uACPI - a portable and easy-to-integrate ACPI implementation | Operating System Development | Page 18

north holly Dec 25, 2024, 7:05 PM

#

oh shit

#

obos is broken on amd

#

📎 obos.iso

calm latch Dec 25, 2024, 7:05 PM

#

I am at home now

north holly Dec 25, 2024, 7:05 PM

#

try it anyway

calm latch Dec 25, 2024, 7:05 PM

#

Where I have 4 more AMD machines

north holly Dec 25, 2024, 7:06 PM

#

meme

calm latch Dec 25, 2024, 7:06 PM

#

From different gens

slim panther Dec 25, 2024, 7:06 PM

#

north holly obos is broken on amd

do you have intel

north holly Dec 25, 2024, 7:06 PM

#

yes

calm latch Dec 25, 2024, 7:06 PM

#

Though I can't find the charger for one of the lapptops sadmeme

north holly Dec 25, 2024, 7:06 PM

#

but only intel

slim panther Dec 25, 2024, 7:06 PM

#

i have only amd

#

i got 3 machines

north holly Dec 25, 2024, 7:07 PM

#

oh funny thing

#

obos stopped booting on my pc without an EC

#

it hangs after trying to init an EC

fiery turtle Dec 25, 2024, 7:07 PM

#

ah no thats unrelated

calm latch Dec 25, 2024, 7:08 PM

#

north holly it hangs after trying to init an EC

Are you sure it doesn't have EC if it tries to init it? spun

north holly Dec 25, 2024, 7:08 PM

#

there's an EC object actually

calm latch Dec 25, 2024, 7:08 PM

#

Try pmOS ultrameme

north holly Dec 25, 2024, 7:08 PM

#

but it's bogus

fiery turtle Dec 25, 2024, 7:09 PM

#

actually nvm it was fine lol

north holly Dec 25, 2024, 7:09 PM

#

it has no _CRS

#

therefore it doesn't exist

calm latch Dec 25, 2024, 7:09 PM

#

My EC driver doesn't use that

#

Afaik

#

iirc

fiery turtle Dec 25, 2024, 7:10 PM

#

it cant not use that

#

it does current_resources() right?

#

thats _CRS

median crest Dec 25, 2024, 7:18 PM

#

Least batshit insane firmeare

north holly Dec 25, 2024, 7:19 PM

#

fiery turtle thats _CRS

there is no _CRS object for the EC controller

#

at all

north holly Dec 25, 2024, 7:20 PM

#

median crest Least batshit insane firmeare

when EC driver in astral

median crest Dec 25, 2024, 7:22 PM

#

north holly when EC driver in astral

When I do my devices

north holly Dec 25, 2024, 7:22 PM

#

it's not really a loadable driver

#

or one that I would make loadable

#

or discoverable through conventional methods

fiery turtle Dec 25, 2024, 7:27 PM

#

north holly there is no _CRS object for the EC controller

yes i remember your pc

north holly Dec 25, 2024, 7:28 PM

#

and so does my other computer's blob

#

has EC object

#

but no _CRS under it

fiery turtle Dec 25, 2024, 7:29 PM

#

could be in other SSDTs under the scope

north holly Dec 25, 2024, 7:29 PM

#

y tho

#

sounds retarted as fuck

#

wait

#

nvm

fiery turtle Dec 25, 2024, 7:30 PM

#

just filled by the bios dynamically

north holly Dec 25, 2024, 7:30 PM

#

I remember on my PC it said no EC

#

so

#

just another case of firmware dev

median crest Dec 25, 2024, 7:33 PM

#

north holly it's not really a loadable driver

Y not

north holly Dec 25, 2024, 7:34 PM

#

well, it intertwines with uacpi init quite a bit

#

afaik you rather initialize it before namespace load or after namespace init, depending on how you discover the EC

#

before namespace load is when you have an ECDT

#

and after namespace init is when you don't have an ECDT

fiery turtle Dec 25, 2024, 7:35 PM

#

yeah

north holly Dec 25, 2024, 7:36 PM

#

if you truly wanted to, you could add a config option to your kernel

#

to compile it out

jaunty fox Dec 25, 2024, 7:38 PM

#

how has my track record been

fiery turtle Dec 25, 2024, 7:38 PM

#

1 for 1 I think?

jaunty fox Dec 25, 2024, 7:38 PM

#

i think i've only reported one or two problems with u.ACPI and both genuine

fiery turtle Dec 25, 2024, 7:38 PM

#

yeah

jaunty fox Dec 25, 2024, 7:39 PM

#

it sounds like a lot of kernels' allocators are stretched to breaking point by uACPI and that provokes false reports

fiery turtle Dec 25, 2024, 7:39 PM

#

and locking primitives lately meme

median crest Dec 25, 2024, 7:39 PM

#

north holly afaik you rather initialize it before namespace load or after namespace init, de...

Sounds stupid

#

Astral survived both teste

#

Tests

north holly Dec 25, 2024, 7:40 PM

#

median crest Sounds stupid

tell that to linux

#

not me

#

if you had a penny for each time obos crashed because of uacpi in my 4th rewrite, you would be a billionaire

slim panther Dec 25, 2024, 7:46 PM

#

just use rust

#

it cannot crash

#

impossible

north holly Dec 25, 2024, 7:46 PM

#

I hate rust

slim panther Dec 25, 2024, 7:46 PM

#

why

vale isle Dec 25, 2024, 7:46 PM

#

slim panther impossible

hahahaha 🤣😂 - i am a rust user

north holly Dec 25, 2024, 7:47 PM

#

slim panther why

it smells stinky

vale isle Dec 25, 2024, 7:47 PM

#

i disagree

slim panther Dec 25, 2024, 7:47 PM

#

wow :(

vale isle Dec 25, 2024, 7:47 PM

#

i use rust because i like how it smells

slim panther Dec 25, 2024, 7:47 PM

#

it's just a tool

vale isle Dec 25, 2024, 7:47 PM

#

not because it's safe

slim panther Dec 25, 2024, 7:47 PM

#

i use rust because i don't have to think about the code i write

north holly Dec 25, 2024, 7:47 PM

#

I use C because I like how it smells

slim panther Dec 25, 2024, 7:48 PM

#

the compiler is my slave

vale isle Dec 25, 2024, 7:48 PM

#

tbh everyone should learn rust just to understand how to write safe code

#

then just use whatever smells the best

slim panther Dec 25, 2024, 7:48 PM

#

everyone should be open to learn new things

#

i tried zig and hated it

#

i tried rust and love it

heady pewter Dec 25, 2024, 7:48 PM

#

D

slim panther Dec 25, 2024, 7:48 PM

#

:D

vale isle Dec 25, 2024, 7:49 PM

#

heady pewter D

the language?

jaunty fox Dec 25, 2024, 7:49 PM

#

vale isle tbh everyone should learn rust just to understand how to write safe code

the question is what you mean by "how to write safe code"

heady pewter Dec 25, 2024, 7:49 PM

#

Yeah

jaunty fox Dec 25, 2024, 7:49 PM

#

you could do the same in Java or Haskell or ML

north holly Dec 25, 2024, 7:49 PM

#

just use kasan and ubsan smh

vale isle Dec 25, 2024, 7:49 PM

#

jaunty fox the question is what you mean by "how to write safe code"

i mean the borrowck mindset

slim panther Dec 25, 2024, 7:49 PM

#

north holly just use kasan and ubsan smh

smh just use runtime checks to fix compile time issues

vale isle Dec 25, 2024, 7:49 PM

#

whenever i write other languages i still think in rust

slim panther Dec 25, 2024, 7:49 PM

#

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

Tenor

north holly Dec 25, 2024, 7:49 PM

#

slim panther smh just use runtime checks to fix compile time issues

yes

#

idk how corrupting memory in my kernel is the compiler's problem tho

vale isle Dec 25, 2024, 7:51 PM

#

vale isle whenever i write other languages i still think in rust

which is why my C++ kernel was pretty much rust-with-templates cuz the stdlib i wrote was very rusty

jaunty fox Dec 25, 2024, 7:51 PM

#

i don't intend to come across arrogant in the slightest with this statement, but i have found obsessing over scalability and good locking in keyronex a tonic that has quite changed how i see memory and safety

vale isle Dec 25, 2024, 7:52 PM

#

sure experience in that will affect how you see this

#

that's not arrogant

#

it's not arrogant to be more experienced

jaunty fox Dec 25, 2024, 7:52 PM

#

the problems of memory reclamation in single-threaded or crudely multithreaded code now seem completely trifling

flat badge Dec 25, 2024, 7:53 PM

#

acutally, having RCU at your disposal sometimes makes lifetimes easier, not harder IMHO

jaunty fox Dec 25, 2024, 7:53 PM

#

flat badge acutally, having RCU at your disposal sometimes makes lifetimes easier, not hard...

you can say that again

flat badge Dec 25, 2024, 7:54 PM

#

because a lot of times you can just say "i know that objects of this type are never going away while I work with them"

jaunty fox Dec 25, 2024, 7:54 PM

#

the first taste of it was the vm_page struct

#

since those always exist that makes life so much easier

#

and now just being able to make anything that exists not go away over some period of time - or even to make anything that exists, if the memory backing it ends up reused, be reused as the same type - that is powerful too (and that way is less overhead since it fits so naturally with a slab allocator anyway - just make the slab be deferred free)

rustic compass Dec 25, 2024, 8:12 PM

#

@fiery turtle what is that? what type does it have exactly?

typedef struct uacpi_namespace_node uacpi_namespace_node;

#

i dont know what rust equivalent i should choose

fiery turtle Dec 25, 2024, 8:13 PM

#

opaque

rustic compass Dec 25, 2024, 8:13 PM

#

i mean is it mutated?

fiery turtle Dec 25, 2024, 8:15 PM

#

thats a tough question

#

not via public api

rustic compass Dec 25, 2024, 8:16 PM

#

is uacpi_namespace_node itself mutated or only to what it points?

#

forgive me rustc

fiery turtle Dec 25, 2024, 8:17 PM

#

The latter

rustic compass Dec 25, 2024, 8:17 PM

#

then opaque should be safe enough

#

is it reference counted?

fiery turtle Dec 25, 2024, 8:18 PM

#

Only internally

#

Aka the user never gets access to temporary nodes

rustic compass Dec 25, 2024, 8:19 PM

#

so its safe to copy around?

fiery turtle Dec 25, 2024, 8:19 PM

#

U cant copy it, its an opaque type

#

Its only ever returned as a pointer

rustic compass Dec 25, 2024, 8:21 PM

#

is the same true for uacpi_object?

fiery turtle Dec 25, 2024, 8:22 PM

#

Yes

rustic compass Dec 25, 2024, 8:22 PM

#

i see that it exposes reference counting, does that mean its allowed to be copied?

fiery turtle Dec 25, 2024, 8:24 PM

#

what do u mean by copied

#

again its a pointer

#

u can copy the pointer yes

#

stuff like uacpi_eval returns a uacpi_object* that is owned by the caller

#

the caller can unref it to delete or ref if it wants to share it

#

same goes for uacpi_object_create_*

rustic compass Dec 25, 2024, 8:25 PM

#

i mean that i would allow the opaque type wrapper of uacpi_object to be cloneable which would increase the ref count

#

basicly Object.clone() would increase the ref count

fiery turtle Dec 25, 2024, 8:26 PM

#

u mean cloning a pointer object?

#

or would that object wrap the pointer

rustic compass Dec 25, 2024, 8:27 PM

#

this is a struct which contains a uacpi_object pointer which is invisible for the user
the user would get object and is free to clone/copy it, the Object implementation would update the refcount acordingly

pub struct Object(pub(crate) *mut uacpi_sys::uacpi_object);

fiery turtle Dec 25, 2024, 8:27 PM

#

sorry no idea what that means LULW

#

like a struct that contains a pointer or what

rustic compass Dec 25, 2024, 8:29 PM

#

i pressed enter before i was finished writing

fiery turtle Dec 25, 2024, 8:29 PM

#

ah ok

#

yeah

#

in that sense u can copy it

rustic compass Dec 25, 2024, 8:31 PM

#

as namespace_node doesnt expose refcounting there would only be one instance per pointer

#

its probably a good idea that i explain the wrapper to you once im finished so that you can catch stuff that i implemented wrong

fiery turtle Dec 25, 2024, 8:33 PM

#

yeah

rustic compass Dec 25, 2024, 8:42 PM

#

is it correct that these two are identical?

