep 7, 2024, 12:27 PM

#

wait it's supposed to map uncached?

fiery turtle Sep 7, 2024, 12:28 PM

#

a) Yes, this is the problem i want to solve iwth uacpi_setup_early_table_access
b) That's the thing, even uACPI doesnt know which cacheability attribute should be used for a mapping, as AML is allowed to map absolutely any physical address. What Linux does is scan the memory map and the decide the attributes based on that. On x86 it just does high half mapping for everything i think

#

So basically if u want b u have to solve it yourself

#

by finding the range AML wants to map in the memory map

#

e.g. if its normal ram u just map cached, if its reserved u map uncached, etc

jaunty fox Sep 7, 2024, 12:29 PM

#

fiery turtle a) Yes, this is the problem i want to solve iwth `uacpi_setup_early_table_access...

b is not unreasonable

#

i can do that, will need to keep the memory map in persisted form but that's sensible anyway

fiery turtle Sep 7, 2024, 12:29 PM

#

yeah i guess

#

For a) i need some elaboration from you as to what specific problem you want solved

jaunty fox Sep 7, 2024, 12:31 PM

#

fiery turtle For a) i need some elaboration from you as to what specific problem you want sol...

will uacpi call uacpi_kernel_map() when dealing in early table access?

#

i can always set a flag to decide between "return hhdm offset" and "map it properly" if need be

#

and set that flag after i'm done with early tables and have set up the memory manager

fiery turtle Sep 7, 2024, 12:35 PM

#

will this work for you @jaunty fox

// ------------ acpi.c
// acpi_early_init
static uint8_t buffer[4096];

// automatically calls uacpi_kernel_get_rsdp() and maps it
uacpi_setup_early_table_access(buffer, sizeof(buffer), false);

// ------------ pmm.c
// pmm_init
uacpi_table_find("SLIT", ...);
// do init
uacpi_table_unref(table);

// ------------ uacpi_glue.c
void *uacpi_kernel_map(addr, ...) {
    if (uacpi_get_init_level() == UACPI_INIT_LEVEL_EARLY)
        return addr + hhdm;

    return vmm_map(addr);
}

void uacpi_kernel_unmap(addr, ...) {
    if (uacpi_get_init_level() == UACPI_INIT_LEVEL_EARLY)
        // no-op, hhdm mapping
        return;
    vmm_unmap(addr);
}

jaunty fox Sep 7, 2024, 12:35 PM

#

fiery turtle will this work for you <@391791686881247234> ```c // ------------ acpi.c // ac...

it certainly will

fiery turtle Sep 7, 2024, 12:35 PM

#

uacpi_get_init_level is already a thing

#

so i think it can be used for this very effectively

fiery turtle Sep 7, 2024, 12:36 PM

#

jaunty fox it certainly will

ok cool

fiery turtle Sep 7, 2024, 12:40 PM

#

jaunty fox i can always set a flag to decide between "return hhdm offset" and "map it prope...

this is what linux does btw, it has a lot of hackish things like that to work around acpica limitations

#

acpica has no notion of init levels, its either early or not

#

and thats a private global variable

#

https://elixir.bootlin.com/linux/v6.10.8/source/drivers/acpi/osl.c#L324

osl.c - drivers/acpi/osl.c - Linux source code v6.10...

Elixir Cross Referencer - source file of Linux (version v6.10.8). Browsed file: /drivers/acpi/osl.c

#

if (!acpi_permanent_mmap)
    return __acpi_map_table((unsigned long)phys, size);

#

on x86 it just does early_remap

#

and then acpi_permanent_mmap is set to true during later init

#

right before the call to acpi_initialize_subsystem();

#

which on uacpi would bump the init level to UACPI_INIT_LEVEL_SUBSYSTEM_INITIALIZED

fiery turtle Sep 8, 2024, 2:49 PM

#

found a very important bug in uacpi which was actually hiding another bug and thus making it not really a bug

#

thanks to this screenshot #1061407633745125397 message

#

what it says in the warning is, under the \_GPE namespace it already found a handler for event 8 called _L08 so it won't rebind it to _L08

#

but thats impossible since you cant have two nodes with the same name

#

under the same namespace

#

that would've been caught at init time earlier

#

Looking at the DSDT dump from that laptop, it indeed has two _L08 methods, but the second one is under a completely unrelated namespace

#

that should've never been reached

#

Anyway long story short

#

uacpi_namespace_for_each_node_depth_first had a bug where it would escape outside of the provided initial node and would iterate its peers

#

the fix was pretty simple

hollow elm Sep 8, 2024, 2:53 PM

#

damn

fiery turtle Sep 8, 2024, 2:54 PM

#

this also uncovered another bug, which is the _REG running code for some reason started iterating at the child node

#

#

this used to work simply because of this bug

#

Now im wondering

#

Would u expect uacpi_namespace_for_each_node_depth_first(node) to also visit node

#

or would u only expect it to visit the children of node

#

im tempted to just not make it visit the node

hollow elm Sep 8, 2024, 2:56 PM

#

if i had to guess i'd say only children?

fiery turtle Sep 8, 2024, 2:56 PM

#

thats what ACPICA does

#

so probably

#

its purely a matter of taste, initially i thought well for each means for each so the node is also one of the each

#

now im not so sure, i dont think it should

#

if u do for_each_node(_GPE) for example, u dont need _GPE visited, its the parent liek u dont really care

#

so ill probably change that, idk

#

fiery turtle Sep 8, 2024, 3:01 PM

#

hollow elm damn

this might be why the power button on your laptop didnt work, maybe we ended up calling some _REG thing multiple times because of this

#

or called something at the wrong time because of this

hollow elm Sep 8, 2024, 3:02 PM

#

yeah i was wondering about it again

#

also out of curiosity had a look if there are any gpes and it seems like there were but the bios patches them out? there's \_GPE.XL{08,0E,19} and the bios patches out nodes by replacing the first underscore with an X for _S3 as well

fiery turtle Sep 8, 2024, 3:04 PM

#

lol

#

usually they just hide them under ifs

#

but not everyone is so smart

hollow elm Sep 8, 2024, 3:05 PM

#

for _S3 i saw the code that patches the dsdt to turn it into XS3 in a 100% legit ami source code leak

fiery turtle Sep 8, 2024, 3:05 PM

#

is it available anywhere meme

#

like seriously they had to make the bios patch it instead of just doing

if (SUPP(3)) {
   Name (_S3, ...)
}

#

like im sure it just reads some io reg or whatever

fiery turtle Sep 8, 2024, 5:23 PM

#

fiery turtle

alright the people have spoken, i changed uacpi_namespace_for_each_node_depth_first to start its iteration at the first child and fixed the bug

#

@hollow elm hey uhh i know its a lot to ask but maybe if u have the time u could rebase managarm to latest uacpi and try running it on your hardware and see if anything changed or regressed (i currently dont have a working build of it), there's a chance that the power button works now

#

im waiting for the CI to finish and ill merge it

hollow elm Sep 8, 2024, 5:25 PM

#

i can try in a moment yeah

north holly Sep 8, 2024, 5:25 PM

#

fiery turtle alright the people have spoken, i changed `uacpi_namespace_for_each_node_depth_f...

does it change anything for the one time I use it in my kernel for pnp:

uacpi_namespace_for_each_node_depth_first(
    uacpi_namespace_root(),
    acpi_enumerate_callback,
    &udata
);```

hollow elm Sep 8, 2024, 5:25 PM

#

i'll also see if my current xhci changes broke it or if it still works

fiery turtle Sep 8, 2024, 5:25 PM

#

hollow elm i can try in a moment yeah

that would be big

#

oh yeah

#

u havent tested your xhci changes on your hardware yet?

hollow elm Sep 8, 2024, 5:26 PM

#

i'm doing some code cleanup now

#

like right now

fiery turtle Sep 8, 2024, 5:26 PM

#

north holly does it change anything for the one time I use it in my kernel for pnp: ```c uac...

previously the callback would receive the root node as the first node

#

now it doesnt receive it at all

#

TLDR: no

fiery turtle Sep 8, 2024, 5:26 PM

#

hollow elm like right now

ohh nice

#

i should recheck on my hw as well

#

maybe my pc/laptop xhci works now

median crest Sep 8, 2024, 5:27 PM

#

managarm xhci works now??

fiery turtle Sep 8, 2024, 5:28 PM

#

well theres a chance

hollow elm Sep 8, 2024, 5:28 PM

#

there is at least one more change i have stashed but it'll require some refactoring (ideally in the generic usb code so all the controllers don't need to reimplement the functionality)

median crest Sep 8, 2024, 5:28 PM

#

if it does I will try managarm on my computers

fiery turtle Sep 8, 2024, 5:28 PM

#

hollow elm there is at least one more change i have stashed but it'll require some refactor...

nice

hollow elm Sep 8, 2024, 5:29 PM

#

i still haven't figured out why the usb hub doesn't work though

fiery turtle Sep 8, 2024, 5:29 PM

#

hubs are the worst part

hollow elm Sep 8, 2024, 5:29 PM

#

it theoretically should work but maybe i'm missing some step that qemu doesn't care about

