uACPI - a portable and easy-to-integrate ACPI implementation | Operating System Development | Page 19

fiery turtle Dec 26, 2024, 1:16 AM

#

Yeah

#

But I think the expensive cases are past the benchmark

#

Because all aml buffers have a corresponding size field etc

#

It’s mostly to do with e.g. generate_absolute_path(), or HID strings or similar

gentle peak Dec 26, 2024, 1:18 AM

#

forgot to respond to this, it probably is because no code is ever truly optimal but i can't see how. maybe there's a better non-simd memcpy and memset than rep movsb and rep stosb? most other things are pretty much as fast as i can get them while also still being thread safe and having all the features i want

#

the one thing that comes to mind is PCI access (currently i scan all ECAM ranges for every UACPI access) but i don't think that gets called by uacpi during the benchmark

fiery turtle Dec 26, 2024, 1:19 AM

#

gentle peak the one thing that comes to mind is PCI access (currently i scan all ECAM ranges...

Yeah don't think so

median crest Dec 26, 2024, 1:19 AM

#

managarm phd memcpy meme

fiery turtle Dec 26, 2024, 1:19 AM

#

Yeah I think its actually faster than rep

#

Although its in c++ lol

#

It does heavy unrolling which I guess really helps for small memcopies

gentle peak Dec 26, 2024, 1:21 AM

#

gentle peak the one thing that comes to mind is PCI access (currently i scan all ECAM ranges...

this is what i mean by this btw, the intended usage is to call get_pci_config once during init and then use that handle (it's actually just a pointer into ecam) using the read/write methods which are all just one memory access. uacpi doesn't have the api to use that properly, so i need to do get_pci_config on every access

fiery turtle Dec 26, 2024, 1:22 AM

#

gentle peak this is what i mean by this btw, the intended usage is to call get_pci_config on...

This is why I thought about introducing a uacpi_kernel_pci_device_open

#

Which would give back some sort of a handle

gentle peak Dec 26, 2024, 1:22 AM

#

that'd be nice yeah

fiery turtle Dec 26, 2024, 1:22 AM

#

Same as all other current api

#

Perhaps I shall add that

gentle peak Dec 26, 2024, 1:23 AM

#

is aml structured in a way that'd make that worthwhile? or would it amount to open+access+close for every pci op?

fiery turtle Dec 26, 2024, 1:23 AM

#

Its only used for opregions so I can optimize this to open once actually

gentle peak Dec 26, 2024, 1:23 AM

#

oh yeah then i think that'd be worth adding

fiery turtle Dec 26, 2024, 1:24 AM

#

Yeah definitely

#

I wanted that for uapi anyway

#

Will add tomorrow I guess, I have a draft PR open for opregion rework

#

Anyways very based work today, thx iretq and monkuous LETSFUCKINGGOOOOOO

#

Im gonna go sleep now

#

Gn

gentle peak Dec 26, 2024, 1:27 AM

#

gn

left orbit Dec 26, 2024, 1:28 AM

#

gn

gentle peak Dec 26, 2024, 1:37 AM

#

fiery turtle But I think the expensive cases are past the benchmark

i tested it just in case and wow that made a much larger difference than i thought it would.

unsized frees, kfree gives empty pages back to pmm, page size is 4k: 4 582 951/s
unsized frees, kfree gives empty pages back to pmm, page size is 16k (this is what's currently on github): 4 781 911/s
unsized frees, kfree keeps empty pages, page size is 4k: 4 830 247/s
unsized frees, kfree keeps empty pages, page size is 16k: 5 090 935/s
sized frees, kfree gives empty pages back to pmm, page size is 4k: 4 996 014/s
sized frees, kfree gives empty pages back to pmm, page size is 16k (this is what's going to be on github): 5 163 677/s
sized frees, kfree keeps empty pages, page size is 4k: 5 112 014/s
sized frees, kfree keeps empty pages, page size is 16k: 5 379 635/s

#

alright i pushed the change and updated the benchmark message

torpid root Dec 26, 2024, 8:32 AM

#

I just remembered that my pmm memsets memory too

#

could've gotten a few more points

#

nooo

median crest Dec 26, 2024, 8:47 AM

#

Rewrite whole kernel for 1000 more points on a fake spreadsheet

#

Do it

torpid root Dec 26, 2024, 8:56 AM

#

yes sir

rustic compass Dec 26, 2024, 9:03 AM

#

Also take qemu and underlying hardware characteristics into account to perfectly optimize for your stack

fiery turtle Dec 26, 2024, 9:40 AM

#

gentle peak i tested it just in case and wow that made a much larger difference than i thoug...

Damn

torpid root Dec 26, 2024, 9:48 AM

#

hmmm could anyone test this? I don't think much has improved though

qemu-system-x86_64 -serial stdio -m 512M -cpu max,migratable=off,+hypervisor,+invtsc,+tsc-deadline -M q35 -accel kvm -cdrom image-x86_64.iso

📎 SiFcPeKL.iso

gentle peak Dec 26, 2024, 9:50 AM

#

1359972/s

torpid root Dec 26, 2024, 9:53 AM

#

I get 1.5mil

#

same as before

left orbit Dec 26, 2024, 9:54 AM

#

i love how this has turned into a competition and it totally dragged everyone away from actually working on their kernels

torpid root Dec 26, 2024, 9:55 AM

#

technically we are still working on kernels and trying to make them faster

left orbit Dec 26, 2024, 10:00 AM

#

i don't count hyperoptimizing your mem* and uacpi kernel api as actual work tho

fiery turtle Dec 26, 2024, 10:02 AM

#

left orbit i don't count hyperoptimizing your mem* and uacpi kernel api as actual work tho ...

i mean u use mem stuff everywhere around the kernel right

#

so its a useful optimization

left orbit Dec 26, 2024, 10:03 AM

#

yeah yeah but it usually doesn't make a big difference

#

or, well, shouldn't

fiery turtle Dec 26, 2024, 10:03 AM

#

who knows

left orbit Dec 26, 2024, 10:03 AM

#

but yeah any work is useful work of course!

fiery turtle Dec 26, 2024, 10:03 AM

#

no one did a generalized benchmark before meme

left orbit Dec 26, 2024, 10:04 AM

#

maybe we could do actual profiling when we replace acpica in linux

#

i know linux can do pgo

#

so there must be a good way to profile it

fiery turtle Dec 26, 2024, 10:04 AM

#

yeah

#

btw did you finish that branch work?

left orbit Dec 26, 2024, 10:04 AM

#

better way than sampling rip outside of qemu lol

fiery turtle Dec 26, 2024, 10:04 AM

#

true

left orbit Dec 26, 2024, 10:04 AM

#

yeah as far as implementing kernel api and initializing it goes

#

i did a submodule + link from drivers/acpi/uacpi/source to ../../../uacpi/source

fiery turtle Dec 26, 2024, 10:05 AM

#

https://github.com/UltraOS/uacpi-linux

GitHub

GitHub - UltraOS/uacpi-linux: uACPI integration for the Linux kernel

uACPI integration for the Linux kernel. Contribute to UltraOS/uacpi-linux development by creating an account on GitHub.

left orbit Dec 26, 2024, 10:05 AM

#

yooo

#

i shall make a pr

fiery turtle Dec 26, 2024, 10:06 AM

#

lol

#

pls do

left orbit Dec 26, 2024, 10:06 AM

#

oh god i have to fork it now too lmaoo

#

how long did this take to fork? meme

fiery turtle Dec 26, 2024, 10:06 AM

#

1 sec

left orbit Dec 26, 2024, 10:06 AM

#

oh wtf

fiery turtle Dec 26, 2024, 10:06 AM

#

yeah it's a lazy copy i guess

left orbit Dec 26, 2024, 10:06 AM

#

its a gigantic repo

#

probably yeah makes sense

#

hacker@raptor:~/sources/linux$ git fetch uacpi master
error: cannot run ssh: No such file or directory
fatal: unable to fork```

#

what the fuck

#

HOW DO I NOT HAVE SSH

fiery turtle Dec 26, 2024, 10:08 AM

#

lmao

left orbit Dec 26, 2024, 10:12 AM

#

https://github.com/UltraOS/uacpi-linux/pull/1

GitHub

drivers/acpi: Basic uACPI support by 48cf · Pull Request #1 · Ultra...

fiery turtle Dec 26, 2024, 10:12 AM

#

LETSFUCKINGGOOOOOO

#

should the upper one be removed?

left orbit Dec 26, 2024, 10:13 AM

#

nope

#

the upper one contains uacpi_libc.h

#

and types i think

fiery turtle Dec 26, 2024, 10:13 AM

#

ah

left orbit Dec 26, 2024, 10:13 AM

#

and the lower one is the actual uacpi submodule

#

its a bit uhh

#

i didnt know how to properly organize it

#

but this keeps most of the actual uacpi stuff under drivers/acpi/uacpi

#

except for the submodule

gentle peak Dec 26, 2024, 10:14 AM

#

fiery turtle should the upper one be removed?

is this really the intended way for drivers to add include paths?

left orbit Dec 26, 2024, 10:14 AM

#

i think submodule at the top level makes sense

#

i dont think so lol

fiery turtle Dec 26, 2024, 10:14 AM

#

gentle peak is this really the intended way for drivers to add include paths?

highly doubt

left orbit Dec 26, 2024, 10:14 AM

#

actually hold on ill see if there isnt a better place for this

fiery turtle Dec 26, 2024, 10:14 AM

#

but idk the right way

gentle peak Dec 26, 2024, 10:15 AM

#

looking at /include, it seems it might be, because headers for all drivers are there

left orbit Dec 26, 2024, 10:15 AM

#

so symlink uacpi/include/uacpi -> include/uacpi?

#

i guess

#

and put the two uacpi headers there maybe meme

fiery turtle Dec 26, 2024, 10:16 AM

#

makes sense

#

acpica includes are there

#

i think?

left orbit Dec 26, 2024, 10:16 AM

#

acpi/*.h

#

yeah

fiery turtle Dec 26, 2024, 10:16 AM

#

ye

left orbit Dec 26, 2024, 10:17 AM

#

do i just put uacpi_{types,libc}.h in include/?

median crest Dec 26, 2024, 10:17 AM

#

left orbit i don't count hyperoptimizing your mem* and uacpi kernel api as actual work tho ...