UACPI_OBJECT_BUFFER_INDEX = 21,
UACPI_OBJECT_MAX_TYPE_VALUE = UACPI_OBJECT_BUFFER_INDEX

fiery turtle Dec 25, 2024, 8:43 PM

#

the last one u probably dont even need to propagate to rust

#

not very useful

rustic compass Dec 25, 2024, 8:43 PM

#

yea rust also doesnt like it if two fields in an enum have the same value

#

if its not needed i will ignore it else i would give it a different value and translate it

fiery turtle Dec 25, 2024, 8:44 PM

#

ye

rustic compass Dec 25, 2024, 8:44 PM

#

so needed or not?

fiery turtle Dec 25, 2024, 8:44 PM

#

not needed

#

only way how its used

rustic compass Dec 25, 2024, 8:45 PM

#

i bet 10 second after uacpi-rs is published someone will complain meme

fiery turtle Dec 25, 2024, 8:45 PM

#

about what?

rustic compass Dec 25, 2024, 8:45 PM

#

that i dropped that enum value

#

max_type_value

fiery turtle Dec 25, 2024, 8:46 PM

#

its a common C construct needed only for C

vale isle Dec 25, 2024, 8:58 PM

#

rustic compass that i dropped that enum value

idk how you structured uacpi-rs but usually the normal thing to do is expose everything the headers expose with a machine generated -sys crate

#

if you do that there's no way anyone would complain but i don't know what that is used for

vale isle Dec 25, 2024, 8:59 PM

#

rustic compass yea rust also doesnt like it if two fields in an enum have the same value

i would expose that as a constant

#

because that's what it really is

#

like const UACPI_OBJECT_MAX_TYPE_VALUE: Foo = Foo::ObjectBufferIndex or something like that

#

if it's needed that is, if it's not needed just expose it in the -sys crate as a constant which is what stuff like bindgen should do automatically

rustic compass Dec 25, 2024, 9:08 PM

#

vale isle idk how you structured uacpi-rs but usually the normal thing to do is expose eve...

the bindgen exposes it as a const

#

using a enum in rust groups the values nicely together

left orbit Dec 25, 2024, 9:42 PM

#

uacpi on linux when pepepray

#

i can't wait to see how linux scores

#

@fiery turtle would you be able to send a patch or your linux work?

#

unless you actually plan on finishing it

#

otherwise i wanted to look at it myself, but it looks painful to say the least

fiery turtle Dec 25, 2024, 9:57 PM

#

sure i can send it

#

why painful?

left orbit Dec 25, 2024, 9:58 PM

#

idk, maybe i'm just unfamiliar with how it's supposed to actually work

#

but the acpi_* functions are ujsed in a lot of places

fiery turtle Dec 25, 2024, 9:58 PM

#

i wasnt planning on actually integrating uacpi

#

that would take years

#

of literal time

left orbit Dec 25, 2024, 9:58 PM

#

oh lol

fiery turtle Dec 25, 2024, 9:58 PM

#

just implement kernel api and initialize it

#

alongside acpica

#

just to see the numbers

#

but yeah there's like a thousand different files using acpica API if u want to fully replace it with uacpi

left orbit Dec 25, 2024, 9:59 PM

#

yeah, i do wonder how painful would it be meme

#

it would be kinda based tho

#

mainline running with uacpi instead of acpica

fiery turtle Dec 25, 2024, 10:00 PM

#

probably a year of work

#

if i had to estimate

left orbit Dec 25, 2024, 10:00 PM

#

yeah... more than i am willing to put in lol

fiery turtle Dec 25, 2024, 10:00 PM

#

maybe less if u can make a smart script or something

left orbit Dec 25, 2024, 10:00 PM

#

by about a year

fiery turtle Dec 25, 2024, 10:00 PM

#

because a lot of apis are translatable 1 for 1

#

some are more complex but not all

#

but for getting the score i was going to just implement kernel api then call uacpi_initialize; uacpi_namespace_load

#

that would be enough

left orbit Dec 25, 2024, 10:01 PM

#

if i had to guess the "easiest" way to replace acpica would be to remove all the implementation and basically implement it with uacpi instead

fiery turtle Dec 25, 2024, 10:02 PM

#

u can mostly blindly copy acpica acpi_os_* implementation for that

left orbit Dec 25, 2024, 10:02 PM

#

but yeah just running using linux kernmel as testbench is good too!

#

i think i can do without patches in that case, unless you think that will greatly reduce the amount of work i have to do

fiery turtle Dec 25, 2024, 10:02 PM

#

maybe once its integrated into reactos and is successful + stable for a while that could be an idea to entertain

left orbit Dec 25, 2024, 10:02 PM

#

but it's not like i know linux at all so i'll be staring at the code for a while anyway

fiery turtle Dec 25, 2024, 10:02 PM

#

i mean full replacement

left orbit Dec 25, 2024, 10:03 PM

#

i wonder if that would actually be upstreamable

#

just imagine

#

booting your favorite linux distro™️

#

dmesg | grep -i uacpi

#

is uacpi compatible with acpica in terms of aml behavior?

#

i wonder if linux relies on the way acpica works, that sounds silly

fiery turtle Dec 25, 2024, 10:04 PM

#

--- a/Makefile
+++ b/Makefile
@@ -570,6 +570,7 @@ LINUXINCLUDE    := \
                -I$(objtree)/arch/$(SRCARCH)/include/generated \
                -I$(srctree)/include \
                -I$(objtree)/include \
+               -I$(srctree)/drivers/acpi/uACPI/include \
                $(USERINCLUDE)

 KBUILD_AFLAGS   := -D__ASSEMBLY__ -fno-PIE
diff --git a/drivers/acpi/Makefile b/drivers/acpi/Makefile
index 40208a0f5dfb..379f6131bd41 100644
--- a/drivers/acpi/Makefile
+++ b/drivers/acpi/Makefile
@@ -31,6 +31,8 @@ acpi-$(CONFIG_ACPI_SYSTEM_POWER_STATES_SUPPORT) += sleep.o
 acpi-y                         += device_sysfs.o device_pm.o
 acpi-$(CONFIG_ACPI_SLEEP)      += proc.o

+obj-y += uACPI/
+

Makefile in uACPI/

ccflags-y           += -DUACPI_OVERRIDE_LIBC
ccflags-y           += -DUACPI_OVERRIDE_TYPES

obj-y += glue.o source/

Makefile in uACPI/source/

ccflags-y           += -DUACPI_OVERRIDE_LIBC
ccflags-y           += -DUACPI_OVERRIDE_TYPES

obj-y += \
        default_handlers.o \
        event.o \
        interpreter.o \
        io.o \
        mutex.o \
        namespace.o \
        notify.o \
        opcodes.o \
        opregion.o \
        osi.o \
        registers.o \
        resources.o \
        shareable.o \
        sleep.o \
        stdlib.o \
        tables.o \
        types.o \
        uacpi.o \
        utilities.o

fiery turtle Dec 25, 2024, 10:05 PM

#

left orbit i wonder if linux relies on the way acpica works, that sounds silly

nah it doesnt

#

uacpi would work just fine

fiery turtle Dec 25, 2024, 10:05 PM

#

left orbit just imagine

that would be really cool ofc

fiery turtle Dec 25, 2024, 10:05 PM

#

left orbit i wonder if that would actually be upstreamable

if we can clearly explain the benefit of that and prove that it's better then definitely

#

include/uacpi_types.h:

#pragma once

#include <linux/kernel.h>

#include <uacpi/helpers.h>

typedef uint8_t uacpi_u8;
typedef uint16_t uacpi_u16;
typedef uint32_t uacpi_u32;
typedef uint64_t uacpi_u64;

typedef int8_t uacpi_i8;
typedef int16_t uacpi_i16;
typedef int32_t uacpi_i32;
typedef int64_t uacpi_i64;

#define UACPI_TRUE true
#define UACPI_FALSE false
typedef bool uacpi_bool;

#define UACPI_NULL NULL

typedef uintptr_t uacpi_uintptr;
typedef uacpi_uintptr uacpi_virt_addr;
typedef size_t uacpi_size;

typedef va_list uacpi_va_list;
#define uacpi_va_start va_start
#define uacpi_va_end va_end
#define uacpi_va_arg va_arg

typedef char uacpi_char;

#define uacpi_offsetof offsetof

/*
 * We use unsignd long long for 64-bit number formatting because 64-bit types
 * don't have a standard way to format them. The inttypes.h header is not
 * freestanding therefore it's not practical to force the user to define the
 * corresponding PRI macros. Moreover, unsignd long long  is required to be
 * at least 64-bits as per C99.
 */
UACPI_BUILD_BUG_ON_WITH_MSG(
    sizeof(unsigned long long) < 8,
    "unsigned long long must be at least 64 bits large as per C99"
);
#define UACPI_PRIu64 "llu"
#define UACPI_PRIx64 "llx"
#define UACPI_PRIX64 "llX"
#define UACPI_FMT64(val) ((unsigned long long)(val))

#

this is basically all I have

#

for uacpi_libc.h you'll wnat to just make sure everything is using the linux's super fast versions

left orbit Dec 25, 2024, 10:09 PM

#

i'll see how far i can get

#

for now i'm just compiling it with those changes and see where it errors out

fiery turtle Dec 25, 2024, 10:10 PM

#

drivers/acpi/osl.c has basically everything in terms of kernel_api.h

left orbit Dec 25, 2024, 10:10 PM

#

no glue yet ofc

#

ooo thanks

fiery turtle Dec 25, 2024, 10:10 PM

#

you just have to copy paste carefully

#

everything that starts with acpi_os_* is uacpi_kernel_ basically

#

well they have custom ones that arent actually referenced by acpica but whatever

left orbit Dec 25, 2024, 10:13 PM

#

also one quick question

#

why did you add the -D flags in both makefiles?

#

is there any reason behind this or just copy paste mistake?

fiery turtle Dec 25, 2024, 10:13 PM

#

apparently if u add it to one its not propagated to the other

#

im not sure why

left orbit Dec 25, 2024, 10:13 PM

#

oh wow that sucks

#

thanks

fiery turtle Dec 25, 2024, 10:14 PM

#

maybe it was a skill issue but i had to add to both

left orbit Dec 25, 2024, 10:40 PM

#

is uacpi event == acpica semaphore

#

ah fuck it what can go wrong meme

fiery turtle Dec 25, 2024, 10:40 PM

#

yeah it is

left orbit Dec 25, 2024, 10:40 PM

#

epic

#

i need to somehow implement mutexes

#

i wonder if i can just do

fiery turtle Dec 25, 2024, 10:41 PM

#

binary semaphore == mutex, semaphore == event

left orbit Dec 25, 2024, 10:41 PM

#

spinlock -> mutex -> semaphore

fiery turtle Dec 25, 2024, 10:41 PM

#

hm?

#

like in your os?

left orbit Dec 25, 2024, 10:41 PM

#

no like in linux

#

acpi_os implements lock in terms of semaphore

#

and a mutex can be imlemented using a semaphore too

fiery turtle Dec 25, 2024, 10:42 PM

#

linux has a mutex primitive lol just use that

left orbit Dec 25, 2024, 10:42 PM

#

oh?

#

hmm

fiery turtle Dec 25, 2024, 10:42 PM

#

static DEFINE_MUTEX(acpi_ioremap_lock);

left orbit Dec 25, 2024, 10:42 PM

#

yeah i have no idea what im doing, ill take a look

fiery turtle Dec 25, 2024, 10:42 PM

#

example

left orbit Dec 25, 2024, 10:42 PM

#

oh wow nice

#

i have never done linux kernel dev

fiery turtle Dec 25, 2024, 10:43 PM

#

same, i just know it has every locking primitive on the planet

left orbit Dec 25, 2024, 10:43 PM

#

so like i actually dont know what im doing

fiery turtle Dec 25, 2024, 10:43 PM

#

https://elixir.bootlin.com/linux/v6.12.6/source/include/linux/mutex.h

mutex.h - include/linux/mutex.h - Linux source code ...

