uACPI - a portable and easy-to-integrate ACPI implementation | Operating System Development | Page 16

gentle peak Dec 18, 2024, 11:42 PM

#

i started this rewrite ~16h ago

jaunty fox Dec 18, 2024, 11:42 PM

#

i will implement a magazines layer for keyronex

median crest Dec 18, 2024, 11:43 PM

#

Same, mine varies a lot

gentle peak Dec 18, 2024, 11:43 PM

#

also worth noting that this is a single-processor kernel, so spinlocks only have to disable irqs instead of having an extra atomic op. don't know if it'd make a difference but well

torpid root Dec 18, 2024, 11:44 PM

#

wow 3m even on my cpu

slim panther Dec 18, 2024, 11:44 PM

#

fiery turtle Dec 18, 2024, 11:44 PM

#

are you using the kvm clock?

gentle peak Dec 18, 2024, 11:45 PM

#

fiery turtle are you using the kvm clock?

not on this build yet but it shouldn't be too hard to add

fiery turtle Dec 18, 2024, 11:45 PM

#

what are you using?

gentle peak Dec 18, 2024, 11:45 PM

#

right now i just ignore the invtsc bit (except for printing a simple warning)

slim panther Dec 18, 2024, 11:45 PM

#

lol 4.7mil on my machine

gentle peak Dec 18, 2024, 11:45 PM

#

tsc, unconditionally, even if the invtsc bit isn't set

#

hpet for calibration

fiery turtle Dec 18, 2024, 11:46 PM

#

interesting

vagrant hull Dec 18, 2024, 11:46 PM

#

are you absolutely positive the timer is fine though lol

north holly Dec 18, 2024, 11:46 PM

#

@hollow crescent you should try

gentle peak Dec 18, 2024, 11:47 PM

#

gentle peak tsc, unconditionally, even if the invtsc bit isn't set

the idea is that invtsc is old enough that i'm not gonna bother supporting anything that doesn't have it, but i didn't want to outright panic if the bit isn't set since you can't set it when using TCG

north holly Dec 18, 2024, 11:47 PM

#

looking at all these reminds me that obos' performance is dogshit

frank canopy Dec 18, 2024, 11:48 PM

#

gentle peak the idea is that invtsc is old enough that i'm not gonna bother supporting anyth...

tbh this is the best way to do it

jaunty fox Dec 18, 2024, 11:50 PM

#

i can leap up to
[uacpi]: successfully loaded 1 AML blob, 1705 ops in 0ms (avg 2620798/s)
with no spinlocking/ipl adjustment around slab allocatio

gentle peak Dec 18, 2024, 11:51 PM

#

vagrant hull are you absolutely positive the timer is fine though lol

pretty sure yeah:

    uint64_t threshold = read_time() + tsc_freq;

    for (int i = 0;; i++) {
        sched_stop(threshold); // sched_stop times out when tsc >= threshold
        threshold += tsc_freq;

        uint64_t cur_time = timeconv_apply(tsc2ns_conv, read_time());
        printk("iter %d, %U.%9U seconds since boot\n", i, cur_time / 1000000000, cur_time % 1000000000);
    }```

fiery turtle Dec 18, 2024, 11:51 PM

#

ngl it is kinda shady that the fastest os is closed source

#

but i believe you

vagrant hull Dec 18, 2024, 11:52 PM

#

i doubt they are cheating

slim panther Dec 18, 2024, 11:52 PM

#

speedrunner caught cheating in uacpi any%

gentle peak Dec 18, 2024, 11:52 PM

#

i can make it open source if you want lol it's nothing special yet, the only reason i haven't is that i usually like to get a bit further than this before going open

vagrant hull Dec 18, 2024, 11:52 PM

#

more so there is a chance they are making a mistake

fiery turtle Dec 18, 2024, 11:52 PM

#

perhaps

#

but sleep looks accurate

jaunty fox Dec 18, 2024, 11:53 PM

#

i think muonkous is in the right

#

though i would see how it would look if it were SMP-safe

fiery turtle Dec 18, 2024, 11:54 PM

#

true

gentle peak Dec 18, 2024, 11:54 PM

#

i do wonder how much that'd change things

vagrant hull Dec 18, 2024, 11:55 PM

#

so just to be clear: https://github.com/mintsuki/cynix/commit/f675511119cd39b5f4379f4b27635759b12e8d89 this is correct now, right?

torpid root Dec 18, 2024, 11:55 PM

#

slim panther speedrunner caught cheating in uacpi any%

wait isn't there speedrun.com or something

#

add uacpi there meme

fiery turtle Dec 18, 2024, 11:56 PM

#

vagrant hull so just to be clear: <https://github.com/mintsuki/cynix/commit/f675511119cd39b5f...

yeah

gentle peak Dec 18, 2024, 11:56 PM

#

it should be mostly SMP-safe already, i'd just need to add spinlocks to some areas that right now only use irq guards (off the top of my head: printk, uacpi deferred work, scheduler, idt, mutex slow path, semaphore, timer events)

vagrant hull Dec 18, 2024, 11:56 PM

#

alright ty

gentle peak Dec 18, 2024, 11:57 PM

#

oh and ioapic state is also only locked by an irq guard

#

but other than that everything should work fine

#

although the scheduler is obviously Not Optimal for an SMP kernel given that it's a global queue

fiery turtle Dec 18, 2024, 11:57 PM

#

torpid root add uacpi there <:meme:575445094589661215>

lmao

#

namespace load any%

#

w/ stubs

jaunty fox Dec 18, 2024, 11:58 PM

#

the hard limit i can get to at the moment by simple optimisations is 2.6mil (monkuous gets 3.6mil on my system)

fiery turtle Dec 18, 2024, 11:59 PM

#

literal beast

vagrant hull Dec 18, 2024, 11:59 PM

#

jaunty fox the hard limit i can get to at the moment by simple optimisations is 2.6mil (mon...

try mine

#

i mean it's probably gonna be less than 2.5mil so don't bother actually

jaunty fox Dec 18, 2024, 11:59 PM

#

vagrant hull i mean it's probably gonna be less than 2.5mil so don't bother actually

i can try anyway if you throw an ISO

#

what i think i can identify in the meantime is that the current limiting factors for me are probably:

IPL management
slab allocator

torpid root Dec 19, 2024, 12:00 AM

#

fadanoid could you test this please? #1217009725711847465 message

jaunty fox Dec 19, 2024, 12:01 AM

#

one paradox, i'm faster with freeing enabled than with no-op freeing!

jaunty fox Dec 19, 2024, 12:01 AM

#

torpid root fadanoid could you test this please? https://discord.com/channels/44044296114719...

testing now

fiery turtle Dec 19, 2024, 12:01 AM

#

jaunty fox one paradox, i'm faster with freeing enabled than with no-op freeing!

how is that even possible

kindred beacon Dec 19, 2024, 12:01 AM

#

function segments for LTO?

north holly Dec 19, 2024, 12:02 AM

#

fiery turtle how is that even possible

maybe freeing makes the allocator less fragmented

#

or smth

fiery turtle Dec 19, 2024, 12:02 AM

#

ah

kindred beacon Dec 19, 2024, 12:02 AM

#

jaunty fox Dec 19, 2024, 12:03 AM

#

fiery turtle how is that even possible

if the pattern of freeing tends to result in lots of allocs and frees being "paired" temporally

#

then a trip to the page allocator is avoided

fiery turtle Dec 19, 2024, 12:04 AM

#

I see

slim panther Dec 19, 2024, 12:04 AM

#

i'm getting 0% benefit by noping free

jaunty fox Dec 19, 2024, 12:04 AM

#

that's my hypothesis anyway

kindred beacon Dec 19, 2024, 12:04 AM

#

fiery turtle ngl it is kinda shady that the fastest os is closed source

IDA my beloved

jaunty fox Dec 19, 2024, 12:04 AM

#

i haven't tested it

#

other possibiltiies are odd changes to code generation

#

different cache characteristics and such

#

it's a promising observation anyway

#

because it means that i am likely to get huge gains from introducing a magazine layer to the slab allocator

#

#

the operations that can be served from the CPU layer can be coded such to avoid the IPL management overhead by simply disabling interrupts instead

loud ice Dec 19, 2024, 12:31 AM

#

kindred beacon IDA my beloved

yeah, with an ida license nothing is truly closed source meme

torpid root Dec 19, 2024, 12:33 AM

#

jaunty fox testing now

:(

gentle peak Dec 19, 2024, 12:43 AM

#

kindred beacon function segments for LTO?

for --gc-sections yeah

#

mostly just because i got into a habit of doing so

gentle peak Dec 19, 2024, 12:47 AM

#

fiery turtle ngl it is kinda shady that the fastest os is closed source

https://github.com/proxima-os/proxima i've polished it up a bit and published the source

#

turns out i did everything except for actually making the repo public, fixed now

fiery turtle Dec 19, 2024, 12:49 AM

#

wow ok

#

very clean implementation

#

do u do this stuff for a living?

gentle peak Dec 19, 2024, 12:50 AM

#

nah, just a hobby

fiery turtle Dec 19, 2024, 12:50 AM

#

but you're still an SE right?

gentle peak Dec 19, 2024, 12:51 AM

#

i'm only 18 lol i'm not anything

fiery turtle Dec 19, 2024, 12:51 AM

#

damn

#

not even a phd

slim panther Dec 19, 2024, 12:51 AM

#

meme

#

i should rethink my allocator approach

#

my slab is very slow

fiery turtle Dec 19, 2024, 12:52 AM

#

do you want to be mentioned now that its open source?

gentle peak Dec 19, 2024, 12:52 AM

#

sure

slim panther Dec 19, 2024, 12:52 AM

#

took the top spot

fiery turtle Dec 19, 2024, 12:52 AM

#

congrats, very talented work

fiery turtle Dec 19, 2024, 12:52 AM

#

gentle peak sure

could you add a comment to the pr please?

gentle peak Dec 19, 2024, 12:53 AM

#

less talent and more "has had a chronic case of way-too-eager-to-rewrite-stuff for years and now knows good approaches for a lot of stuff"

fiery turtle Dec 19, 2024, 12:53 AM

#

nice

slim panther Dec 19, 2024, 12:53 AM

#

extremely relatable

#

(minus the now knows good)

vagrant hull Dec 19, 2024, 12:54 AM

#

huh

#

ok so, you have mutexes? what do they do?

gentle peak Dec 19, 2024, 12:55 AM

#

fiery turtle could you add a comment to the pr please?

done

fiery turtle Dec 19, 2024, 12:56 AM

#

very nice

vagrant hull Dec 19, 2024, 12:58 AM

#

so yeah i do about 4 million now

fiery turtle Dec 19, 2024, 12:59 AM

#

wait what

#

what did u do

vagrant hull Dec 19, 2024, 12:59 AM

#

i enfattened my slab

fiery turtle Dec 19, 2024, 12:59 AM

#

what does that mean

vagrant hull Dec 19, 2024, 1:00 AM

#

made slabs larger, allowed for larger allocations through slabs, but mainly made slabs larger

fiery turtle Dec 19, 2024, 1:00 AM

#

thats it?

vagrant hull Dec 19, 2024, 1:01 AM

#

yup

fiery turtle Dec 19, 2024, 1:01 AM

#

nice

#

im not sure if I should include it because your mutexes are a no-op right

gentle peak Dec 19, 2024, 1:02 AM

#

vagrant hull ok so, you have mutexes? what do they do?

lock: spin for 40 iters (cmpxchg MUTEX_UNLOCKED->MUTEX_LOCKED_NOQUEUE), disable irqs, check state again to see if it was just unlocked, set state to MUTEX_LOCKED_QUEUED, puts the task on the list of waiting tasks, stops the task
unlock: cmpxchg MUTEX_LOCKED_NOQUEUE->MUTEX_UNLOCKED, if successful early exit, disables irqs, pops the first waiting task off the list, sets state to MUTEX_LOCKED_NOQUEUE if the queue is now empty, starts the task

fiery turtle Dec 19, 2024, 1:02 AM

#

so u avoid work that other implementations do

vagrant hull Dec 19, 2024, 1:02 AM

#

fiery turtle im not sure if I should include it because your mutexes are a no-op right

yes it's fine, you don't need to include it

gentle peak Dec 19, 2024, 1:02 AM

#

the fast path for mutex lock/unlock is a few nanoseconds i believe

vagrant hull Dec 19, 2024, 1:02 AM

#

my mutexes are no-ops, but i don't foresee them being such a major issue

#

in terms of knocking down the score i mean

fiery turtle Dec 19, 2024, 1:03 AM

#

probably, should just be a few atomic ops if not contended technically

vagrant hull Dec 19, 2024, 1:03 AM

#

yeah

fiery turtle Dec 19, 2024, 1:04 AM

#

gentle peak lock: spin for 40 iters (cmpxchg MUTEX_UNLOCKED->MUTEX_LOCKED_NOQUEUE), disable ...

what is you overall plan for the kernel

#

linux compat or just posix layer or

gentle peak Dec 19, 2024, 1:05 AM

#

posix compatibility (well, mostly - the goal is to implement enough that doing the rest in userspace has 0 performance penalty)

fiery turtle Dec 19, 2024, 1:05 AM

#

any goals for ports?