It made me realize how shit my mutexes are and that I need to use a real lock

fiery turtle Dec 26, 2024, 10:17 AM

#

I guess

median crest Dec 26, 2024, 10:17 AM

#

Once I swtich over I can probably break 4 mil on qwincis computer

fiery turtle Dec 26, 2024, 10:18 AM

#

i wonder if linux semaphores are just crap

gentle peak Dec 26, 2024, 10:19 AM

#

i used linux's normal mutex type and i believe my result was similar to iretq's

#

so i doubt it's due to the semaphores

fiery turtle Dec 26, 2024, 10:19 AM

#

ah

#

then yeah

left orbit Dec 26, 2024, 10:20 AM

#

ok building again

gentle peak Dec 26, 2024, 10:21 AM

#

left orbit i don't count hyperoptimizing your mem* and uacpi kernel api as actual work tho ...

the only uacpi-based optimization i've done that i can think of is the sized frees thing, other than that all improvement in proxima so far has just been due to allocator improvements i would've made anyway

left orbit Dec 26, 2024, 10:24 AM

#

@fiery turtle pushed, should be better now

fiery turtle Dec 26, 2024, 10:25 AM

#

big

left orbit Dec 26, 2024, 10:25 AM

#

next step is probaly putting it behind a config option

#

also, do you want to replace acpica completely or make it an option to choose between these two?

fiery turtle Dec 26, 2024, 10:26 AM

#

aight merged

left orbit Dec 26, 2024, 10:26 AM

#

lfg

fiery turtle Dec 26, 2024, 10:27 AM

#

left orbit also, do you want to replace acpica completely or make it an option to choose be...

nah, it would be a maintenance hell to support both, since linux doesnt abstract it away in any way unlike nt for example

median crest Dec 26, 2024, 10:27 AM

#

2025 will be the year of the uacpi desktop

left orbit Dec 26, 2024, 10:27 AM

#

doesn it?

fiery turtle Dec 26, 2024, 10:27 AM

#

so only full replace is realistic

left orbit Dec 26, 2024, 10:27 AM

#

i thought drivers/acpi was a wrapper around acpica

fiery turtle Dec 26, 2024, 10:28 AM

#

its acpi-specific drivers that all work on top of direct calls to acpica api

left orbit Dec 26, 2024, 10:28 AM

#

oh ew

fiery turtle Dec 26, 2024, 10:28 AM

#

yeah

#

e.g. ec.c or battery.c or button.c etc

left orbit Dec 26, 2024, 10:28 AM

#

yeah i guess replacing it then is the only sensible option

fiery turtle Dec 26, 2024, 10:28 AM

#

think so

gentle peak Dec 26, 2024, 10:29 AM

#

huh, when i compile proxima with -march=native -mtune=native i get a KVM emulation failure

#

that's unexpected

left orbit Dec 26, 2024, 10:30 AM

#

classic kvm error

fiery turtle Dec 26, 2024, 10:31 AM

#

gentle peak huh, when i compile proxima with `-march=native -mtune=native` i get a KVM emula...

do u do -cpu host?

gentle peak Dec 26, 2024, 10:31 AM

#

yeah

fiery turtle Dec 26, 2024, 10:31 AM

#

hmm

gentle peak Dec 26, 2024, 10:31 AM

#

and if i didn't i would've expected #UD instead of an emulation failure

fiery turtle Dec 26, 2024, 10:31 AM

#

depends

#

there are probably edge cases that cause a kvm error

#

like setting some msr bits that dont exist

#

idk

gentle peak Dec 26, 2024, 10:32 AM

#

rip points to bextr %eax,0x10(%r12,%rbx,1),%edx

flat badge Dec 26, 2024, 10:32 AM

#

this can be caused by e.g. instructions that KVM cannot emulate doing MMIO

gentle peak Dec 26, 2024, 10:32 AM

#

oh

#

let me check what that's pointing to

fiery turtle Dec 26, 2024, 10:33 AM

#

O3 doing funny stuff?

gentle peak Dec 26, 2024, 10:33 AM

#

it's using bextr on io apic registers

#

that'd explain it

calm latch Dec 26, 2024, 10:33 AM

#

fiery turtle i mean u use mem stuff everywhere around the kernel right

We need uDRM urgently

fiery turtle Dec 26, 2024, 10:34 AM

#

do u have special compile flags that cause that?

fiery turtle Dec 26, 2024, 10:34 AM

#

calm latch We need uDRM urgently

im not the author of udrm sooo

gentle peak Dec 26, 2024, 10:34 AM

#

don't think so, my flags are just -march=native -mtune=native -O3 -DNDEBUG -std=gnu11 -flto=thin -fno-common -Wall -Wextra -ffreestanding -fno-pie -fno-stack-check -fno-stack-protector -mcmodel=kernel -mgeneral-regs-only -mno-red-zone

fiery turtle Dec 26, 2024, 10:34 AM

#

yeah thats probably why linux always does O2

#

O3 is not tested nor supported

calm latch Dec 26, 2024, 10:34 AM

#

calm latch We need uDRM urgently

So that every hobby kernel suddenly gets good PCI infrastructure

flat badge Dec 26, 2024, 10:39 AM

#

you should access MMIO through inline assembly helpers

#

to avoid this

gentle peak Dec 26, 2024, 10:39 AM

#

yeah i'll probably do that

fiery turtle Dec 26, 2024, 10:39 AM

#

in your kernel?

flat badge Dec 26, 2024, 10:39 AM

#

yeah

#

that's what linux + managarm do as well

fiery turtle Dec 26, 2024, 10:40 AM

#

oh i completely missed that it was proxima dying lol

#

i thought it was linux

fiery turtle Dec 26, 2024, 10:40 AM

#

flat badge that's what linux + managarm do as well

linux eventually gets into a function that does a switch on a width and volatile loads or stores

#

there isnt assembly there

#

or well on x86 it does actually

#

my bad

#

#define build_mmio_read(name, size, type, reg, barrier) \
static inline type name(const volatile void __iomem *addr) \
{ type ret; asm volatile("mov" size " %1,%0":reg (ret) \
:"m" (*(volatile type __force *)addr) barrier); return ret; }

#define build_mmio_write(name, size, type, reg, barrier) \
static inline void name(type val, volatile void __iomem *addr) \
{ asm volatile("mov" size " %0,%1": :reg (val), \
"m" (*(volatile type __force *)addr) barrier); }

build_mmio_read(readb, "b", unsigned char, "=q", :"memory")
build_mmio_read(readw, "w", unsigned short, "=r", :"memory")
build_mmio_read(readl, "l", unsigned int, "=r", :"memory")

build_mmio_read(__readb, "b", unsigned char, "=q", )
build_mmio_read(__readw, "w", unsigned short, "=r", )
build_mmio_read(__readl, "l", unsigned int, "=r", )

build_mmio_write(writeb, "b", unsigned char, "q", :"memory")
build_mmio_write(writew, "w", unsigned short, "r", :"memory")
build_mmio_write(writel, "l", unsigned int, "r", :"memory")

build_mmio_write(__writeb, "b", unsigned char, "q", )
build_mmio_write(__writew, "w", unsigned short, "r", )
build_mmio_write(__writel, "l", unsigned int, "r", )

#

idk what the switch was for i cant find it

gentle peak Dec 26, 2024, 10:50 AM

#

this should probably be made into a kernel api then

left orbit Dec 26, 2024, 10:50 AM

#

it used to be

flat badge Dec 26, 2024, 10:55 AM

#

or it should use inline asm within uacpi

left orbit Dec 26, 2024, 10:56 AM

#

uacpi would have to provide inline assembly for all platforms

#

and i think thats what infy doesn't want to do

fiery turtle Dec 26, 2024, 10:57 AM

#

^

#

but also i think this should be fine?

slim panther Dec 26, 2024, 10:58 AM

#

calm latch So that every hobby kernel suddenly gets good PCI infrastructure

true and real

flat badge Dec 26, 2024, 10:59 AM

#

left orbit and i think thats what infy doesn't want to do

why not? how many supported platforms are there?

median crest Dec 26, 2024, 11:00 AM

#

Only 2 real platforms exist anyways anything else is mental illness

flat badge Dec 26, 2024, 11:00 AM

#

*3 meme

median crest Dec 26, 2024, 11:00 AM

#

Which platforms I mean is left as an exercise to the reader

gentle peak Dec 26, 2024, 11:00 AM

#

fiery turtle but also i think this should be fine?

seems like it works fine rn, but apparently compilers just use non-mmio instructions for volatile ptrs whenever they feel like it, so it seems a bit fragile

fiery turtle Dec 26, 2024, 11:00 AM

#

gentle peak this should probably be made into a kernel api then

ACPICA does it like this, I think i should be fine

flat badge Dec 26, 2024, 11:00 AM

#

fwiw i don't think volatile guarantees that the compiler doesn't use insane access modes that the memory controller doesn't support

slim panther Dec 26, 2024, 11:00 AM

#

median crest Only 2 real platforms exist anyways anything else is mental illness

3

flat badge Dec 26, 2024, 11:00 AM

#

for MMIO

#

volatile is fine if you know it's backed by RAM

fiery turtle Dec 26, 2024, 11:01 AM

#

flat badge why not? how many supported platforms are there?

x86, arm, itanium

left orbit Dec 26, 2024, 11:01 AM

#

no riscv?

gentle peak Dec 26, 2024, 11:01 AM

#

uacpi itself is completely platform independent, no?

fiery turtle Dec 26, 2024, 11:01 AM

#

riscv yeah

fiery turtle Dec 26, 2024, 11:01 AM

#

gentle peak uacpi itself is completely platform independent, no?

it is

slim panther Dec 26, 2024, 11:01 AM

#

based

fiery turtle Dec 26, 2024, 11:01 AM

#

so im definitely not adding arch specific stuff tehre

#

its not aware of what arch its executing on

flat badge Dec 26, 2024, 11:02 AM

#

then i'd suggest to turn this into a kernel API

fiery turtle Dec 26, 2024, 11:02 AM

#

perhaps yeah

#

that works on a virtual address

gentle peak Dec 26, 2024, 11:02 AM

#

maybe you can make it a weakly linked kernel api and provide a default impl that does volatile?

fiery turtle Dec 26, 2024, 11:02 AM

#

uacpi_kernel_system_memory_read

gentle peak Dec 26, 2024, 11:02 AM

#

then again it's not a particularly hard kernel api to implement

fiery turtle Dec 26, 2024, 11:02 AM

#

most osses will just do this volatile switch

#

so probably fine to have a default

#

idk

left orbit Dec 26, 2024, 11:03 AM

#

@fiery turtle whats next for uacpi-linux?

gentle peak Dec 26, 2024, 11:03 AM

#

anyway proxima now only uses volatile pointers for framebuffer and uacpi so that's nice

fiery turtle Dec 26, 2024, 11:04 AM

#

left orbit <@353593948523986945> whats next for uacpi-linux?

code drivers/acpi/*.c then manually edit acpica calls and replace with uacpi

#

first probably bus.c and more important drivers

#

i guess im doing the thing that korona suggested now

#

every read and write, be it io, pci, or mem is done on a previously mapped handle instead of directly

#

in case of virtual memory ofc the handle is expected to be the actual virtual address tho

#

for non-aml io

#

like tables etc

gentle peak Dec 26, 2024, 11:06 AM

#

makes sense

fiery turtle Dec 26, 2024, 11:07 AM

#

whether im adding defaults for memory r/w im not sure yet

#

probably yes

gentle peak Dec 26, 2024, 11:07 AM

#

oh wow the -march=native -mtune=native actually had a noticeable impact on speed, i didn't expect that (5.1M/s -> 5.3M/s)

fiery turtle Dec 26, 2024, 11:07 AM

#

because it works everywhere atm

fiery turtle Dec 26, 2024, 11:07 AM

#

gentle peak oh wow the `-march=native -mtune=native` actually had a noticeable impact on spe...

interesting

left orbit Dec 26, 2024, 11:09 AM

#

ok i guess i'll take a look at drivers/acpi/*.c then

fiery turtle Dec 26, 2024, 11:09 AM

#

damn you're actually doing it?