Elixir Cross Referencer - source code of Linux v6.12.6: include/linux/mutex.h

left orbit Dec 25, 2024, 10:43 PM

#

but i hope i can make it through

fiery turtle Dec 25, 2024, 10:47 PM

#

i didnt find a mutex acquire with timeout api

#

there was a patch series in 2020

#

#

maybe u should use semaphores instead then lmao

left orbit Dec 25, 2024, 10:48 PM

#

oops

#

yeah i will

#

mutex api is weird

fiery turtle Dec 25, 2024, 10:49 PM

#

https://elixir.bootlin.com/linux/v6.12.6/source/drivers/acpi/osl.c#L1194

osl.c - drivers/acpi/osl.c - Linux source code v6.12...

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

#

this basically already is a mutex

#

because they have ifs everywhere checking that unit count is always 1

#

    if (!sem || (units < 1))
        return AE_BAD_PARAMETER;

    if (units > 1)
        return AE_SUPPORT;

#

as long as initial_units is 1 this is a mutex

left orbit Dec 25, 2024, 10:51 PM

#

can i use pthread apis in the kernel ☠️

#

uacpi_thread_id uacpi_kernel_get_thread_id(void)
{
    return (uacpi_thread_id)pthread_self();
}```

#

i was planning for that to be my get_thread_id

#

but i am afraid that might not work ugh

calm latch Dec 25, 2024, 10:52 PM

#

That's what I do

left orbit Dec 25, 2024, 10:53 PM

#

yeah but ur not linux

fiery turtle Dec 25, 2024, 10:55 PM

#

left orbit ```c uacpi_thread_id uacpi_kernel_get_thread_id(void) { return (uacpi_thread...

return current?

left orbit Dec 25, 2024, 10:56 PM

#

what the heck is a current

#

is that really a thing?

fiery turtle Dec 25, 2024, 10:56 PM

#

current task

#

yeah

#

wait lemme see

left orbit Dec 25, 2024, 10:56 PM

#

damn lol thats crazy

fiery turtle Dec 25, 2024, 10:56 PM

#

acpica has that kernel api as well

left orbit Dec 25, 2024, 10:56 PM

#

fiery turtle Dec 25, 2024, 10:57 PM

#

static inline acpi_thread_id acpi_os_get_thread_id(void)
{
    return (acpi_thread_id) (unsigned long)current;
}

left orbit Dec 25, 2024, 10:57 PM

#

☠️

fiery turtle Dec 25, 2024, 10:57 PM

#

yup

left orbit Dec 25, 2024, 10:57 PM

#

even copilot knows

hollow elm Dec 25, 2024, 10:57 PM

#

those who know

left orbit Dec 25, 2024, 11:00 PM

#

does timeout = 0 work the same for events as it does for mutexes?

slim panther Dec 25, 2024, 11:00 PM

#

cursed

left orbit Dec 25, 2024, 11:00 PM

#

the kernel api docs dont explain that

fiery turtle Dec 25, 2024, 11:00 PM

#

yeah

#

same thing

left orbit Dec 25, 2024, 11:01 PM

#

cool ty

fiery turtle Dec 25, 2024, 11:01 PM

#

0 => try_acquire

left orbit Dec 25, 2024, 11:05 PM

#

yeah, question still stands

#

i guess i can read the acpica os deps

fiery turtle Dec 25, 2024, 11:05 PM

#

they already do

left orbit Dec 25, 2024, 11:05 PM

#

well i was asking because uacpi kernel api returns the flags

#

and there's raw_spin_lock and raw_spin_lock_irqsave

#

i was wondering which ones i should use ;D

fiery turtle Dec 25, 2024, 11:06 PM

#

for whatever reason i think acpica spinlocks dont??

#

u want lock_irqsave

left orbit Dec 25, 2024, 11:06 PM

#

makes sense

#

now the "worst" kernel apis are left uacpi_kernel_handle_firmware_request uacpi_kernel_install_interrupt_handler uacpi_kernel_uninstall_interrupt_handler uacpi_kernel_schedule_work uacpi_kernel_wait_for_work_completion

#

i wonder if i should try running without those

#

i mean i don't think they are needed during namespace load

gentle peak Dec 25, 2024, 11:07 PM

#

i think of those only install_interrupt_handler is ran by uacpi_namespace_load on qemu/kvm

#

and i think it'll still work without it

fiery turtle Dec 25, 2024, 11:07 PM

#

https://github.com/torvalds/linux/commit/8e57de43076477c5cce113f2579bef02ce3e8b27

GitHub

ACPI: OSL: Use spin locks without disabling interrupts · torvalds/l...

After commit 7a36b901a6eb ("ACPI: OSL: Use a threaded interrupt handler
for SCI") any ACPICA code never runs in a hardirq handler, so it need
not dissable interrupts on the local ...

#

very interesting

left orbit Dec 25, 2024, 11:08 PM

#

also, where should i call uacpi_initialize? :^)

#

or where did you call it

gentle peak Dec 25, 2024, 11:08 PM

#

i'd do so wherever acpi_subsystem_initialize is currently being called

fiery turtle Dec 25, 2024, 11:08 PM

#

https://elixir.bootlin.com/linux/v6.12.6/source/drivers/acpi/bus.c#L1424

bus.c - drivers/acpi/bus.c - Linux source code v6.12...

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

#

here

fiery turtle Dec 25, 2024, 11:08 PM

#

gentle peak and i think it'll still work without it

yeah

#

acpica has exact same api, u could just steal osl.c for it

#

with workqueues and stuff

left orbit Dec 25, 2024, 11:09 PM

#

effort

gentle peak Dec 25, 2024, 11:09 PM

#

meanwhile i'm just trying to figure out how to get linux to compile with -march=x86-64-v2 to get rid of the popcount libgcc reference

left orbit Dec 25, 2024, 11:09 PM

#

and probably not needed meme

fiery turtle Dec 25, 2024, 11:09 PM

#

gentle peak meanwhile i'm just trying to figure out how to get linux to compile with -march=...

wait u doing uacpi/linux as well?

gentle peak Dec 25, 2024, 11:09 PM

#

trying to yeah

left orbit Dec 25, 2024, 11:09 PM

#

fiery turtle here

sorry i might sound dense, but where? lol

#

like before bus_register would be good?

fiery turtle Dec 25, 2024, 11:10 PM

#

like literally anywhere in that function at the bottom

#

this is right after acpica init

left orbit Dec 25, 2024, 11:10 PM

#

ok ill do it after the bus_register then

fiery turtle Dec 25, 2024, 11:10 PM

#

gentle peak trying to yeah

daamn

#

cool

left orbit Dec 25, 2024, 11:10 PM

#

hopefully it wont breka nything

gentle peak Dec 25, 2024, 11:10 PM

#

this is what i'm doing for init (bus.c)

#

figured i'd make acpica init fail since idk if they'd try to double init hardware

fiery turtle Dec 25, 2024, 11:11 PM

#

on qemu shouldnt matter

#

(probably)

#

i dont think QEMU's aml has global state that would interfere, but im not 100% confident

gentle peak Dec 25, 2024, 11:12 PM

#

probably not but better safe than sorry

fiery turtle Dec 25, 2024, 11:12 PM

#

yeah

gentle peak Dec 25, 2024, 11:15 PM

#

wonder how long linux compile would take with LTO

fiery turtle Dec 25, 2024, 11:15 PM

#

depends how many modules u have as -y

gentle peak Dec 25, 2024, 11:15 PM

#

defconfig

fiery turtle Dec 25, 2024, 11:15 PM

#

quite a lot probably lol

#

inb4 linux scores less than your kernel

gentle peak Dec 25, 2024, 11:16 PM

#

idk if linux parallelizes init but if it does hydrogen will probably outperform it since on there uacpi has the whole vcpu to itself

#

during init at least

fiery turtle Dec 25, 2024, 11:17 PM

#

nah, this is fairly early init as part of subsystem initcalls

#

i think its single threaded

left orbit Dec 25, 2024, 11:18 PM

#

ld: vmlinux.o: in function uacpi_popcount': (.text+0x5de955): undefined reference to __popcountdi2'

#

kms

gentle peak Dec 25, 2024, 11:18 PM

#

yeah i just edited one of the makefiles to pass -march=x86-64-v2

fiery turtle Dec 25, 2024, 11:18 PM

#

i should really get rid of that __builtin_popcount

gentle peak Dec 25, 2024, 11:19 PM

#

yeahh

fiery turtle Dec 25, 2024, 11:19 PM

#

its needed only for the aml countbits i think

gentle peak Dec 25, 2024, 11:19 PM

#

proxima isn't affected by it since it uses clang for compilation but still

fiery turtle Dec 25, 2024, 11:19 PM

#

clang ships its own implementation for it?

left orbit Dec 25, 2024, 11:19 PM

#

gentle peak yeah i just edited one of the makefiles to pass -march=x86-64-v2

lol thanks works :D

gentle peak Dec 25, 2024, 11:19 PM

#

i think clang just doesn't use a builtin

left orbit Dec 25, 2024, 11:19 PM

#

it linked! 🎉

gentle peak Dec 25, 2024, 11:20 PM

#

if popcount isn't available for the target it does it inline

fiery turtle Dec 25, 2024, 11:20 PM

#

qemu-system-x86_64 -kernel vmlinuz should work i think?

gentle peak Dec 25, 2024, 11:20 PM

#

the command i'm using for testing is make -j12 && qemu-system-x86_64 -accel kvm -cpu host -kernel arch/x86/boot/bzImage -serial stdio -append "earlyprintk=ttyS0 console=ttyS0" -display none

fiery turtle Dec 25, 2024, 11:20 PM

#

nice

#

@left orbit u did the libc overrides and stuff right?

#

to make sure it uses the linux stuff

left orbit Dec 25, 2024, 11:21 PM

#

nope lol but i just want ot make it works first

#

i forgot

fiery turtle Dec 25, 2024, 11:21 PM

#

that would probably really bump the perf

left orbit Dec 25, 2024, 11:22 PM

#

i dont see any uacpi logs

#

how did i fuck it up

fiery turtle Dec 25, 2024, 11:22 PM

#

lol

left orbit Dec 25, 2024, 11:22 PM

#

ill try to bug_on(false) in print

#

👍

fiery turtle Dec 25, 2024, 11:23 PM

#

u mean bug_on(true)?

gentle peak Dec 25, 2024, 11:23 PM

#

where is it getting this oom from, i'm logging all uacpi_kernel_alloc calls

fiery turtle Dec 25, 2024, 11:23 PM

#

mutex or spinlock

gentle peak Dec 25, 2024, 11:23 PM

#

ohh

fiery turtle Dec 25, 2024, 11:23 PM

#

or event

gentle peak Dec 25, 2024, 11:23 PM

#

right

left orbit Dec 25, 2024, 11:24 PM

#

it never gets to my prints

#

rip

fiery turtle Dec 25, 2024, 11:24 PM

#

lmaoo

#

u did bug_on(true) right?

left orbit Dec 25, 2024, 11:24 PM

#

yeah

fiery turtle Dec 25, 2024, 11:24 PM

#

because false means not a bug

left orbit Dec 25, 2024, 11:25 PM

#

ill call uacpi_init at the start of acpi_bus_init

fiery turtle Dec 25, 2024, 11:25 PM

#

or just do WARN_ON(1)

left orbit Dec 25, 2024, 11:25 PM

#

it didnt get called

fiery turtle Dec 25, 2024, 11:25 PM

#

how did u even manage to fuck that part up 💀

left orbit Dec 25, 2024, 11:25 PM

#

[ 0.426670] RSDP 0x00000000000F52B0 00000000 v00 (BOCHS )
[ 0.428618] ioremap error for 0x7fe2000-0x7fe3000, requested 0x2, got 0x0
[ 0.429566] ACPI: Unable to initialize uACPI

#

yoooo

fiery turtle Dec 25, 2024, 11:25 PM

#

nice

#

what was it?