#

it's like really close to working too, i start initializing the device behind the hub and the first control transfer fails with a usb transaction error

fiery turtle Sep 8, 2024, 5:30 PM

#

its so close i can feel my hubs working already

hollow elm Sep 8, 2024, 5:31 PM

#

:^)

fiery turtle Sep 8, 2024, 5:31 PM

#

meme

#

would it support hubs behind hubs as well?

hollow elm Sep 8, 2024, 5:32 PM

#

it should yeah

fiery turtle Sep 8, 2024, 5:32 PM

#

nice

hollow elm Sep 8, 2024, 5:33 PM

#

theoretically hubs with xhci are easy enough, when you set up a new device there are some fields that point to the hub device and which hub port the device is connected to and that should be it

fiery turtle Sep 8, 2024, 5:33 PM

#

theoretically caret

hollow elm Sep 8, 2024, 6:20 PM

#

@fiery turtle do you have implementations for the new kernel functions that uacpi needs?

fiery turtle Sep 8, 2024, 6:23 PM

#

i think its get_thread_id, which i just returned the pointer to current coroutine i think

hollow elm Sep 8, 2024, 6:23 PM

#

also spinlock functions

fiery turtle Sep 8, 2024, 6:25 PM

#

that one should be identical to mutex helpers, just return frg::construct<spinlock>(*kernelAlloc)

hollow elm Sep 8, 2024, 6:26 PM

#

makes sense

fiery turtle Sep 8, 2024, 6:26 PM

#

even simpler because acquire is not timed

#

i think i had the patch for that but mightve lost it

hollow elm Sep 8, 2024, 6:27 PM

#

yeah i recall you sent some patch and i'm looking rn :^)

fiery turtle Sep 8, 2024, 6:31 PM

#

i remember literally posting the implementation of get_thread_id

#

for managarm

#

i cant find it lmao

hollow elm Sep 8, 2024, 6:31 PM

#

yeah same

fiery turtle Sep 8, 2024, 6:32 PM

#

might've been return getCpuData()->activeFiber

#

or rather

#

return thisFiber()

hollow elm Sep 8, 2024, 6:33 PM

#

sounds likely yeah

#

i'll just do that ig

fiery turtle Sep 8, 2024, 6:33 PM

#

i think its correct

hollow elm Sep 8, 2024, 6:34 PM

#

fiery turtle that one should be identical to mutex helpers, just `return frg::construct<spinl...

well with the exception that locking is supposed to turn off interrupts and return the previous state

fiery turtle Sep 8, 2024, 6:34 PM

#

there are 4 helpers

#

create
lock
unlock
free

#

this is create, free is the same but destruct

#

lock does return previous state

#

but i think the state is stored internally in managarm spinlock so u can just return 0

hollow elm Sep 8, 2024, 6:36 PM

#

well depends on what type i use

fiery turtle Sep 8, 2024, 6:37 PM

#

oh yeah

#

in my impl i just used the irq spinlock thats used everywhere else

hollow elm Sep 8, 2024, 6:37 PM

#

actually, we don't have a combined spinlock + irq disabling

#

nvm it's litereally called IrqSpinlock

fiery turtle Sep 8, 2024, 6:37 PM

#

wdym

#

yes lmao

#

thats the one

hollow elm Sep 8, 2024, 6:37 PM

#

i remembered IrqMutex and looked at that

#

but that only handles recursively turning irqs on/off

fiery turtle Sep 8, 2024, 6:38 PM

#

ohh

hollow elm Sep 8, 2024, 6:38 PM

#

IrqSpinlock combines a spinlock and a global irqMutex

#

anyway

fiery turtle Sep 8, 2024, 6:39 PM

#

yeah that makes sense

hollow elm Sep 8, 2024, 6:40 PM

#

btw why is the naming not consistent?

#

{acquire,release}_mutex vs spinlock_{lock,unlock}

fiery turtle Sep 8, 2024, 6:40 PM

#

acquire vs lock?

hollow elm Sep 8, 2024, 6:40 PM

#

that and the order of subject and verb

fiery turtle Sep 8, 2024, 6:41 PM

#

the latter is definitely wrong

#

acquire vs lock, idk
lock because it does so unconditionally, whereas acquire is more of a try to acquire based on the timeout

hollow elm Sep 8, 2024, 6:41 PM

#

makes sense

fiery turtle Sep 8, 2024, 6:41 PM

#

but i should fix that, thanks for spotting

#

break every implementation so far to keep the naming consistent meme

hollow elm Sep 8, 2024, 6:42 PM

#

do it while you still have time to do so meme

fiery turtle Sep 8, 2024, 6:42 PM

#

yeah i mean i have a warning in my readme

#

that things will break as this is pre 1.0

hollow elm Sep 8, 2024, 6:43 PM

#

seems to still work in qemu

#

i'll write the image to a usb stick and try it

fiery turtle Sep 8, 2024, 6:43 PM

#

nice

#

thanks for rebasing!

#

do u have just the one laptop?

hollow elm Sep 8, 2024, 6:44 PM

#

i have some more

#

one ivb and 3 with some older cpus (around 2008 i think?)

#

one of the older ones has managarm installed on the hard drive and it actually works :^)

fiery turtle Sep 8, 2024, 6:45 PM

#

does managarm boot there?

#

oh nice

hollow elm Sep 8, 2024, 6:45 PM

#

on the ivb laptop i think it should boot

#

it was the first machine i got xhci working on :^)

mild patrol Sep 8, 2024, 6:46 PM

#

Please inform me in this project you have here! Tail wags

fiery turtle Sep 8, 2024, 6:46 PM

#

Sorry?

fiery turtle Sep 8, 2024, 6:46 PM

#

hollow elm it was the first machine i got xhci working on :^)

nice

dusky glade Sep 8, 2024, 6:47 PM

#

fiery turtle Sorry?

dw abt him

#

😭

hollow elm Sep 8, 2024, 6:47 PM

#

oh i actually have one more laptop i forgot about

fiery turtle Sep 8, 2024, 6:47 PM

#

lol

#

laptops appearing out of nowehre

hollow elm Sep 8, 2024, 6:48 PM

#

i don't rememeber what cpu gen it is but it's a bit newer than the ivb laptop

dusky glade Sep 8, 2024, 6:48 PM

#

can i have your laptop

#

actually im gonna buy all the laptops

hollow elm Sep 8, 2024, 6:48 PM

#

hollow elm i don't rememeber what cpu gen it is but it's a bit newer than the ivb laptop

maybe haswell?

dusky glade Sep 8, 2024, 6:48 PM

#

anyways im procastinating on nyaux

hollow elm Sep 8, 2024, 6:49 PM

#

while waiting for the image to write to the usb stick i can think about how i'm gonna refactor the usb code :^)

fiery turtle Sep 8, 2024, 6:49 PM

#

does this image already contain usb refactors as well?

hollow elm Sep 8, 2024, 6:50 PM

#

it has the code cleanup but it's gonna fail with a babble error i'm pretty sure

fiery turtle Sep 8, 2024, 6:50 PM

#

ah

hollow elm Sep 8, 2024, 6:50 PM

#

because we assume that for usb fs the ep0 max packet size is 8, but it's 64, so the device sends an overlong packet

fiery turtle Sep 8, 2024, 6:51 PM

#

i see

hollow elm Sep 8, 2024, 6:51 PM

#

i plan on getting that fixed along with the refactor

#

which is basically making more of the enumeration logic generic

#

and doing what linux does when addressing the device

fiery turtle Sep 8, 2024, 6:52 PM

#

i hope u figure out the hub stuff eventually as well

hollow elm Sep 8, 2024, 6:52 PM

#

so: read device descriptor, and only after that issue SET_ADDRESS

hollow elm Sep 8, 2024, 6:53 PM

#

fiery turtle i hope u figure out the hub stuff eventually as well

yeah it'd be great to have them working as well

#

perhaps i just need to stare at the linux code harder

fiery turtle Sep 8, 2024, 6:53 PM

#

indeed

hollow elm Sep 8, 2024, 7:21 PM

#

@fiery turtle unfortunately no change in the power button not working

fiery turtle Sep 8, 2024, 7:21 PM

#

oh well

#

no other regressions noticed there?

hollow elm Sep 8, 2024, 7:22 PM

#

everything else seemed the same

fiery turtle Sep 8, 2024, 7:22 PM

#

well thanks for testing anyway

#

maybe its the next bug i find meme

deft canopy Sep 8, 2024, 8:39 PM

#

fiery turtle maybe its the next bug i find <:meme:575445094589661215>

I believe in u

fiery turtle Sep 8, 2024, 8:40 PM

#

tbh i dont think its a bug in uacpi, it probably isnt

#

either init order incorrect, or some driver managarm is missing that firmware expects to have been initialized

#

because everything initializes and works on that laptop, e.g. lid events arrive via EC just fine etc

#

but for whatever reason it probably assumes we're an old os and dont support ec power button events

fiery turtle Sep 9, 2024, 3:00 PM

#

fiery turtle

poll_question_text

Should uacpi_namespace_for_each_node_depth_first(node) also visit node

victor_answer_votes

6

total_votes

11

victor_answer_id

2

victor_answer_text

No, it should start iteration at first child of node

fiery turtle Sep 10, 2024, 1:03 PM

#

fiery turtle

interesting how much divide there is

#

i think logically yes makes more sense, but in reality u dont need it

hasty plinth Sep 10, 2024, 1:16 PM

#

fiery turtle i think logically `yes` makes more sense, but in reality u dont need it