#

nice

left orbit Dec 26, 2024, 11:09 AM

#

converting driver code sounds like pain

#

i mean i'm not sure what i'll do but i'll take a look at the very least

fiery turtle Dec 26, 2024, 11:09 AM

#

its mostly convert acpi_x to uacpi_x

#

some are literally like that, some are harder probably

left orbit Dec 26, 2024, 11:10 AM

#

converting the types everywhere is gonna be fun

#

but yeah i see what you mean

#

its mostly just tedious work

fiery turtle Dec 26, 2024, 11:12 AM

#

yup

#

but careful so its annoying

#

also they have patterns where they create an object on the stack then shove it into a method as an argument

#

thats just not something u can do safely

flat badge Dec 26, 2024, 11:13 AM

#

Can't you make a acpica compat lib on top of uacpi?

fiery turtle Dec 26, 2024, 11:14 AM

#

you can of course, technically

#

it would be very slow and insane

#

like implementing dx12 in terms of vulkan

#

every struct would have to be converted, a lot of hidden allocations, etc

#

i dont think an abomination like that would be upstreamable ever

#

thats why e.g. NT draws the line of abstraction higher

#

so it expects acpi.sys to export more higher level helpers, that don't have shit like eval_aml()

gentle peak Dec 26, 2024, 11:18 AM

#

is this due to my lack of an EC driver orr

#

oh wow that quality is terrible wtf

#

whatever

fiery turtle Dec 26, 2024, 11:19 AM

#

can I have your blob

left orbit Dec 26, 2024, 11:19 AM

#

lmao

gentle peak Dec 26, 2024, 11:20 AM

#

what's the command for that again

fiery turtle Dec 26, 2024, 11:20 AM

#

acpidump > dump.txt

fiery turtle Dec 26, 2024, 11:21 AM

#

gentle peak is this due to my lack of an EC driver orr

do u install a fixed event handler?

calm latch Dec 26, 2024, 11:21 AM

#

try pmos risc

fiery turtle Dec 26, 2024, 11:22 AM

#

so far this looks bogus af lol

#

like a handler was uninstalled midway

#

switch (event->handler_type) {
case GPE_HANDLER_TYPE_AML_HANDLER: {
    uacpi_object *method_obj;

    method_obj = uacpi_namespace_node_get_object_typed(
        event->aml_handler, UACPI_OBJECT_METHOD_BIT
    );
    if (uacpi_unlikely(method_obj == UACPI_NULL)) {
        uacpi_error("GPE(%02X) AML handler gone\n", event->idx);
        break;

#

like i cant even imagine how to trigger this error

gentle peak Dec 26, 2024, 11:23 AM

#

fiery turtle `acpidump > dump.txt`

📎 dump.txt

gentle peak Dec 26, 2024, 11:24 AM

#

fiery turtle do u install a fixed event handler?

no, right now i don't do anything with uacpi except the init sequence

fiery turtle Dec 26, 2024, 11:24 AM

#

very interesting

#

can u try pmos real quick as well?

gentle peak Dec 26, 2024, 11:25 AM

#

what's the repo for that so i can compile it

fiery turtle Dec 26, 2024, 11:25 AM

#

there's an iso

#

@calm latch can u link it real quick

calm latch Dec 26, 2024, 11:25 AM

#

https://gitlab.com/mishakov/pmos

GitLab

Mikhail Kovalev / pmOS · GitLab

GitLab.com

fiery turtle Dec 26, 2024, 11:25 AM

#

no i mean the iso lol

calm latch Dec 26, 2024, 11:26 AM

#

#1236769772805554206 message

fiery turtle Dec 26, 2024, 11:26 AM

#

gentle peak is this due to my lack of an EC driver orr

the fact that u even get a fixed event here is bogus, because its explicitly disabled by uacpi init code

#

i wonder whose bug it is LULW

gentle peak Dec 26, 2024, 11:28 AM

#

PMOS hangs on a blank screen at boot

fiery turtle Dec 26, 2024, 11:28 AM

#

inb4 some uacpi skill issue again

fiery turtle Dec 26, 2024, 11:28 AM

#

gentle peak PMOS hangs on a blank screen at boot

nooo lol

calm latch Dec 26, 2024, 11:28 AM

#

Does it turn off if you press power button?

#

After 3 seconds

gentle peak Dec 26, 2024, 11:29 AM

#

let me try

#

yes

fiery turtle Dec 26, 2024, 11:30 AM

#

bruh

gentle peak Dec 26, 2024, 11:30 AM

#

what the fuck

fiery turtle Dec 26, 2024, 11:30 AM

#

this is why u dont make a recursive aml interpreter

#

this is all encoded as part of 1 opcode

gentle peak Dec 26, 2024, 11:31 AM

#

i wonder if the issue is fixed if i add a power button handler

fiery turtle Dec 26, 2024, 11:31 AM

#

probably, I still dont understand the AML handler gone thing

#

any way u can enable tracing as well?

gentle peak Dec 26, 2024, 11:31 AM

#

yeah 2s

#

thats just uacpi_context_set_log_level before doing anything with uacpi right

fiery turtle Dec 26, 2024, 11:32 AM

#

anywhere yeah

gentle peak Dec 26, 2024, 11:32 AM

#

here should be fine

fiery turtle Dec 26, 2024, 11:32 AM

#

sure

fiery turtle Dec 26, 2024, 11:34 AM

#

gentle peak is this due to my lack of an EC driver orr

does it say VALZ, 0x8E there? the quality is really bad lol

gentle peak Dec 26, 2024, 11:34 AM

#

right I'm gonna have to delay every print because this just scrolls past all useful info before it can be recorded

gentle peak Dec 26, 2024, 11:34 AM

#

fiery turtle does it say VALZ, 0x8E there? the quality is really bad lol

it does

calm latch Dec 26, 2024, 11:34 AM

#

fiery turtle nooo lol

I think my framebuffer driver is seriously broken

gentle peak Dec 26, 2024, 11:34 AM

#

and yeah idk what happened to the quality, it's fine locally on my phone so something happened during upload to discord

fiery turtle Dec 26, 2024, 11:35 AM

#

gentle peak it does

hmm so the GPE handler for _L02 does fire once, but then it just somehow gets corrupted

#

wait

#

fuck

#

--- a/source/event.c
+++ b/source/event.c
@@ -445,7 +445,6 @@ static void async_run_gpe_handler(uacpi_handle opaque)
                 uacpi_status_to_string(ret)
             );
         }
-        uacpi_object_unref(method_obj);

Can u do this real quick

fiery turtle Dec 26, 2024, 11:40 AM

#

fiery turtle bruh

btw this is like 100 memory reads just to evaluate this while

#

least insane AML

fiery turtle Dec 26, 2024, 11:41 AM

#

gentle peak PMOS hangs on a blank screen at boot

the cool news is your blob seems to be old school in that it does notifications via GPE events and not the embedded controller

#

u can try plugging/unplugging the charger and might see something

#

same goes for lid stuff

gentle peak Dec 26, 2024, 11:42 AM

#

alright uploading the trace log on youtube now

fiery turtle Dec 26, 2024, 11:42 AM

#

did u apply the patch?

gentle peak Dec 26, 2024, 11:42 AM

#

fiery turtle u can try plugging/unplugging the charger and might see something

this laptop doesn't have a battery (it was Expanding a few years ago so it is now Gone from my home)

fiery turtle Dec 26, 2024, 11:42 AM

#

ah well

gentle peak Dec 26, 2024, 11:42 AM

#

fiery turtle did u apply the patch?

when i recorded the video no, i'm applying and testing the patch now

fiery turtle Dec 26, 2024, 11:42 AM

#

the lid u can still open and close

#

i refactored something in a retarded way so it was a UAF at best

gentle peak Dec 26, 2024, 11:43 AM

#

https://youtu.be/1Hcl8JgaHqg?si=PiPIvOyjjFwTxnD6

YouTube

monkuous

26 December 2024

▶ Play video

fiery turtle Dec 26, 2024, 11:44 AM

#

gentle peak https://youtu.be/1Hcl8JgaHqg?si=PiPIvOyjjFwTxnD6

oh yeah u see those GRSV reads? thats the while being evaluated lmao

gentle peak Dec 26, 2024, 11:46 AM

#

the patch seems to have fixed it

rn_image_picker_lib_temp_bc67d43a-cc70-4d20-98d4-6003ce5f1ee5.jpg

#

each of those later ignoring firmware ... prints is one press of the power button

fiery turtle Dec 26, 2024, 11:46 AM

#

what about the lid

gentle peak Dec 26, 2024, 11:47 AM

#

uacpi: warn: ignoring firmware Notify(\_SB_.LID_, 0x80) request, no listeners
uacpi: warn: ignoring firmware Notify(\_SB_.VALZ, 0x8F) request, no listeners
uacpi: warn: ignoring firmware Notify(\_SB_.LID_, 0x80) request, no listeners

fiery turtle Dec 26, 2024, 11:47 AM

#

oh damn

#

even lid works

#

0x80 is status change so its correct

#

if u hook up notifications to the lid device u can now detect when the lid is closed or opened

gentle peak Dec 26, 2024, 11:48 AM

#

awesome

fiery turtle Dec 26, 2024, 11:48 AM

#

gentle peak the patch seems to have fixed it

whats strange is fixed event 2 doesnt fire after its disabled again by uacpi

#

but literally all events are disabled on init

#

i wonder if your firmware toggles it

#

lmao

gentle peak Dec 26, 2024, 11:49 AM

#

wouldn't be surprised

#

it was made by firmware devs after all

fiery turtle Dec 26, 2024, 11:49 AM

#

with that 100 conditions while i wouldnt be surprised

#

but its a cosmetic problem so whatever

#

in a real os you would ofc have a power button handler installed

#

(which is fixed event 2)