left orbit Dec 25, 2024, 11:26 PM

#

oh i am retarded

#

result = bus_register(&acpi_bus_type);
if (!result)
    return 0;```

#

i put it

#

after the return 0

fiery turtle Dec 25, 2024, 11:26 PM

#

lmao

left orbit Dec 25, 2024, 11:26 PM

#

well no shit it never got called

#

time to figure out why my kernel api is fucked

fiery turtle Dec 25, 2024, 11:27 PM

#

maybe we should fork linux and make a branch with uacpi integration since thats pretty interesting

#

could start slowly carving away at acpica calls

left orbit Dec 25, 2024, 11:29 PM

#

that would be cool as fuck

fiery turtle Dec 25, 2024, 11:30 PM

#

yes LETSFUCKINGGOOOOOO

left orbit Dec 25, 2024, 11:30 PM

#

📎 1735169419118.txt

#

got somewhere

#

1.7m

fiery turtle Dec 25, 2024, 11:30 PM

#

thats with my shitty libc right?

left orbit Dec 25, 2024, 11:30 PM

#

i'd assume so, yes

fiery turtle Dec 25, 2024, 11:30 PM

#

nice

hollow elm Dec 25, 2024, 11:30 PM

#

oddly precise 1705000/s

fiery turtle Dec 25, 2024, 11:30 PM

#

exactly 1ms i geuss

#

or is the timer resolution just garbage

left orbit Dec 25, 2024, 11:31 PM

#

i copied that from acpica

#

uacpi_u64 uacpi_kernel_get_nanoseconds_since_boot(void)
{
    return (get_jiffies_64() - INITIAL_JIFFIES) * (ACPI_NSEC_PER_SEC / HZ);
}```

fiery turtle Dec 25, 2024, 11:31 PM

#

looks correct

#

how did u manage to invoke an invalid opcode error in spinlock_free LULW

slim panther Dec 25, 2024, 11:34 PM

#

i just realized the fn is called "since boot"

left orbit Dec 25, 2024, 11:34 PM

#

because my BUG_ON was inverted lmao

slim panther Dec 25, 2024, 11:34 PM

#

is it fine to just pass any timer

left orbit Dec 25, 2024, 11:34 PM

#

i did BUG_ON(!locked) in spinlock_free

fiery turtle Dec 25, 2024, 11:34 PM

#

slim panther is it fine to just pass any timer

no, uacpi will find out it's actually not since boot but since earlier boot and come to your house

left orbit Dec 25, 2024, 11:34 PM

#

[    0.436317] RSDP 0x00000000000F52C0 00000000 v00 (BOCHS )
[    0.438221] RSDT 0x0000000007FE22FC 00000038 v01 (BOCHS  BXPC    )
[    0.440229] DSDT 0x0000000007FE0040 000020B4 v01 (BOCHS  BXPC    )
[    0.441216] FACP 0x0000000007FE20F4 000000F4 v03 (BOCHS  BXPC    )
[    0.444218] APIC 0x0000000007FE21E8 00000078 v03 (BOCHS  BXPC    )
[    0.446232] HPET 0x0000000007FE2260 00000038 v01 (BOCHS  BXPC    )
[    0.448240] MCFG 0x0000000007FE2298 0000003C v01 (BOCHS  BXPC    )
[    0.450249] WAET 0x0000000007FE22D4 00000028 v01 (BOCHS  BXPC    )
[    0.452570] FACS 0x0000000007FE0000 00000040
[    0.455686] successfully loaded 1 AML blob, 1705 ops in 2ms (avg 852500/s)
[    0.456373] unable to install SCI interrupt handler: unimplemented
[    0.457216] event initialization failed: unimplemented
[    0.461014] ACPI: Unable to load uACPI namespace```

#

damn, it's slower with linux ones

fiery turtle Dec 25, 2024, 11:34 PM

#

huh what lol

left orbit Dec 25, 2024, 11:35 PM

#

#pragma once

#include <asm/string.h>

#define uacpi_memcpy memcpy
#define uacpi_memset memset
#define uacpi_memcmp memcmp
#define uacpi_strcmp strcmp
#define uacpi_memmove memmove
#define uacpi_strlen strlen
#define uacpi_snprintf snprintf
#define uacpi_vsnprintf vsnprintf

#

i did that

#

basically

fiery turtle Dec 25, 2024, 11:35 PM

#

wtf

#

how is that possible LULW

#

my impl is literally crap

left orbit Dec 25, 2024, 11:35 PM

#

great question

fiery turtle Dec 25, 2024, 11:36 PM

#

or that timer is bogus

#

but looks correct to me idk

left orbit Dec 25, 2024, 11:36 PM

#

ok yeah that timer is shit

#

[ 0.454548] successfully loaded 1 AML blob, 1705 ops in 0ms (avg 1705000000000/s)

#

fastest uacpi implementation

fiery turtle Dec 25, 2024, 11:37 PM

#

what did u change?

left orbit Dec 25, 2024, 11:37 PM

#

nothing

#

i just reran it

slim panther Dec 25, 2024, 11:37 PM

#

fiery turtle what did u change?

lies

left orbit Dec 25, 2024, 11:37 PM

#

¯_(ツ)_/¯

fiery turtle Dec 25, 2024, 11:37 PM

#

ah yeah its crap

#

i guess

left orbit Dec 25, 2024, 11:37 PM

#

[ 0.458990] successfully loaded 1 AML blob, 1705 ops in 2ms (avg 852500/s)
[ 0.450287] successfully loaded 1 AML blob, 1705 ops in 3ms (avg 568333/s)
[ 0.452721] successfully loaded 1 AML blob, 1705 ops in 1ms (avg 1705000/s)

#

three consecutive runs

#

makes no sense for there to be so much variation

fiery turtle Dec 25, 2024, 11:38 PM

#

yeah not to mention its slow as shit

left orbit Dec 25, 2024, 11:38 PM

#

i mean

#

honestly i didnt expect it to be particularly fast

winter orbit Dec 25, 2024, 11:39 PM

#

fiery turtle maybe we should fork linux and make a branch with uacpi integration since thats ...

uACPI the new reference ACPI implementation

fiery turtle Dec 25, 2024, 11:39 PM

#

i kinda did because their mem* stuff is very fast

left orbit Dec 25, 2024, 11:39 PM

#

i dont think the speed of linux comes from it being really well optimized and shit

#

maybe it matters more in a hobby os? idk

slim panther Dec 25, 2024, 11:39 PM

#

left orbit [ 0.458990] successfully loaded 1 AML blob, 1705 ops in 2ms (avg 852500/s) [ ...

damn linux slower than menix

fiery turtle Dec 25, 2024, 11:39 PM

#

if u look at the leaderboard thats like menix speed lol

calm latch Dec 25, 2024, 11:39 PM

#

fiery turtle ah yeah its crap

Can I claim I have better timers than Linux? sunflush

slim panther Dec 25, 2024, 11:39 PM

#

fiery turtle if u look at the leaderboard thats like menix speed lol

KEKW

left orbit Dec 25, 2024, 11:39 PM

#

left orbit [ 0.454548] successfully loaded 1 AML blob, 1705 ops in 0ms (avg 170500000000...

aboslutely not slower

fiery turtle Dec 25, 2024, 11:40 PM

#

maybe monkuous will have better results

left orbit Dec 25, 2024, 11:40 PM

#

linux is the fastest out there

fiery turtle Dec 25, 2024, 11:40 PM

#

with that magic touch

slim panther Dec 25, 2024, 11:40 PM

#

maybe menix slab is the same as the linux impl

fiery turtle Dec 25, 2024, 11:40 PM

#

left orbit aboslutely not slower

this just means get_nanoseconds_since_boot returned the exact same value twice

left orbit Dec 25, 2024, 11:40 PM

#

ikik

#

i can tell lmao

fiery turtle Dec 25, 2024, 11:40 PM

#

lmao

gentle peak Dec 25, 2024, 11:42 PM

#

fiery turtle maybe monkuous will have better results

i'm trying to fix get_nanoseconds_since_boot since for whatever reason ktime_get_ns is returning 0

#

and so is get_jiffies_64

fiery turtle Dec 25, 2024, 11:42 PM

#

ohh

#

like the fact that overriding libc made no difference does sound sketchy

gentle peak Dec 25, 2024, 11:42 PM

#

i know the kernel has a time reference at time of init already because printk is printing time

fiery turtle Dec 25, 2024, 11:43 PM

#

yeah

left orbit Dec 25, 2024, 11:43 PM

#

ill try ktime_get

#

oh

#

well ill try anyway meme

gentle peak Dec 25, 2024, 11:43 PM

#

yknow what i'm just gonna return __builtin_ia32_rdtsc()

fiery turtle Dec 25, 2024, 11:43 PM

#

lool

left orbit Dec 25, 2024, 11:43 PM

#

same shit

fiery turtle Dec 25, 2024, 11:43 PM

#

are u sure 1 tick == 1 ns?

left orbit Dec 25, 2024, 11:44 PM

#

as inaccurate as get_jiffies_64

gentle peak Dec 25, 2024, 11:44 PM

#

fiery turtle are u sure 1 tick == 1 ns?

i can tell qemu to make it so

fiery turtle Dec 25, 2024, 11:44 PM

#

ohh

#

i wonder if kvm clock is still not initialized at this point

gentle peak Dec 25, 2024, 11:44 PM

#

that might be it

fiery turtle Dec 25, 2024, 11:44 PM

#

hmm

gentle peak Dec 25, 2024, 11:44 PM

#

although i wonder what magical time api printk is using then

#

anyway: [ 0.070157] uacpi: 3: successfully loaded 1 AML blob, 1705 ops in 1ms (avg 1512953/s)

fiery turtle Dec 25, 2024, 11:45 PM

#

thats with libc overrides right

gentle peak Dec 25, 2024, 11:45 PM

#

yeah

fiery turtle Dec 25, 2024, 11:45 PM

#

L

gentle peak Dec 25, 2024, 11:45 PM

#

#pragma once

#define memcmp __builtin_memcmp
#define memcpy __builtin_memcpy
#define memmove __builtin_memmove
#define memset __builtin_memset

#define strcmp __builtin_strcmp
#define strlen __builtin_strlen
#define strnlen __builtin_strnlen

fiery turtle Dec 25, 2024, 11:45 PM

#

menix > linux

left orbit Dec 25, 2024, 11:45 PM

#

kvm clock is initialized at that point

#

[ 0.000000] kvm-clock: Using msrs 4b564d01 and 4b564d00
[ 0.000002] kvm-clock: using sched offset of 435683178 cycles
[ 0.001871] clocksource: kvm-clock: mask: 0xffffffffffffffff max_cycles: 0x1cd42e4dffb, max_idle_ns: 881590591483 ns

fiery turtle Dec 25, 2024, 11:45 PM

#

gentle peak ```c #pragma once #define memcmp __builtin_memcmp #define memcpy __builtin_memc...

could also do vsnprintf etc

gentle peak Dec 25, 2024, 11:45 PM

#

wasn't sure whether the linux versions were fully conforming

fiery turtle Dec 25, 2024, 11:45 PM

#

they are i think

#

with some extensions

#

is it possible that override libc is not propagated everywhere?

left orbit Dec 25, 2024, 11:46 PM

#

it is

fiery turtle Dec 25, 2024, 11:46 PM

#

like the define

#

beacuse i had issues with that

left orbit Dec 25, 2024, 11:46 PM

#

adding the c flag in the makefile should be enough

fiery turtle Dec 25, 2024, 11:47 PM

#

should be

fiery turtle Dec 25, 2024, 11:47 PM

#

gentle peak yeah

and if u dont do that?

gentle peak Dec 25, 2024, 11:47 PM

#

oh wait i didn't prefix the things with uacpi_

fiery turtle Dec 25, 2024, 11:47 PM

#

lol

gentle peak Dec 25, 2024, 11:47 PM

#

brain not working full speed today

fiery turtle Dec 25, 2024, 11:48 PM

#

1.5 with naive memcpy is fair

gentle peak Dec 25, 2024, 11:48 PM

#

[ 0.067545] uacpi: 3: successfully loaded 1 AML blob, 1705 ops in 0ms (avg 3104283/s)