I agree.
Maybe change the name to for_each_child if it doesn't visit the initial node?

fiery turtle Sep 10, 2024, 1:17 PM

#

yeah that makes sense

#

i really want to iron out the api and names before 1.0

jaunty fox Sep 10, 2024, 1:48 PM

#

hasty plinth I agree. Maybe change the name to `for_each_child` if it doesn't visit the init...

i agree

#

my feeling is the function should probably start visiting from the node itself

#

i would have a look at what acpica's analogous api does and imitate that

fiery turtle Sep 10, 2024, 1:50 PM

#

jaunty fox i would have a look at what acpica's analogous api does and imitate that

AcpiNsWalkNamespace

#

and that visits only children

jaunty fox Sep 10, 2024, 1:51 PM

#

odd, netbsd doesn't seem to call it

fiery turtle Sep 10, 2024, 1:52 PM

#

AcpiWalkNamespace is probably the public one

#

the Ns one is internal i guess

dusky glade Sep 10, 2024, 1:55 PM

#

uacpi integrated with linux when

#

memedown

jaunty fox Sep 10, 2024, 1:55 PM

#

yes, that one is called

#

most notably here

fiery turtle Sep 10, 2024, 1:55 PM

#

dusky glade uacpi integrated with linux when

that would be easy but there are thousands of usages of public acpica api, so too many places to change

dusky glade Sep 10, 2024, 1:55 PM

#

fiery turtle that would be easy but there are thousands of usages of public acpica api, so to...

fair

fiery turtle Sep 10, 2024, 1:56 PM

#

jaunty fox yes, that one is called

makes sense

jaunty fox Sep 10, 2024, 1:56 PM

#

something i want to do in the future

#

as well as the devicekit tree proper, a second plane for a tree of acpi device nodes

#

this will help clean up the pci routing logic a little

fiery turtle Sep 10, 2024, 1:58 PM

#

sounds interesting, maybe something to steal for my kernel if i ever get to it with all my plans meme

jaunty fox Sep 10, 2024, 2:00 PM

#

fiery turtle sounds interesting, maybe something to steal for my kernel if i ever get to it w...

i stole the concept from iokit

#

fiery turtle Sep 10, 2024, 2:01 PM

#

interesting

jaunty fox Sep 10, 2024, 4:57 PM

#

uacpi_kernel_create_mutex being called very early might be problematic

#

(i am looking for table access before i can allocate memory)

#

i don't have an immediate proposal on a good alternative though

fiery turtle Sep 10, 2024, 5:17 PM

#

jaunty fox (i am looking for table access before i can allocate memory)

I think I wont be constructing one if early table access was requested all the way until uacpi_initialize

fiery turtle Sep 10, 2024, 5:18 PM

#

jaunty fox (i am looking for table access before i can allocate memory)

Do you need that like right now?

jaunty fox Sep 10, 2024, 5:19 PM

#

fiery turtle Do you need that like right now?

if you're offering to facilitate it then i certainly won't turn you down

#

i can always hack around it myself if i must

fiery turtle Sep 10, 2024, 5:31 PM

#

jaunty fox i can always hack around it myself if i must

Out of curiosity what do you need it for right now?

jaunty fox Sep 10, 2024, 5:31 PM

#

fiery turtle Out of curiosity what do you need it for right now?

identify numa topology prior to initialising pmm

fiery turtle Sep 10, 2024, 5:32 PM

#

Are you actually implementing numa support for keyronex? Wow

fiery turtle Sep 10, 2024, 5:43 PM

#

jaunty fox if you're offering to facilitate it then i certainly won't turn you down

It was the next thing on my list so yeah, if you can hack around it for now that'd be cool, although im not sure how you would do it

deft canopy Sep 10, 2024, 5:46 PM

#

fiery turtle Are you actually implementing numa support for keyronex? Wow

as goddess intended tbh

jaunty fox Sep 10, 2024, 7:08 PM

#

come to think of it i'll need to be able to support allocating before numa setup if i want to support openfirmware

fiery turtle Sep 10, 2024, 7:09 PM

#

jaunty fox come to think of it i'll need to be able to support allocating before numa setup...

why is that

jaunty fox Sep 10, 2024, 7:10 PM

#

fiery turtle why is that

not a must-must i don't think but usually people find it less obnoxious to reconstitute the blob into a tree

fiery turtle Sep 10, 2024, 7:10 PM

#

i wonder how linux handles it

#

anyways i am going to work on early table access rn, not sure when ill get it done, but soon hopefully

jaunty fox Sep 10, 2024, 7:16 PM

#

fiery turtle i wonder how linux handles it

looks like it instantiates nodes by allocating with what it calls the MemBlock allocator

fiery turtle Sep 10, 2024, 7:16 PM

#

jaunty fox looks like it instantiates nodes by allocating with what it calls the MemBlock a...

ah, im guessing that one is not numa aware?

fiery turtle Sep 10, 2024, 7:17 PM

#

jaunty fox come to think of it i'll need to be able to support allocating before numa setup...

also im guessing this is not the only problem with mutexes at early init, since that stuff needs a working scheduler as well

#

being able to allocate them is only part of the problem

jaunty fox Sep 10, 2024, 7:18 PM

#

fiery turtle ah, im guessing that one is not numa aware?

it just bumps the offset stored in the structs describing the available physical memory

fiery turtle Sep 10, 2024, 7:18 PM

#

jaunty fox it just bumps the offset stored in the structs describing the available physical...

makes sense

jaunty fox Sep 10, 2024, 7:18 PM

#

fiery turtle also im guessing this is not the only problem with mutexes at early init, since ...

in my case no, it will still work

fiery turtle Sep 10, 2024, 7:19 PM

#

huh really?

#

even get_thread_id would work?

jaunty fox Sep 10, 2024, 7:19 PM

#

only the blocking case is a problem (i don't smp or multithread that early so that can only be erroneous)

#

but i can't say whether it would be the same on other kernels

fiery turtle Sep 10, 2024, 7:20 PM

#

actually it wont use get_thread_id, thats only needed for recursive mutexes which are only for AML

#

so maybe u would be fine

#

ill implement early tables anyway i think, seems like a useful feature, and all kernels are different

fiery turtle Sep 10, 2024, 10:17 PM

#

started working on this, will continue tomorrow

fiery turtle Sep 19, 2024, 6:45 PM

#

Alright uhh.. back from the oss so hopefully back to work soon

#

Although I have tons of shit to do for work next week

north holly Sep 19, 2024, 6:45 PM

#

fiery turtle Alright uhh.. back from the oss so hopefully back to work soon

Open source software?

fiery turtle Sep 19, 2024, 6:46 PM

#

north holly Open source software?

Open source summit

north holly Sep 19, 2024, 6:46 PM

#

O

fiery turtle Sep 19, 2024, 6:47 PM

#

Should've gone to plumbers tbh

hollow elm Sep 19, 2024, 6:48 PM

#

should've gone to gnu cauldron :^)

fiery turtle Sep 19, 2024, 7:21 PM

#

hollow elm should've gone to gnu cauldron :^)

Lmao

hollow elm Sep 19, 2024, 7:21 PM

#

would've had a chance to meet arsen

fiery turtle Sep 21, 2024, 12:44 PM

#

Would it be useful to add a pageable/nonpageable hint to uacpi_kernel_alloc?

#

most allocations should be pageable but not all of them

north holly Sep 21, 2024, 12:44 PM

#

yes

leaden fox Sep 21, 2024, 12:44 PM

#

yes

fiery turtle Sep 21, 2024, 12:44 PM

#

leaden fox yes

are u a uacpi user

leaden fox Sep 21, 2024, 12:46 PM

#

If I stop procastinating I will be

fiery turtle Sep 21, 2024, 12:46 PM

#

lol

leaden fox Sep 21, 2024, 12:46 PM

#

but I can't imagine it not being useful

fiery turtle Sep 21, 2024, 12:46 PM

#

do u have a progress report thing

hasty plinth Sep 21, 2024, 12:46 PM

#

fiery turtle Would it be useful to add a pageable/nonpageable hint to uacpi_kernel_alloc?

If it's not too much trouble, it would be nice. Even if most don't use the extra info it'd be handy to a few.

leaden fox Sep 21, 2024, 12:46 PM

#

if you have no demand paging in your kernel just ignore the flag

fiery turtle Sep 21, 2024, 12:47 PM

#

ill make it optional at compile time same as sized frees ofc

leaden fox Sep 21, 2024, 12:47 PM

#

fiery turtle do u have a progress report thing

not yet :(

north holly Sep 21, 2024, 12:47 PM

#

currently obos allocates all uacpi memory as pageable

#

which I think is around a megabyte?

#

idk

fiery turtle Sep 21, 2024, 12:47 PM

#

this might be correct tbh, im not sure if uh

#

as long as all events and spinlocks are non pageable

#

should be fine i think

hasty plinth Sep 21, 2024, 12:48 PM

#

north holly which I think is around a megabyte?