#

maybe they just havent tested with that event disabled lmao

#

so the firmware ignores the enable bit

gentle peak Dec 26, 2024, 11:50 AM

#

maybe

#

but then why does the 2nd disable get respected

fiery turtle Dec 26, 2024, 11:51 AM

#

maybe its not, usually there's a timeout between how often power button presses can be generated

#

as to not confuse the host os

#

if u wait a bit it might get fired again

#

but im glad you tested on your laptop, very few actually use this notification mechanism so it was barely tested

#

its an old one right?

gentle peak Dec 26, 2024, 11:52 AM

#

fiery turtle if u wait a bit it might get fired again

tried waiting 20 seconds between presses and still didn't fire the event again so

fiery turtle Dec 26, 2024, 11:52 AM

#

guess its my former guess then

gentle peak Dec 26, 2024, 11:52 AM

#

fiery turtle its an old one right?

pretty old yeah, portege r30-a, came with windows 8 (not 8.1) so i believe its from 2012

fiery turtle Dec 26, 2024, 11:52 AM

#

or maybe need to wait longer :^)

fiery turtle Dec 26, 2024, 11:53 AM

#

gentle peak pretty old yeah, portege r30-a, came with windows 8 (not 8.1) so i believe its f...

cool cool

#

        Device (VALZ)
        {
            Name (_HID, EisaId ("TOS6208"))  // _HID: Hardware ID
            Name (_DDN, "VALZeneral")  // _DDN: DOS Device Name

#

I wonder what this VALZ thing is that keeps getting notified

#

ACPI-COMPLIANT VALUE ADDED LOGICAL AND GENERAL PURPOSE DEVICE

#

proprietary chipset specific crap

calm latch Dec 26, 2024, 11:54 AM

#

ACPI compliant == horribly proprietary and probably not compliant

fiery turtle Dec 26, 2024, 11:55 AM

#

its so proprietary there isnt even a _CRS

#

so u have no idea what ports its using

gentle peak Dec 26, 2024, 11:56 AM

#

huh, i just noticed, linux complains about the aml during init

ACPI Error: AE_NOT_FOUND, While resolving a named reference package element - \_SB_.PCI0.EHC1.HUB0.RMH0.PRT6 (20240827/dspkginit-438)

#

is this an acpica skill issue?

fiery turtle Dec 26, 2024, 11:57 AM

#

lets see

#

TLDR: yes

#

it does nt incompatible things

gentle peak Dec 26, 2024, 11:59 AM

#

gentle peak tried waiting 20 seconds between presses and still didn't fire the event again s...

anyway this firmware's pretty fucked regardless. if the device loses power or it's force powered off (aka power button held down while in acpi mode) it seems to triple fault during POST

fiery turtle Dec 26, 2024, 11:59 AM

#

yeah pretty shitty firmware

#

does it power on correctly after uacpi shutdown?

gentle peak Dec 26, 2024, 11:59 AM

#

yeah

fiery turtle Dec 26, 2024, 11:59 AM

#

nice

fiery turtle Dec 26, 2024, 12:15 PM

#

ill cherry pick the bug fixes & windows clang support i guess

#

since that opregion extension stuff is gonna take a while

fiery turtle Dec 26, 2024, 12:38 PM

#

waiting for large tests to finish then ill merge

#

next on the list:

uacpi_kernel_pci_device_open
make pci r/w use the handle
optional uacpi_kernel_memory_{read/write}
new address space support

gentle peak Dec 26, 2024, 12:40 PM

#

oh right i don't think i ever notified you, the acpi dumps of that laptop and my main dev machine are in https://github.com/UltraOS/ACPIDumps/pull/3

GitHub

Add Gigabyte B550M S2H and Toshiba Portege R30-A dumps by monkuous ...

fiery turtle Dec 26, 2024, 12:41 PM

#

Oh damn, really cool, thanks for improving the collection

#

how did you come up with these ids?

gentle peak Dec 26, 2024, 12:42 PM

#

i think they were somewhere in the hw-probe output?

fiery turtle Dec 26, 2024, 12:42 PM

#

ah

#

i think last people to submit these just did random numbers lol

gentle peak Dec 26, 2024, 12:43 PM

#

yeah they're the value of HWid

fiery turtle Dec 26, 2024, 12:43 PM

#

ah

fiery turtle Dec 26, 2024, 12:52 PM

#

gentle peak the patch seems to have fixed it

alright merged the bugfixes for GPE handlers and clang for windows builds (cc @rustic compass )

fiery turtle Dec 26, 2024, 12:53 PM

#

gentle peak oh right i don't think i ever notified you, the acpi dumps of that laptop and my...

merged this as well, thanks a lot

north holly Dec 26, 2024, 1:17 PM

#

fiery turtle `uacpi_kernel_system_memory_read`

removes raw memory helpers
adds raw memory helpers back, rebranded

fiery turtle Dec 26, 2024, 1:17 PM

#

there's a huge difference

#

their purpose was performing io on a physical address

#

the new ones will be for MMIO abstraction

#

and they will have a default implementation

#

(and obviously work on virtual addresses instead)

north holly Dec 26, 2024, 1:37 PM

#

fiery turtle alright merged the bugfixes for GPE handlers and clang for windows builds (cc <@...

time to bump my uacpi commit and hope nothing goes to shit

fiery turtle Dec 26, 2024, 1:38 PM

#

barely anything changed lol

north holly Dec 26, 2024, 1:39 PM

#

reminder, this is obos

#

ok it boots on qmeu, I couldn't be bothered to test anything else atm

fiery turtle Dec 26, 2024, 1:42 PM

#

@north holly thoughts on this aml

While ((GP50 || (GP51 || (GP52 || (GP53 || (GP54 || (
    GP55 || (GP56 || (GP57 || (GP60 || (GP61 || (GP62 || (GP63 || 
    (GP66 || (GP67 || (GP70 || (GP71 || (GP93 || (GP94 || (
    GPAP || (GPBP || (GPCP || (GPDP || (GPEP || (GPFP || (GPGP || 
    (GPHP || (FCDP || (FTVP || (FQSP || (GP96 || (GP97 || (
    GPD0 || (GPD1 || (GPD2 || (GPD3 || (GPD4 || (BPFE || (TPBP || 
    (ECDS || (B1ST || B2ST)))))))))))))))))))))))))))))))))))))))))

slim panther Dec 26, 2024, 1:43 PM

#

fent

fiery turtle Dec 26, 2024, 1:43 PM

#

if you make a recursive aml interpreter the call stack would be

parse_logical_or
parse_logical_or
parse_logical_or
..
parse_logical_or
parse_if

#

pretty confident this would exhaust any kernel stack

fiery turtle Dec 26, 2024, 1:46 PM

#

fiery turtle <@841762704766140486> thoughts on this aml ``` While ((GP50 || (GP51 || (GP52 ||...

@pine yew FYI, here's a real hw example

north holly Dec 26, 2024, 1:46 PM

#

exception thrown: ur aml is dogshit (so is your interpreter) and it caused your interpreter to stack overflow

north holly Dec 26, 2024, 1:46 PM

#

fiery turtle <@128111853304807425> FYI, here's a real hw example

was that some random guy who wanted to make an AML interpreter

fiery turtle Dec 26, 2024, 1:46 PM

#

This is the person behind the rust ACPI crate

north holly Dec 26, 2024, 1:46 PM

#

ah

fiery turtle Dec 26, 2024, 1:48 PM

#

whats interesting is monkuous' hw also tells you the physical location of every xHCI port

north holly Dec 26, 2024, 1:49 PM

#

yeah because that's obviously needed for power mangement

fiery turtle Dec 26, 2024, 1:49 PM

#

lol

#

its nice for drawing diagrams in userspace etc

#

that extra integration

deft canopy Dec 26, 2024, 2:11 PM

#

fiery turtle if you make a recursive aml interpreter the call stack would be ``` parse_logica...

Nice lmao

left orbit Dec 26, 2024, 2:12 PM

#

fiery turtle <@128111853304807425> FYI, here's a real hw example

what machine did that aml come from

fiery turtle Dec 26, 2024, 2:14 PM

#

monkuous' laptop

left orbit Dec 26, 2024, 2:16 PM

#

someone should lose their job for that

fiery turtle Dec 26, 2024, 2:16 PM

#

https://laptopmedia.com/review/toshiba-portege-r30-a-review-a-reliable-and-affordable-business-companion/

LaptopMedia.com

LaptopMedia Team

Toshiba Portege R30-A review - a reliable and affordable business c...

Every now and then a perfectly good devic...