#

that seems more reasonable

fiery turtle Dec 25, 2024, 11:48 PM

#

indeed

gentle peak Dec 25, 2024, 11:48 PM

#

i also wonder whether telling qemu to override the tsc frequency degrades speed

fiery turtle Dec 25, 2024, 11:48 PM

#

right behind managarm

fiery turtle Dec 25, 2024, 11:49 PM

#

gentle peak i also wonder whether telling qemu to override the tsc frequency degrades speed

probably

gentle peak Dec 25, 2024, 11:49 PM

#

so i am going to get the real tsc freq from proxima and hardcode that instead

left orbit Dec 25, 2024, 11:49 PM

#

damn that rdtsc is nice

gentle peak Dec 25, 2024, 11:49 PM

#

3792.961870 MHz

fiery turtle Dec 25, 2024, 11:49 PM

#

do u also build with LTO?

gentle peak Dec 25, 2024, 11:49 PM

#

not yet

#

i do with proxima

fiery turtle Dec 25, 2024, 11:49 PM

#

ah

#

that might give it the final bump LULW

gentle peak Dec 25, 2024, 11:50 PM

#

probably but i figured i wouldn't wait >1h for it to compile before making sure the kernel api works

fiery turtle Dec 25, 2024, 11:50 PM

#

fair fair

gentle peak Dec 25, 2024, 11:50 PM

#

actually i might try and figure out how much i need to enable starting from tinyconfig to get the output

fiery turtle Dec 25, 2024, 11:50 PM

#

what is that?

gentle peak Dec 25, 2024, 11:51 PM

#

tinyconfig is a preset that's just the minimum to get a working kernel

fiery turtle Dec 25, 2024, 11:51 PM

#

ohh

#

i wonder if a lot of other allocations including acpica are fragmenting the heap causing degraded perf

left orbit Dec 25, 2024, 11:51 PM

#

pretty sure linux does lto with default config

#

when i did BUG_ON in print i got a lot of noreturn warnings

gentle peak Dec 25, 2024, 11:52 PM

#

fiery turtle i wonder if a lot of other allocations including acpica are fragmenting the heap...

i commented out acpica init

#

so i doubt it

fiery turtle Dec 25, 2024, 11:52 PM

#

maybe other subsystems as well ig

gentle peak Dec 25, 2024, 11:52 PM

#

with hardcoded tsc frequency so that qemu can just pass through rdtsc to the host cpu:
[ 0.065107] uacpi: 3: successfully loaded 1 AML blob, 1705 ops in 0ms (avg 3120527/s)

#

didn't make a huge difference

fiery turtle Dec 25, 2024, 11:52 PM

#

yeah i guess its cheap then

gentle peak Dec 25, 2024, 11:52 PM

#

makes sense considering there were a grand total of 2 calls

fiery turtle Dec 25, 2024, 11:52 PM

#

lol

gentle peak Dec 25, 2024, 11:53 PM

#

tinyconfig time

fiery turtle Dec 25, 2024, 11:54 PM

#

left orbit pretty sure linux does lto with default config

dont break my hopes Sadge

gentle peak Dec 25, 2024, 11:54 PM

#

left orbit pretty sure linux does lto with default config

i really doubt it considering LTO is horribly slow with gcc

#

and linking didn't take >1h

#

...huh

#

with tinyconfig there's a significant performance regression

#

uacpi: 3: successfully loaded 1 AML blob, 1705 ops in 0ms (avg 2374152/s)

#

wonder what that's caused by

fiery turtle Dec 25, 2024, 11:55 PM

#

btw you might also be building with unconditional memory zeroing and stuff

#

or sanitizers

gentle peak Dec 25, 2024, 11:55 PM

#

maybe

fiery turtle Dec 25, 2024, 11:55 PM

#

a lot of stuff that could be on by default

#

i know ubsan and unconditional zeroing is enabled for ubuntus

gentle peak Dec 25, 2024, 11:56 PM

#

i really have no idea where to look in menuconfig for those things

fiery turtle Dec 25, 2024, 11:56 PM

#

ubsan is just CONFIG_UBSAN_*

#

zeroing is somewhere in the security section

gentle peak Dec 25, 2024, 11:56 PM

#

ah this is enabled by default with tinyconfig

fiery turtle Dec 25, 2024, 11:56 PM

#

dont remember the exact option

#

ah the crappy slab

left orbit Dec 25, 2024, 11:57 PM

#

just use the arch config

#

its pretty good minus the terrible amount of drivers you have to build

gentle peak Dec 25, 2024, 11:57 PM

#

gentle peak ah this is enabled by default with tinyconfig

didn't have a significant impact

left orbit Dec 25, 2024, 11:57 PM

#

but u can turn those off

#

wtf

gentle peak Dec 25, 2024, 11:57 PM

#

left orbit its pretty good minus the terrible amount of drivers you have to build

yeah that's the problem, i'm trying to minimize the amount of stuff to build so that i can try LTO

fiery turtle Dec 25, 2024, 11:57 PM

#

uacpi is not allocator bound proved meme

gentle peak Dec 25, 2024, 11:58 PM

#

ok so that's not it

#

stack vars were zero-inited by default i imagine that had an impact

#

but none of the other zero-init stuff was enabled

fiery turtle Dec 25, 2024, 11:59 PM

#

config INIT_ON_ALLOC_DEFAULT_ON
    bool "Enable heap memory zeroing on allocation by default"
    depends on !KMSAN
    help
      This has the effect of setting "init_on_alloc=1" on the kernel
      command line. This can be disabled with "init_on_alloc=0".
      When "init_on_alloc" is enabled, all page allocator and slab
      allocator memory will be zeroed when allocated, eliminating
      many kinds of "uninitialized heap memory" flaws, especially
      heap content exposures. The performance impact varies by
      workload, but most cases see <1% impact. Some synthetic
      workloads have measured as high as 7%.

#

this one maybe

gentle peak Dec 25, 2024, 11:59 PM

#

yeah that one's disabled

gentle peak Dec 25, 2024, 11:59 PM

#

gentle peak stack vars were zero-inited by default i imagine that had an impact

disabling this got it to 2.7M

fiery turtle Dec 25, 2024, 11:59 PM

#

makes sense

left orbit Dec 26, 2024, 12:00 AM

#

im compiling with -O3 -march=native -pipe with clang thinlto

gentle peak Dec 26, 2024, 12:00 AM

#

i really have no idea what the rest could be since the only kernel api functions i had to implement are incredibly basic

left orbit Dec 26, 2024, 12:00 AM

#

lets see how bad it is

gentle peak Dec 26, 2024, 12:00 AM

#

oh god does linux even work with -O3?

fiery turtle Dec 26, 2024, 12:00 AM

#

does linux even support O3

#

i dont think so

left orbit Dec 26, 2024, 12:00 AM

#

@gentle peak can i have the command line you're using to do rdtsc?

#

we're about to fdind out meme

gentle peak Dec 26, 2024, 12:00 AM

#

-cpu host,tsc-frequency=1000000000

left orbit Dec 26, 2024, 12:00 AM

#

god my pc is dying

fiery turtle Dec 26, 2024, 12:00 AM

#

linux doesnt work with anything other than O2

left orbit Dec 26, 2024, 12:00 AM

#

thanks'

median crest Dec 26, 2024, 12:01 AM

#

fiery turtle linux doesnt work with anything other than O2

Astral better than linux astral number one best os

fiery turtle Dec 26, 2024, 12:01 AM

#

welllll

#

if u look at the leaderboard...

gentle peak Dec 26, 2024, 12:02 AM

#

yknow what i'm just gonna remove stuff from defconfig instead of adding to tinyconfig

fiery turtle Dec 26, 2024, 12:02 AM

#

fair

slim panther Dec 26, 2024, 12:02 AM

#

fiery turtle linux doesnt work with anything other than O2

lmao

median crest Dec 26, 2024, 12:03 AM

#

fiery turtle if u look at the leaderboard...

Soon

left orbit Dec 26, 2024, 12:03 AM

#

any bets on how long it'll take to link with lld? :^)

#

thinlto

#

idk waht that means but

#

it definitely is doing something

#

judging by it hogging up an entire code for whatever it's doing

#

mining crypto or whatever

gentle peak Dec 26, 2024, 12:04 AM

#

oh god the final stage of lto is single threaded?

left orbit Dec 26, 2024, 12:04 AM

#

i think so lmao

fiery turtle Dec 26, 2024, 12:04 AM

#

median crest Soon

i believe 🙏

left orbit Dec 26, 2024, 12:05 AM

#

it went full blast for like 20 seconds and now its stuck on a single thread

#

IT DID IT

gentle peak Dec 26, 2024, 12:05 AM

#

fiery turtle if u look at the leaderboard...

oh btw proxima was up to a consistent 5.0-5.2M/s at some point but it's now around 4.8-4.9M/s

left orbit Dec 26, 2024, 12:05 AM

#

link takes about 2 mintues meme

#

[    0.360596] RSDP 0x00000000000F52C0 00000000 v00 (BOCHS )
[    0.362528] RSDT 0x0000000007FE22FC 00000038 v01 (BOCHS  BXPC    )
[    0.364499] DSDT 0x0000000007FE0040 000020B4 v01 (BOCHS  BXPC    )
[    0.365495] FACP 0x0000000007FE20F4 000000F4 v03 (BOCHS  BXPC    )
[    0.369574] APIC 0x0000000007FE21E8 00000078 v03 (BOCHS  BXPC    )
[    0.370500] HPET 0x0000000007FE2260 00000038 v01 (BOCHS  BXPC    )
[    0.372585] MCFG 0x0000000007FE2298 0000003C v01 (BOCHS  BXPC    )
[    0.374644] WAET 0x0000000007FE22D4 00000028 v01 (BOCHS  BXPC    )
[    0.375497] FACS 0x0000000007FE0000 00000040
[    0.378258] successfully loaded 1 AML blob, 1705 ops in 0ms (avg 2857324/s)
[    0.380480] unable to install SCI interrupt handler: unimplemented
[    0.380496] event initialization failed: unimplemented
[    0.383523] ACPI: Unable to load uACPI namespace```

#

meh that's shit

fiery turtle Dec 26, 2024, 12:06 AM

#

gentle peak oh btw proxima was up to a consistent 5.0-5.2M/s at some point but it's now arou...

daamn, what happened tho?

gentle peak Dec 26, 2024, 12:06 AM

#

really not sure

fiery turtle Dec 26, 2024, 12:06 AM

#

left orbit ``` [ 0.360596] RSDP 0x00000000000F52C0 00000000 v00 (BOCHS ) [ 0.362528] ...

well u were at 1.5 before right

left orbit Dec 26, 2024, 12:06 AM

#

also , it works with -O3? :^)

gentle peak Dec 26, 2024, 12:06 AM

#

i didn't pay attention to perf for a while because i use tcg while developing

#

and like 20 commits and a few uacpi updates later i checked again

#

don't know what exactly caused the regression

left orbit Dec 26, 2024, 12:07 AM

#

📎 bzImage

#

if anybody wants to give the image a go

#

qemu-system-x86_64 -M q35 -kernel arch/x86/boot/bzImage -enable-kvm -serial stdio -append "earlyprintk=ttyS0 console=ttyS0" -cpu host,tsc-frequency=1000000000

fiery turtle Dec 26, 2024, 12:07 AM

#

gentle peak don't know what exactly caused the regression

perhaps just me adding missing locking? meme

fiery turtle Dec 26, 2024, 12:08 AM

#

left orbit also , it works with -O3? :^)

big

gentle peak Dec 26, 2024, 12:08 AM

#

left orbit

i'll test later but how do you get the kernel to compile with lto and -o3?