It depends on the system, I've seen it use 300MiB on my laptop lol

north holly Sep 21, 2024, 12:49 PM

#

well on qemu

hasty plinth Sep 21, 2024, 12:49 PM

#

Granted it's a fairly new system, I imagine there's a lot going on

north holly Sep 21, 2024, 12:49 PM

#

it's a megabyte

#

at least on my qemu config

fiery turtle Sep 21, 2024, 12:49 PM

#

leaden fox not yet :(

im currently working on UACPI_BAREBONES_MODE or such, this will only leave logging and map/unmap callbacks as host dependency and will allow to utilize table search api without aml execution etc

#

this might be useful for u

#

so u can start transitioning to full uacpi when able to but without having to roll your own skeleton table parsing

north holly Sep 21, 2024, 12:50 PM

#

will uacpi have its own scheduler or smth

#

with that mode

fiery turtle Sep 21, 2024, 12:50 PM

#

like i said only the table subsystem

north holly Sep 21, 2024, 12:50 PM

#

oh

#

I thought it would implement the kernel api for you

fiery turtle Sep 21, 2024, 12:50 PM

#

nah

leaden fox Sep 21, 2024, 12:50 PM

#

fiery turtle im currently working on `UACPI_BAREBONES_MODE` or such, this will only leave log...

I'm currently planning on switching to Rust, how far are the rust bindings?

north holly Sep 21, 2024, 12:50 PM

#

and I was like "uKernel?!?!?1"

fiery turtle Sep 21, 2024, 12:51 PM

#

it will straight up strip most of uacpi out

#

and leave only the table stuff

#

without mutexes and such

fiery turtle Sep 21, 2024, 12:51 PM

#

leaden fox I'm currently planning on switching to Rust, how far are the rust bindings?

last time i checked they worked fine albeit not fully implemented

north holly Sep 21, 2024, 12:51 PM

#

still don't know why people can't just parse the tables themself

leaden fox Sep 21, 2024, 12:51 PM

#

fiery turtle last time i checked they worked fine albeit not fully implemented

were they manual ones?

leaden fox Sep 21, 2024, 12:52 PM

#

north holly still don't know why people can't just parse the tables themself

code duplication

fiery turtle Sep 21, 2024, 12:52 PM

#

pinned message

north holly Sep 21, 2024, 12:52 PM

#

leaden fox code duplication

fair

fiery turtle Sep 21, 2024, 12:52 PM

#

north holly still don't know why people can't just parse the tables themself

why write boileplate yourself if you're going to switch later on anyway

jaunty fox Sep 21, 2024, 12:52 PM

#

fiery turtle Would it be useful to add a pageable/nonpageable hint to uacpi_kernel_alloc?

it could be useful

north holly Sep 21, 2024, 12:52 PM

#

I just don't look back at the boilerplate code I write

fiery turtle Sep 21, 2024, 12:52 PM

#

especially since uacpi has a robust impl for that that handles firmware quirks and stuff

jaunty fox Sep 21, 2024, 12:53 PM

#

the exact semantics of what can and can't be pageable will vary between kernels (keyronex mutexes, semaphores, etc can't be - but when keyronex has turnstiles, turnstile-backed rwlocks might be)

fiery turtle Sep 21, 2024, 12:53 PM

#

fiery turtle pinned message

or actually these bindings are outdated i think, qwinci had a newer fork

#

@mortal yoke can u link those pls

mortal yoke Sep 21, 2024, 12:54 PM

#

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

fiery turtle Sep 21, 2024, 12:54 PM

#

jaunty fox the exact semantics of what can and can't be pageable will vary between kernels ...

thats true, thats why event/mutex etc allocation is not done via uacpi_kernel_alloc

fiery turtle Sep 21, 2024, 12:54 PM

#

mortal yoke https://github.com/Qwinci/uacpi-rs

thanks!

#

updated the pins with your link

jaunty fox Sep 21, 2024, 12:55 PM

#

fiery turtle thats true, thats why event/mutex etc allocation is not done via uacpi_kernel_al...

yes, the current abstraction for it is very good

fiery turtle Sep 21, 2024, 12:56 PM

#

thanks

jaunty fox Sep 21, 2024, 12:57 PM

#

in general i have no objections to the uacpi api as it is today, pageable allocations could be a nice addition, and of course early table access

torpid ferry Sep 21, 2024, 1:00 PM

#

how much memory is it used anyways on average?

deft canopy Sep 21, 2024, 1:03 PM

#

fiery turtle Would it be useful to add a pageable/nonpageable hint to uacpi_kernel_alloc?

I'd say so
worst case scenario if someone doesn't want potential paging in uacpi they can just not have the kernel API include paged memory

fiery turtle Sep 21, 2024, 1:05 PM

#

jaunty fox in general i have no objections to the uacpi api as it is today, pageable alloca...

im working hard on the latter atm!

fiery turtle Sep 21, 2024, 1:05 PM

#

torpid ferry how much memory is it used anyways on average?

good question, idk
really need to measure that

#

i know most of it is never accessed again so its nice to be able to page it

fiery turtle Sep 22, 2024, 11:54 AM

#

So close to having table refcounting work

#

the only thing left is now solving test runner mental gymnastics around whats a real physical address and whats a fake one

#

previously it was derivable from just the point at which kernel_map is called during init

#

now it may be called to map a table way later on

#

as long as its not referenced

#

rn the way it works is kinda dumb

void* uacpi_kernel_map(uacpi_phys_addr addr, uacpi_size size)
{
    if (!g_expect_virtual_addresses) {
        auto it = phys_to_virt.find(addr);
        if (it != phys_to_virt.end()) {
            auto mapping_it = it->second.find({ nullptr, size });

            if (mapping_it != it->second.end()) {
                virt_to_phys_and_refcount[mapping_it->virt].second++;
                return mapping_it->virt;
            }

            std::printf("WARN: remapping physical 0x%016" PRIX64
                        " with size %zu\n", addr, size);
        }

        void *virt = std::calloc(size, 1);
        mapping m = { virt, size };

        virt_to_phys_and_refcount[virt] = { addr, 1 };
        phys_to_virt[addr].insert(m);
        return virt;
    }

    return reinterpret_cast<void*>(addr);
}

#

ill just take the easy route and make the test runner bump refcount to the DSDT/SSDT itself so they're not unmapped

#

this should cut it

#

few new apis and some behavior change

#

(this is all work towards early table access)

fiery turtle Sep 22, 2024, 1:29 PM

#

alright here it is

#

very close now

fiery turtle Sep 22, 2024, 2:46 PM

#

early tables managed to break exactly 2 completely random tests

#

not bad

#

and ofc why would asan say anything

#

nvm its not a memory bug

#

DSDT revision was getting overwritten by late init code, since its now initialized earlier

fiery turtle Sep 22, 2024, 3:20 PM

#

Alright, early table access is basically done (all tests pass on all platforms)! I will do a bit of additional testing before merging it ofc, but please be my guest to test it if you have a use case for it: https://github.com/UltraOS/uACPI/pull/92
Feedback also welcome since i changed quite a bit of other related API
CC: @jaunty fox

GitHub

Early/heapless table management by d-tatianin · Pull Request #92 · ...

reef plover Sep 23, 2024, 9:44 AM

#

This project is pretty nice, I'm gonna integrate it into my OS, especially for the device self-discovery.

#

My old code xd:

RSDT* rsdt = (RSDT*) memmgr_get_from_physical(rsdp->RsdtAddress);
    if(!doChecksum(&rsdt->h)) {
        printf("Warning: ACPI XSDT wrong checksum");

        return;
    }

    uintptr_t entries = (rsdt->h.Length - sizeof(rsdt->h)) / 4;
    for(uintptr_t i = 0; i < entries; i++) {
        ACPISDTHeader *h = (ACPISDTHeader *) memmgr_get_from_physical(rsdt->PointerToOtherSDT[i]);
        if (!strncmp(h->Signature, "MCFG", 4)) {
            MCFG* mcfg = (MCFG*) memmgr_get_from_physical(rsdt->PointerToOtherSDT[i]);

            if(!doChecksum(&mcfg->h)) {
                printf("Warning: ACPI MCFG wrong checksum");

                return;
            }

            uintptr_t mcfgEntries = (mcfg->h.Length - 44) / 16;
            for(uintptr_t j = 0; j < mcfgEntries; j++) {
                printf("PCIe base address: %x\n", mcfg->csba[j].base);

                pciBase = (uint8_t *) memmgr_get_from_physical(mcfg->csba[j].base);
            }
        }
    }

fiery turtle Sep 23, 2024, 9:47 AM

#

thanks

fiery turtle Sep 23, 2024, 9:47 AM

#

reef plover My old code xd: ```c RSDT* rsdt = (RSDT*) memmgr_get_from_physical(rsdp->RsdtAdd...

yeah uacpi has helpers for this ofc and structs for most used tables

rustic compass Sep 23, 2024, 5:10 PM

#

fiery turtle Alright, early table access is basically done (all tests pass on all platforms)!...

what happens if the allocated buffer is to small?