gentle peak Dec 19, 2024, 1:05 AM

#

for the os as a whole, NIH all the way to desktop, but that's a very far out goal that will likely never happen (i'm way too eager to rewrite things)

#

ports not sure, maybe java to run a minecraft server or something, and doom ofc

#

oh and i'd like to be able to do the whole self-hosting thing but that's honestly the easiest of those goals

fiery turtle Dec 19, 2024, 1:06 AM

#

does that mean we'll see your own ACPI subsystem meme

jaunty fox Dec 19, 2024, 1:06 AM

#

torpid root :(

i did test

#

1.1mil or so

#

[00:00:00.011864] [info] uacpi: successfully loaded 1 AML blob, 1705 ops in 1ms (avg 1125196/s)

gentle peak Dec 19, 2024, 1:07 AM

#

fiery turtle does that mean we'll see your own ACPI subsystem <:meme:575445094589661215>

maybe eventually, for now i decided to treat it the same as the bootloader (i eventually want to add a custom implementation of the limine boot protocol too)

fiery turtle Dec 19, 2024, 1:07 AM

#

makes sense

slim panther Dec 19, 2024, 1:08 AM

#

okay i've done the same and made my slabs larger

#

now i get 1.5m on avg

jaunty fox Dec 19, 2024, 1:09 AM

#

the big future gains for me will be whenever i add a magazines frontend to the slab allocator

#

that's something i'd want primarily for SMP to decontend the allocator, but it should help here too

median crest Dec 19, 2024, 1:16 AM

#

honestly I probably need to revisit my mutexes and actually make a proper implementation

#

this is replacing it with a spinlock_acqurie and spinlock_release

fiery turtle Dec 19, 2024, 1:19 AM

#

https://github.com/UltraOS/uACPI/tree/leaderboard?tab=readme-ov-file#notable-projects-using-uacpi--performance-leaderboards

#

Please take a look before I merge

#

Did i get everyone

north holly Dec 19, 2024, 1:22 AM

#

fiery turtle https://github.com/UltraOS/uACPI/tree/leaderboard?tab=readme-ov-file#notable-pro...

funny how my processor is like the oldest there

fiery turtle Dec 19, 2024, 1:22 AM

#

get qwinci to run your stuff

north holly Dec 19, 2024, 1:22 AM

#

when I finally optimize obos I can make a pr to change that value, right

fiery turtle Dec 19, 2024, 1:22 AM

#

yeah ofc

north holly Dec 19, 2024, 1:22 AM

#

fiery turtle get qwinci to run your stuff

nah it's fine

fiery turtle Dec 19, 2024, 1:22 AM

#

i will add changes in batches most likely

#

as new submissions if any come in

north holly Dec 19, 2024, 1:23 AM

#

I am waiting for the submission to put obos in the 3rd last position meme

#

but seriously I gotta optimize that

fiery turtle Dec 19, 2024, 1:23 AM

#

lol

#

just wait for tons of submissions and it will look like you're at the top

north holly Dec 19, 2024, 1:24 AM

#

eggsactly

#

or at the bottom

gentle peak Dec 19, 2024, 1:24 AM

#

north holly funny how my processor is like the oldest there

wanted to try it on my machine, is there a step missing from the build instructions in the readme?

north holly Dec 19, 2024, 1:24 AM

#

you're on master right?

gentle peak Dec 19, 2024, 1:24 AM

#

yeah

north holly Dec 19, 2024, 1:24 AM

#

clear cache and switch to userspace-work

#

master is months old

gentle peak Dec 19, 2024, 1:24 AM

#

ah

fiery turtle Dec 19, 2024, 1:24 AM

#

yeah btw your cmake is bogus, idk why you're calling that command

#

its a uacpi-internal cmake function to add sources lol

north holly Dec 19, 2024, 1:25 AM

#

wait it is

#

I have had that for months lol

fiery turtle Dec 19, 2024, 1:26 AM

#

lmao

gentle peak Dec 19, 2024, 1:26 AM

#

..huh

north holly Dec 19, 2024, 1:26 AM

#

wut the hell

fiery turtle Dec 19, 2024, 1:26 AM

#

can't u just send over an iso?

north holly Dec 19, 2024, 1:26 AM

#

I can

gentle peak Dec 19, 2024, 1:26 AM

#

that'd work yeah

north holly Dec 19, 2024, 1:27 AM

#

gentle peak ..huh

https://github.com/UltraOS/uACPI/commit/cc2ac5f788d81874b9c677c763b803aa375e9395
ur git is cursed

GitHub

opregion: make concurrent install & uninstall thread-safe · UltraOS...

Signed-off-by: Daniil Tatianin [email protected]

#

that commit exists

fiery turtle Dec 19, 2024, 1:27 AM

#

This commit does not belong to any branch on this repository, and may belong to a fork outside of the repository.

north holly Dec 19, 2024, 1:27 AM

#

wait infy

fiery turtle Dec 19, 2024, 1:27 AM

#

did u checkout a wip-pr?

north holly Dec 19, 2024, 1:27 AM

#

wtf

fiery turtle Dec 19, 2024, 1:27 AM

#

lol

north holly Dec 19, 2024, 1:27 AM

#

fiery turtle did u checkout a wip-pr?

idk maybe

#

I indeed did

#

was testing something out for you

fiery turtle Dec 19, 2024, 1:28 AM

#

b8a2a35fb60719641adefbdb945006cb673c

#

this hash is latest

north holly Dec 19, 2024, 1:28 AM

#

wha

#

I don't see it on the repo

fiery turtle Dec 19, 2024, 1:29 AM

#

wait no lmao

#

1721b8a2a35fb60719641adefbdb945006cb673c

north holly Dec 19, 2024, 1:29 AM

#

yeah

fiery turtle Dec 19, 2024, 1:29 AM

#

alright i merged the leaderboard for now, will update as people optimize their stuff further

north holly Dec 19, 2024, 1:29 AM

#

damn it infy

#

as soon as I change obos to use master's commit

#

and am about to commit the change

#

you do that

fiery turtle Dec 19, 2024, 1:30 AM

#

That shouldn't change your numbers lol

north holly Dec 19, 2024, 1:30 AM

#

ik

#

it's just a minor inconvinence

#

@gentle peak I fixed it

#

git pull && rm dependencies/needs_download.cmake

gentle peak Dec 19, 2024, 1:32 AM

#

one second, gotta install docker so that i don't clog up my path with x86_64-elf-*

north holly Dec 19, 2024, 1:32 AM

#

you do you

fiery turtle Dec 19, 2024, 1:37 AM

#

Just send over the damn ISO lol

north holly Dec 19, 2024, 1:39 AM

#

📎 obos.iso

#

@gentle peak here iso

gentle peak Dec 19, 2024, 1:39 AM

#

alright ty

fiery turtle Dec 19, 2024, 1:40 AM

#

Inb4 5 mil

north holly Dec 19, 2024, 1:40 AM

#

yes 100%

gentle peak Dec 19, 2024, 1:40 AM

#

38761/s

north holly Dec 19, 2024, 1:41 AM

#

woo

#

📎 obos.iso

#

try this one

#

it uses a bump allocator

gentle peak Dec 19, 2024, 1:42 AM

#

465911/s

fiery turtle Dec 19, 2024, 1:43 AM

#

Not bad

north holly Dec 19, 2024, 1:43 AM

#

fiery turtle Not bad

yeah because it uses a bump allocator and sets the irql to dispatch during uacpi init

#

otherwise it would be the same old

#

obos

fiery turtle Dec 19, 2024, 1:44 AM

#

Take some inspiration from monkuous

#

If you know what I mean meme

north holly Dec 19, 2024, 1:45 AM

#

mv proxima.iso obos.iso

#

soon, I will port mimalloc to obos

fiery turtle Dec 19, 2024, 1:46 AM

#

Find every string that says proxima and replace with obos

north holly Dec 19, 2024, 1:46 AM

#

north holly soon, I will port mimalloc to obos

and that will hopefully boost perf

gentle peak Dec 19, 2024, 1:46 AM

#

i'm pretty sure the kernel image actually doesn't have the string 'proxima' at all

#

since the kernel itself is called hydrogen

fiery turtle Dec 19, 2024, 1:46 AM

#

Fuck

north holly Dec 19, 2024, 1:46 AM

#

north holly and that will hopefully boost perf

and also optimize my memory mapping

#

I feel like that could be made faster

median crest Dec 19, 2024, 1:47 AM

#

gentle peak since the kernel itself is called hydrogen

what happens if I take two of them and push them together with a lot of force

gentle peak Dec 19, 2024, 1:47 AM

#

explosion

north holly Dec 19, 2024, 1:47 AM

#

memory mapping is bottlenecked by the allocator, and the allocator is probably bottlenecked by the memory mapping

gentle peak Dec 19, 2024, 1:47 AM

#

or well

fiery turtle Dec 19, 2024, 1:47 AM

#

Uranium kernel when

gentle peak Dec 19, 2024, 1:47 AM

#

not explosion on our scale

fiery turtle Dec 19, 2024, 1:47 AM

#

I have an idea

gentle peak Dec 19, 2024, 1:47 AM

#

release of a significant amount of energy relative to the energy of the resulting helium nucleus

#

how about that

north holly Dec 19, 2024, 1:47 AM

#

if nyaux ever surpasses me in perf
I will rewrite obos /j

fiery turtle Dec 19, 2024, 1:47 AM

#

Bump the number for every rewrite

#

What comes after hydrogen

fiery turtle Dec 19, 2024, 1:48 AM

#

north holly if nyaux ever surpasses me in perf I will rewrite obos /j

Lmao

gentle peak Dec 19, 2024, 1:48 AM

#

fiery turtle Bump the number for every rewrite

oh god if i did that i'd be somewhere around vanadium right now

fiery turtle Dec 19, 2024, 1:48 AM

#

Now thats a based kernel name

median crest Dec 19, 2024, 1:48 AM

#

my biggest trouble is my slow ass mutex

north holly Dec 19, 2024, 1:49 AM

#

north holly if nyaux ever surpasses me in perf I will rewrite obos /j

36.6k loc down the drain because of one silly kernel

median crest Dec 19, 2024, 1:49 AM

#

which is why I am looking into getting a real lock

north holly Dec 19, 2024, 1:49 AM

#

I have 7x more code than comments

median crest Dec 19, 2024, 1:49 AM

#

cant believe I'm trying this hard to optimise for a fucking fake number on a table

#

probably not that different from my corporate job

north holly Dec 19, 2024, 1:49 AM

#

north holly I have 7x more code than comments

or a 1831:261 ratio

fiery turtle Dec 19, 2024, 1:49 AM

#

Still the only hobby os that can build gcc

fiery turtle Dec 19, 2024, 1:50 AM

#

median crest probably not that different from my corporate job

Lmao

north holly Dec 19, 2024, 1:50 AM

#

when will astral run minecraft servers

#

or even better

#

minecraft

gentle peak Dec 19, 2024, 1:50 AM

#

median crest cant believe I'm trying this hard to optimise for a fucking fake number on a tab...

tbf considering the main performance impact comes from allocator/string funcs any improvement here will probably result in noticeable performance improvement kernel-wide

median crest Dec 19, 2024, 1:50 AM

#

fiery turtle Still the only hobby os that can build gcc

once I rewrite my scheduler I will use it as a stress test

fiery turtle Dec 19, 2024, 1:50 AM

#

Good idea

fiery turtle Dec 19, 2024, 1:51 AM

#

gentle peak tbf considering the main performance impact comes from allocator/string funcs an...