#

reliable

#

not with that aml

pine yew Dec 26, 2024, 2:17 PM

#

fiery turtle <@128111853304807425> FYI, here's a real hw example

seeing that in the wild does make me lol at the recursive approach yeah

#

if NT's interpreter was less resilient the firmware devs wouldn't be able to get away with that dogshit though

fiery turtle Dec 26, 2024, 2:17 PM

#

true

pine yew Dec 26, 2024, 2:18 PM

#

rewriting the rust crate is on my todo list after feb-ish

fiery turtle Dec 26, 2024, 2:18 PM

#

big

dusky glade Dec 26, 2024, 3:25 PM

#

fiery turtle <@841762704766140486> thoughts on this aml ``` While ((GP50 || (GP51 || (GP52 ||...

fein

#

i wanna die from this aml

#

thanks

fiery turtle Dec 26, 2024, 4:29 PM

#

❯ grep -rni '= Timer' -l
Convertible/ASUSTek Computer/Q325/Q325UAR/7E4A9E65FDE9
Convertible/ASUSTek Computer/ZenBook/ZenBook UX563FD_UX563FD/4965BD15F4B6
Convertible/Dell/Latitude/Latitude 7400 2-in-1/5DA0C196CB26
Convertible/Hewlett-Packard/ENVY/ENVY x360 2-in-1 Laptop 15-ew0xxx/211A1085E85B
Convertible/Hewlett-Packard/ENVY/ENVY x360 Convertible 15-dr1xxx/6FE4CE9270F1
Convertible/Lenovo/Yoga/Yoga C640-13IML 81UE/91A793680B40
Notebook/ASUSTek Computer/ASUS/ASUS TUF Gaming F15 FX506HC_FX566HC/2418D4F13406
Notebook/ASUSTek Computer/VivoBook_ASUSLaptop/VivoBook_ASUSLaptop K6500ZE_K6500ZE/674A4C791E8B
Notebook/ASUSTek Computer/ZenBook/Zenbook UX5401ZAS_UX5401ZAS/297345551BD5
Notebook/Avell High Performance/A70/A70 MOB/16859F37F144
Notebook/Dell/Latitude/Latitude 5511/A37FB9368F2A
Notebook/Dell/Precision/Precision 3571/AD37470CEC0D
Notebook/Dell/Precision/Precision 7550/FE48AAC0D405
Notebook/Gigabyte Technology/AORUS/AORUS 15G KB/A7DC3F8318A8
Notebook/Hewlett-Packard/EliteBook/EliteBook 840 G7 Notebook PC/05F5CA53EEBB
Notebook/Hewlett-Packard/EliteBook/EliteBook 850 G8 Notebook PC/5BEFC8C16D82
Notebook/Hewlett-Packard/ENVY/ENVY Laptop 13-aq0xxx/A34009921DC1
Notebook/Hewlett-Packard/ZHAN/ZHAN 99 Mobile Workstation G1/CB1EEAC36F91
Notebook/HUAWEI/CREF-XX/CREF-XX/651E7E8E95D2
Notebook/Lenovo/IdeaPad/IdeaPad L340-15IRH Gaming 81LK/5B5E032BF08D
Notebook/Lenovo/Legion/Legion 7 16IAX7 82TD/23401686E604
Notebook/Lenovo/Legion/Legion Y540-15IRH-PG0 81SY/6658FB40979F
Notebook/Lenovo/Legion/Legion Y7000 81FW/814C5DAC1BEE
Notebook/Lenovo/ThinkPad/ThinkPad E15 Gen 2 20TD0005MH/188EB681251A
Notebook/Lenovo/ThinkPad/ThinkPad L15 Gen 2 20X4S27200/CCA34DA7A4E9
Notebook/Lenovo/ThinkPad/ThinkPad P1 Gen 4i 20Y4S0KS00/4012A98BFAA3
Notebook/Lenovo/ThinkPad/ThinkPad P15s Gen 2i 20W6001VUS/C49AF6F23385
Notebook/Lenovo/ThinkPad/ThinkPad T14s Gen 1 20T1S2WF00/35B0F227248A
Notebook/Lenovo/ThinkPad/ThinkPad T490s 20NX000DRT/14F305E2ED3B
Notebook/Lenovo/ThinkPad/ThinkPad X1 Carbon Gen 8 20UAS89500/282408B7FDEC
Notebook/Lenovo/ThinkPad/ThinkPad X1 Carbon Gen 9 20XW007YPB/C06A0E31B5D6
Notebook/Lenovo/Yoga/Yoga Slim 7 ProX 14IAH7 82TK/BD603C52BF3B
Notebook/Notebook/NLx0/NLx0MU/17E69D6CB4D0
Notebook/PC Specialist/NS50/NS50MU/30E58AFD506A
Notebook/Teclast/F15Plus/F15Plus 2/5D2F5E56BCAB
Notebook/Timi/RedmiBook/RedmiBook Pro 15S/3509D1F696E7
Tablet/Microsoft/Surface/Surface Pro 3/074C5C2E091B

Stats about AML u didnt care about

#

about 10% of blobs use TimerOp

#

Notebook/Dell/Precision/Precision 3571/AD37470CEC0D:103573:    Debug = Timer
Notebook/Dell/Precision/Precision 3571/AD37470CEC0D:103623:    Debug = Timer
Notebook/Dell/Precision/Precision 3571/AD37470CEC0D:139046:    Debug = Timer
Notebook/Dell/Precision/Precision 3571/AD37470CEC0D:139749:    Debug = Timer

Some literally spam it to stdout

rustic compass Dec 26, 2024, 5:06 PM

#

@fiery turtle uacpi-rs builds now on uacpi master branch, you just gotta ignore the billion warnings meme

fiery turtle Dec 26, 2024, 5:07 PM

#

lemme guess, the stupid uninitialized struct thing?

vale isle Dec 26, 2024, 5:13 PM

#

wow linux w/uacpi is actually in the works now???

fiery turtle Dec 26, 2024, 5:15 PM

#

it boots but there's still thousands of acpica api calls that must be changed

vale isle Dec 26, 2024, 5:16 PM

#

omg

#

is there a repository

#

i can't believe this happened overnight

slim panther Dec 26, 2024, 5:16 PM

#

wait what

#

linux uses acpica?

vale isle Dec 26, 2024, 5:17 PM

#

i will have roughly 5 days of free time in the semester break

#

i would love to try uacpi linux on my hardware

#

and maybe fix a few bugs

hollow elm Dec 26, 2024, 5:17 PM

#

slim panther linux uses acpica?

what else would it be using

slim panther Dec 26, 2024, 5:17 PM

#

idk something in house

vale isle Dec 26, 2024, 5:18 PM

#

hollow elm what else would it be using

uacpi ofc

fiery turtle Dec 26, 2024, 5:19 PM

#

vale isle i can't believe this happened overnight

like, we just implemented kernel api and called uacpi_initialize lol

#

its not that big of a deal

#

all drivers are still calling into acpica

#

they work alongside

north holly Dec 26, 2024, 5:20 PM

#

I wonder if uacpi linux will fix the bug where my friend's laptop starts smoking meme

vale isle Dec 26, 2024, 5:20 PM

#

oh but that simply means the work was done way before this with uacpi's kernel api being similar to acpica i think

vale isle Dec 26, 2024, 5:20 PM

#

north holly I wonder if uacpi linux will fix the bug where my friend's laptop starts smoking...

i wonder if the acpi errors on boot on my laptop is a linux/acpica issue or broken fw

fiery turtle Dec 26, 2024, 5:21 PM

#

well if u give me the dump and errors i can tell u

vale isle Dec 26, 2024, 5:21 PM

#

my dump is on the repo already

#

the errors are gibberish to me

fiery turtle Dec 26, 2024, 5:21 PM

#

oh it is?

#

what are the errors

vale isle Dec 26, 2024, 5:21 PM

#

i do not remember the errors at all

fiery turtle Dec 26, 2024, 5:22 PM

#

ah sad

vale isle Dec 26, 2024, 5:22 PM

#

will check when I'm on my laptop sometime in the next 3 montjs

#

if i don't forget

fiery turtle Dec 26, 2024, 5:22 PM

#

btw linux scores 3rd on the performance leaderboard

#

behind managarm

#

or 4th

vale isle Dec 26, 2024, 5:22 PM

#

wow that's interesting

#

what affects the performance btw?

#

other than hardware

#

locking and stuff?

#

and allocation

fiery turtle Dec 26, 2024, 5:23 PM

#

yeah, locks, mem* functions, allocator performance

vale isle Dec 26, 2024, 5:23 PM

#

ah

#

i would except linux to do better lol

fiery turtle Dec 26, 2024, 5:23 PM

#

i think best we got out of it is 3.4M ops/s, whereas e.g. proxima gets 5.2M at peak

slim panther Dec 26, 2024, 5:23 PM

#

turns out a complex project is not as fast as a really simple one

left orbit Dec 26, 2024, 5:24 PM

#

thats what i was talking about yesterday 😅

#

it is still pretty fast though

vale isle Dec 26, 2024, 5:24 PM

#

i have always wanted to host a build server

#

remind me in a year

#

to set up something

#

and benchmark homebrew oses

#

on every commit maybe

#

or every release if they do releases

fiery turtle Dec 26, 2024, 5:25 PM

#

lol

#

sure thatd be cool

gentle peak Dec 26, 2024, 5:25 PM

#

I got proxima to 5.398M by making my heap leak memory (ish, since new allocations can still use the "leaked" memory, it's just not returned by the pmm)

#

unfortunately I could not get it to 5.4M

left orbit Dec 26, 2024, 5:26 PM

#

margin of error meme

#

5.4M it is

fiery turtle Dec 26, 2024, 5:27 PM

#

lol

#

leaking memory is still kinda unfair tho lol

gentle peak Dec 26, 2024, 5:27 PM

#

what I mean by "leaking memory" is pretty much the same thing most heaps do

#

most heaps never shrink and return memory to the os (pmm in this case), only ever expanding

#

that's what I mean here

fiery turtle Dec 26, 2024, 5:28 PM

#

ah

gentle peak Dec 26, 2024, 5:30 PM

#

but returning memory to the pmm is a property I'd really like to have, which is why I'm not removing the shrinking and why I'm not counting that result or submitting it for the leaderboard

fiery turtle Dec 26, 2024, 5:30 PM

#

fair

rustic compass Dec 26, 2024, 5:48 PM

#

fiery turtle lemme guess, the stupid uninitialized struct thing?

Mostly

flat badge Dec 26, 2024, 6:11 PM

#

meme frigg outperforms linux, confirmed meme

fiery turtle Dec 26, 2024, 6:14 PM

#

yeah frigg + phd memcpy do a solid job

median crest Dec 26, 2024, 6:17 PM

#

Year of managarm desktop soon

kindred beacon Dec 26, 2024, 6:17 PM

#

yes

left orbit Dec 26, 2024, 7:24 PM

#

damn @fiery turtle i get a weird crash in uacpi and i'm not sure why, you think you could help me figure it out? i'm pretty out of date but i don't think that matters, i crash on interpreter.c:66 which is c DYNAMIC_ARRAY_WITH_INLINE_STORAGE_IMPL( op_context_array, struct op_context, static )

and the cr2 is 0x905cf which is even more confusing lol

fiery turtle Dec 26, 2024, 7:24 PM

#

allocator corruption probably

left orbit Dec 26, 2024, 7:24 PM

#

my uacpi_kernel_alloc returns addresses with hhdm applied

#

and it's a bump allocator

fiery turtle Dec 26, 2024, 7:24 PM

#

alignment bug?

left orbit Dec 26, 2024, 7:24 PM

#

size = align_up(size, alignof(max_align_t));

fiery turtle Dec 26, 2024, 7:24 PM

#

OR

#

you applying some uacpi option only to some source files

left orbit Dec 26, 2024, 7:25 PM

#

nah i apply it to the entire project

#

ill log my return values

#

from the kernel apis

fiery turtle Dec 26, 2024, 7:25 PM

#

nyaux had a weird crash in dynamic array release was related to that

left orbit Dec 26, 2024, 7:25 PM

#

kernel_c_args = [
  '-DUACPI_FORMATTED_LOGGING=1',
  # ...