#

and why does: static uacpi_status initialize_from_rxsdt(uacpi_phys_addr rxsdt_addr, uacpi_size entry_size) take a size parameter? rxsdt_addr is the pointer to the rsdt so a size parameter shouldnt be needed?

fiery turtle Sep 23, 2024, 5:42 PM

#

rustic compass and why does: ```static uacpi_status initialize_from_rxsdt(uacpi_phys_addr rxsdt...

Its the size of an individual entry

#

4 or 8 bytes

#

Xsdt or rsdt

fiery turtle Sep 23, 2024, 5:46 PM

#

rustic compass what happens if the allocated buffer is to small?

Warning and OUT_OF_MEMORY

fiery turtle Sep 23, 2024, 5:46 PM

#

rustic compass and why does: ```static uacpi_status initialize_from_rxsdt(uacpi_phys_addr rxsdt...

This is very old code btw, just moved to a different file

rustic compass Sep 23, 2024, 6:38 PM

#

fiery turtle Its the size of an individual entry

but shouldnt uacpi figure that out?

fiery turtle Sep 23, 2024, 6:41 PM

#

this is an internal function called by uacpi

rustic compass Sep 23, 2024, 6:41 PM

#

woops

fiery turtle Sep 24, 2024, 9:27 AM

#

fiery turtle Alright, early table access is basically done (all tests pass on all platforms)!...

nvm decided to go even further here and optimize it so uacpi maps as little as possible when initializing the tables

#

will probably make checksum validation lazy

#

unless overridden by a flag

#

now it should be mapping one page at a time at max

#

at least at init time

#

also found a bug in that PR anyway

rustic compass Sep 24, 2024, 10:33 AM

#

What is the intended path to recover from a OOM buffer error in the pre heap Phase?

fiery turtle Sep 24, 2024, 10:38 AM

#

none really, because its a very very unlikely path

#

since user controls the buffer size, they can increase it

#

in the edge case that its needed

#

but i expect like a page to be enough

#

4096 byte buffer can currently store 73 table descriptors

#

i dont think ive even seen a machine with that many tables

rustic compass Sep 24, 2024, 10:39 AM

#

I mean can i reset uacpi and try again with more memory?

fiery turtle Sep 24, 2024, 10:40 AM

#

yeah ofc

#

it will auto-reset if init fails

#

but u can do uacpi_state_reset manually

rustic compass Sep 24, 2024, 10:42 AM

#

👍

fiery turtle Sep 24, 2024, 10:42 AM

#

https://elixir.bootlin.com/linux/v6.11/source/drivers/acpi/tables.c#L36

tables.c - drivers/acpi/tables.c - Linux source code...

Elixir Cross Referencer - source file of Linux (version v6.11). Browsed file: /drivers/acpi/tables.c

#

seems like 128 is enough as a hardcoded value for linux

#

if u give uacpi a page for early table heap u get 90 tables

#

also added this flag

#

previosuly this was the default behavior

fiery turtle Sep 25, 2024, 12:24 PM

#

Code for retrieving battery info using uACPI: https://pastebin.com/PKVNpSGA
TODO: convert to a proper example to use for wiki

kindred beacon Sep 26, 2024, 9:23 AM

#

have you tried setting up a handler for battery notifications?

fiery turtle Sep 26, 2024, 9:24 AM

#

yeah

#

it worked

kindred beacon Sep 26, 2024, 9:27 AM

#

did you also get a distinction between Notify(BAT0, 0x01) and Notify(BAT0, 0x80) et al

fiery turtle Sep 26, 2024, 9:29 AM

#

wdym by that?

#

0x80 is status change iirc, so u should re-read it

#

0x01 idk to be honest

kindred beacon Sep 26, 2024, 9:29 AM

#

yes, the problem is I ever only get 0x01

#

even when plugging/unplugging

fiery turtle Sep 26, 2024, 9:30 AM

#

maybe linux has a define for it

#

does your blob do 0x80 at all?

kindred beacon Sep 26, 2024, 9:31 AM

#

yep

#

fiery turtle Sep 26, 2024, 9:31 AM

#

what is this

kindred beacon Sep 26, 2024, 9:31 AM

#

qemu meme

fiery turtle Sep 26, 2024, 9:31 AM

#

lol

#

i mean

#

its easy to see when qemu generates that

kindred beacon Sep 26, 2024, 9:32 AM

#

how would I log that tho

fiery turtle Sep 26, 2024, 9:32 AM

#

log what?

#

how would u install a notify handler?

kindred beacon Sep 26, 2024, 9:33 AM

#

when qemu generated a notify

fiery turtle Sep 26, 2024, 9:33 AM

#

like in qemu or managarm

kindred beacon Sep 26, 2024, 9:33 AM

#

uacpi_install_notify_handler(node, handleBattery, nullptr);?

#

in mammogram

fiery turtle Sep 26, 2024, 9:33 AM

#

yup

#

same way we do power button

kindred beacon Sep 26, 2024, 9:36 AM

#

is the u64 value I get in the notification handler the 0x01/0x80/0x81?

fiery turtle Sep 26, 2024, 9:36 AM

#

yup

kindred beacon Sep 26, 2024, 9:36 AM

#

hmm

#

guess I'll try to debug mammogram on my test machine and try it there

#

might just be a qemu limitation

fiery turtle Sep 26, 2024, 9:37 AM

#

which values do u get?

kindred beacon Sep 26, 2024, 9:37 AM

#

only 0x01

fiery turtle Sep 26, 2024, 9:37 AM

#

i mean

#

u have the patch code right

#

u can see when it generates it

kindred beacon Sep 26, 2024, 9:37 AM

#

yes and I do not understand how notifications are generated lol

#

I tried looking at a DSDT dump too, no idea

fiery turtle Sep 26, 2024, 9:38 AM

#

let me see

kindred beacon Sep 26, 2024, 9:38 AM

#

https://github.com/blochl/qemu/pull/1/commits

GitHub

Qemu acpi devs by blochl · Pull Request #1 · blochl/qemu

This PR is for ease of review only. Please refer to the mailing list for the actual request.

#

this is what I get when plugging and unplugging

fiery turtle Sep 26, 2024, 9:41 AM

#

its a bug in this guys code

kindred beacon Sep 26, 2024, 9:41 AM

#

ah ok, good

fiery turtle Sep 26, 2024, 9:41 AM

#

he sends 128 aka bit 7

#

so its GPE 7

#

which does Notify(1)

wind fiber Sep 26, 2024, 9:41 AM

#

kindred beacon this is what I get when plugging and unplugging

ahem, in correct dutch it's balkanker, not ballkanker

fiery turtle Sep 26, 2024, 9:41 AM

#

kindred beacon ah ok, good

but i mean maybe linux does a re-read for 0x01 as well

kindred beacon Sep 26, 2024, 9:41 AM

#

prolly does, same for windoze

fiery turtle Sep 26, 2024, 9:42 AM

#

it updates the status if u boot into linux?

kindred beacon Sep 26, 2024, 9:42 AM

#

wind fiber ahem, in correct dutch it's balkanker, not bal**l**kanker

https://tenor.com/view/opinion-gif-20305030

Tenor

wind fiber Sep 26, 2024, 9:42 AM

#

kindred beacon https://tenor.com/view/opinion-gif-20305030

https://tenor.com/view/crane-brazil-grab-poor-car-gif-17372636

Tenor

fiery turtle Sep 26, 2024, 9:43 AM

#

kindred beacon prolly does, same for windoze

https://elixir.bootlin.com/linux/v6.11/source/drivers/acpi/battery.c#L1063

battery.c - drivers/acpi/battery.c - Linux source co...

Elixir Cross Referencer - source file of Linux (version v6.11). Browsed file: /drivers/acpi/battery.c

#

looks like it does it unconditionally

kindred beacon Sep 26, 2024, 9:43 AM

#

fiery turtle it updates the status if u boot into linux?

yeah, the mailing list post says it updates correctly on linux and windoze

fiery turtle Sep 26, 2024, 9:43 AM

#

    if (event == ACPI_BATTERY_NOTIFY_INFO)
        acpi_battery_refresh(battery);

#

it checks 0x80 only for this

kindred beacon Sep 26, 2024, 9:43 AM

#

fiery turtle he sends 128 aka bit 7

where is that

#

and how do I fix it meme

fiery turtle Sep 26, 2024, 9:43 AM

#

looks like its correct

#

0x80 refreshes literally everything about the battery

#

0x01 would just re-read its state

kindred beacon Sep 26, 2024, 9:44 AM

#

ok

fiery turtle Sep 26, 2024, 9:45 AM

#

u can just look at that code i linked

#

acpi_battery_update unconditionally on notify

#

acpi_battery_refresh on 0x80

fiery turtle Sep 26, 2024, 9:47 AM

#

kindred beacon where is that

battery.c:284

kindred beacon Sep 26, 2024, 9:47 AM

#

alright

fiery turtle Sep 26, 2024, 9:47 AM

#

(in qemu)

fiery turtle Sep 26, 2024, 10:01 AM

#

kindred beacon guess I'll try to debug mammogram on my test machine and try it there

did u try

kindred beacon Sep 26, 2024, 10:14 AM

#

I tried, I can enumerate the battery but before I can get any more info it hangs on some unrelated shizz

fiery turtle Sep 26, 2024, 10:23 AM

#

F

median crest Sep 26, 2024, 10:26 AM

#

Cant wait to have a more proper fleshed out interface to give uacpi some more attention in astral

fiery turtle Sep 26, 2024, 10:32 AM

#

that would be based

#

i need to rebase either managarm or astral to test the latest uacpi pr

#

and also wait until fadanoid does numa stuff

north holly Sep 26, 2024, 11:00 AM

#

fiery turtle i need to rebase either managarm or astral to test the latest uacpi pr

I can test obos if you want

#

Not now though, as I need to get ready

#

Actually I could test it

#

Rn

#

Pretty quick

fiery turtle Sep 26, 2024, 11:06 AM

#

sure

north holly Sep 26, 2024, 11:11 AM

#

bruh uacpi_initialize api changes

fiery turtle Sep 26, 2024, 11:11 AM

#

simplifications yes

#

just uacpi_initialize(0)

#

and ill probably make it take in a void

#

sooner rather than later

north holly Sep 26, 2024, 11:14 AM

#

and how do I give it my rsdp?