Very true yeah

median crest Dec 19, 2024, 1:51 AM

#

my og stress test was neofetch, then running configure

fiery turtle Dec 19, 2024, 1:51 AM

#

Lol

north holly Dec 19, 2024, 1:51 AM

#

median crest my og stress test was neofetch, then running configure

how is neofetch a stress test?

fiery turtle Dec 19, 2024, 1:51 AM

#

Neofetch stress test

north holly Dec 19, 2024, 1:51 AM

#

what does it do that's so stressful

fiery turtle Dec 19, 2024, 1:51 AM

#

Exists

median crest Dec 19, 2024, 1:51 AM

#

north holly how is neofetch a stress test?

my kernel was so buggy it didnt handle opening a lot of processes like neofetch does

north holly Dec 19, 2024, 1:52 AM

#

trust, obos definitely won't die when you blow on it as soon as I get userspace

median crest Dec 19, 2024, 1:52 AM

#

now its stable but slow

#

which I will fix

north holly Dec 19, 2024, 1:52 AM

#

(+2 processes has not been tested ever)

#

((in obos))

median crest Dec 19, 2024, 1:52 AM

#

(my manhood has been challenged

north holly Dec 19, 2024, 1:52 AM

#

north holly ((in obos))

(((unless you don't count the kernel as a process)))

fiery turtle Dec 19, 2024, 1:52 AM

#

median crest now its stable but slow

Stability first then speed is the right approach

north holly Dec 19, 2024, 1:53 AM

#

north holly (((unless you don't count the kernel as a process)))

((((then only one process has ever ran in obos))))

fiery turtle Dec 19, 2024, 1:53 AM

#

north holly (+2 processes has not been tested ever)

Port doom

north holly Dec 19, 2024, 1:53 AM

#

good idea

jaunty fox Dec 19, 2024, 1:53 AM

#

who's doing the benchmarking now?

north holly Dec 19, 2024, 1:53 AM

#

monkuous

fiery turtle Dec 19, 2024, 1:53 AM

#

Monkuous is our leader now

strong heath Dec 19, 2024, 1:54 AM

#

fiery turtle Still the only hobby os that can build gcc

havent tried, but I dont see why Ironclad wouldnt be able to

fiery turtle Dec 19, 2024, 1:54 AM

#

strong heath havent tried, but I dont see why Ironclad wouldnt be able to

Configure scripts

strong heath Dec 19, 2024, 1:54 AM

#

it configures fine itself, I can configure ironclad inside ironclad

fiery turtle Dec 19, 2024, 1:54 AM

#

Oh nice

#

Then you might be able to as well

#

But your os is not hobby

#

You get paid for it

median crest Dec 19, 2024, 1:55 AM

#

you should do it I wanna see the funny seal do things

strong heath Dec 19, 2024, 1:55 AM

#

is this what it feels to be denied hobbyhood

fiery turtle Dec 19, 2024, 1:55 AM

#

Yup

strong heath Dec 19, 2024, 1:55 AM

#

I feel empty

fiery turtle Dec 19, 2024, 1:55 AM

#

Youre a corporation now

north holly Dec 19, 2024, 1:55 AM

#

strong heath I feel empty

when will ironclad run on airplanes?

#

considering it uses ada

median crest Dec 19, 2024, 1:56 AM

#

@strong heath hire me I will accept very low pay and write ada

north holly Dec 19, 2024, 1:56 AM

#

should be pretty safe

strong heath Dec 19, 2024, 1:56 AM

#

I wish a foundating didnt need 30k starting capital to be funded, else I might have had an ironclad fundation

median crest Dec 19, 2024, 1:56 AM

#

just ahve to learn it fist

fiery turtle Dec 19, 2024, 1:56 AM

#

median crest <@275574198968123393> hire me I will accept very low pay and write ada

Lmao

vagrant hull Dec 19, 2024, 1:56 AM

#

hiring mathew is a good idea

#

i vouch for it

fiery turtle Dec 19, 2024, 1:56 AM

#

True

strong heath Dec 19, 2024, 1:56 AM

#

I cannot hire anyone without a legal organization

fiery turtle Dec 19, 2024, 1:56 AM

#

Hire nyaux as well

north holly Dec 19, 2024, 1:56 AM

#

fiery turtle Hire nyaux as well

ironclad stability 📉

vagrant hull Dec 19, 2024, 1:56 AM

#

strong heath I cannot hire anyone without a legal organization

hire with an illegal organization

strong heath Dec 19, 2024, 1:57 AM

#

fair enough

fiery turtle Dec 19, 2024, 1:57 AM

#

north holly ironclad stability 📉

Lmao

north holly Dec 19, 2024, 1:57 AM

#

somehow nyaux managed to make something less stable than obos

jaunty fox Dec 19, 2024, 1:57 AM

#

one underrated development of tonight

#

cynix is publicised now

median crest Dec 19, 2024, 1:57 AM

#

true

north holly Dec 19, 2024, 1:57 AM

#

north holly somehow nyaux managed to make something less stable than obos

(obos' fourth rewrite, the worst one)

fiery turtle Dec 19, 2024, 1:57 AM

#

And hydrogen

north holly Dec 19, 2024, 1:57 AM

#

((I should remove it from existance, it was terrible))

fiery turtle Dec 19, 2024, 1:57 AM

#

It was gonna be closed source

fiery turtle Dec 19, 2024, 1:58 AM

#

north holly ((I should remove it from existance, it was terrible))

Lmao

jaunty fox Dec 19, 2024, 1:58 AM

#

fiery turtle And hydrogen

the proxima one?

strong heath Dec 19, 2024, 1:58 AM

#

imagine calling an OS after the lightest element

fiery turtle Dec 19, 2024, 1:58 AM

#

jaunty fox the proxima one?

Yeah

slim panther Dec 19, 2024, 1:58 AM

#

north holly somehow nyaux managed to make something less stable than obos

real

strong heath Dec 19, 2024, 1:58 AM

#

if you are gonna name an OS, name it after lead

jaunty fox Dec 19, 2024, 1:58 AM

#

strong heath imagine calling an OS after the lightest element

it seems a reasonable name when you put it that way

#

floats like a butterfly, stings like a bee

fiery turtle Dec 19, 2024, 1:58 AM

#

^

north holly Dec 19, 2024, 1:59 AM

#

I should name my os after alexander fleming

fiery turtle Dec 19, 2024, 1:59 AM

#

Aml is scared of this os

jaunty fox Dec 19, 2024, 2:00 AM

#

north holly I should name my os after alexander fleming

does it exterminate bacteria?

north holly Dec 19, 2024, 2:00 AM

#

yes

median crest Dec 19, 2024, 2:00 AM

#

I named astral that because space is fucking cool

#

you cannot argue with this logic

north holly Dec 19, 2024, 2:00 AM

#

I named obos because well

fiery turtle Dec 19, 2024, 2:00 AM

#

I like astrals theme

jaunty fox Dec 19, 2024, 2:00 AM

#

it's incredible btw how we invented almost everything

#

scots on top non stop

fiery turtle Dec 19, 2024, 2:01 AM

#

LETSFUCKINGGOOOOOO

north holly Dec 19, 2024, 2:01 AM

#

best human invention: cheese

fiery turtle Dec 19, 2024, 2:01 AM

#

north holly I named obos because well

Oberrow os

north holly Dec 19, 2024, 2:02 AM

#

yeah

#

creative name

median crest Dec 19, 2024, 2:02 AM

#

but yeah, I know how I can (probably) improve astral's ops/sec by 20%-50%

fiery turtle Dec 19, 2024, 2:02 AM

#

I know obos lore

median crest Dec 19, 2024, 2:02 AM

#

now into the grind

fiery turtle Dec 19, 2024, 2:02 AM

#

median crest now into the grind

chad

north holly Dec 19, 2024, 2:05 AM

#

fiery turtle I know obos lore

the stone age of obos: first and second rewrite
iron age: third rewrite
middle ages: fourth rewrite
modern time: current rewrite
future: next rewrite??

fiery turtle Dec 19, 2024, 2:06 AM

#

Pls no

north holly Dec 19, 2024, 2:06 AM

#

jk

#

crazy how this obos rewrite started -1 days ago

#

(/j)

#

-1 days ago means
in one day thonk

median crest Dec 19, 2024, 2:07 AM

#

in 3 rewrites you will do it in astral because it will already be the year of the astral desktop

gentle peak Dec 19, 2024, 2:07 AM

#

2029 will be the year of the obos desktop

north holly Dec 19, 2024, 2:07 AM

#

yes.

#

my christmas break starts next week

#

so I can rather go outside

#

or smth

#

or work on obos

median crest Dec 19, 2024, 2:10 AM

#

https://tenor.com/bNIhD.gif

Tenor

hollow crescent Dec 19, 2024, 6:16 AM

#

north holly <@700663172226482254> you should try

what should I try?

vast kestrel Dec 19, 2024, 7:05 AM

#

north holly the stone age of obos: first and second rewrite iron age: third rewrite middle a...

But do you completely change the goal and design of the kernel each time you rewrite :^)

flat badge Dec 19, 2024, 7:06 AM

#

the latency of rep movsb is still pretty high (at least on some CPUs), so you should only do that for large regions

torpid root Dec 19, 2024, 7:53 AM

#

jaunty fox [00:00:00.011864] [info] uacpi: successfully loaded 1 AML blob, 1705 ops in 1ms ...

that's slow, even my CPU gets 1.5 mil
what flags for qemu?

dusky glade Dec 19, 2024, 8:19 AM

#

gentle peak 2029 will be the year of the obos desktop

no 2029 is the year of the nyaux desktop

calm latch Dec 19, 2024, 9:17 AM

#

pmOS desktop

#

(maybe I'll finally have a working shell)

#

and with no rewrites

kindred beacon Dec 19, 2024, 9:51 AM

#

2024 was the year of getting managarm desktop somewhat daily drivable

#

2025 will be the year of the managarm desktop

#

1000000000 % confirmed

fiery turtle Dec 19, 2024, 10:00 AM

#

should i pin this

torpid root Dec 19, 2024, 10:05 AM

#

kmanagarm with plasma 6 in 2025

median crest Dec 19, 2024, 10:29 AM

#

Managarm will replace windows

fiery turtle Dec 19, 2024, 10:58 AM

#

/*
 * Safely mask the event before we modify its handlers.
 *
 * This makes sure we can't get an IRQ in the middle of modifying this
 * event's structures.
 */
static uacpi_bool gpe_mask_safe(struct gp_event *event)
{
    uacpi_u8 this_mask;
    uacpi_bool was_masked;

    this_mask = gpe_get_mask(event);

    // No need to flush or do anything if it's not currently enabled
    if (!(event->reg->current_mask & this_mask))
        return UACPI_FALSE;

    if (gpe_masked(event)) {
        /*
         * This GPE is masked, but there still might be pending work involving
         * its handlers. Wait in case it was masked very recently.
         */
        uacpi_kernel_wait_for_work_completion();
        return UACPI_FALSE;
    }

    // 1. Mask the GPE, this makes sure its state is no longer modifyable
    gpe_mask_unmask(event, UACPI_TRUE);

    /*
     * 2. Wait for in-flight work & IRQs to finish, these might already
     *    be past the respective "if (masked)" check and therefore may
     *    try to re-enable a masked GPE.
     */
    uacpi_kernel_wait_for_work_completion();

    /*
     * 3. Now that this GPE's state is unmodifyable and we know that currently
     *    in-flight IRQs will see the masked state, we can safely disable this
     *    event knowing it won't be re-enabled by a racing IRQ.
     */
    set_gpe_state(event, GPE_STATE_DISABLED);

    /*
     * 4. Wait for the last possible IRQ to finish, now that this event is
     *    disabled.
     */
    uacpi_kernel_wait_for_work_completion();

    return UACPI_TRUE;
}

Mental gymnastics to make GPE modification thread-safe

#

/*
 * Waits for two types of work to finish:
 * 1. All in-flight interrupts installed via uacpi_kernel_install_interrupt_handler
 * 2. All work scheduled via uacpi_kernel_schedule_work
 *
 * Note that the waits must be done in this order specifically.
 */
uacpi_status uacpi_kernel_wait_for_work_completion(void);

And this is what is now expected from wait_for_work_completion

#

linux already does this for acpica as well:

void acpi_os_wait_events_complete(void)
{
    /*
     * Make sure the GPE handler or the fixed event handler is not used
     * on another CPU after removal.
     */
    if (acpi_sci_irq_valid())
        synchronize_hardirq(acpi_sci_irq);
    flush_workqueue(kacpid_wq);
    flush_workqueue(kacpi_notify_wq);
}

#

although it doesnt bother masking it because they dont care i guess

north holly Dec 19, 2024, 12:14 PM

#

hollow crescent what should I try?

Doing the uacpi benchmark thing

fiery turtle Dec 19, 2024, 12:22 PM

#

hollow crescent what should I try?

we have this leaderboard now: https://github.com/UltraOS/uACPI?tab=readme-ov-file#notable-projects-using-uacpi--performance-leaderboards

hollow crescent Dec 19, 2024, 12:33 PM

#

north holly Doing the uacpi benchmark thing

I don't even have userspace why would you want me to port uACPI

north holly Dec 19, 2024, 12:36 PM

#

Wait you don't have kt

#

*it

#

Nvm then

north holly Dec 19, 2024, 12:37 PM

#

fiery turtle we have this leaderboard now: https://github.com/UltraOS/uACPI?tab=readme-ov-fil...