#

unfortunately that ain't it

fiery turtle Dec 26, 2024, 7:26 PM

#

logging is probably not related to that

left orbit Dec 26, 2024, 7:26 PM

#

i hope my prints will reveal something

#

yeah i know

#

i'm just saying that i apply it project-wide

fiery turtle Dec 26, 2024, 7:26 PM

#

ye

#

also if u have the entire backtrace that would be useful

#

and inspect the array state with gdb as well

left orbit Dec 26, 2024, 7:27 PM

#

i don't do backtraces unfortunately, sorry

#

i'' hack that together if prints don't reveal anything

fiery turtle Dec 26, 2024, 7:27 PM

#

just connect gdb lol

fiery turtle Dec 26, 2024, 7:29 PM

#

left orbit `size = align_up(size, alignof(max_align_t));`

btw that doesnt guarantee anything

left orbit Dec 26, 2024, 7:29 PM

#

well prints don't really tell me much so i'll attach gdb

fiery turtle Dec 26, 2024, 7:29 PM

#

is your bump allocator storage just a char array?

left orbit Dec 26, 2024, 7:29 PM

#

nope

#

i allocate memory from my pmm

fiery turtle Dec 26, 2024, 7:30 PM

#

ah

left orbit Dec 26, 2024, 7:30 PM

#

ill try a static array

fiery turtle Dec 26, 2024, 7:30 PM

#

one crash we had was because zig was adding extra alignment checks to c code lol

#

via some clang flags or whatever

left orbit Dec 26, 2024, 7:31 PM

#

well i am using C so that shouldn't affect me meme

#

but i do compile using clang

fiery turtle Dec 26, 2024, 7:31 PM

#

nah its not that, that was dying on a ud2

left orbit Dec 26, 2024, 7:31 PM

#

ah i see

#

yeah zig likes to do that

#

also i think it might have been the allocator lol

#

using a static char array it gets much further it seems

#

slow as balls because i print in every kernel api

fiery turtle Dec 26, 2024, 7:32 PM

#

how do u fuck up a bump allocator 💀

left orbit Dec 26, 2024, 7:32 PM

#

?????????

#

IDK

fiery turtle Dec 26, 2024, 7:32 PM

#

lmao

left orbit Dec 26, 2024, 7:32 PM

#

maybe my memory manager is fucked up?

fiery turtle Dec 26, 2024, 7:32 PM

#

oberrow had this problem as well in rewrite 22

left orbit Dec 26, 2024, 7:32 PM

#

wouldn't be surprised

#

i never tested it properly

#

also uacpi is freeing lots of 0x0's

#

that normal?

fiery turtle Dec 26, 2024, 7:33 PM

#

maybe you're returning some sort of reserved memory

#

yeah

#

its a no-op

left orbit Dec 26, 2024, 7:33 PM

#

yay

#

time to update

#

so i can get the cool thing

fiery turtle Dec 26, 2024, 7:34 PM

#

wait uh

#

where are the numbers

left orbit Dec 26, 2024, 7:34 PM

#

yeah im on an old commit

fiery turtle Dec 26, 2024, 7:34 PM

#

damn, this is like really old

left orbit Dec 26, 2024, 7:34 PM

#

yeahhh

#

20554599e65b3a98d1d77fa2a88b9af49815da5b

fiery turtle Dec 26, 2024, 7:34 PM

#

left orbit yay

what was the bug?

left orbit Dec 26, 2024, 7:34 PM

#

¯_(ツ)_/¯

fiery turtle Dec 26, 2024, 7:34 PM

#

just fixed itself?

left orbit Dec 26, 2024, 7:34 PM

#

im running with the shitty bump™️

#

and that works

#

so i'll investigate

#

but first uacpi update

fiery turtle Dec 26, 2024, 7:35 PM

#

hmm

#

so like u did nothing and now it works?

left orbit Dec 26, 2024, 7:35 PM

#

no no i replaced my "proper" bump with one that uses a static array

#

it's probably something funky going on with my memory mnaager

#

or maybe my bump just has some stupid bug lol

fiery turtle Dec 26, 2024, 7:36 PM

#

ahh

#

show it meme

left orbit Dec 26, 2024, 7:36 PM

#

https://github.com/48cf/osdev/blob/main/kernel/src/alloc.c

fiery turtle Dec 26, 2024, 7:37 PM

#

yeah its definitely a pmm bug

#

logic lgtm

#

unless your align_up is bugged idk

left orbit Dec 26, 2024, 7:38 PM

#

nah it's not

#

pmm bug but i will investigate in a little bit

#

currently fixing up kernel api

fiery turtle Dec 26, 2024, 7:38 PM

#

i dont even know what changed since that commit

left orbit Dec 26, 2024, 7:39 PM

#

no raw_io/mem_read/write

#

thats for sure

fiery turtle Dec 26, 2024, 7:39 PM

#

btw u even have realloc, u actually use that api in the kernel?

left orbit Dec 26, 2024, 7:39 PM

#

also ticks->ns ☠️

fiery turtle Dec 26, 2024, 7:39 PM

#

💀

left orbit Dec 26, 2024, 7:39 PM

#

fiery turtle btw u even have realloc, u actually use that api in the kernel?

i dont think so

fiery turtle Dec 26, 2024, 7:39 PM

#

linux has like 50 i think

#

in the entire kernel

#

compare to idk how many normal kmalloc calls

#

or no, 135 files vs 2684

left orbit Dec 26, 2024, 7:41 PM

#

FACS 0x0000000007FE0000 00000040
successfully loaded 1 AML blob, 1705 ops in 6ms (avg 277825/s)
unable to install SCI interrupt handler: unimplemented```

#

wow that's shit

fiery turtle Dec 26, 2024, 7:41 PM

#

hmm lol

#

well if thats with tcg

left orbit Dec 26, 2024, 7:41 PM

#

nope kvm lmaoo

fiery turtle Dec 26, 2024, 7:41 PM

#

💀

left orbit Dec 26, 2024, 7:41 PM

#

half of the kernel api stubbed out

#

and it's still 270k

fiery turtle Dec 26, 2024, 7:41 PM

#

literally how ☠️

left orbit Dec 26, 2024, 7:42 PM

#

😃

fiery turtle Dec 26, 2024, 7:42 PM

#

is your hhdm mapped cached?

left orbit Dec 26, 2024, 7:42 PM

#

no optimizations

#

and uh i dont think so

fiery turtle Dec 26, 2024, 7:42 PM

#

so uncached?

left orbit Dec 26, 2024, 7:42 PM

#

but getting rid of assert(bump_offset + size <= PAGE_SIZE_L2); in alloc gets it up to 580k lol

gentle peak Dec 26, 2024, 7:42 PM

#

is that built in debug mode or release mode

left orbit Dec 26, 2024, 7:42 PM

#

im pretty sure debug

fiery turtle Dec 26, 2024, 7:42 PM

#

ohh

#

well now do O2

left orbit Dec 26, 2024, 7:42 PM

#

-O0 -g

#

yeaaa lmao

#

let me try

fiery turtle Dec 26, 2024, 7:43 PM

#

inb4 6 mil

left orbit Dec 26, 2024, 7:43 PM

#

triple faults

#

fiery turtle Dec 26, 2024, 7:43 PM

#

lmao

left orbit Dec 26, 2024, 7:45 PM

#

hacker@raptor:~/sources/osdev$ objdump -d build/kernel | grep -B 10 ffffffff8001c6cc
ffffffff8001c6c1 <_write_cr0>:
ffffffff8001c6c1:       0f 22 c7                mov    %rdi,%cr0
ffffffff8001c6c4:       c3                      ret

ffffffff8001c6c5 <_write_cr3>:
ffffffff8001c6c5:       0f 22 df                mov    %rdi,%cr3
ffffffff8001c6c8:       c3                      ret

ffffffff8001c6c9 <_write_cr4>:
ffffffff8001c6c9:       0f 22 e7                mov    %rdi,%cr4
ffffffff8001c6cc:       c3                      ret```

#

the crash rip points at the ret in _write_cr4

#

it's a pf with e=0000

#

how does writing to cr4 generate a page fault

fiery turtle Dec 26, 2024, 7:46 PM

#

what did u do 💀

gentle peak Dec 26, 2024, 7:46 PM

#

then the cr4 write probably changed the expected page table format

hollow elm Dec 26, 2024, 7:46 PM

#

inb4 NX kernel .text

fiery turtle Dec 26, 2024, 7:47 PM

#

left orbit the crash rip points at the ret in _write_cr4

whats the cr2 value tho

left orbit Dec 26, 2024, 7:47 PM

#

CR2=ffffffff8001c6cc

#

so the rip

fiery turtle Dec 26, 2024, 7:47 PM

#

oh u either changed the PT format or NXed yourself

left orbit Dec 26, 2024, 7:47 PM

#

but if i comment out the _write_cr4 at the end of init_cpu_features

#

it works?

fiery turtle Dec 26, 2024, 7:48 PM

#

do u have uninitialized variables?

left orbit Dec 26, 2024, 7:48 PM

#

nope

#

not that i know of

fiery turtle Dec 26, 2024, 7:48 PM

#

and do u read it before updating

left orbit Dec 26, 2024, 7:48 PM

#

https://github.com/48cf/osdev/blob/7a656b03c17661f6f427bc834423dba3237a28c6/kernel/src/cpu.c#L107

#

this _write_cr4 is faulting

#

and it doesn't set any of these bits above, i checked

#

OH WOW IT'S TRYING TO WRITE 0

fiery turtle Dec 26, 2024, 7:50 PM

#

hmm

left orbit Dec 26, 2024, 7:51 PM

#

ok, so i print cr4 after reading it initially, after writing it the first time and before writing it the 2nd time

#

i get 0x20, 0x220 and 0x0??