#

@fiery turtle

fiery turtle Sep 26, 2024, 11:15 AM

#

uacpi_kernel_get_rsdp

north holly Sep 26, 2024, 11:15 AM

#

does it magically know the rsdp?

#

oh wait

#

that's kernel api

fiery turtle Sep 26, 2024, 11:15 AM

#

lol

#

the nice thing aobut this api is retrieving the rsdp is arch specific anyway

#

so u can implement that part of the glue in arch specific place

median crest Sep 26, 2024, 11:16 AM

#

limine<3 handles it for me

fiery turtle Sep 26, 2024, 11:16 AM

#

u have arch_get_rsdp as well btw

#

now u dont even need that anymore

median crest Sep 26, 2024, 11:19 AM

#

It just returns the limine provided ptr meme

fiery turtle Sep 26, 2024, 11:19 AM

#

yeah

#

but if u were to implement e.g. itanium it would be different

north holly Sep 26, 2024, 11:22 AM

#

@fiery turtle obos boots with the new limine revision on qemu

fiery turtle Sep 26, 2024, 11:22 AM

#

okay nice that tells me at least i didnt fuck it up too much

#

i basically rewrote all of table init

north holly Sep 26, 2024, 11:22 AM

#

I'm testing real hw

fiery turtle Sep 26, 2024, 11:22 AM

#

u can try uacpi_setup_early_table_access as well if u want

north holly Sep 26, 2024, 11:22 AM

#

I need to go soon

fiery turtle Sep 26, 2024, 11:23 AM

#

aight thanks for testing

north holly Sep 26, 2024, 11:23 AM

#

np

fiery turtle Sep 26, 2024, 11:23 AM

#

north holly <@353593948523986945> obos boots with the new limine revision on qemu

also

#

limine revision?

north holly Sep 26, 2024, 11:23 AM

#

version

#

commit

#

whatever

fiery turtle Sep 26, 2024, 11:23 AM

#

limine?

north holly Sep 26, 2024, 11:23 AM

#

uacpi

fiery turtle Sep 26, 2024, 11:23 AM

#

lmao

north holly Sep 26, 2024, 11:23 AM

#

sametyhing for me

fiery turtle Sep 26, 2024, 11:23 AM

#

limine and uacpi are the same thing, yes

median crest Sep 26, 2024, 11:24 AM

#

ubootloadwe

fiery turtle Sep 26, 2024, 11:24 AM

#

anyways thanks for testing

north holly Sep 26, 2024, 11:24 AM

#

fiery turtle anyways thanks for testing

Works on real hw

#

Or the one real hw I've

fiery turtle Sep 26, 2024, 11:24 AM

#

nice

north holly Sep 26, 2024, 11:24 AM

#

Atm

fiery turtle Sep 26, 2024, 11:25 AM

#

the init for kernels that don't use the hh for mapping tables (like astral) should be faster now too

north holly Sep 26, 2024, 11:25 AM

#

hh?

fiery turtle Sep 26, 2024, 11:25 AM

#

hhdm

north holly Sep 26, 2024, 11:25 AM

#

Yeah I realized that

torpid root Sep 26, 2024, 11:30 AM

#

uacpi stable api when?

fiery turtle Sep 26, 2024, 11:30 AM

#

after 1.0

#

soon hopefully

torpid root Sep 26, 2024, 11:30 AM

#

noice

fiery turtle Sep 28, 2024, 11:33 AM

#

aight time to merge early table support, i tested it in astral and oberrow tested in in his kernel

#

then ill proceed to the thing ive been procrastinating on

#

which is thread safety

#

@vagrant hull sorry for ping but is there any way i can detect whether __popcountdi2 is natively available at build time?

vast kestrel Sep 28, 2024, 12:26 PM

#

Isn't there a builtin popcount?

fiery turtle Sep 28, 2024, 12:27 PM

#

yeah i use it atm

#

but the problem is it generates undefined references to that thing for the default amd64 target

vast kestrel Sep 28, 2024, 12:27 PM

#

I mean, you can maybe make a weak impl for it or something

fiery turtle Sep 28, 2024, 12:27 PM

#

i can but msvc doesnt support that

#

or well, this is gcc specific anyway

vast kestrel Sep 28, 2024, 12:28 PM

#

It also doesn't support builtin_popcount lel

fiery turtle Sep 28, 2024, 12:28 PM

#

uacpi_u8 uacpi_popcount(uacpi_u64 value)
{
#ifdef _MSC_VER

#ifdef _WIN64
    return __popcnt64(value);
#else
    return __popcnt(value) + __popcnt(value >> 32);
#endif

#else
    return __builtin_popcountll(value);
#endif
}

#

yeah i do this atm

vast kestrel Sep 28, 2024, 12:29 PM

#

Looks good enough tbh

#

If you need a compiler rt then link against it

#

Same for all the 64bit math things for 32bit

fiery turtle Sep 28, 2024, 12:29 PM

#

yeah like i can do that fine, its just extra annoyance for the users of uacpi ig

vast kestrel Sep 28, 2024, 12:29 PM

#

It's the exact same for 64 math tho, unless you don't use it ig

fiery turtle Sep 28, 2024, 12:30 PM

#

i had to do this in astral https://github.com/Mathewnd/Astral/blob/167a0dc2ba74b0f3d6d1429c89ac54ae97b71d3f/kernel-src/io/acpi/glue.c#L168

GitHub

Astral/kernel-src/io/acpi/glue.c at 167a0dc2ba74b0f3d6d1429c89ac54a...

x86-64 Operating System. Contribute to Mathewnd/Astral development by creating an account on GitHub.

fiery turtle Sep 28, 2024, 12:30 PM

#

vast kestrel It's the exact same for 64 math tho, unless you don't use it ig

its not a problem since 64-bit math is natively available for the default x86_64 thing, but popcount is x86_64-v1 or whatever

vast kestrel Sep 28, 2024, 12:31 PM

#

64/64 is not available, it needs a function

#

For the x86 target at least

fiery turtle Sep 28, 2024, 12:31 PM

#

oh yeah

vast kestrel Sep 28, 2024, 12:31 PM

#

Like I feel like the correct thing is to link against the compiler rt if it's needed Shrug

#

GCC/clang assumes they exist

fiery turtle Sep 28, 2024, 12:31 PM

#

its probably the correct thing yeah

vast kestrel Sep 28, 2024, 12:32 PM

#

Exactly the same as memset/memcpy

fiery turtle Sep 28, 2024, 12:32 PM

#

i guess i can just add a weak implementation under an ifdef GCC

torpid ferry Sep 28, 2024, 3:59 PM

#

check that if you build with x86_64-v1 it turns into a single popcount instruction automatically

vagrant hull Sep 28, 2024, 10:42 PM

#

fiery turtle <@379163868364275712> sorry for ping but is there any way i can detect whether _...

define "natively available"?

#

i can think of running autoconf tests or whatever the cmake equivalent is

vagrant hull Sep 28, 2024, 10:43 PM

#

vast kestrel Like I feel like the correct thing is to link against the compiler rt if it's ne...

all my kernels link against https://github.com/osdev0/cc-runtime

#

compiler-rt as provided by llvm is annoying to use and not gcc friendly

#

this is just dump in the source tree and forget

fiery turtle Sep 29, 2024, 6:40 AM

#

vagrant hull define "natively available"?

Like would be expanded into a popcnt instruction

#

Instead of a call to this

vagrant hull Sep 29, 2024, 9:02 AM

#

yeah i am not sure if there is a way to do that with gcc natively that isn't a hack

fiery turtle Sep 29, 2024, 9:08 AM

#

unlucky ig

fiery turtle Sep 29, 2024, 9:08 AM

#

vagrant hull yeah i am not sure if there is a way to do that with gcc natively that isn't a h...

but if i do like __attribute__((weak)) __popcountdi2 that would be fine?

#

like ship my own

vagrant hull Sep 29, 2024, 9:31 AM

#

fiery turtle like ship my own

why?

#

what is the use case exactly?