Basically we have this to compare kernels' speeds

#

Using a new log added to uacpi

north holly Dec 19, 2024, 12:37 PM

#

hollow crescent I don't even have userspace why would you want me to port uACPI

Also I don't see how uacpi and userspace go together

hollow crescent Dec 19, 2024, 12:42 PM

#

They don't, I just have a set of priorities

#

First, finish the userspace memory manager, then start userspace

#

Then worry about other things like uACPI

#

Although a lot of things that are requisites of user space, are also not done

fiery turtle Dec 19, 2024, 12:43 PM

#

if u only want table parsing then uacpi doesnt need anything besides map/unmap

#

so technically doesnt matter

hollow crescent Dec 19, 2024, 12:54 PM

#

practically there are bigger fish to fry

#

the fact I'm not frying any of them is part 2

fiery turtle Dec 19, 2024, 12:57 PM

#

i will integrate it first thing so that i dont have to duplicate table parsing for stuff like hpet or madt

rustic compass Dec 19, 2024, 1:01 PM

#

does the benchmark requires uacpi to run in kernelspace or can it run in userspace?

fiery turtle Dec 19, 2024, 1:02 PM

#

i mean you're putting yourself at a disadvantage

#

and pmos already runs it in userspace

#

so no doesnt matter at all

torpid root Dec 19, 2024, 5:46 PM

#

wow I also get less points with no free()

wind fiber Dec 19, 2024, 6:23 PM

#

torpid root kmanagarm with plasma 6 in 2025

you laugh but large parts of KDE6 are already ported

torpid root Dec 19, 2024, 6:24 PM

#

I was 70% serious tbh

calm latch Dec 19, 2024, 10:24 PM

#

Managram in mainline clang when

slim panther Dec 19, 2024, 10:25 PM

#

it's a WIP pr

#

cc @kindred beacon meme

#

i think we just have to ping maintainers again

calm latch Dec 19, 2024, 10:28 PM

#

What do you do without swap though

flat badge Dec 19, 2024, 10:33 PM

#

Hm?

slim panther Dec 19, 2024, 10:34 PM

#

wat

fiery turtle Dec 19, 2024, 11:08 PM

#

Alright the final uACPI thread safety (part 3) PR is basically done, waiting for CI and merging

slim panther Dec 19, 2024, 11:08 PM

#

chad

north holly Dec 19, 2024, 11:08 PM

#

fiery turtle Alright the final uACPI thread safety (part 3) PR is basically done, waiting for...

Big

fiery turtle Dec 19, 2024, 11:08 PM

#

This means that uACPI is basically ready for 1.0, although I still wanted to add support for exotic opregions before declaring it

slim panther Dec 19, 2024, 11:09 PM

#

release 1.0 and then immediately api break with uapi ultrameme

fiery turtle Dec 19, 2024, 11:09 PM

#

lol

#

if anything uapi will be an optional layer

fiery turtle Dec 19, 2024, 11:37 PM

#

#

Merged & issue closed

#

I will push to implement support for exotic opregions in the coming days and release uACPI 1.0

#

that shouldnt be too hard I think

#

that should put me mostly at feature parity with acpica

hollow elm Dec 20, 2024, 12:06 AM

#

what opregions do you want to support?

loud ice Dec 20, 2024, 12:07 AM

#

im guessing all of them

fiery turtle Dec 20, 2024, 12:08 AM

#

UACPI_ADDRESS_SPACE_SMBUS = 4,
UACPI_ADDRESS_SPACE_IPMI = 7,
UACPI_ADDRESS_SPACE_GENERAL_PURPOSE_IO = 8,
UACPI_ADDRESS_SPACE_GENERIC_SERIAL_BUS = 9,
UACPI_ADDRESS_SPACE_PCC = 0x0A,
UACPI_ADDRESS_SPACE_PRM = 0x0B,
UACPI_ADDRESS_SPACE_FFIXEDHW = 0x7F,

#

all of these require special handling

hollow elm Dec 20, 2024, 12:09 AM

#

i assume they also require special os support?

fiery turtle Dec 20, 2024, 12:09 AM

#

yeah, via uacpi_install_address_space_handler

#

no hobby os with such drivers yet

#

but still important to get these done

#

linux does support all of them

fiery turtle Dec 20, 2024, 12:10 AM

#

hollow elm i assume they also require special os support?

basically all of these are similar to EC

north holly Dec 20, 2024, 12:11 AM

#

What's generic serial? Com ports?

fiery turtle Dec 20, 2024, 12:11 AM

#

you wish

#

special controllers i think

north holly Dec 20, 2024, 12:12 AM

#

Sounds boring and painful

#

Which is why obos will be the first to have them

fiery turtle Dec 20, 2024, 12:14 AM

#

I would tell u which driver linux uses for it but

#

bootlin decides to die

fiery turtle Dec 20, 2024, 12:15 AM

#

north holly Which is why obos will be the first to have them

its used by surface3_power.c lol

#

status = acpi_install_address_space_handler(adev->handle,
                                            ACPI_ADR_SPACE_GSBUS,
                                            &mshw0011_space_handler,
                                            NULL,
                                            data);
if (ACPI_FAILURE(status)) {
        dev_err(&client->dev, "Error installing i2c space handler\n");
        acpi_bus_detach_private_data(adev->handle);
        kfree(data);
        return -ENOMEM;
}

#

yeah thats the only user in the entire kernel

#

fiery turtle Dec 20, 2024, 12:24 AM

#

fiery turtle

OperationRegion (OR01, GenericSerialBus, Zero, 0x0100)
Field (OR01, BufferAcc, NoLock, Preserve)
{
    Connection (I2Z0), 
    AccessAs (BufferAcc, AttribRawProcessBytes (0x02)), 
    SAN0,   8
}

fiery turtle Dec 20, 2024, 11:30 AM

#

    /*
     * _REG methods for global address space handlers (installed to root)
     * get called during the namespace initialization, no reason
     * to call them here manually as that will be done later by init code
     * anyway. Just delay that work until later.
     */
    if (device_node == uacpi_namespace_root() &&
        g_uacpi_rt_ctx.init_level == UACPI_INIT_LEVEL_NAMESPACE_LOADED)
        goto out;

Least dumb uACPI code

#

I think this breaks early EC registration

#

in theory

fiery turtle Dec 20, 2024, 11:55 AM

#

i wonder if i should get rid of uacpi_kernel_raw_* helpers

#

and just map io/mem manually

#

the idea there was that the kernel can keep a cache like linux

#

but i think its just annoying to have 9999 extra helpers

#

to implement in the kernel

#

so like uacpi_gas_read currently does this:

if (gas->address_space_id == UACPI_ADDRESS_SPACE_SYSTEM_IO) {
    ret = uacpi_kernel_raw_io_read(
        address, access_byte_width, &data
    );
} else {
    ret = uacpi_kernel_raw_memory_read(
        address, access_byte_width, &data
    );
}

#

if i get rid of raw stuff it will call map, read, unmap

#

there's already a precedent for this anyway since i handle SystemMemory opregions myself and not via kernel raw_read/write

calm latch Dec 20, 2024, 11:58 AM

#

You can make it optional (?)

fiery turtle Dec 20, 2024, 12:00 PM

#

#polls message

#

lets see

fiery turtle Dec 20, 2024, 12:01 PM

#

calm latch You can make it optional (?)

im tired of introducing entropy for no reason

fiery turtle Dec 20, 2024, 12:01 PM

#

fiery turtle https://discord.com/channels/440442961147199490/1227738482152181761/131963559353...

pls vote

slim panther Dec 20, 2024, 12:01 PM

#

fiery turtle im tired of introducing entropy for no reason

this

#

less shit to implement = better

fiery turtle Dec 20, 2024, 12:01 PM

#

if these are removed uacpi will instead do map(), io_read/write, unmap() and for memory it will do the IO itself completely

#

I initially stole the raw* helpers idea from ACPICA but now i think it just sucks

#

ok looks like people hate the raw api anyway lmao

#

but ill wait for the vote to finish

hollow elm Dec 20, 2024, 12:49 PM

#

this makes me wonder, what caching requirements do opregions have

#

it can either be mmio or just ram (e.g. nvs)

#

and on arm etc it matters since there arent mtrrs to set it for you

gentle peak Dec 20, 2024, 12:50 PM

#

i believe the intended way to determine caching mode is to check the memory map, but on x86 at least it's fine to rely on mtrrs

fiery turtle Dec 20, 2024, 12:52 PM

#

^

hollow elm Dec 20, 2024, 12:54 PM

#

how does it help? acpi reclaimable & nvs => normal ram, reserved,not in mmap => device memory?

gentle peak Dec 20, 2024, 12:54 PM

#

yeah basically

hollow elm Dec 20, 2024, 12:55 PM

#

feels a bit iffy since reserved does not imply !ram (to me at least)

#

but if thats the intended way then so be it

gentle peak Dec 20, 2024, 12:55 PM

#

there aren't really better options, i don't think aml specifies the caching mode so it's not like uacpi can tell you how to map it

fiery turtle Dec 20, 2024, 12:56 PM

#

yup

#

thats how linux does it at least

flat badge Dec 20, 2024, 12:58 PM

#

map + write + unmap is slow af

#

It's probably two orders of magnitude slower than a raw write

#

Especially for proper SMP OSes that implement TLB shootdown

fiery turtle Dec 20, 2024, 1:00 PM

#

but a raw write is also done on a physical address

#

so u kinda have to map it no?

flat badge Dec 20, 2024, 1:00 PM

#

unmap requires an IPI to all CPUs and blocking for the other cpus to react

#

No, a raw write can use a direct mapping

mortal yoke Dec 20, 2024, 1:00 PM

#

you can use direct mapping with the map fn too

flat badge Dec 20, 2024, 1:01 PM

#

If you get rid of the helpers, you could keep the cache in uacpi such that you only ever map once

flat badge Dec 20, 2024, 1:01 PM

#

mortal yoke you can use direct mapping with the map fn too

Not necessarily

mortal yoke Dec 20, 2024, 1:02 PM

#

well not on all platforms but with like x86_64 or whatever

flat badge Dec 20, 2024, 1:02 PM

#

A direct mapping for raw writes must be contiguous at the granularity of the largest possible write

#

which is 8 in practice

#

Mapping via direct map must be contiguous at an arbitrary granularity

flat badge Dec 20, 2024, 1:04 PM

#

mortal yoke well not on all platforms but with like x86_64 or whatever

Well on ARM you can map as nGnEnR

jaunty fox Dec 20, 2024, 1:04 PM

#

hollow elm feels a bit iffy since reserved does not imply !ram (to me at least)

i don't like it either

#

not one bit

#

but since linux is doing it i assume it's the only way

flat badge Dec 20, 2024, 1:04 PM

#

And on rv the memory controller is supposed to handle cacheability transparently

fiery turtle Dec 20, 2024, 1:18 PM

#

flat badge A direct mapping for raw writes must be contiguous at the granularity of the lar...

what does that mean

#

kernel_map will be called with the size of the write

#

so i dont see any difference in this case

flat badge Dec 20, 2024, 1:22 PM

#

I guess you could distinguish between single page and multi page mappings, that's true

#

Then it's not that bad

rustic compass Dec 20, 2024, 2:03 PM

#

fiery turtle if these are removed uacpi will instead do `map(), io_read/write, unmap()` and f...

does uacpi unmap the io even if it needs it in the future again or will it only unmap if it doesnt need it anymore?

mortal yoke Dec 20, 2024, 2:06 PM

#

it can't really know whether it needs it in the future

#

what it does rn afaik for eg. op regions is mapping it once

rustic compass Dec 20, 2024, 2:09 PM

#

good enough

left orbit Dec 20, 2024, 2:24 PM

#

did i see right that you were building linux with uacpi yesterday? or am i tripping

#

i saw some messages with linker error that referenced uacpi and vmlinuz.o meme

#

i was wondering how that went/how it's going

fiery turtle Dec 20, 2024, 2:26 PM

#

i was yeah

#

i got too lazy to implement kernel api lamo

#

i might finish it later

fiery turtle Dec 20, 2024, 2:27 PM

#

rustic compass does uacpi unmap the io even if it needs it in the future again or will it only ...

on opregion destruction, or in case of FADT register writes, right away

frank canopy Dec 20, 2024, 7:19 PM

#

in case im misunderstanding, is there any realistic thing for the map/unmap io functions to do on x86? id assumed they were mostly there for mmio on other architectures etc

fiery turtle Dec 20, 2024, 7:20 PM

#

frank canopy in case im misunderstanding, is there any realistic thing for the map/unmap io f...

yeah, for microkernels to request io permissions from the kernel

frank canopy Dec 20, 2024, 7:21 PM

#

ahhh ok

fiery turtle Dec 21, 2024, 11:25 AM

#

@@ -564,13 +581,30 @@ uacpi_status uacpi_gas_write(const struct acpi_gas *gas, uacpi_u64 in_value)
             uacpi_u64 address = gas->address + (index * access_byte_width);

             if (gas->address_space_id == UACPI_ADDRESS_SPACE_SYSTEM_IO) {
-                ret = uacpi_kernel_raw_io_write(
-                    address, access_byte_width, data
-                );
+                uacpi_handle handle;
+
+                ret = uacpi_kernel_io_map(address, access_byte_width, &handle);
+                if (uacpi_unlikely_error(ret))
+                    return ret;
+
+                ret = uacpi_kernel_io_write(handle, 0, access_byte_width, data);
+                uacpi_kernel_io_unmap(handle);
             } else {
-                ret = uacpi_kernel_raw_memory_write(
-                    address, access_byte_width, data
-                );
+                void *virt;
+
+                virt = uacpi_kernel_map(address, access_byte_width);
+                if (uacpi_unlikely(virt == UACPI_NULL))
+                    return UACPI_STATUS_MAPPING_FAILED;
+
+                ret = uacpi_system_memory_write(virt, access_byte_width, data);
+                uacpi_kernel_unmap(virt, access_bit_width);
             }

#

Now that raw helpers are gone, uacpi_gas_write is going to look like this

#

kinda bulky but at least its less stuff for kernel to implement

fiery turtle Dec 21, 2024, 11:46 AM

#

actually I just made an abstraction around io r/w

uacpi_status uacpi_system_io_read(
    uacpi_io_addr address, uacpi_u8 width, uacpi_u64 *out
);
uacpi_status uacpi_system_io_write(
    uacpi_io_addr address, uacpi_u8 width, uacpi_u64 in
);

Maybe I could implement a transparent LRU cache around it if the need arises

#

(this is internal api)

#

that's 4 less kernel_api functions to implement

#

I think that's pretty good

dusky glade Dec 21, 2024, 12:11 PM

#

hold on i think nyaux can do faster now

#

nvm

kindred beacon Dec 21, 2024, 12:25 PM

#

Ah ok I see

#

Windows skill issue

dusky glade Dec 21, 2024, 12:33 PM

#

kindred beacon Windows skill issue

yes nooo

torpid root Dec 21, 2024, 1:46 PM

#

switch to mamogram

sterile egret Dec 21, 2024, 9:39 PM

#

torpid root switch to mamogram

ASSERT FAILED

calm latch Dec 21, 2024, 9:56 PM

#

assertos

slim panther Dec 22, 2024, 1:33 AM

#

made some small opts and i'm now on 1.5 mil in the benchmark

fiery turtle Dec 22, 2024, 1:38 AM

#

Nice, but did you fix the stack corruption

slim panther Dec 22, 2024, 1:38 AM

#

no 😢

#

i have genuinely no clue what it could be

#

see the latest messages

fiery turtle Dec 22, 2024, 1:38 AM

#

Yeah idk, id printf the shit out of it

slim panther Dec 22, 2024, 1:38 AM

#

#1287456798407790684 message

#

maybe i can clear the TSS?