#

wanna test with my kernel api impl

left orbit Dec 26, 2024, 12:08 AM

#

make LLVM=1 KCFLAGS="-O3 -march=native -pipe" -j26

fiery turtle Dec 26, 2024, 12:08 AM

#

left orbit if anybody wants to give the image a go

qemu-system-x86_64: warning: TSC frequency mismatch between VM (1000000 kHz) and host (3600010 kHz), and TSC scaling unavailable
qemu-system-x86_64: kvm_init_vcpu: kvm_arch_init_vcpu failed (0): Operation not supported

left orbit Dec 26, 2024, 12:08 AM

#

also make LLVM=1 menuconfig

#

then go into general architectue blah blah blah

#

and enable lto

#

i dont think it can do lto with gcc, there were no options in menuconfig without LLVM=1

#

so you'll have to recompile with llvm

fiery turtle Dec 26, 2024, 12:09 AM

#

WSL2 loses again i guess

left orbit Dec 26, 2024, 12:09 AM

#

oh, im on wsl2

fiery turtle Dec 26, 2024, 12:10 AM

#

hmm then idk lol

left orbit Dec 26, 2024, 12:10 AM

#

median crest Dec 26, 2024, 12:10 AM

#

Its some cpu specific bullshit

gentle peak Dec 26, 2024, 12:10 AM

#

i'm gonna use -j20 even though i only have 16 lcpus wish me luck

fiery turtle Dec 26, 2024, 12:10 AM

#

im on i9 9900K

median crest Dec 26, 2024, 12:10 AM

#

Hyperv likes some cpus and dislikes others

left orbit Dec 26, 2024, 12:10 AM

#

i do -j26 and i have 24 threads

#

should i try linux zen config?

#

or arch maybe

gentle peak Dec 26, 2024, 12:11 AM

#

oh great this isn't valid on clang

left orbit Dec 26, 2024, 12:11 AM

#

-fno-builtin?

gentle peak Dec 26, 2024, 12:12 AM

#

-fno-builtin just means that it doesn't translate x() to __builtin_x()

left orbit Dec 26, 2024, 12:12 AM

#

oh i see

gentle peak Dec 26, 2024, 12:13 AM

#

that linked really quickly wtf is it even using lto

#

it's now 3.5M/s vs 3.2M/s so maybe?

#

well whatever

#

i'll test your image

fiery turtle Dec 26, 2024, 12:14 AM

#

seems fair

#

btw does the score change if u change -smp?

gentle peak Dec 26, 2024, 12:15 AM

#

left orbit

this image is only 600k-800k on my machine somehow

fiery turtle Dec 26, 2024, 12:15 AM

#

magic mokuous touch is real

#

it boosts your points by 3mil at least

gentle peak Dec 26, 2024, 12:16 AM

#

fiery turtle btw does the score change if u change `-smp`?

if it does it's negligible

#

-smp 4 is still in the 3.4-3.6M/s range

fiery turtle Dec 26, 2024, 12:16 AM

#

i guess doesnt matter

north holly Dec 26, 2024, 12:16 AM

#

fiery turtle it boosts your points by 3mil at least

I need proof

fiery turtle Dec 26, 2024, 12:17 AM

#

would be fun to do isolcpus=1,2,3,4 and then pin qemu vcpus there

#

see how perf changes with exclusive vcpu pinning

gentle peak Dec 26, 2024, 12:17 AM

#

i wonder...

left orbit Dec 26, 2024, 12:17 AM

#

gentle peak this image is only 600k-800k on my machine somehow

tf

#

oh god turning lto off is a full rebuild?

#

wtf

fiery turtle Dec 26, 2024, 12:18 AM

#

gentle peak i wonder...

but then again you didnt measure proxima that way and got 5m anyway

fiery turtle Dec 26, 2024, 12:18 AM

#

left orbit oh god turning lto off is a full rebuild?

because object files are llvm bytecode for LTO

slim panther Dec 26, 2024, 12:18 AM

#

left orbit

[ 0.114300] successfully loaded 1 AML blob, 1705 ops in 2ms (avg 783966/s)

#

lol

slim panther Dec 26, 2024, 12:18 AM

#

fiery turtle i guess doesnt matter

actually, smp on my end is super influential

fiery turtle Dec 26, 2024, 12:18 AM

#

slim panther `[ 0.114300] successfully loaded 1 AML blob, 1705 ops in 2ms (avg 783966/s)`

pmos speeds

gentle peak Dec 26, 2024, 12:19 AM

#

fiery turtle but then again you didnt measure proxima that way and got 5m anyway

oh no i wasn't wondering about that, what i was wondering about is that proxima's perf actually slightly degrades when i'm using -accel kvm -cpu host,migratable=off because it can then get the tsc frequency from cpuid, skip calibration and get to uacpi init so quickly that linux hasn't increased the cpu freq of whatever it's running on or whatever

slim panther Dec 26, 2024, 12:19 AM

#

-smp 1 yields 1.8mil on menix, but -smp 2 yields 700k

gentle peak Dec 26, 2024, 12:19 AM

#

whereas if i just use normal -accel kvm -cpu host it calibrates for 500ms during which, from linux's perspective, qemu's utilizing 100% cpu

north holly Dec 26, 2024, 12:19 AM

#

fiery turtle pmos speeds

better than obos speeds

#

which tbf without smp enabled are a lot faster than what they were before

gentle peak Dec 26, 2024, 12:20 AM

#

i was wondering whether something similar could be applied to linux here

fiery turtle Dec 26, 2024, 12:20 AM

#

gentle peak whereas if i just use normal `-accel kvm -cpu host` it calibrates for 500ms duri...

ohhh

#

yeah

slim panther Dec 26, 2024, 12:20 AM

#

okay this is weird

#

[ 0.116306] successfully loaded 1 AML blob, 1705 ops in 0ms (avg 2601524/s)

#

rerun

fiery turtle Dec 26, 2024, 12:20 AM

#

are u on a laptop?

gentle peak Dec 26, 2024, 12:20 AM

#

nah, ryzen 5800x

fiery turtle Dec 26, 2024, 12:20 AM

#

i mean marvin

gentle peak Dec 26, 2024, 12:20 AM

#

oh

slim panther Dec 26, 2024, 12:20 AM

#

7700X desktop

fiery turtle Dec 26, 2024, 12:20 AM

#

because laptops do this scaling thing

#

maybe desktops too lmao idk

slim panther Dec 26, 2024, 12:21 AM

#

it's only on the first ever boot up

gentle peak Dec 26, 2024, 12:21 AM

#

let's see

slim panther Dec 26, 2024, 12:21 AM

#

[ 0.115882] successfully loaded 1 AML blob, 1705 ops in 0ms (avg 2621600/s)

#

yea seems stable

#

2.6m

fiery turtle Dec 26, 2024, 12:22 AM

#

gentle peak let's see

lmao

#

secret uacpi benchmarking tricks they dont tell u about

slim panther Dec 26, 2024, 12:22 AM

#

gentle peak let's see

what is blessing my eyes rn

fiery turtle Dec 26, 2024, 12:23 AM

#

cpu warming up before the benchmark

gentle peak Dec 26, 2024, 12:23 AM

#

okay i'm pretty sure this is getting optimized away even though the asm is volatile

#

either that or my threshold is wrong

fiery turtle Dec 26, 2024, 12:24 AM

#

::: "memory"?

gentle peak Dec 26, 2024, 12:24 AM

#

multiplying the threshold by 10 first because there's no way i'm missing a 5s delay

#

yeah ok i just wasn't noticing it

#

didn't have a noticeable effect on uacpi perf though, damn

fiery turtle Dec 26, 2024, 12:25 AM

#

damn

gentle peak Dec 26, 2024, 12:27 AM

#

for reference: proxima when doing tsc calibration (aka 500ms of 100% cpu usage) gets 4.8-4.9M/s. when not doing tsc calibration, it gets 3.7-3.8M/s

fiery turtle Dec 26, 2024, 12:27 AM

#

very curious

#

i wonder if other projects are affected by this LULW

left orbit Dec 26, 2024, 12:27 AM

#

@gentle peak do you do irq handlers?

gentle peak Dec 26, 2024, 12:27 AM

#

left orbit <@938223886439436320> do you do irq handlers?

in the linux kernel api? no, but i print whenever the function gets called, and it isn't called until after the benchmark

#

in proxima yeah

left orbit Dec 26, 2024, 12:28 AM

#

i was curious about your linux api

gentle peak Dec 26, 2024, 12:28 AM

#

https://hst.sh/opewupavaw.cpp

fiery turtle Dec 26, 2024, 12:28 AM

#

your malloc is so fucking simple lol

#

i guess struct page helps

gentle peak Dec 26, 2024, 12:28 AM

#

fiery turtle your malloc is so fucking simple lol

the key to improving performance is to just do less

#

it's simple, and it's enough for my purposes

fiery turtle Dec 26, 2024, 12:29 AM

#

yeah

#

i wonder how it scales tho

#

im definitely stealing this when i get back to actual osdev

gentle peak Dec 26, 2024, 12:29 AM

#

like, smp scaling? probably similar to any other allocator that doesn't have magazines

fiery turtle Dec 26, 2024, 12:29 AM

#

or even threads, since it doesnt have lockless fast paths etc

gentle peak Dec 26, 2024, 12:30 AM

#

true but it doesn't do much work while holding a lock

fiery turtle Dec 26, 2024, 12:30 AM

#

yeah

gentle peak Dec 26, 2024, 12:30 AM

#

iirc lockless stuff tends to be slower on single-cpu machines

fiery turtle Dec 26, 2024, 12:30 AM

#

why does it need a mutex there, wouldnt a spinlock be enough?

median crest Dec 26, 2024, 12:30 AM

#

fiery turtle i wonder if other projects are affected by this <:LULW:543420553021882371>

astral uses kvm clock by default sooo meme

gentle peak Dec 26, 2024, 12:30 AM

#

fiery turtle why does it need a mutex there, wouldnt a spinlock be enough?

mutexes are faster than spinlocks

#

so if it doesn't need a spinlock it gets a mutex

fiery turtle Dec 26, 2024, 12:31 AM

#

hmm

#

wouldnt a spinlock that doesnt disable irqs be faster

gentle peak Dec 26, 2024, 12:31 AM

#

same speed actually

fiery turtle Dec 26, 2024, 12:31 AM

#

or are we talking uncontended cases

median crest Dec 26, 2024, 12:31 AM

#

same speed yeah

#

its all a CAS at the end

gentle peak Dec 26, 2024, 12:31 AM

#

spinlock without disabling irqs is 1 CAS, mutex is 1 CAS

#

if it's contended any kind of spinlock is usually slower especially on single cpu machines

fiery turtle Dec 26, 2024, 12:31 AM

#

i guess spinlock would be even slower

#

because it cant reschedule when it waits

gentle peak Dec 26, 2024, 12:32 AM

#

yeah

fiery turtle Dec 26, 2024, 12:32 AM

#

whereas for mutexes it would just block the thread

gentle peak Dec 26, 2024, 12:32 AM

#

a spinlock that does sched_yield instead of cpu_relax would be somewhere in the middle i believe

#

but still worse than a normal mutex

fiery turtle Dec 26, 2024, 12:32 AM

#

makes sense

#

yeah im definitely stealing a lot of fundamentals from your os

#

meme

gentle peak Dec 26, 2024, 12:32 AM

#

no problem, that's why it's MIT licensed

fiery turtle Dec 26, 2024, 12:32 AM

#

it's mit too which is nice

#

yeah

median crest Dec 26, 2024, 12:33 AM

#

median crest its all a CAS at the end

which is why my mutex sucks, its based out of the semaphore 💀

#

and why I am gonna use something else soon enough

fiery turtle Dec 26, 2024, 12:33 AM

#

any plans to make it numa aware?

gentle peak Dec 26, 2024, 12:33 AM

#

not really

left orbit Dec 26, 2024, 12:33 AM

#

[    0.926961] RSDP 0x00000000000F52C0 00000000 v00 (BOCHS )
[    0.927269] RSDT 0x0000000007FE22FC 00000038 v01 (BOCHS  BXPC    )
[    0.928269] DSDT 0x0000000007FE0040 000020B4 v01 (BOCHS  BXPC    )
[    0.929272] FACP 0x0000000007FE20F4 000000F4 v03 (BOCHS  BXPC    )
[    0.930273] APIC 0x0000000007FE21E8 00000078 v03 (BOCHS  BXPC    )
[    0.931267] HPET 0x0000000007FE2260 00000038 v01 (BOCHS  BXPC    )
[    0.932269] MCFG 0x0000000007FE2298 0000003C v01 (BOCHS  BXPC    )
[    0.933266] WAET 0x0000000007FE22D4 00000028 v01 (BOCHS  BXPC    )
[    0.934288] FACS 0x0000000007FE0000 00000040
[    0.936003] successfully loaded 1 AML blob, 1705 ops in 0ms (avg 2834533/s)```

gentle peak Dec 26, 2024, 12:33 AM

#

at least not in this rewrite

left orbit Dec 26, 2024, 12:33 AM

#

final score, i am not going to try anymore

fiery turtle Dec 26, 2024, 12:33 AM

#

close enough

#

got the image?