#

the correct solution would be for whatever kernel using uACPI to use cc-runtime or link with libgcc/compiler-rt

fiery turtle Sep 29, 2024, 9:31 AM

#

i dont want people to copy paste this cc-runtime thing every time

#

its very annoying

vagrant hull Sep 29, 2024, 9:31 AM

#

as i said

#

or libgcc/compiler-rt

fiery turtle Sep 29, 2024, 9:32 AM

#

does the default libgcc for 64 bit ship it?

vagrant hull Sep 29, 2024, 9:32 AM

#

it should

fiery turtle Sep 29, 2024, 9:32 AM

#

hmm

vagrant hull Sep 29, 2024, 9:32 AM

#

maybe it depends

#

on whether the target needs it

#

for base x86-64 i assume it has it

#

since popcount isn't guaranteed to exist

#

take libgcc.a and check

#

i checked

#

it has it

#

_popcountsi2.o:     file format elf64-x86-64


Disassembly of section .text:

0000000000000000 <__popcountdi2>:
   0:    f3 0f 1e fa              endbr64
   4:    48 ba 55 55 55 55 55     movabs rdx,0x5555555555555555
   b:    55 55 55 
   e:    48 89 f8                 mov    rax,rdi
  11:    48 d1 e8                 shr    rax,1
  14:    48 21 d0                 and    rax,rdx
  17:    48 29 c7                 sub    rdi,rax
  1a:    48 b8 33 33 33 33 33     movabs rax,0x3333333333333333
  21:    33 33 33 
  24:    48 89 fa                 mov    rdx,rdi
  27:    48 c1 ef 02              shr    rdi,0x2
  2b:    48 21 c2                 and    rdx,rax
  2e:    48 21 c7                 and    rdi,rax
  31:    48 01 fa                 add    rdx,rdi
  34:    48 89 d0                 mov    rax,rdx
  37:    48 c1 e8 04              shr    rax,0x4
  3b:    48 01 d0                 add    rax,rdx
  3e:    48 ba 0f 0f 0f 0f 0f     movabs rdx,0xf0f0f0f0f0f0f0f
  45:    0f 0f 0f 
  48:    48 21 d0                 and    rax,rdx
  4b:    48 ba 01 01 01 01 01     movabs rdx,0x101010101010101
  52:    01 01 01 
  55:    48 0f af c2              imul   rax,rdx
  59:    48 c1 e8 38              shr    rax,0x38
  5d:    c3                       ret```

fiery turtle Sep 29, 2024, 9:34 AM

#

i see

vagrant hull Sep 29, 2024, 9:35 AM

#

fwiw mlibc also depends on a few libgcc functions and we are certainly not having it ship weak versions of them just to attend those who do not wanna link against a libgcc equivalent

torpid ferry Sep 30, 2024, 2:58 PM

#

just only have the naive one, popcount won't ever be a perf bottleneck there

fiery turtle Sep 30, 2024, 3:03 PM

#

Fair as well

vast kestrel Sep 30, 2024, 4:05 PM

#

If the compiler is smart enough it will recognize what you do and optimize it :^)

rustic compass Sep 30, 2024, 5:12 PM

#

also if it becomes a problem maybe use a function pointer stored in a mut static and update it at first run to call the software emulation or the hardware instruction

#

unless the call overhead becomes even more problematic

fiery turtle Sep 30, 2024, 5:20 PM

#

Call overhead is definitely more expensive

fiery turtle Oct 2, 2024, 7:45 AM

#

Okay here are my plans for thread safety of (namespace vs interpreter):

the namespace and modification of nodes is guarded by a rw mutex lock
interpreter always takes the write lock (because it's allowed to modify the namespace and randomly update nodes from AML)
(public api for) iterating the namespace, finding a node etc takes a read lock
iterating the namespace publically will skip temporary nodes (created inside methods)
getting the object from a namespace node directly is not allowed. uacpi_eval() must be used, which takes the read lock, makes a deep copy, then returns the object
updating a namespace node from public api will be done via uacpi_namespace_node_update(), which will take the write lock

#

this is the biggest part that needed thread safety, the rest should be easier

#

getting the object from a namespace node directly is not allowed. uacpi_eval() must be used, which takes the read lock, makes a deep copy, then returns the object
Although this could be fine as long as reference chain is incremented with the write lock taken

#

it is theoretically possible to make it way more granular and get rid of the GIL, but that would increase memory overhead per node 2x and code complexity by an order of magnitude

#

so im definitely not doing it now, not for 1.0 anyway

#

maybe if i have infinite time on my hands at some point

wind fiber Oct 3, 2024, 1:23 PM

#

ret = uacpi_reboot();
if(uacpi_unlikely_error(ret))
    infoLogger() << "thor: ACPI reset failed: " << uacpi_status_to_string(ret) << frg::endlog;

returning ACPI reset failed: not found in qemu. Known? @fiery turtle

#

so that's https://github.com/UltraOS/uACPI/blob/9036f48cc29c4c4f75661f093311dcb69414d1fa/source/sleep.c#L532-L544C5 this case then

GitHub

uACPI/source/sleep.c at 9036f48cc29c4c4f75661f093311dcb69414d1fa · ...

A portable and easy-to-integrate implementation of the Advanced Configuration and Power Interface (ACPI) - UltraOS/uACPI

#

but I'd assume this should just work™️ in qemu

fiery turtle Oct 3, 2024, 1:28 PM

#

wind fiber ```c ret = uacpi_reboot(); if(uacpi_unlikely_error(ret)) infoLogger() << "th...

not known, no. did u pass -no-reboot to qemu by any chance?

wind fiber Oct 3, 2024, 1:28 PM

#

nope

#

tools/host-qemu/bin/qemu-system-x86_64 -s -m 8192 -name 'Managarm x86_64' -enable-kvm -debugcon stdio -cpu host,migratable=no -smp 4 -device piix3-usb-uhci,id=uhci -device usb-ehci,id=ehci -device qemu-xhci,id=xhci -drive id=boot-drive,file=image,format=raw,if=none -device virtio-blk-pci,drive=boot-drive -netdev user,id=net0 -device virtio-net,disable-modern=on,netdev=net0 -vga vmware -device usb-kbd,bus=xhci.0 -device usb-tablet,bus=xhci.0 -chardev file,id=ostrace,path=ostrace.bin -device dmalog,chardev=ostrace,tag=ostrace -chardev file,id=kernel-profile,path=kernel-profile.bin -device dmalog,chardev=kernel-profile,tag=kernel-profile -chardev file,id=kernel-alloc-trace,path=kernel-alloc-trace.bin -device dmalog,chardev=kernel-alloc-trace,tag=kernel-alloc-trace -chardev socket,id=gdbsocket,host=0.0.0.0,port=5678,server=on,wait=no -device dmalog,chardev=gdbsocket,tag=kernel-gdbserver -chardev socket,id=posix-gdbsocket,host=0.0.0.0,port=5679,server=on,wait=off -serial chardev:posix-gdbsocket

#

that's the full qemu command

fiery turtle Oct 3, 2024, 1:29 PM

#

i just booted astral like yesterday and reboot definitely worked

#

so it must be one of these command line options

wind fiber Oct 3, 2024, 1:30 PM

#

oh it works (with the ps2 reset meme )

#

shutdown works fine btw

#

and the prepare to enter S5 thing no issues on that either

fiery turtle Oct 3, 2024, 1:31 PM

#

yeah definitely works on astral

#

can u try trimming this command line

wind fiber Oct 3, 2024, 1:32 PM

#

sure

fiery turtle Oct 3, 2024, 1:32 PM

#

wind fiber sure

https://github.com/qemu/qemu/blob/423be09ab9492735924e73a2d36069784441ebc6/hw/i386/acpi-build.c#L219

GitHub

qemu/hw/i386/acpi-build.c at 423be09ab9492735924e73a2d36069784441eb...

Official QEMU mirror. Please see https://www.qemu.org/contribute/ for how to submit changes to QEMU. Pull Requests are ignored. Please only use release tarballs from the QEMU website. - qemu/qemu

wind fiber Oct 3, 2024, 1:32 PM

#

trimmed to nuke all the gdbsocket memes

#

tools/host-qemu/bin/qemu-system-x86_64 -s -m 8192 -name 'Managarm x86_64' -enable-kvm -debugcon stdio -cpu host,migratable=no -smp 4 -device piix3-usb-uhci,id=uhci -device usb-ehci,id=ehci -device qemu-xhci,id=xhci -drive id=boot-drive,file=image,format=raw,if=none -device virtio-blk-pci,drive=boot-drive -netdev user,id=net0 -device virtio-net,disable-modern=on,netdev=net0 -vga vmware -device usb-kbd,bus=xhci.0 -device usb-tablet,bus=xhci.0

fiery turtle Oct 3, 2024, 1:33 PM

#

in qemu its iffed under LPC

wind fiber Oct 3, 2024, 1:33 PM

#

are we missing some -m flag?

fiery turtle Oct 3, 2024, 1:33 PM

#

good question

wind fiber Oct 3, 2024, 1:34 PM

#

yes

#

adding -M q35 makes the issue go away

fiery turtle Oct 3, 2024, 1:34 PM

#

https://github.com/qemu/qemu/blob/423be09ab9492735924e73a2d36069784441ebc6/hw/i386/pc_q35.c#L234

GitHub

qemu/hw/i386/pc_q35.c at 423be09ab9492735924e73a2d36069784441ebc6 ·...