#

not sure if that does anything

fiery turtle Dec 22, 2024, 1:39 AM

#

Doubt it

#

Only stack pointer is used from it

#

I bet it's a one line fix

slim panther Dec 22, 2024, 1:40 AM

#

probably

#

nobody can help me though nooo

fiery turtle Dec 22, 2024, 1:40 AM

#

The Gnu debugger of course

slim panther Dec 22, 2024, 1:41 AM

#

it can't

#

this is kvm only

fiery turtle Dec 22, 2024, 1:41 AM

#

Kvm is debuggable under gdb tho

slim panther Dec 22, 2024, 1:41 AM

#

didnt work when i tried

#

it just fell through my breakpoints

fiery turtle Dec 22, 2024, 1:41 AM

#

I think it should

#

Not sure

hollow elm Dec 22, 2024, 1:41 AM

#

slim panther it just fell through my breakpoints

hbreak

slim panther Dec 22, 2024, 1:42 AM

#

i keep forgetting

slim panther Dec 22, 2024, 2:17 AM

#

fiery turtle I bet it's a one line fix

you were correct

fiery turtle Dec 22, 2024, 9:40 AM

#

slim panther you were correct

Why didnt it break on tcg

north holly Dec 22, 2024, 12:56 PM

#

fiery turtle Why didnt it break on tcg

tcg doesn't check some stuff

slim panther Dec 22, 2024, 1:04 PM

#

^

fiery turtle Dec 22, 2024, 1:05 PM

#

there should be something like -accel tcg,fortify=on or something

#

thats slow but actually checks stuff

slim panther Dec 22, 2024, 1:39 PM

#

qemu-system-x86_64: -accel tcg,fortify=on: Property 'tcg-accel.fortify' not found

#

meme

fiery turtle Dec 22, 2024, 2:00 PM

#

should be => I want that feature

slim panther Dec 22, 2024, 2:10 PM

#

oh

#

i thought you meant "i'm not certain but there must be a feature like that"

#

meme

fiery turtle Dec 22, 2024, 2:10 PM

#

i worder that poorly

sterile egret Dec 22, 2024, 11:45 PM

#

fiery turtle should be => I want that feature

Add it meme

slim panther Dec 22, 2024, 11:51 PM

#

@fiery turtle 😢

north holly Dec 22, 2024, 11:51 PM

#

no -Werror for uacpi

#

is what I do

slim panther Dec 22, 2024, 11:51 PM

#

is this another gcc false alarm

#

fucking gnu tools

#

trash

north holly Dec 22, 2024, 11:52 PM

#

yes

fiery turtle Dec 22, 2024, 11:52 PM

#

it is yeah

#

ill try to fix it, but i have an old gcc lol

slim panther Dec 22, 2024, 11:52 PM

#

woe

#

i'm on latest latest

fiery turtle Dec 22, 2024, 11:53 PM

#

i have whatever ubuntu 20 ships with

slim panther Dec 22, 2024, 11:54 PM

#

damn

#

build from source galaxybrain

#

oh damn why does -smp tank so much performance on the uacpi bench

#

woe

#

okay new record is 1.5mil

#

it's bottlenecked by the not fair scheduler and slab of shit

fiery turtle Dec 23, 2024, 12:21 AM

#

slim panther <@353593948523986945> 😢

can u copy paste all warnings and send a pastebin or something

slim panther Dec 23, 2024, 12:22 AM

#

i did already

fiery turtle Dec 23, 2024, 12:22 AM

#

hm?

slim panther Dec 23, 2024, 12:22 AM

#

i need to find it

#

i posted it days ago

fiery turtle Dec 23, 2024, 12:22 AM

#

oh u did?

#

i guess i missed it somehow

slim panther Dec 23, 2024, 12:23 AM

#

.

fiery turtle Dec 23, 2024, 12:23 AM

#

thx

#

I literally have a comment right above it lmao

#

this is literally what gcc is complaning about

#

oh well

slim panther Dec 23, 2024, 12:25 AM

#

welp

fiery turtle Dec 23, 2024, 12:28 AM

#

nice, its forcing me to zero a 64-byte struct

slim panther Dec 23, 2024, 12:28 AM

#

= {0} meme

fiery turtle Dec 23, 2024, 12:28 AM

#

i know

#

but its still not free

slim panther Dec 23, 2024, 12:28 AM

#

true

#

muh safety

kind mantle Dec 23, 2024, 12:29 AM

#

I think there was an attribute for this?

fiery turtle Dec 23, 2024, 12:30 AM

#

ill think of a solution

kind mantle Dec 23, 2024, 12:32 AM

#

The other option is to init to 0 on error just to shut GCC up, while not impacting performance too much because it's unlikely to happen.

fiery turtle Dec 23, 2024, 12:41 AM

#

There's already a return on error above it in the switch, but apparently thats not enough

slim panther Dec 23, 2024, 12:58 AM

#

@fiery turtle is it normal to get a shit ton of warnings on real hw?

#

something about a scope not found

fiery turtle Dec 23, 2024, 12:58 AM

#

Do you implement all kernel api?

slim panther Dec 23, 2024, 12:59 AM

#

no

fiery turtle Dec 23, 2024, 12:59 AM

#

Then yes

#

Try pmos and see if it produces any warnings like what u saw

vast kestrel Dec 23, 2024, 8:22 AM

#

How does windows treat the scope not found and alike?

fiery turtle Dec 23, 2024, 9:05 AM

#

I think skips as well but id have to double check

fiery turtle Dec 23, 2024, 10:15 AM

#

I might have to pre-map all IO registers from FADT

#

otherwise i leak a mapping if it shuts down or suspends too fast

#

which is probably a good idea anyway

vast kestrel Dec 23, 2024, 10:17 AM

#

Lmao

fiery turtle Dec 23, 2024, 10:18 AM

#

or option 2

#

remove io mapping abstractions entirely

#

because so far everyone just stubs them out anyway

#

since im doing kernel_api simplifications anyway..

#

ill make a poll

vast kestrel Dec 23, 2024, 10:21 AM

#

I mean, the io port abstraction is only relevant for non-x86

#

And micro kernels

fiery turtle Dec 23, 2024, 10:22 AM

#

#polls message

fiery turtle Dec 23, 2024, 10:22 AM

#

vast kestrel And micro kernels

yeah

vast kestrel Dec 23, 2024, 10:22 AM

#

Even if stubbed I think it's still the correct and most versatile abstraction

fiery turtle Dec 23, 2024, 10:22 AM

#

thing is

vast kestrel Dec 23, 2024, 10:23 AM

#

Could also make it optional

fiery turtle Dec 23, 2024, 10:23 AM

#

u can still keep a cache internally and map if needed, like for the raw helper before

vast kestrel Dec 23, 2024, 10:23 AM

#

Like you have for the printf

#

Yeah but you don't know when to unmap

fiery turtle Dec 23, 2024, 10:23 AM

#

thing is this abstraction adds a ton of extra complexity

vast kestrel Dec 23, 2024, 10:23 AM

#

Even tho for io it probably doesn't matter

fiery turtle Dec 23, 2024, 10:23 AM

#

i cant make it optional and not leak at the same time for example

#

or I can but its not trivial

#

idk

vast kestrel Dec 23, 2024, 10:23 AM

#

Tbh

#

Because of how acpi is

#

It's probably fine to keep the mapping forever

#

Even tho 32bit might not like it

fiery turtle Dec 23, 2024, 10:24 AM

#

probably

#

32-bit arm u mean?

vast kestrel Dec 23, 2024, 10:24 AM

#

32bit address space

#

You have limited space for mapping

fiery turtle Dec 23, 2024, 10:24 AM

#

io is port-io

#

SystemIO not SystemMemory

#

or do u mean in general

vast kestrel Dec 23, 2024, 10:25 AM

#

Ooooo

#

Yeah for port io it probably doesn't matter

#

Or well, like you said 32bit arm

#

Hmmmm

fiery turtle Dec 23, 2024, 10:26 AM

#

then again i think SystemIO is used only on x86 even though technically they could use it for arm

vast kestrel Dec 23, 2024, 10:27 AM

#

Yeah for io I think lazy mapping for whoever needs it is enough

#

And for the once in a blue moon it happens on arm they can handle it

calm latch Dec 23, 2024, 10:27 AM

#

fiery turtle then again i think SystemIO is used only on x86 even though technically they cou...

mmio?

fiery turtle Dec 23, 2024, 10:27 AM

#

inb/outb etc

vast kestrel Dec 23, 2024, 10:28 AM

#

SystemMemory includes MMIO

fiery turtle Dec 23, 2024, 10:28 AM

#

^

vast kestrel Dec 23, 2024, 10:28 AM

#

SystemIO is purely io ports

#

On arm/riscv they will be memory mapped too because no instructions

fiery turtle Dec 23, 2024, 10:28 AM

#

ACPICA has a check for SystemIO like > 0xFFFF die() or something

#

so its probably not used for arm

vast kestrel Dec 23, 2024, 10:29 AM

#

Other than very legacy devices it's not really needed

fiery turtle Dec 23, 2024, 10:29 AM

#

yeah

vast kestrel Dec 23, 2024, 10:29 AM

#

Like, sure alot of devices have port io for simpler reg access, but no one actually is going to access them like that since you have a full address space anyways

#

It's mostly meant for legacy bioses running real mode and don't want unreal mode

fiery turtle Dec 23, 2024, 10:30 AM

#

true

#

acpi_status acpi_os_read_port(acpi_io_address port, u32 *value, u32 width)
{
    u32 dummy;

    if (!IS_ENABLED(CONFIG_HAS_IOPORT)) {
        /*
         * set all-1 result as if reading from non-existing
         * I/O port
         */
        *value = GENMASK(width, 0);
        return AE_NOT_IMPLEMENTED;
    }

    if (value)
        *value = 0;
    else
        value = &dummy;

    if (width <= 8) {
        *value = inb(port);
    } else if (width <= 16) {
        *value = inw(port);
    } else if (width <= 32) {
        *value = inl(port);
    } else {
        pr_debug("%s: Access width %d not supported\n", __func__, width);
        return AE_BAD_PARAMETER;
    }

    return AE_OK;
}

torpid root Dec 23, 2024, 10:32 AM

#

<= there is weird

fiery turtle Dec 23, 2024, 10:33 AM

#

acpica passes misaligned bit count unlike uacpi

#

uacpi aligns up to byte count to make it less confusing

#

but whats funny is aarch64 selects CONFIG_HAS_IOPORT LULW

gentle peak Dec 23, 2024, 10:36 AM

#

imo the pio map/unmap should stay, they're helpful for microkernels that use iopb and for monolithic ones and microkernels that use iopl you can just cast the base address to a pointer and back

vast kestrel Dec 23, 2024, 10:37 AM

#

Yeah but you can also iopb on demand

fiery turtle Dec 23, 2024, 10:37 AM

#

we have pmos as an example and it just toggles the IOPL bit LULW

fiery turtle Dec 23, 2024, 10:37 AM

#

vast kestrel Yeah but you can also iopb on demand

also this

#

or if you're lazy u can map() do_io() unmap()

vast kestrel Dec 23, 2024, 10:38 AM

#

Tbf, given you are going to give all ports to acpi anyways

#

Why not iopl

gentle peak Dec 23, 2024, 10:38 AM

#

that's true

fiery turtle Dec 23, 2024, 10:38 AM

#

true actually

vast kestrel Dec 23, 2024, 10:38 AM

#

Or premap everything you allow and catch the fault

#

/verify with whitelist

fiery turtle Dec 23, 2024, 10:38 AM

#

yeah thats actually a smart thought

#

because the firmware might write to any io port

#

u dont know which one does what

vast kestrel Dec 23, 2024, 10:39 AM

#

Like, Linux only has a tiny blacklist iirc

fiery turtle Dec 23, 2024, 10:39 AM

#

yeah i think cmos is masked

#

and something else

vast kestrel Dec 23, 2024, 10:39 AM

#

CMOS, PCI and some others

fiery turtle Dec 23, 2024, 10:39 AM

#

the blacklist is actually built into acpica

vast kestrel Dec 23, 2024, 10:39 AM

#

oh

fiery turtle Dec 23, 2024, 10:40 AM

#

and I think windows also has a blacklist for ports

vast kestrel Dec 23, 2024, 10:40 AM

#

Does windows have a blacklist?