left orbit Dec 26, 2024, 12:34 AM

#

me?

fiery turtle Dec 26, 2024, 12:34 AM

#

yeah

left orbit Dec 26, 2024, 12:34 AM

#

📎 bzImage

gentle peak Dec 26, 2024, 12:34 AM

#

fiery turtle any plans to make it numa aware?

this rewrite is basically meant to just be "get me a simple kernel that lets me develop userspace, won't even support smp, i'll switch it out for a proper one later"

left orbit Dec 26, 2024, 12:34 AM

#

it does the funny rdtsc warmup

fiery turtle Dec 26, 2024, 12:34 AM

#

maybe it would be faster on monkuous' cpu

left orbit Dec 26, 2024, 12:34 AM

#

idk if it actually makes any difference

gentle peak Dec 26, 2024, 12:34 AM

#

left orbit idk if it actually makes any difference

seems like it doesn't because linux apparently does enough work pre-acpi init that it effectively does the same thing as warmup

left orbit Dec 26, 2024, 12:34 AM

#

https://gist.github.com/48cf/32736b0a6e01ec7510454c39acf3e711 my kernel api for anyone curious

fiery turtle Dec 26, 2024, 12:34 AM

#

gentle peak this rewrite is basically meant to just be "get me a simple kernel that lets me ...

what are your plans as far as userspace?

gentle peak Dec 26, 2024, 12:35 AM

#

nothing concrete, probably just "indistinguishable from standard linux except NIH"

fiery turtle Dec 26, 2024, 12:35 AM

#

left orbit <https://gist.github.com/48cf/32736b0a6e01ec7510454c39acf3e711> my kernel api fo...

you also do malloc for io_map

#

which probably takes a bit of your points

fiery turtle Dec 26, 2024, 12:35 AM

#

gentle peak nothing concrete, probably just "indistinguishable from standard linux except NI...

like even abi stuff?

gentle peak Dec 26, 2024, 12:35 AM

#

oh no not abi

fiery turtle Dec 26, 2024, 12:36 AM

#

ah

left orbit Dec 26, 2024, 12:36 AM

#

yeah true, i was thinking about properly implementing a bounds check but i gave up because i was like "meh"

#

ill change that real quick and see if it makes a difference

gentle peak Dec 26, 2024, 12:36 AM

#

the kernel won't even have a true stable abi

#

my plan is to make a vdso-like thing be the only stable way to do syscalls

fiery turtle Dec 26, 2024, 12:36 AM

#

left orbit ill change that real quick and see if it makes a difference

wait unmap is a no-op 💀

fiery turtle Dec 26, 2024, 12:36 AM

#

gentle peak my plan is to make a vdso-like thing be the only stable way to do syscalls

nice

#

a jit-like vdso

north holly Dec 26, 2024, 12:37 AM

#

gentle peak my plan is to make a vdso-like thing be the only stable way to do syscalls

that's what windows does right

#

ntdll.dll

gentle peak Dec 26, 2024, 12:37 AM

#

yeah

left orbit Dec 26, 2024, 12:37 AM

#

fiery turtle wait unmap is a no-op 💀

memory unmap? yeah

fiery turtle Dec 26, 2024, 12:37 AM

#

ntdll is not exactly a vdso

gentle peak Dec 26, 2024, 12:37 AM

#

fiery turtle maybe it would be faster on monkuous' cpu

1.7M/s

fiery turtle Dec 26, 2024, 12:37 AM

#

oh

north holly Dec 26, 2024, 12:37 AM

#

fiery turtle ntdll is not exactly a vdso

meh

left orbit Dec 26, 2024, 12:37 AM

#

it does this weird thing where its either hhdm or actual mapping

#

and i didnt know how to properly do unmap for that lol

#

its probably easy but yeah idk

fiery turtle Dec 26, 2024, 12:37 AM

#

lol

#

its probably always hhdm

left orbit Dec 26, 2024, 12:38 AM

#

nope it does get to the ioremap

fiery turtle Dec 26, 2024, 12:38 AM

#

@gentle peak can you dump your api impl as well?

gentle peak Dec 26, 2024, 12:38 AM

#

https://hst.sh/opewupavaw.cpp

fiery turtle Dec 26, 2024, 12:38 AM

#

left orbit nope it does get to the ioremap

oh its because of your is_ram check probably

left orbit Dec 26, 2024, 12:39 AM

#

3271792/s with proper iomap :D

left orbit Dec 26, 2024, 12:39 AM

#

fiery turtle oh its because of your is_ram check probably

yeah, acpica does that too

fiery turtle Dec 26, 2024, 12:39 AM

#

ah

fiery turtle Dec 26, 2024, 12:39 AM

#

left orbit 3271792/s with proper iomap :D

nice

left orbit Dec 26, 2024, 12:39 AM

#

static void acpi_unmap(acpi_physical_address pg_off, void __iomem *vaddr)
{
    unsigned long pfn;

    pfn = pg_off >> PAGE_SHIFT;
    if (should_use_kmap(pfn))
        kunmap(pfn_to_page(pfn));
    else
        iounmap(vaddr);
}```

#

their unmap can do that

#

because it takes a pg_off

#

whatever that means

gentle peak Dec 26, 2024, 12:40 AM

#

i think i indirectly use that exact method

#

since my uacpi_kernel_{map,unmap} just call acpi_os_*

fiery turtle Dec 26, 2024, 12:40 AM

#

loool

gentle peak Dec 26, 2024, 12:40 AM

#

not the most familiar with linux's kernel internals and i figured that that'd be the easiest way to make sure it worked

fiery turtle Dec 26, 2024, 12:41 AM

#

lmao

#

probably true

left orbit Dec 26, 2024, 12:41 AM

#

what the hell did u pass to unmap as the physical address

#

u dont even get one from uacpi

gentle peak Dec 26, 2024, 12:41 AM

#

the function you sent is an internal one for the osl layer

fiery turtle Dec 26, 2024, 12:41 AM

#

LULW

gentle peak Dec 26, 2024, 12:41 AM

#

the actual function - acpi_os_unmap_memory - takes a pointer

#

that's the one i call

fiery turtle Dec 26, 2024, 12:42 AM

#

left orbit what the hell did u pass to unmap as the physical address

https://hst.sh/opewupavaw.cpp

gentle peak Dec 26, 2024, 12:42 AM

#

and i assume it eventually calls acpi_unmap

left orbit Dec 26, 2024, 12:42 AM

#

ah they just keep a list of mappings

fiery turtle Dec 26, 2024, 12:42 AM

#

they have a cache yeah

gentle peak Dec 26, 2024, 12:48 AM

#

gentle peak oh btw proxima was up to a consistent 5.0-5.2M/s at some point but it's now arou...

tracked down part of the cause: if i comment out the part of my allocator code that gives free pages back to the pmm it gets back up to 4.9-5.1M/s

fiery turtle Dec 26, 2024, 12:49 AM

#

I can update the score later if you'd like meme it’s still at 4.6

left orbit Dec 26, 2024, 12:49 AM

#

back to 2.8-2.9M wtf

#

i did nothing 😭

gentle peak Dec 26, 2024, 12:49 AM

#

too small of a difference to make me really care about it, and the giving empty pages back to pmm thing is desirable behavior, so i won't be permanently removing it, but nice to know where it came from

fiery turtle Dec 26, 2024, 12:49 AM

#

Do you think it’s possible to squeeze out more?

gentle peak Dec 26, 2024, 12:50 AM

#

fiery turtle I can update the score later if you'd like <:meme:575445094589661215> it’s still...

if there are more scores to be updated or something feel free to update mine as well but i don't really care about whether such a small difference is recorded

north holly Dec 26, 2024, 12:50 AM

#

fiery turtle Do you think it’s possible to squeeze out more?

infy can you update my score:

[uACPI][INFO]: successfully loaded 1 AML blob, 1705 ops in 1ms (avg 1371085/s)

fiery turtle Dec 26, 2024, 12:50 AM

#

gentle peak if there are more scores to be updated or something feel free to update mine as ...

Im recording them for later to update in batches

gentle peak Dec 26, 2024, 12:50 AM

#

ah ok

fiery turtle Dec 26, 2024, 12:50 AM

#

north holly infy can you update my score: ``` [uACPI][INFO]: successfully loaded 1 AML blob,...

Sure

gentle peak Dec 26, 2024, 12:50 AM

#

2s let me get a real sample

north holly Dec 26, 2024, 12:50 AM

#

I'm allowed to do --no-smp right

#

to the kernel

fiery turtle Dec 26, 2024, 12:51 AM

#

The rules are all API must be implemented fairly, q35 + kvm

#

Everything else is up to u

dusky glade Dec 26, 2024, 12:51 AM

#

after i get this bug fixed i might make uacpi run in a thread trl

#

kentry thread anyway

#

i have no way of communicating with other threads atm

north holly Dec 26, 2024, 12:51 AM

#

dusky glade i have no way of communicating with other threads atm

morse code

dusky glade Dec 26, 2024, 12:51 AM

#

trl trl

north holly Dec 26, 2024, 12:51 AM

#

block threads for specific periods of time

fiery turtle Dec 26, 2024, 12:52 AM

#

Kid named Morse

north holly Dec 26, 2024, 12:52 AM

#

fiery turtle The rules are all API must be implemented fairly, q35 + kvm

the cpu is Intel i5-4590T

#

btw

fiery turtle Dec 26, 2024, 12:52 AM

#

Noted

north holly Dec 26, 2024, 12:52 AM

#

although I still don't beat anyone yet

#

I will at least be far from being beaten by nyaux

fiery turtle Dec 26, 2024, 12:52 AM

#

north holly infy can you update my score: ``` [uACPI][INFO]: successfully loaded 1 AML blob,...

north holly Dec 26, 2024, 12:53 AM

#

north holly I will at least be far from being beaten by nyaux

(the total disgrace in osdev)

fiery turtle Dec 26, 2024, 12:53 AM

#

Inb4 nyaux fixes skill issues and gets 10M

north holly Dec 26, 2024, 12:53 AM

#

https://cdn.discordapp.com/emojis/816178972672131083.gif?size=64

north holly Dec 26, 2024, 12:53 AM

#

north holly https://cdn.discordapp.com/emojis/816178972672131083.gif?size=64

this would be me if nyaux passes me /j

fiery turtle Dec 26, 2024, 12:53 AM

#

fiery turtle Do you think it’s possible to squeeze out more?

But im curious about this

left orbit Dec 26, 2024, 12:53 AM

#

left orbit <https://gist.github.com/48cf/32736b0a6e01ec7510454c39acf3e711> my kernel api fo...

updated my kernel api gist, no unimplemented stuff

#

but im not too sure about how correct it is

fiery turtle Dec 26, 2024, 12:54 AM

#

Nice, we should really make a fork

left orbit Dec 26, 2024, 12:54 AM

#

but not like it matters

fiery turtle Dec 26, 2024, 12:54 AM

#

To at least store it for testing

left orbit Dec 26, 2024, 12:54 AM

#

that would definitely be cool

north holly Dec 26, 2024, 12:54 AM

#

is it finally happening

fiery turtle Dec 26, 2024, 12:54 AM

#

Ill make one

north holly Dec 26, 2024, 12:54 AM

#

is linux getting uacpi support

fiery turtle Dec 26, 2024, 12:54 AM

#

Kinda

fiery turtle Dec 26, 2024, 12:54 AM

#

left orbit that would definitely be cool

Can you upload everything you have to a branch in your local fork perhaps?

left orbit Dec 26, 2024, 12:55 AM

#

i can make a patch one sec

gentle peak Dec 26, 2024, 12:55 AM

#

fiery turtle Im recording them for later to update in batches

with the iso image from https://github.com/proxima-os/proxima/actions/runs/12537199433

uacpi: info: successfully loaded 1 AML blob, 1705 ops in 0ms (avg 6406643/s)

fiery turtle Dec 26, 2024, 12:55 AM

#

left orbit i can make a patch one sec

Or I guess u have makefiles in a submodule dir hmm

fiery turtle Dec 26, 2024, 12:55 AM

#

gentle peak with the iso image from https://github.com/proxima-os/proxima/actions/runs/12537...

left orbit Dec 26, 2024, 12:55 AM

#

oh i don't have a submodule meme

fiery turtle Dec 26, 2024, 12:55 AM

#

gentle peak with the iso image from https://github.com/proxima-os/proxima/actions/runs/12537...