#

that is some mad stack corruption i am smelling

fiery turtle Dec 26, 2024, 7:51 PM

#

cpuid corrupting the stack?

left orbit Dec 26, 2024, 7:51 PM

#

yueah

#

it actualy is

#

printf("before cpuid cr4: %llx\n", cr4);
_cpuid(0x7, 0, &cpuid);
printf("after cpuid cr4: %llx\n", cr4);

#

before cpuid cr4: 220
after cpuid cr4: 0```

#

that is crazy

fiery turtle Dec 26, 2024, 7:52 PM

#

wait you arent preserving any regs there, are they all scratch?

left orbit Dec 26, 2024, 7:53 PM

#

what the fuck am i looking at??????

#

WHAT IS THIS CFG

fiery turtle Dec 26, 2024, 7:53 PM

#

lol

left orbit Dec 26, 2024, 7:53 PM

#

i have a feeling debugoptimized with lto is not gonna be fun to debug

fiery turtle Dec 26, 2024, 7:55 PM

#

2. Clobbering of %ebx: The cpuid instruction modifies the %eax, %ebx, %ecx, and %edx registers. If this code is expected to adhere to System V AMD64 ABI calling conventions, keep in mind that %rbx is a callee-saved register. If %rbx is used in the caller, it should be preserved. You might need to save and restore %rbx around the cpuid instruction.

Idk chat gpt said this

#

lmao

#

is this true?

left orbit Dec 26, 2024, 7:55 PM

#

Functions preserve the registers rbx, rsp, rbp, r12, r13, r14, and r15; while rax, rdi, rsi, rdx, rcx, r8, r9, r10, r11 are scratch registers.

#

hmmm

#

i might be dumb

fiery turtle Dec 26, 2024, 7:55 PM

#

looks like u should preserve it

left orbit Dec 26, 2024, 7:56 PM

#

guess whose cr4 isn't getting corrupted anymore

#

(thank you)

fiery turtle Dec 26, 2024, 7:56 PM

#

np lol

left orbit Dec 26, 2024, 7:56 PM

#

successfully loaded 1 AML blob, 1705 ops in 0ms (avg 2418107/s)

#

yep as expected

fiery turtle Dec 26, 2024, 7:56 PM

#

nice

left orbit Dec 26, 2024, 7:56 PM

#

with half of the kernel api stubbed out

#

still worse than linux

#

crazy

fiery turtle Dec 26, 2024, 7:57 PM

#

somehow 3 times slower than proxima with everything implemented

left orbit Dec 26, 2024, 7:57 PM

#

probably due to "shitty" mem* functions

fiery turtle Dec 26, 2024, 7:57 PM

#

oh yeah

fiery turtle Dec 26, 2024, 7:57 PM

#

left orbit still worse than linux

yeah they have very optimized mem

gentle peak Dec 26, 2024, 7:58 PM

#

just tested, proxima with uacpi's builtin mem* funcs is 3.4M/s

left orbit Dec 26, 2024, 7:59 PM

#

HOOOOOOOOOW

#

WHAT

#

successfully loaded 1 AML blob, 1705 ops in 0ms (avg 2884452/s)

#

with inline asm memcpy/memset

hollow elm Dec 26, 2024, 8:02 PM

#

just need a fast cpu :^)

left orbit Dec 26, 2024, 8:02 PM

#

successfully loaded 1 AML blob, 1705 ops in 0ms (avg 3045507/s)

#

random improvement because rerun

#

this largely assumes my time keeping is good™️

fiery turtle Dec 26, 2024, 8:02 PM

#

gentle peak just tested, proxima with uacpi's builtin mem* funcs is 3.4M/s

Who said built-in stuff was crap KEKW

left orbit Dec 26, 2024, 8:02 PM

#

fiery turtle Who said built-in stuff was crap <:KEKW:913708372329644042>

-# you

fiery turtle Dec 26, 2024, 8:02 PM

#

(I did)

gentle peak Dec 26, 2024, 8:02 PM

#

(it is)

fiery turtle Dec 26, 2024, 8:02 PM

#

left orbit -# you

Lmao

gentle peak Dec 26, 2024, 8:03 PM

#

gentle peak just tested, proxima with uacpi's builtin mem* funcs is 3.4M/s

if i replace #define memset __builtin_memset with size_t memset(void *dest, int value, size_t n); it goes down to 2.8M/s

fiery turtle Dec 26, 2024, 8:03 PM

#

gentle peak just tested, proxima with uacpi's builtin mem* funcs is 3.4M/s

I dont understand what it is you do that it's so fast when iretq literally has everything stubbed AND fast mem AND a bump allocator

left orbit Dec 26, 2024, 8:04 PM

#

ok when i initialize uacpi after i switch to and from a test thread that sleeps for 1s i get consistent 3.02M

gentle peak Dec 26, 2024, 8:04 PM

#

mostly because calloc doesn't use uacpi builtins

fiery turtle Dec 26, 2024, 8:04 PM

#

Maybe bump allocator is worse because cold cache

gentle peak Dec 26, 2024, 8:04 PM

#

gentle peak if i replace `#define memset __builtin_memset` with `size_t memset(void *dest, i...

and if i use uacpi_memset for calloc it goes back up to 3.4M/s

#

wtf?

fiery turtle Dec 26, 2024, 8:05 PM

#

Uacpi has PhD memset

gentle peak Dec 26, 2024, 8:05 PM

#

calling a function that uses rep stosb is apparently slower than calling uacpi builtin memset

#

probably due to LTO and clang turning it into PhD memset

fiery turtle Dec 26, 2024, 8:05 PM

#

Lmao

gentle peak Dec 26, 2024, 8:05 PM

#

because calling rep stosb can't be LTO'd since that's implemented in asm

fiery turtle Dec 26, 2024, 8:05 PM

#

Makes sense

left orbit Dec 26, 2024, 8:06 PM

#

i should implement it in assembly too

#

wait wdym it can't be LTOd

#

really?

fiery turtle Dec 26, 2024, 8:06 PM

#

Whats needed to support LTO in a kernel

gentle peak Dec 26, 2024, 8:06 PM

#

nothing

#

well, patience when compiling

#

and not having UB

#

but other than that nothing

fiery turtle Dec 26, 2024, 8:06 PM

#

Hmm

median crest Dec 26, 2024, 8:06 PM

#

I should try astral with it

left orbit Dec 26, 2024, 8:06 PM

#

it's great

gentle peak Dec 26, 2024, 8:06 PM

#

left orbit wait wdym it can't be LTOd

LTO is basically just that the compiler outputs IR instead of machine code

left orbit Dec 26, 2024, 8:06 PM

#

uncovered the first bug in my kernel 👍

hollow elm Dec 26, 2024, 8:06 PM

#

left orbit wait wdym it can't be LTOd

how could it be? LTO is done by compiling c/c++ into ir and writing the ir into object files

gentle peak Dec 26, 2024, 8:06 PM

#

and that IR is turned into machine code by the linker which has global knowledge

left orbit Dec 26, 2024, 8:07 PM

#

hollow elm how could it be? LTO is done by compiling c/c++ into ir and writing the ir into ...

yeah i didn't know

gentle peak Dec 26, 2024, 8:07 PM

#

assembly can't be turned into IR because that's already effectively machine code

left orbit Dec 26, 2024, 8:07 PM

#

i thought it was some linker magic™️

fiery turtle Dec 26, 2024, 8:07 PM

#

Wait so if I have standalone asm files I cant lto anything?

left orbit Dec 26, 2024, 8:07 PM

#

well, it is, but not as magic as i thought meme

left orbit Dec 26, 2024, 8:07 PM

#

fiery turtle Wait so if I have standalone asm files I cant lto anything?

can but not the symbols in them

hollow elm Dec 26, 2024, 8:07 PM

#

fiery turtle Wait so if I have standalone asm files I cant lto anything?

lto only affects the c/c++ files built with lto

gentle peak Dec 26, 2024, 8:07 PM

#

fiery turtle Wait so if I have standalone asm files I cant lto anything?

you can still LTO but if you call an asm function it won't be inlined

#

that kinda thing

fiery turtle Dec 26, 2024, 8:07 PM

#

Ah

#

Thats fine then

hollow elm Dec 26, 2024, 8:08 PM

#

just like you can lto a program that calls a shared library function

gentle peak Dec 26, 2024, 8:08 PM

#

oh btw

#

when building in debug mode (-O0 no LTO) proxima gets 1M/s

fiery turtle Dec 26, 2024, 8:08 PM

#

Lmao

#

Still somehow 3x from iretq

left orbit Dec 26, 2024, 8:09 PM

#

Sadge

fiery turtle Dec 26, 2024, 8:09 PM

#

Bump allocator fault

#

Every allocation is cold memory

gentle peak Dec 26, 2024, 8:10 PM

#

the github source seems outdated but from what i'm seeing there the main slowdown might be that you implement things like pio and mmio in standalone asm files instead of with inline asm

median crest Dec 26, 2024, 8:10 PM

#

That memory needs some empathy

left orbit Dec 26, 2024, 8:10 PM

#

void*
alloc(size_t size)
{
   size = align_up(size, alignof(max_align_t));

   void* ptr = (void*)(bump_buffer + bump_offset);

   bump_offset += size;

   memset(ptr, 0, size);

   return ptr;
}```

vast kestrel Dec 26, 2024, 8:10 PM

#

gentle peak because calling rep stosb can't be LTO'd since that's implemented in asm

Inline asm?

left orbit Dec 26, 2024, 8:10 PM

#

how can that be at fault :(

vast kestrel Dec 26, 2024, 8:11 PM

#

I am pretty sure inline asm can be inlined

gentle peak Dec 26, 2024, 8:11 PM

#

vast kestrel Inline asm?

if it were inline asm it could be lto'd yeah but mine are in a standalone .S file

vast kestrel Dec 26, 2024, 8:11 PM

#

Then don't :^)

#

Also something something small and unaligned buffers being slow with the string instructions on older machines

gentle peak Dec 26, 2024, 8:12 PM

#

fiery turtle Every allocation is cold memory

not quite the same thing but if i make kfree a no-op proxima goes from 5.1M/s to 3.5M/s

#

and if i make the allocator unconditionally zero memory it goes from 5.1M/s to 4.5M/s

#

if i do both it goes from 5.1M/s to 3.1M/s

#

so stuff like that is pretty significant

median crest Dec 26, 2024, 8:15 PM

#

Bruh enabling lto in astral gives a missing popcount refence

#

Ill look into that in a bit

fiery turtle Dec 26, 2024, 8:15 PM

#

gentle peak not quite the same thing but if i make `kfree` a no-op proxima goes from 5.1M/s ...