fiery turtle Dec 23, 2024, 10:40 AM

#

ill have to implement that in uacpi as well probably

fiery turtle Dec 23, 2024, 10:40 AM

#

vast kestrel Does windows have a blacklist?

ye

#

/*
 * Protected I/O ports. Some ports are always illegal, and some are
 * conditionally illegal. This table must remain ordered by port address.
 *
 * The table is used to implement the Microsoft port access rules that
 * first appeared in Windows XP. Some ports are always illegal, and some
 * ports are only illegal if the BIOS calls _OSI with nothing newer than
 * the specific _OSI strings.
 *
 * This provides ACPICA with the desired port protections and
 * Microsoft compatibility.
 *
 * Description of port entries:
 *  DMA:   DMA controller
 *  PIC0:  Programmable Interrupt Controller (8259A)
 *  PIT1:  System Timer 1
 *  PIT2:  System Timer 2 failsafe
 *  RTC:   Real-time clock
 *  CMOS:  Extended CMOS
 *  DMA1:  DMA 1 page registers
 *  DMA1L: DMA 1 Ch 0 low page
 *  DMA2:  DMA 2 page registers
 *  DMA2L: DMA 2 low page refresh
 *  ARBC:  Arbitration control
 *  SETUP: Reserved system board setup
 *  POS:   POS channel select
 *  PIC1:  Cascaded PIC
 *  IDMA:  ISA DMA
 *  ELCR:  PIC edge/level registers
 *  PCI:   PCI configuration space
 */
static const ACPI_PORT_INFO     AcpiProtectedPorts[] =
{
    {"DMA",     0x0000, 0x000F, ACPI_OSI_WIN_XP},
    {"PIC0",    0x0020, 0x0021, ACPI_ALWAYS_ILLEGAL},
    {"PIT1",    0x0040, 0x0043, ACPI_OSI_WIN_XP},
    {"PIT2",    0x0048, 0x004B, ACPI_OSI_WIN_XP},
    {"RTC",     0x0070, 0x0071, ACPI_OSI_WIN_XP},
    {"CMOS",    0x0074, 0x0076, ACPI_OSI_WIN_XP},
    {"DMA1",    0x0081, 0x0083, ACPI_OSI_WIN_XP},
    {"DMA1L",   0x0087, 0x0087, ACPI_OSI_WIN_XP},
    {"DMA2",    0x0089, 0x008B, ACPI_OSI_WIN_XP},
    {"DMA2L",   0x008F, 0x008F, ACPI_OSI_WIN_XP},
    {"ARBC",    0x0090, 0x0091, ACPI_OSI_WIN_XP},
    {"SETUP",   0x0093, 0x0094, ACPI_OSI_WIN_XP},
    {"POS",     0x0096, 0x0097, ACPI_OSI_WIN_XP},
    {"PIC1",    0x00A0, 0x00A1, ACPI_ALWAYS_ILLEGAL},
    {"IDMA",    0x00C0, 0x00DF, ACPI_OSI_WIN_XP},
    {"ELCR",    0x04D0, 0x04D1, ACPI_ALWAYS_ILLEGAL},
    {"PCI",     0x0CF8, 0x0CFF, ACPI_OSI_WIN_XP}
};

vast kestrel Dec 23, 2024, 10:41 AM

#

I mean, tbh it's mostly superficial and to discourage bugs

fiery turtle Dec 23, 2024, 10:42 AM

#

PCI is technically legal even

vast kestrel Dec 23, 2024, 10:42 AM

#

Damn

fiery turtle Dec 23, 2024, 10:42 AM

#

u just have to _OSI("Windows XP")

vast kestrel Dec 23, 2024, 10:42 AM

#

Lol

fiery turtle Dec 23, 2024, 10:43 AM

#

thats kinda strange but okay i guess

#

earlier versions of aml might do nasty stuff ig?

vast kestrel Dec 23, 2024, 10:43 AM

#

I wonder if windows does the same

fiery turtle Dec 23, 2024, 10:43 AM

#

it says so in the comment above at least

vast kestrel Dec 23, 2024, 10:44 AM

#

Yeah but is windows doing the same check or is just that winxp didn't check

fiery turtle Dec 23, 2024, 10:45 AM

#

The table is used to implement the Microsoft port access rules that
first appeared in Windows XP
Ig?

vast kestrel Dec 23, 2024, 10:45 AM

#

Hmmmm

fiery turtle Dec 23, 2024, 10:46 AM

#

/* Port illegality may depend on the _OSI calls made by the BIOS */
if (PortInfo->OsiDependency == ACPI_ALWAYS_ILLEGAL ||
    AcpiGbl_OsiData == PortInfo->OsiDependency)
{
    ACPI_DEBUG_PRINT ((ACPI_DB_VALUES,
        "Denied AML access to port 0x%8.8X%8.8X/%X (%s 0x%.4X-0x%.4X)\n",
        ACPI_FORMAT_UINT64 (Address), ByteWidth, PortInfo->Name,
        PortInfo->Start, PortInfo->End));

    return_ACPI_STATUS (AE_AML_ILLEGAL_ADDRESS);
}

#

this check looks bogus as well

#

AcpiGbl_OsiData == PortInfo->OsiDependency since its an ==

#

esentially if latest_requested_win ==

#

so disallowed for XP

#

but allowed for e.g. windows vista?

#

how does this make sense

#

@gentle peak btw did we convince you to vote for option 1 instead?

gentle peak Dec 23, 2024, 10:49 AM

#

you can change your vote?

fiery turtle Dec 23, 2024, 10:49 AM

#

think so?

#

"remove vote"

gentle peak Dec 23, 2024, 10:49 AM

#

anyway yeah I didn't think it through fully, didn't realize you'd need the equivalent of iopl3 even in microkernels

fiery turtle Dec 23, 2024, 10:52 AM

#

nice

#

actually im not sure how u could even run it in userspace and support suspend for example

#

like it wakes up in real mode

#

then you have to do full hw re-init

#

or i guess u would have to split responsibilities somehow

#

idk

fiery turtle Dec 23, 2024, 11:10 AM

#

Alright time to implement support for these

    UACPI_ADDRESS_SPACE_SMBUS = 4,
    UACPI_ADDRESS_SPACE_SYSTEM_CMOS = 5,
    UACPI_ADDRESS_SPACE_PCI_BAR_TARGET = 6,
    UACPI_ADDRESS_SPACE_IPMI = 7,
    UACPI_ADDRESS_SPACE_GENERAL_PURPOSE_IO = 8,
    UACPI_ADDRESS_SPACE_GENERIC_SERIAL_BUS = 9,
    UACPI_ADDRESS_SPACE_PCC = 0x0A,
    UACPI_ADDRESS_SPACE_PRM = 0x0B,
    UACPI_ADDRESS_SPACE_FFIXEDHW = 0x7F,

#

i'll start with PCC

#

it should be simple

#

each opregion represents a buffer, writes copy into it, reads copy from it, write at a specific offset invokes the user handler

left orbit Dec 23, 2024, 11:11 AM

#

what the fuck is a pcc

fiery turtle Dec 23, 2024, 11:11 AM

#

left orbit what the fuck is a pcc

https://elixir.bootlin.com/linux/v6.12.6/source/drivers/acpi/acpi_pcc.c

acpi_pcc.c - drivers/acpi/acpi_pcc.c - Linux source ...

Elixir Cross Referencer - source code of Linux v6.12.6: drivers/acpi/acpi_pcc.c

#

 * The PCC Address Space also referred as PCC Operation Region pertains to the
 * region of PCC subspace that succeeds the PCC signature. The PCC Operation
 * Region works in conjunction with the PCC Table(Platform Communications
 * Channel Table). PCC subspaces that are marked for use as PCC Operation
 * Regions must not be used as PCC subspaces for the standard ACPI features
 * such as CPPC, RASF, PDTT and MPST. These standard features must always use
 * the PCC Table instead.
 *
 * This driver sets up the PCC Address Space and installs an handler to enable
 * handling of PCC OpRegion in the firmware.

idk

calm latch Dec 23, 2024, 11:44 AM

#

PCC is PCC what don't you understand

vast kestrel Dec 23, 2024, 12:31 PM

#

Platform communication channel

#

https://uefi.org/htmlspecs/ACPI_Spec_6_4_html/14_Platform_Communications_Channel/Platform_Comm_Channel.html

#

Tbh it seems like an even more generic EC

calm latch Dec 23, 2024, 12:34 PM

#

Do a lot of PCs use it?

left orbit Dec 23, 2024, 12:35 PM

#

judging by the fact it wasn't an immediate priority for uacpi, my guess would be no

calm latch Dec 23, 2024, 12:36 PM

#

I mean GPIO spec says it's for reduced hardware platforms, and it turned out that new laptops wire power buttons through them

fiery turtle Dec 23, 2024, 12:36 PM

#

calm latch Do a lot of PCs use it?

it was invented in 2021

#

it barely even exists

calm latch Dec 23, 2024, 12:44 PM

#

Btw, I think the macbook is doing some stuff with AHCI controller when turning off

fiery turtle Dec 23, 2024, 12:44 PM

#

i mean yeah that makes sense

#

like disabling it?

vast kestrel Dec 23, 2024, 12:45 PM

#

I recently noticed that if I force shutdown my PC using the power button some of the LEDs stay on, while with normal acpi shutdown they get turned off

calm latch Dec 23, 2024, 12:46 PM

#

fiery turtle like disabling it?

Idk, but my AHCI driver gets a bunch of interrupts when it turns off

fiery turtle Dec 23, 2024, 12:46 PM

#

possible

fiery turtle Dec 23, 2024, 12:46 PM

#

vast kestrel I recently noticed that if I force shutdown my PC using the power button some of...

L firmware i guess

calm latch Dec 23, 2024, 12:47 PM

#

fiery turtle Alright time to implement support for these ``` UACPI_ADDRESS_SPACE_SMBUS = ...

Can ACPI write to GPIO?

fiery turtle Dec 23, 2024, 12:47 PM

#

yes

vast kestrel Dec 23, 2024, 12:47 PM

#

Fun fact, it used to be that acpi shutdown from windows also left the LEDs on, but after a bios update it was fixed :^)

fiery turtle Dec 23, 2024, 12:47 PM

#

lol

vast kestrel Dec 23, 2024, 12:48 PM

#

GPIO doesn't have to be done from the GPIO address space, it might be accessed by acpi using the pch directly

calm latch Dec 23, 2024, 12:48 PM

#

fiery turtle yes

I can try implementing that in my driver then

vast kestrel Dec 23, 2024, 12:49 PM

#

that's what I seen a lot of server hardware do at least, accessing random GPIO regs in the pch when doing stuff

fiery turtle Dec 23, 2024, 12:49 PM

#

vast kestrel GPIO doesn't have to be done from the GPIO address space, it might be accessed b...

yeah

#

thats what they do for SMBus

#

yet they dont do it for EC for whatever reason

#

even though there's an SMBUS address space opregion

#

(technically supported)

calm latch Dec 23, 2024, 12:50 PM

#

Could they be doing that if there's no driver?

vast kestrel Dec 23, 2024, 12:50 PM

#

I think it's just that they use the EC driver from one of their drivers while not needing the smbus for one of their drivers

#

Or they just race :^)

fiery turtle Dec 23, 2024, 12:51 PM

#

calm latch Could they be doing that if there's no driver?

yeah

#

some firmware drops into SMM if EC is not available

#

and reads EC via that

calm latch Dec 23, 2024, 12:51 PM

#

Lol

#

You gotta support Windows XP

vast kestrel Dec 23, 2024, 12:51 PM

#

Tbh shit like smbus is annoying to have drivers for

#

Cause it's chipset dependent

fiery turtle Dec 23, 2024, 12:52 PM

#

is it?

vast kestrel Dec 23, 2024, 12:52 PM

#

I think so, the Intel smbus controlled changed between chipsets

#

Same for the Intel GPIO controller

#

Especially with some of the Intel socs

fiery turtle Dec 23, 2024, 12:52 PM

#

L

vast kestrel Dec 23, 2024, 12:53 PM

#

Where they also have GPIO/smbus as a feature for the board

#

Some even have uart dma :^)

fiery turtle Dec 23, 2024, 12:54 PM

#

bruh

fiery turtle Dec 23, 2024, 1:43 PM

#

switch (region->space) {
case UACPI_ADDRESS_SPACE_PCC: {
    uacpi_u8 *cursor;

    cursor = region->internal_buffer + offset;