Crazy numbers

left orbit Dec 26, 2024, 12:55 AM

#

i just downloaded a tarball of uacpi and extracted it

fiery turtle Dec 26, 2024, 12:56 AM

#

Bruh

#

That works ig KEKW

left orbit Dec 26, 2024, 12:56 AM

#

lemme nuke test cases tho

#

and everything that isnt source or a header lol

gentle peak Dec 26, 2024, 12:57 AM

#

fiery turtle Crazy numbers

with my current unpushed code i get 5,063,043/s but i feel like that's a bit cheating since the heap shrinking is commented out

#

(aka stuff <PAGE_SIZE doesn't get returned to pmm)

fiery turtle Dec 26, 2024, 12:57 AM

#

Sort of I guess, since u dont wanna be leaking pages

fiery turtle Dec 26, 2024, 12:58 AM

#

gentle peak (aka stuff <PAGE_SIZE doesn't get returned to pmm)

U mean >?

gentle peak Dec 26, 2024, 12:58 AM

#

no, <

fiery turtle Dec 26, 2024, 12:58 AM

#

Oh

gentle peak Dec 26, 2024, 12:58 AM

#

PAGE_SIZE gets handled by a different allocator, that one still gives back memory

left orbit Dec 26, 2024, 12:58 AM

#

@fiery turtle is a patch okay?

fiery turtle Dec 26, 2024, 12:58 AM

#

I see

fiery turtle Dec 26, 2024, 12:58 AM

#

left orbit <@353593948523986945> is a patch okay?

Sure

left orbit Dec 26, 2024, 12:58 AM

#

📎 0001-drivers-acpi-Basic-uACPI-implementation.patch

#

lots of uacpi :^)

fiery turtle Dec 26, 2024, 12:58 AM

#

Lol

gentle peak Dec 26, 2024, 12:59 AM

#

gentle peak >PAGE_SIZE gets handled by a different allocator, that one still gives back memo...

then again i don't think that allocator gets used by uacpi at all since my PAGE_SIZE is 16k

fiery turtle Dec 26, 2024, 12:59 AM

#

gentle peak then again i don't think that allocator gets used by uacpi at all since my PAGE_...

Waaait

#

Why 16k

gentle peak Dec 26, 2024, 12:59 AM

#

wanted to make sure everything could handle PAGE_SHIFT not matching what the architecture expects so i just picked something at random

left orbit Dec 26, 2024, 12:59 AM

#

for "our" usecase it's probably okay to add a submodule in the root of the repo and symlink source + include into drivers/acpi/uacpi

fiery turtle Dec 26, 2024, 12:59 AM

#

Ah good idea

gentle peak Dec 26, 2024, 12:59 AM

#

i can set it to 12 rq and see how it affects benchmarks

#

page fault.

fiery turtle Dec 26, 2024, 1:00 AM

#

Lol

fiery turtle Dec 26, 2024, 1:00 AM

#

left orbit for "our" usecase it's probably okay to add a submodule in the root of the repo ...

Perhaps

left orbit Dec 26, 2024, 1:00 AM

#

that gets rid of the "makefile in submodule" problem

fiery turtle Dec 26, 2024, 1:01 AM

#

Yea

left orbit Dec 26, 2024, 1:01 AM

#

at the expense of "fuck windows users" but not like ur compiling linux on windows anyway

fiery turtle Dec 26, 2024, 1:01 AM

#

U cant even clone it on windows

left orbit Dec 26, 2024, 1:01 AM

#

actually?

#

lmao

fiery turtle Dec 26, 2024, 1:01 AM

#

Yeah

#

Same name different case

#

Doesnt work on windows

left orbit Dec 26, 2024, 1:01 AM

#

good

#

meme

#

anyway if you do fork linux i can do the dirty work of submodule + adapting this patch for that i guess

fiery turtle Dec 26, 2024, 1:02 AM

#

left orbit anyway if you do fork linux i can do the dirty work of submodule + adapting this...

Where shall I put it

left orbit Dec 26, 2024, 1:02 AM

#

shouldn't be that bad but yeah

#

ultraos org maybe?

#

makes sense

fiery turtle Dec 26, 2024, 1:02 AM

#

Sure ill do that first thing tomorrow

left orbit Dec 26, 2024, 1:02 AM

#

ultraos/linux-uacpi or something

#

sounds good

fiery turtle Dec 26, 2024, 1:03 AM

#

left orbit anyway if you do fork linux i can do the dirty work of submodule + adapting this...

U can make a local fork right now with those changes then pr into my fork if you wish

gentle peak Dec 26, 2024, 1:03 AM

#

gentle peak page fault.

figured it out:

KERNEL_STACK_SIZE is still 16k, so the kernel stack of the init task was allocated with vmalloc_large instead of kalloc as it was when PAGE_SIZE was 16k
the size given to vmfree when freeing said stack was the size of my task_t struct instead of that of the stack itself, so it was freed with kfree instead of vmfree_large

left orbit Dec 26, 2024, 1:03 AM

#

i will do that yes

fiery turtle Dec 26, 2024, 1:03 AM

#

left orbit i will do that yes

LETSFUCKINGGOOOOOO

gentle peak Dec 26, 2024, 1:03 AM

#

this worked before because kfree unlike vmfree doesn't take size as param

#

anyway has no bearing on uacpi benchmark thankfully

fiery turtle Dec 26, 2024, 1:04 AM

#

gentle peak figured it out: - `KERNEL_STACK_SIZE` is still 16k, so the kernel stack of the i...

Oh nice

#

So you've uncovered a bug?

gentle peak Dec 26, 2024, 1:04 AM

#

yeah

#

well a page fault is a bug regardless

#

but still

fiery turtle Dec 26, 2024, 1:04 AM

#

I should adapt a dumb page size idea as well

#

For my stuff

#

Like Linux does with jiffies

#

Setting initial to random values

gentle peak Dec 26, 2024, 1:06 AM

#

uacpi benchmarks

PAGE_SIZE 4k: 4 560 359
PAGE_SIZE 16k: 4 794 861

#

not too significant of an impact

fiery turtle Dec 26, 2024, 1:07 AM

#

Probably because u get buckets for smaller allocations instead of going into the pmm?

gentle peak Dec 26, 2024, 1:07 AM

#

going into the pmm is pretty fast actually, but for >PAGE_SIZE i don't just go into the pmm, i go off of the hhdm entirely so i need to allocate a virtual address and map it

fiery turtle Dec 26, 2024, 1:08 AM

#

Ah

#

Vmalloc basically?

gentle peak Dec 26, 2024, 1:08 AM

#

yeah

#

fiery turtle Dec 26, 2024, 1:08 AM

#

Makes sense

#

Do u prefault those

#

I guess u do

gentle peak Dec 26, 2024, 1:09 AM

#

demand paging isn't used for kernel memory so doesn't really apply

#

but yeah it's the equivalent of prefaulting

left orbit Dec 26, 2024, 1:09 AM

#

ok epic back to where we were but now with a submodule
[ 0.376604] successfully loaded 1 AML blob, 1705 ops in 0ms (avg 2818167/s)

fiery turtle Dec 26, 2024, 1:10 AM

#

Very nice

fiery turtle Dec 26, 2024, 1:10 AM

#

gentle peak demand paging isn't used for kernel memory so doesn't really apply

Wellll nt does that a lot right

gentle peak Dec 26, 2024, 1:10 AM

#

nt yeah but hydrogen doesn't and won't

fiery turtle Dec 26, 2024, 1:10 AM

#

Fair lol

gentle peak Dec 26, 2024, 1:10 AM

#

the idea is that if you're calling into vmalloc_large a lot you're probably doing something wrong anyway

fiery turtle Dec 26, 2024, 1:11 AM

#

Why _large?

gentle peak Dec 26, 2024, 1:11 AM

#

it's only used for >PAGE_SIZE

fiery turtle Dec 26, 2024, 1:11 AM

#

Is there a small equivalent

gentle peak Dec 26, 2024, 1:11 AM

#

there's normal vmalloc which decides whether to use kalloc or vmalloc_large based on the size given

fiery turtle Dec 26, 2024, 1:11 AM

#

Ohh

#

Smart

gentle peak Dec 26, 2024, 1:11 AM

#

basically:

you use kalloc when you need power-of-two alignment or easy mapping to paddr
you use vmalloc anywhere else

fiery turtle Dec 26, 2024, 1:12 AM

#

How does vfree know where to go

gentle peak Dec 26, 2024, 1:12 AM

#

it's a sized free

fiery turtle Dec 26, 2024, 1:12 AM

#

Or well struct page I guess

fiery turtle Dec 26, 2024, 1:12 AM

#

gentle peak it's a sized free

Ah

gentle peak Dec 26, 2024, 1:12 AM

#

nah struct page is only used for kfree since with vmfree the pointer might not be in hhdm

fiery turtle Dec 26, 2024, 1:12 AM

#

U can vfree less than u malloced?

gentle peak Dec 26, 2024, 1:12 AM

#

no the size still has to match

left orbit Dec 26, 2024, 1:12 AM

#

hacker@raptor:~/sources/linux$ git pull origin master --depth=1
remote: Enumerating objects: 39, done.
remote: Counting objects: 100% (39/39), done.
remote: Compressing objects: 100% (14/14), done.
remote: Total 20 (delta 19), reused 6 (delta 6), pack-reused 0
Unpacking objects: 100% (20/20), 2.76 KiB | 201.00 KiB/s, done.
From https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux
 * branch                master     -> FETCH_HEAD
fatal: mmap failed, check sys.vm.max_map_count and/or RLIMIT_DATA: Cannot allocate memory```

#

holy 🔥

gentle peak Dec 26, 2024, 1:12 AM

#

i just figured that for a lot of purposes the size is known at compile time even for free

fiery turtle Dec 26, 2024, 1:12 AM

#

Why require that?

gentle peak Dec 26, 2024, 1:13 AM

#

so making it a sized free allows for LTO to optimize vmfree into a direct call to kfree

fiery turtle Dec 26, 2024, 1:13 AM

#

left orbit ``` hacker@raptor:~/sources/linux$ git pull origin master --depth=1 remote: Enum...

Lmao

fiery turtle Dec 26, 2024, 1:13 AM

#

gentle peak so making it a sized free allows for LTO to optimize vmfree into a direct call t...

Ohh

gentle peak Dec 26, 2024, 1:13 AM

#

i haven't actually measured what impact uacpi's sized frees option has on perf

#

since my uacpi glue code emulates a non-sized free

fiery turtle Dec 26, 2024, 1:13 AM

#

Its probably worse because uacpi has to calculate it in some cases

#

Even strlen sometimes

gentle peak Dec 26, 2024, 1:14 AM

#

oh wow

fiery turtle Dec 26, 2024, 1:14 AM

#

Yeah

#

Most are hardcoded tho

gentle peak Dec 26, 2024, 1:14 AM

#

i might add another allocator that delegates to vmalloc/vmfree but doesn't have a sized free in case things like that pop up elsewhere in the kernel

#

won't do it until i need it but still

fiery turtle Dec 26, 2024, 1:15 AM

#

Fair

gentle peak Dec 26, 2024, 1:16 AM

#

fiery turtle Most are hardcoded tho

hm so sized frees+LTO would speed up most cases a tiny bit and slow down rare cases a lot

#

probably not worth it

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