Official QEMU mirror. Please see https://www.qemu.org/contribute/ for how to submit changes to QEMU. Pull Requests are ignored. Please only use release tarballs from the QEMU website. - qemu/qemu

#

yeah its only created for Q35

wind fiber Oct 3, 2024, 1:34 PM

#

aha

#

good to know

#

(also good to know that ahci breaks on q35 meme )

#

not that that's used

#

but the driver dies

fiery turtle Oct 3, 2024, 1:35 PM

#

i dont know why tho

#

maybe backwards compat

wind fiber Oct 3, 2024, 1:35 PM

#

yeah maybe

#

astral indeed passes -M q35

fiery turtle Oct 3, 2024, 1:35 PM

#

ye

wind fiber Oct 3, 2024, 1:35 PM

#

so I'll propose that Managarm does the same

fiery turtle Oct 3, 2024, 1:36 PM

#

Looks like that enables hotplug and all sorts of things on qemu

#

ECAM as well

wind fiber Oct 3, 2024, 1:36 PM

#

spicy

fiery turtle Oct 3, 2024, 1:36 PM

#

i think we should pass it in managarm ye

#

CPU hotplug support in managarm when

wind fiber Oct 3, 2024, 1:36 PM

#

meme

fiery turtle Oct 3, 2024, 1:37 PM

#

tbh that shouldnt even be that difficult

#

hotunplug on the other hand..

wind fiber Oct 3, 2024, 1:37 PM

#

just yeet it

#

meme

fiery turtle Oct 3, 2024, 1:38 PM

#

what could go wrong

wind fiber Oct 3, 2024, 1:38 PM

#

worst case it asserts somewhere

#

sounds like average AssertOS thing

fiery turtle Oct 3, 2024, 1:38 PM

#

lol

median crest Oct 3, 2024, 1:38 PM

#

Assert(!unplug);

fiery turtle Oct 3, 2024, 1:38 PM

#

this is probably true unironically

#

it would assert that some task hung or whatever

wind fiber Oct 3, 2024, 1:41 PM

#

just unplug the CPU with the unplug thread running 5head

#

can't assert if the thread that's checking it is gone

kind mantle Oct 3, 2024, 4:23 PM

#

When would CPU hotplug even happen?

#

What's it used for?

deft canopy Oct 3, 2024, 4:25 PM

#

kind mantle When would CPU hotplug even happen?

I think it might be used in large clusters? but I'm not really sure
my plan for OpenItanic, and subsequently IFOS, is to provide a CPU hotswap mechanism if ACPI properly allows it (I am yet to fully understand ACPI, but one day I will)
the hotswap would be primarily for testing the limits of the processor interconnect, but might be useful if I'm running tests and have a part fail

kind mantle Oct 3, 2024, 4:26 PM

#

Interesting

#

I've thought of what BadgerOS might do about it and I'm not supporting it for now but...

#

... the current codebase is very much lenient to such shenanigans happening, if you use the proper mutexes.

deft canopy Oct 3, 2024, 4:28 PM

#

kind mantle I've thought of what BadgerOS might do about it and I'm not supporting it for no...

Smhsmh

kind mantle Oct 3, 2024, 4:28 PM

#

lol

vast kestrel Oct 3, 2024, 5:41 PM

#

kind mantle When would CPU hotplug even happen?

Virtual machines

#

Alot of auto-scaling will allow for cpu hot plugging

#

Sometimes even supporting transferring the VM to a larger machine, which then will add it's cores as well

fiery turtle Oct 3, 2024, 5:45 PM

#

wind fiber and the prepare to enter S5 thing no issues on that either

also i just remembered there isnt even a \_PTS on QEMU so its a no-op there anyway lol

wind fiber Oct 3, 2024, 5:46 PM

#

Lol

#

Still good to do correctly regardless

fiery turtle Oct 3, 2024, 5:46 PM

#

yeah this mostly matters for quirky laptops

median crest Oct 3, 2024, 5:48 PM

#

In vms at least you can use the hypervisors for some services like monotonic clocks that dont get out of sync with the hot plug (cough tsc sucks)

vast kestrel Oct 3, 2024, 6:20 PM

#

You don't bad mouth tsc it's awesome

fiery turtle Oct 5, 2024, 10:24 AM

#

finally went ahead with this change

#

also decided i'd expand on how to treat timeout

#

(well this change is part of a giant thread safety pr im working on)

craggy epoch Oct 6, 2024, 11:06 AM

#

How usable is uACPI as of now? I saw that the repo says it isn't production ready because of some thread safety stuff but is it close if it's only used in a single thread?

hasty plinth Oct 6, 2024, 11:09 AM

#

Its in a good enough state that plenty people here have used it for various things

fiery turtle Oct 6, 2024, 11:10 AM

#

craggy epoch How usable is uACPI as of now? I saw that the repo says it isn't production read...

other than thread safety its production quality

#

everything is implemented and stable

craggy epoch Oct 6, 2024, 11:11 AM

#

fiery turtle other than thread safety its production quality

That's perfect! My kernel is in Rust but I want to use it via FFI since it looks like a better solution than any other I could find and certainly more so than anything I could spin up from scratch.

fiery turtle Oct 6, 2024, 11:12 AM

#

there are rust bindings in pins

#

albeit a bit unfinished in some places

craggy epoch Oct 6, 2024, 11:14 AM

#

I'll take a look and maybe see if I can help with that as way to give back.

torpid ferry Oct 6, 2024, 11:14 AM

#

craggy epoch How usable is uACPI as of now? I saw that the repo says it isn't production read...

more than acpica in a lot of ways

fiery turtle Oct 6, 2024, 11:15 AM

#

lol thanks saga

craggy epoch Oct 6, 2024, 11:15 AM

#

torpid ferry more than acpica in a lot of ways

That's kind of crazy to think about. That you guys made something that's better than the reference implementation.

fiery turtle Oct 6, 2024, 11:16 AM

#

thats because the reference implementation doesnt try to follow the real reference implementation which is NT

craggy epoch Oct 6, 2024, 11:16 AM

#

How can you when it's closed source?

fiery turtle Oct 6, 2024, 11:16 AM

#

and uacpi is built as if NT is the reference implementation

fiery turtle Oct 6, 2024, 11:17 AM

#

craggy epoch How can you when it's closed source?

#1217009725711847465 message

craggy epoch Oct 6, 2024, 11:17 AM

#

That makes sense. Since that's what it looks like the FW vendors seem to target. I honestly hate ACPI but what I hate more is all the ARM and RV boards that don't provide it and go off and use other things like SMC and SBI.

fiery turtle Oct 6, 2024, 11:18 AM

#

TLDR: feed arbitrary AML to windows and see how it works with it

#

windbg has a way to enable tracing for every aml opcode that is executed

#

so u can see what it outputs at any point

#

!amli set traceon dbgbrkon errbrkon spewon;g

fiery turtle Oct 6, 2024, 11:19 AM

#

craggy epoch That makes sense. Since that's what it looks like the FW vendors seem to target....

yeah

craggy epoch Oct 6, 2024, 11:20 AM

#

That's interesting. I assume that's all done in a VM so the firmware can be messed around with. So just a question if you don't mind, how does Linux handle firmware that's made for Windows if it uses ACPICA which doesn't follow Windows?

torpid ferry Oct 6, 2024, 11:21 AM

#

poorly, or not at all

fiery turtle Oct 6, 2024, 11:21 AM

#

acpica had to adapt a lot of workarounds throughout its lifetime

#

some things still dont work

torpid ferry Oct 6, 2024, 11:21 AM

#

generally it just screams loudly on dmesg but it works anyways

fiery turtle Oct 6, 2024, 11:21 AM

#

it tries to pretend to do things like nt with hacks

#

but its fundumentally incompatible

fiery turtle Oct 6, 2024, 11:21 AM

#

torpid ferry generally it just screams loudly on dmesg but it works anyways

pretty much

craggy epoch Oct 6, 2024, 11:22 AM

#

Thanks for answering all my dumb questions and you guys are all awesome for making this project! It is much needed.

fiery turtle Oct 6, 2024, 11:23 AM

#

people have collected tons and tons of bug reports throughout the years where most important incompatibilities have been fixed with hacks in acpica, the rest are small things that generally are not that imporant

fiery turtle Oct 6, 2024, 11:23 AM

#

craggy epoch Thanks for answering all my dumb questions and you guys are all awesome for maki...

thanks and np!

#

if ur interested to see some examples look at my readme

craggy epoch Oct 6, 2024, 11:24 AM

#

Will do!

vale isle Oct 6, 2024, 11:27 AM

#

torpid ferry poorly, or not at all

i wonder if it's possible to shove uacpi in linux

#

maybe in the userspace

fiery turtle Oct 6, 2024, 11:28 AM

#

its 100% possible to shove it into the kernel directly, especially since a lot of public uacpi api mimics acpica, but there are thousands of files that use acpica api directly, so all of those would need to be rewritten

#

so its just a lot of work

north holly Oct 6, 2024, 11:29 AM

#

What if you just

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