    /*
     * Reads from PCC just return the current contents of the internal
     * buffer.
     */
    if (op == UACPI_REGION_OP_READ) {
        uacpi_memcpy(in_out, cursor, byte_width);
        goto out;
    }

    uacpi_memcpy(cursor, in_out, byte_width);

    // Dispatch a PCC send command if this was a write to the command field
    if (offset >= 12 && offset < 16) {
        uacpi_region_pcc_send_data data = {
            .region_context = region->user_context,
            .handler_context = handler->user_context,
            .buffer = region->internal_buffer,
            .length = region->length,
        };

        op = UACPI_REGION_OP_PCC_SEND;

        uacpi_namespace_write_unlock();
        ret = handler->callback(op, &data);
        uacpi_namespace_write_lock();
    }

    break;
}
default: {
    uacpi_region_rw_data data = {
        .region_context = region->user_context,
        .handler_context = handler->user_context,
        .byte_width = byte_width,
        .offset = offset,
    };

    if (op == UACPI_REGION_OP_WRITE) {
        data.value = *in_out;
        uacpi_trace_region_io(
            region_node, space, op, abs_offset,
            byte_width, *in_out
        );
    }

    uacpi_namespace_write_unlock();
    ret = handler->callback(op, &data);
    uacpi_namespace_write_lock();

    if (uacpi_unlikely_error(ret))
        break;

    if (op == UACPI_REGION_OP_READ) {
        *in_out = data.value;
        uacpi_trace_region_io(
            region_node, space, op, data.offset,
            byte_width, data.value
        );
    }

    break;
}
}

Alright I guess this is the framework of how I will be adding support for other address spaces

fiery turtle Dec 23, 2024, 1:56 PM

#

fiery turtle ```cpp switch (region->space) { case UACPI_ADDRESS_SPACE_PCC: { uacpi_u8 *cu...

the only thing that worries me is I think GCC would allocate all stack space needed for all switch cases in debug mode

#

causing a larger stack usage

#

which is annoying

vast kestrel Dec 23, 2024, 2:20 PM

#

Use alloca :^)

#

Tbf alloca could get the same treatment depending on impl

fiery turtle Dec 23, 2024, 2:21 PM

#

on gcc it probably would

#

since its implemented in IR

vast kestrel Dec 23, 2024, 2:21 PM

#

Especially if it's constant size and not in a loop

fiery turtle Dec 23, 2024, 2:21 PM

#

yup

rustic compass Dec 23, 2024, 2:29 PM

#

Infy your uacpi api isn't liked by the rust chat KEKW
#592856094527848449 message

fiery turtle Dec 23, 2024, 2:31 PM

#

here we go with the vtable stuff again

#

not biased opinion at all

#

look ill add vtable mode some time in the future

rustic compass Dec 23, 2024, 2:32 PM

#

The current api is fine I just had some trouble with the rust-rust interface

slim panther Dec 23, 2024, 2:32 PM

#

rustic compass Infy your uacpi api isn't liked by the rust chat<:KEKW:913708372329644042> http...

#592856094527848449 message

#

you mean it's not liked by one person

#

meme

rustic compass Dec 23, 2024, 2:34 PM

#

I would have used the linker for rust-rust too but extern rust isn't exactly stable, apparently even when just using a crate from crates.io

slim panther Dec 23, 2024, 2:35 PM

#

that's an issue with the rust toolchain though

#

not with uacpi

fiery turtle Dec 23, 2024, 2:35 PM

#

thats not how the rust mind works LULW

slim panther Dec 23, 2024, 2:36 PM

#

imo maintaining two apis is unnecessary for basically no added benefit

#

either we only do vtables or only linker

fiery turtle Dec 23, 2024, 2:36 PM

#

i agree but as long as it's easy to make it work with uacpi and some people benefit from it, wahtever

#

and as long as its optional ofc

slim panther Dec 23, 2024, 2:37 PM

#

hm

fiery turtle Dec 23, 2024, 2:37 PM

#

ill just make something like #define uacpi_kernel_alloc(size) g_kapi->alloc(size)

slim panther Dec 23, 2024, 2:39 PM

#

in the case of uacpi, fair

#

did you intend of moving it to uapi?

fiery turtle Dec 23, 2024, 2:39 PM

#

in linker-only mode for now probably

#

in a similar fashion

#

just using macros

slim panther Dec 23, 2024, 2:39 PM

#

okay

fiery turtle Dec 23, 2024, 2:39 PM

#

do we even want a vtable mode for uapi?

slim panther Dec 23, 2024, 2:40 PM

#

i don't mind

fiery turtle Dec 23, 2024, 2:40 PM

#

same problem as uapi_status

#

we need an accumulating struct

slim panther Dec 23, 2024, 2:40 PM

#

okay period. can we just summarize in one message what the upsides and downsides are for both approaches. i hate having to scroll between 5 discussions

fiery turtle Dec 23, 2024, 2:41 PM

#

u can summarize in udrm for now sure

gentle peak Dec 23, 2024, 2:41 PM

#

i personally prefer the way it currently works because it allows LTO to inline kernel APIs

fiery turtle Dec 23, 2024, 2:41 PM

#

same

#

i didnt want to pay any perf price just because i decided to make a generic library

gentle peak Dec 23, 2024, 2:45 PM

#

the main argument I've seen in favor of vtables is that it allows you to easily swap implementations depending on what's available, but I don't see why you couldn't just do that yourself in the kernel api impls

north holly Dec 23, 2024, 2:45 PM

#

what's stopping the rusties from just having a c shim (which is actually the kernel api) that calls into their rust vtables

#

or smth

gentle peak Dec 23, 2024, 2:45 PM

#

yeah exactly

fiery turtle Dec 23, 2024, 2:45 PM

#

yup

gentle peak Dec 23, 2024, 2:47 PM

#

that kinda thing is a bit more annoying due to rust's whole "static mut is unsafe" thing but I'm pretty sure it has atomic pointers

fiery turtle Dec 23, 2024, 3:04 PM

#

fuck, implementing support for GPIO and other crap will be difficult

#

since they change the meaning of every field

#

bit offset is now the pin number, length doesnt matter whatsoever, Connection attribute resets pin offset

#

why couldnt they invent a different mechanism for this...

#

OperationRegion (GPOP, GeneralPurposeIo, Zero, 0x0C)
Field (GPOP, ByteAcc, NoLock, Preserve)
{
    Connection (
        GpioIo (Exclusive, PullDefault, 0x0000, 0x0000, IoRestrictionOutputOnly,
            "\\_SB.GPO0", 0x00, ResourceConsumer, ,
            )
            {   // Pin list
                0x0035
            }
    ), 
    BTD3,   1, 
    Connection (
        GpioIo (Exclusive, PullDefault, 0x0000, 0x0000, IoRestrictionOutputOnly,
            "\\_SB.GPO0", 0x00, ResourceConsumer, ,
            )
            {   // Pin list
                0x003A
            }
    ), 
    SUSB,   1, 
    Connection (
        GpioIo (Exclusive, PullDefault, 0x0000, 0x0000, IoRestrictionOutputOnly,
            "\\_SB.GPO0", 0x00, ResourceConsumer, ,
            )
            {   // Pin list
                0x003B
            }
    ), 
    SHD3,   1, 
    Connection (
        GpioIo (Exclusive, PullDefault, 0x0000, 0x0000, IoRestrictionOutputOnly,
            "\\_SB.GPO0", 0x00, ResourceConsumer, ,
            )
            {   // Pin list
                0x003C
            }
    ), 
    TCD3,   1
}
}

Example

#

Desktop/Dell/G5/G5 5090/67C667DAE80A
Desktop/Dell/XPS/XPS 8940/9B79A7439313
Mini Pc/AMI/Aptio/Aptio CRB/259F9FDD46E2
Notebook/Aava Mobile Oy/INARI8/INARI8-LTBN-1/A658B5B6E0CB
Notebook/BBEN/z/z10/65E372FCAFD3
Notebook/Fujitsu/LIFEBOOK/LIFEBOOK U938/71F4D08280F4
Notebook/Hewlett-Packard/Pavilion/Pavilion x2 Detachable/5850EA215153
Notebook/ilife/S/S806/7AF027A015B2
Notebook/Insyde/BayTrail/BayTrail/9F287627613B
Notebook/Lenovo/Flex/Flex 2-15/8F84E14F5157
Notebook/Lenovo/IdeaPad/IdeaPad 100S-11IBY 80R2/EE707040AC1A
Notebook/Others/Others/Others/A2330B5932DB
Tablet/Hewlett-Packard/Pavilion/Pavilion x2 Detachable/3FE302D11C0A
Tablet/Lenovo/MIIX/MIIX 3-1030 80HV/04FF5A51E4B0
Tablet/Microsoft/Surface/Surface Laptop 3/4E426AB8062D
Tablet/Microsoft/Surface/Surface Pro/773032238BB5
Tablet/Microsoft/Surface/Surface Pro 3/074C5C2E091B

#

quite a lot of users of this crap too

north holly Dec 23, 2024, 3:09 PM

#

others/others/others

fiery turtle Dec 23, 2024, 3:10 PM

#

the acpi spec really wants to delay 1.0 as far as possible LULW

calm latch Dec 23, 2024, 3:12 PM

#

Does my laptop have this?

fiery turtle Dec 23, 2024, 3:13 PM

#

now I have to carefully port the collection of ACPICA hacks to make this work in a clean way

fiery turtle Dec 23, 2024, 3:13 PM

#

calm latch Does my laptop have this?

lets see

calm latch Dec 23, 2024, 3:13 PM

#

Also

#

Maybe I need to dump macbook's aml

fiery turtle Dec 23, 2024, 3:14 PM

#

❯ grep 'Connection (' *.dsl

#

no ur saved

calm latch Dec 23, 2024, 3:14 PM

#

That's boring

fiery turtle Dec 23, 2024, 3:14 PM

#

get one of those laptops

calm latch Dec 23, 2024, 3:14 PM

#

I'm sad my spectre x360 died

fiery turtle Dec 23, 2024, 3:15 PM

#

calm latch Maybe I need to dump macbook's aml

why?

calm latch Dec 23, 2024, 3:15 PM

#

It's probably interesting

fiery turtle Dec 23, 2024, 3:15 PM

#

yeah

calm latch Dec 23, 2024, 3:15 PM

#

fiery turtle ``` Desktop/Dell/G5/G5 5090/67C667DAE80A Desktop/Dell/XPS/XPS 8940/9B79A7439313 ...

My sister had dell xps

#

But she lives on a different continent now...

fiery turtle Dec 23, 2024, 3:21 PM

#

unlucky

calm latch Dec 23, 2024, 3:21 PM

#

I have stuff like this ```
Method (_CRS, 0, NotSerialized) // _CRS: Current Resource Settings
{
Name (RBUF, ResourceTemplate ()
{
I2cSerialBusV2 (0x0014, ControllerInitiated, 0x00061A80,
AddressingMode7Bit, "\_SB.I2CB",
0x00, ResourceConsumer, , Exclusive,
)
GpioInt (Level, ActiveLow, Exclusive, PullUp, 0x0000,
"\SB.GPIO", 0x00, ResourceConsumer, ,
)
{ // Pin list
0x0005
}
})
Return (RBUF) /* _SB.I2CB.TPNL._CRS.RBUF */
}

#

Maybe I want to write i2c driver...

fiery turtle Dec 23, 2024, 3:24 PM

#

yeah

flat badge Dec 23, 2024, 3:27 PM

#

vtables are slightly easier if ops are optional

#

Does uacpi have any optional bindings?

vast kestrel Dec 23, 2024, 3:27 PM

#

You are supposed to just stub and return not implemented error

fiery turtle Dec 23, 2024, 3:30 PM

#

flat badge Does uacpi have any optional bindings?

nope

vast kestrel Dec 23, 2024, 3:32 PM

#

Tbh linker gives the most flexibility, and it's like a 20 line file to make a vtable shim

fiery turtle Dec 23, 2024, 3:33 PM

#

well if u count "can gracefully handle lack of" an optional binding then yes

fiery turtle Dec 23, 2024, 3:33 PM

#

vast kestrel Tbh linker gives the most flexibility, and it's like a 20 line file to make a vt...

yeah

slim panther Dec 23, 2024, 3:39 PM

#

gentle peak that kinda thing is a bit more annoying due to rust's whole "static mut is unsaf...

it would be unsafe either way

#

because the vtable would contain extern "C" fn()s

gentle peak Dec 23, 2024, 3:41 PM

#

good point

frank canopy Dec 23, 2024, 4:09 PM

#

the binding difficulty is pretty much the same for zig with either vtable or linker so I'm not impacted by that debate at least

#

maybe a bit less work to make linker bindings but not easier per se just less typing

#

just wish I could propogate error returns from callbacks but that's whatever tbh

fiery turtle Dec 23, 2024, 4:53 PM

#

frank canopy just wish I could propogate error returns from callbacks but that's whatever tbh

what would need to be done to allow that

left orbit Dec 23, 2024, 4:53 PM

#

it's zig abi

#

so i guess you would need to allow for custom return types for the kernel api functions

#

but i don't think you can pass error unions across C boundary anyway

#

you couldn't on may 10 and i don't think it's gonna change any time soon https://stackoverflow.com/a/78459524

#

best you can do is wrap the actual implementation in a function returning an error union and then c export fn uacpi_some_kernel_api(...) uacpi_status { some_kernel_api_wrapped(...) catch |e| return error_to_uacpi_status(e); return UACPI_STATUS_SUCCESS; // or whatever it is }

#

or just catch return UACPI_SOME_ERROR; instead of trying in the actual kernel api

frank canopy Dec 23, 2024, 5:25 PM

#

left orbit but i don't think you can pass error unions across C boundary anyway

what ive got rn is functions that convert between a zig error set and the uacpi_status enum and i use the user context pointer to store the actual zig error

#

and then whatever calls the extern function can detect that error being in the context later and return it instead of the default InternalError i use

#

also in this case youd be able to use the int-error conversion because while the numeric mapping of zig errors isnt guaranteed across recompiles its guaranteed within the same zig compilation unit and everything is statically linked here

#

the biggest issue is not being able to make error return traces marshal through but theres nothing that can be done about that given how those are implemented in zig

#

for reference, every zig function that returns an error takes a secret extra parameter that is an optional pointer to the stack trace which it then uses a builtin function to add its address to whenever it calls an error-returning function or returns an error directly

#

and the size of the buffer used for the traces there is determined either by just giving it a default or by static analysis

#

but afaik you cant (or shouldnt) overwrite the pointer to marshal it across so i dont think i can marshal it properly

#

what ive currently got is just a wrapper around a callback type that wraps the user context pointer in another struct that contains an optional error and the first pointer and then pass that around and use some other wrapper code. it works but its ugly af

#

(it returns iteration decision break for iteration callbacks instead of internal error for status-returning callbacks)

#

i even found a zig compiler bug in that process lmao

#

but yeah no zig error returns are part of zig's own ABI and cant be done directly through C, and zig's ABI is not meant to work across link-boundaries because zig libraries are meant to be distributed as source and included directly rather than as libraries to be linked

fiery turtle Dec 23, 2024, 6:23 PM

#

@hasty plinth could u elaborate a bit on your poll vote

flat badge Dec 23, 2024, 7:08 PM

#

getting rid of io mapping makes it impossible to use uacpi in userspace

#

imho aiming for minimal API is not a good idea if it impacts the functionality

vast kestrel Dec 23, 2024, 7:08 PM

#

we said that iopl makes more sense for acpi runtime anyways, either that or premapping everything or lazy mapping as needed

flat badge Dec 23, 2024, 7:10 PM

#

i think that's a decision that the OS should do

#

i don't buy that getting rid of these functions makes anything easier (just stub them to do nothing)

#

while it definitely makes some stuff harder for some ppl

vast kestrel Dec 23, 2024, 7:12 PM

#

tbh the stub is pretty simple, *handle = (uacpi_handle_t)base, and then just (size_t)*handle + offset when accessing

fiery turtle Dec 23, 2024, 7:14 PM

#

flat badge getting rid of io mapping makes it impossible to use uacpi in userspace

currently there's one such user and they do IOPL

#

also definitely not impossible, just less convenient

vast kestrel Dec 23, 2024, 7:15 PM

#

tbh given the stub is pretty simple, I don't think its that big of a deal

fiery turtle Dec 23, 2024, 7:15 PM

#

that stub adds a ton of extra complexity to uacpi is the issue tbh

vast kestrel Dec 23, 2024, 7:15 PM

#

not needing both the raw and mapping is a big enough of a simplification

#

I mean, isn't it similar to the system memory tho?

#

and for facs you can premap

fiery turtle Dec 23, 2024, 7:17 PM

#

i can premap all fadt registers yes technically

#

or I could just not map at all :^)

vast kestrel Dec 23, 2024, 7:17 PM

#

ig gas is annoying tho

#

even tho there you already do map unmap for system memory

fiery turtle Dec 23, 2024, 7:18 PM

#

flat badge i think that's a decision that the OS should do

i dont buy this as well tbh, the os doesnt get to control what firmware does, it must blindly perform wahtever IO it's told, besides a few blacklisted registers

vast kestrel Dec 23, 2024, 7:19 PM

#

and these are blacklisted mostly by windows and not by spec :^)

fiery turtle Dec 23, 2024, 7:19 PM

#

yeah

flat badge Dec 23, 2024, 7:20 PM

#

why is io special though?

#

i'd expect that it's just like mmio

fiery turtle Dec 23, 2024, 7:22 PM

#

well its not i guess

frank canopy Dec 23, 2024, 7:25 PM

#

mmio would be the same if you were guaranteed a certain mapping from phys->virtual address

#

or at least thats my impression

vast kestrel Dec 23, 2024, 7:25 PM

#

well, on arm io is mapped as mmio

frank canopy Dec 23, 2024, 7:25 PM

#

true

#

though acpi tables might handle that already

vast kestrel Dec 23, 2024, 7:26 PM

#

ig the main diff is that io is pretty much always limited to a 64k address space (I think infy said acpica even asserts that)

#

so even on 32bit you can fully mmap it

frank canopy Dec 23, 2024, 7:29 PM

#

ig my question is if arm acpi tables still use system io address space or just already do the mmio conversion

#

because if they already convert then the argument about arm port io being mmio doesnt really apply here imo

#

since this is for when acpi tables want to do io

vast kestrel Dec 23, 2024, 7:29 PM

#

io ports on arm are mainly for pci compatibility

#

so acpi hardware on arm won't use it (if you eve have non-acpi reduced hardware)

fiery turtle Dec 23, 2024, 7:30 PM

#

I think SystemIO is an x86 thing on acpi

vast kestrel Dec 23, 2024, 7:30 PM

#

technically

#

because of pci bar type

frank canopy Dec 23, 2024, 7:30 PM

#

and therefore the systemio map/unmap handlers being discussed here will never be called on arm and thus the in/out instructions are always a valid implementation, no?

vast kestrel Dec 23, 2024, 7:30 PM

#

you don't even need io ports on arm

fiery turtle Dec 23, 2024, 7:31 PM

#

frank canopy and therefore the systemio map/unmap handlers being discussed here will never be...

thats kind of my point

#

but then microkernels and stuff etc etc

#

was koronas argument

frank canopy Dec 23, 2024, 7:31 PM

#

yeah

#

so its only about microkernels and caching and etc etc

#

makes sense

calm latch Dec 23, 2024, 7:33 PM

#

Another reason I've used iopl is that my io ports stuff is per thread, and I didn't know if io map functions would get called from gpe handlers

fiery turtle Dec 23, 2024, 7:34 PM

#

its unspecified

frank canopy Dec 23, 2024, 7:40 PM

#

tbh my opinion is that stuff like the io map functions etc - stuff only or primarily meant to be called to implement AML operations - should have a correspondence to a thing aml does. if aml has a thing that corresponds to needing a function like that then let the os implement it, if not then dont

#

basically if the obvious 99% implementation is a no-op then why bother

fiery turtle Dec 23, 2024, 7:43 PM

#

thats basically my thinking

frank canopy Dec 23, 2024, 7:45 PM

#

actually, a compromise idea here:
when does acpi logically ever unmap an io region? encountering a systemio region is enough to say it might be used and you arent likely to ever stop needing it anyway, therefore:
why not provide an optional callback when a systemio region is discovered for the first time?

#

basically go hey fyi we encountered an io region of port 0x348..0x350

fiery turtle Dec 23, 2024, 7:46 PM

#

tbh its not about regions

#

its about FADT registers for the most part

#

regions are mapped on first access and umapped when destroyed

#

so basically either never or when function exits

frank canopy Dec 23, 2024, 7:47 PM

#

i feel like FADT registers should be mapped once on initial load and unmapped once on uninitialize

fiery turtle Dec 23, 2024, 7:47 PM

#

perhaps

frank canopy Dec 23, 2024, 7:47 PM

#

the F stands for fixed for a reason, no?

#

idk

fiery turtle Dec 23, 2024, 7:48 PM

#

like yes i can premap

#

but its extra logic and stuff that must be handled with care

frank canopy Dec 23, 2024, 7:48 PM

#

mhm

fiery turtle Dec 23, 2024, 7:48 PM

#

for abstraction that on 100% current implementations does nothing

frank canopy Dec 23, 2024, 7:48 PM

#

yep

#

i voted for remove it so idk why im coming up with compromises lol

calm latch Dec 23, 2024, 7:49 PM

#

calm latch Another reason I've used iopl is that my io ports stuff is per thread, and I did...

I can try implementing this to see what happens

fiery turtle Dec 23, 2024, 7:49 PM

#

acpica doesnt have such abstractions for pio for example

vast kestrel Dec 23, 2024, 7:52 PM

#

big brain

#

have uacpi always call the io map function once

#

mapping an entire 64k range

#

:^)

fiery turtle Dec 23, 2024, 7:52 PM

#

lmao

rustic compass Dec 23, 2024, 8:05 PM

#

uacpi is compileable by msvc and clang right?

fiery turtle Dec 23, 2024, 8:05 PM

#

ye

rustic compass Dec 23, 2024, 8:06 PM

#

ok currently having trouble compiling uacpi-rs as it aparently aborts as it cant find gcc

rustic compass Dec 23, 2024, 8:22 PM

#

what compiler flags do you use to compile uacpi with msvc?

#

uacpi-rs doesnt build as either the wrapper.h or the compiler flags are incompatible with msvc

fiery turtle Dec 23, 2024, 8:24 PM

#

rustic compass what compiler flags do you use to compile uacpi with msvc?

tests/CMakeLists.txt

rustic compass Dec 23, 2024, 8:24 PM

#

https://github.com/Shinribo/uacpi-rs/tree/1.0-dev

GitHub

GitHub - Shinribo/uacpi-rs at 1.0-dev

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

mortal yoke Dec 23, 2024, 8:24 PM

#

the current flags in uacpi-rs are gcc/clang specific yeah

rustic compass Dec 23, 2024, 8:26 PM

#

fiery turtle tests/CMakeLists.txt

i cant find it

fiery turtle Dec 23, 2024, 8:26 PM

#

/runner

rustic compass Dec 23, 2024, 8:28 PM

#

which set of these do i need to have the equivalent to the clang ones? /W3 /WX
/wd4200 /wd4267 /wd4244

mortal yoke Dec 23, 2024, 8:29 PM

#

I don't think any of those are equivalent for them, those are for warnings

#

idk what would be the msvc equivalents of the flags used in uacpi-rs, you'd have to look them up (but also F msvc, who uses that for a kernel when you have clang/mingw gcc if you want to target pe)

calm latch Dec 23, 2024, 8:29 PM

#

I compile pmOS with clang

fiery turtle Dec 23, 2024, 8:29 PM

#

these are just warnings yeah

rustic compass Dec 23, 2024, 8:45 PM

#

what msvc flags do i need to compile uacpi like the flags used for clang?

fiery turtle Dec 23, 2024, 8:46 PM

#

default ones

slim panther Dec 23, 2024, 8:46 PM

#

rustic compass which set of these do i need to have the equivalent to the clang ones? /W3 /WX ...

are these msvc flags

#

https://tenor.com/view/cat-despair-meme-atone-gif-18083281511005463831

Tenor

rustic compass Dec 23, 2024, 8:49 PM

#

fiery turtle default ones

so you mean i can just ignore these flags when compiling uacpi with msvc?
"-fno-stack-protector"
"-mgeneral-regs-only"
"-nostdlib"
"-ffreestanding"

#

i dont think that that will work properly

mortal yoke Dec 23, 2024, 8:49 PM

#

for a kernel I don't think it would work yeah

#

or well ig it depends on the kernel

fiery turtle Dec 23, 2024, 8:50 PM

#

oh u mean for a kernel

#

no idea

rustic compass Dec 23, 2024, 8:57 PM

#

i found a solution for msvc incompetence to support gcc/clangs compiler flags:

if cc.get_compiler().is_like_msvc() {
  compile_error!("uACPI does not support MSVC");
}

fiery turtle Dec 23, 2024, 8:58 PM

#

not true btw

rustic compass Dec 23, 2024, 8:59 PM

#

well

cc.files(sources)
        .include(format!("{uacpi_path_str}/include"))
        .define("UACPI_SIZED_FREES", "1")
        .flag("-fno-stack-protector")
        .flag("-mgeneral-regs-only")
        .flag("-nostdlib")
        .flag("-ffreestanding");

this doesnt work with msvc, it fails to build and ignores all flags

frank canopy Dec 23, 2024, 8:59 PM

#

just need to actually find the right flags

#

because msvc should work

#

you probably shouldnt use it for a kernel if you can help it

#

unless you want a PE executable for some reason

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