Makes sense

#

I guess uacpi has a pattern of freeing and reallocating quickly

#

Which makes memory reuse fast

gentle peak Dec 26, 2024, 8:16 PM

#

i wonder how fast it'd get if i made mutexes noops

#

6.4M/s

fiery turtle Dec 26, 2024, 8:18 PM

#

Bruh

left orbit Dec 26, 2024, 8:22 PM

#

.global memcpy
.type memcpy,"function"
memcpy:
   cld
   push %rdi
   mov %rdx, %rcx
   rep movsb
   pop %rax
   ret
.size memcpy, .-memcpy

.global memset
.type memset,"function"
memset:
   cld
   push %rdi
   mov %rdx, %rcx
   mov %sil, %al
   rep stosb
   pop %rax
   ret
.size memset, .-memset

anything wrong with these?

gentle peak Dec 26, 2024, 8:23 PM

#

cld is unnecessary since sysv abi requires it upon function call

left orbit Dec 26, 2024, 8:23 PM

#

fair

gentle peak Dec 26, 2024, 8:23 PM

#

and the pushing can be replaced with copying to a scratch register that movsb/stosb don't touch

#

which removes an extra memory access

left orbit Dec 26, 2024, 8:23 PM

#

perfect

#

but still, the semantics?

mortal yoke Dec 26, 2024, 8:23 PM

#

you can just mov rdi to rax at the start

gentle peak Dec 26, 2024, 8:24 PM

#

these are mine for reference ```
memcpy:
mov %rdi, %rax
mov %rdx, %rcx
rep movsb
ret

memset:
mov %rdi, %r8
mov %rsi, %rax
mov %rdx, %rcx
rep stosb
mov %r8, %rax
ret```

left orbit Dec 26, 2024, 8:24 PM

#

i am getting weird uacpi errors

fiery turtle Dec 26, 2024, 8:24 PM

#

You seem to know your optimizations chad

gentle peak Dec 26, 2024, 8:24 PM

#

left orbit but still, the semantics?

semantically those should be fine yeah

left orbit Dec 26, 2024, 8:25 PM

#

FACS 0x0000000007FE0000 00000040
unable to lookup named object 'W1�N' within (or above) scope '\_SB_.PCI0': invalid AML name string
aborting table load due to previous error: invalid AML name string
    #0 in \()
failed to load 'DSDT' (OEM ID 'BOCHS ' OEM Table ID 'BXPC    '): invalid AML name string
loaded 1 AML blob in 7ms (1 error)```

#

like wha

hollow elm Dec 26, 2024, 8:25 PM

#

W1�N yum

left orbit Dec 26, 2024, 8:25 PM

#

bringing the old C versions of those functions makes it work again

#

ill try one at a time

fiery turtle Dec 26, 2024, 8:26 PM

#

gentle peak these are mine for reference ``` memcpy: mov %rdi, %rax mov %rdx, %rcx ...

Any reason why u do a separate file and not inline

gentle peak Dec 26, 2024, 8:26 PM

#

honestly idk

#

probably doesn't matter since the usages are via __builtin_X and the actual functions only get called if it wouldn't have been inlined anyway

#

i could try and see if it'd be better with inline asm

left orbit Dec 26, 2024, 8:27 PM

#

lol it works now? wtf

fiery turtle Dec 26, 2024, 8:28 PM

#

I wonder how the perf would change if every memcpy call is inlined to that asm

left orbit Dec 26, 2024, 8:28 PM

#

OH FUCK ME I KNOW WHY

#

I WAS USING ; FOR COMMENTS

#

IN A GNU AS SOURCE FILE

#

😭

gentle peak Dec 26, 2024, 8:28 PM

#

right now i also have memcmp implemented in ASM (though not using rep cmpsb ofc) and i wonder whether a C implementation of it would perform better

fiery turtle Dec 26, 2024, 8:28 PM

#

left orbit IN A GNU AS SOURCE FILE

What does that do

left orbit Dec 26, 2024, 8:29 PM

#

; is a semicolon

#

like in a C source file

#

// is a comment

fiery turtle Dec 26, 2024, 8:29 PM

#

Oh

left orbit Dec 26, 2024, 8:29 PM

#

so i had a bunch of random code "commented out" with semicolons

fiery turtle Dec 26, 2024, 8:29 PM

#

Lol

median crest Dec 26, 2024, 8:29 PM

#

lto does not like my pmm

#

wait I think it might be due to flanterm defining a memset too

#

#

idfk

gentle peak Dec 26, 2024, 8:31 PM

#

gentle peak i could try and see if it'd be better with inline asm

the performance is Exactly The Same!

north holly Dec 26, 2024, 8:33 PM

#

Exactly The Same™️

gentle peak Dec 26, 2024, 8:35 PM

#

median crest

your memset prototype is incorrect. i don't think it matters in this case since the call convention doesn't change but value is supposed to be int not ulong

median crest Dec 26, 2024, 8:35 PM

#

ah

#

how would you memset a region over 2gb tho

gentle peak Dec 26, 2024, 8:35 PM

#

the size is still 64 bit

#

since it's size_t

#

value is supposed to be truncated to the lowest CHAR_BIT bits by the actual memset anyway

fiery turtle Dec 26, 2024, 8:36 PM

#

median crest how would you memset a region over 2gb tho

The second param is what to memset to 💀

median crest Dec 26, 2024, 8:36 PM

#

Wait I had a brainfart nvm

#

Lmao

#

My guess from it dying that early is something relating to the memset and the page struct not being zero

dusky glade Dec 26, 2024, 8:39 PM

#

left orbit ?????????

im fucking nyaux and even i dont do that

#

💀

north holly Dec 26, 2024, 8:40 PM

#

fiery turtle oberrow had this problem as well in rewrite 22

wut

#

I don't believe I have ever fucked up a bump allocator

fiery turtle Dec 26, 2024, 8:43 PM

#

U did I think

#

Id have to search for it

hollow elm Dec 26, 2024, 8:44 PM

#

median crest

why is your memset taking an unsigned long?

north holly Dec 26, 2024, 8:44 PM

#

#1141057599584878645 message

#

found it

hollow elm Dec 26, 2024, 8:45 PM

#

hollow elm why is your memset taking an unsigned long?

memset is supposed to truncate the given value to a byte and fill the range with that so an int is more than adequate

median crest Dec 26, 2024, 8:48 PM

#

hollow elm why is your memset taking an unsigned long?

I didnt look at the type when I implemented it but its literally uust changinng to an int

fiery turtle Dec 26, 2024, 8:50 PM

#

north holly https://discord.com/channels/440442961147199490/1141057599584878645/127654618257...

Lmao

north holly Dec 26, 2024, 8:51 PM

#

fiery turtle Lmao

were you first saying it jokingly not expecting it to be true

#

lol

fiery turtle Dec 26, 2024, 8:51 PM

#

No I remembered u fucking it up lol

torpid root Dec 26, 2024, 9:00 PM

#

dusky glade im fucking nyaux and even i dont do that

you are doing WHAT to nyaux?

dusky glade Dec 26, 2024, 9:00 PM

#

torpid root you are doing WHAT to nyaux?

WHAT

torpid root Dec 26, 2024, 9:01 PM

#

https://cdn.discordapp.com/emojis/1233846746493222964.webp?size=64

kind mantle Dec 26, 2024, 9:21 PM

#

what?

calm latch Dec 26, 2024, 9:29 PM

#

fleshed

median crest Dec 26, 2024, 9:45 PM

#

what

fiery turtle Dec 26, 2024, 10:21 PM

#

+#ifdef UACPI_NATIVE_CALLOC
 /*
  * Allocate a block of memory of 'count' * 'size' bytes.
  * The returned memory block is expected to be zero-filled.
  */
 void *uacpi_kernel_calloc(uacpi_size count, uacpi_size size);
+#endif

Fuck it

#

Less potential host bugs to worry about

#

i'll just have a default implementation in terms of alloc+memset

rustic compass Dec 26, 2024, 10:31 PM

#

how about uacpi_kernel_calloc_zeroed? lets be honest, there will be at least one person who forgets to memset to zero otherwise (probably still at least 1 person even if its in the function name)

#

also is const uacpi_char *uacpi_status_to_string(uacpi_status); backed by a predefined list of names for all status variants?

#

found the big switch statement

#

not needing to cast c strings into rust strings makes it so much easier

left orbit Dec 26, 2024, 10:44 PM

#

calloc by definition zeroes the memory

slim panther Dec 26, 2024, 10:44 PM

#

rustic compass how about uacpi_kernel_calloc_zeroed? lets be honest, there will be at least one...

calloc is always zeroed

left orbit Dec 26, 2024, 10:44 PM

#

also the comment says "The returned memory block is expected to be zero-filled."

#

if you don't read the comments it's your fault

#

:)

fiery turtle Dec 26, 2024, 11:09 PM

#

like calloc is the classic C/posix api for zeroed allocation

#

so calloc_zeroed is kinda redundant lol

#

in hindsight i should've just called it alloc_zeroed of course

#

but i thought calloc was apparent enough

frank canopy Dec 26, 2024, 11:12 PM

#

i only knew the zeroed part from the comment but im also coming from zig without much of a direct C background

#

but the comment was there so it was fine

fiery turtle Dec 26, 2024, 11:13 PM

#

yeah my fault for assuming everyone knows C or has that background

frank canopy Dec 26, 2024, 11:16 PM

#

imo that - in osdev especially - if people cant read the comments/docs thats their problem. ~~maybe im just tired of fielding questions in other places from people who wont bother to read the intel x86 manual though~~

fiery turtle Dec 26, 2024, 11:16 PM

#

fair i guess

fiery turtle Dec 26, 2024, 11:53 PM

#

might just do this honestly

#

no api break for existing users as a bonus i guess

frank canopy Dec 26, 2024, 11:54 PM

#

i think my calloc shim actually does something that im not 100% sure is required that a direct alloc(count * size) + memset wouldnt which is alignment

fiery turtle Dec 26, 2024, 11:55 PM

#

does it do some special alignment tricks?

#

definitely not required by uacpi

frank canopy Dec 26, 2024, 11:55 PM

#

calloc in theory requires fundamental alignment rather than arbitrary alignment

#

and im not sure if normal alloc does

fiery turtle Dec 26, 2024, 11:56 PM

#

i think malloc is just guaranteed to return whatever the maximum alignment needed for fundamental types

#

which is enough for most cases

frank canopy Dec 26, 2024, 11:56 PM

#

yeah i think same for calloc but i didnt see it in the malloc bit lol

left orbit Dec 26, 2024, 11:56 PM

#

fiery turtle might just do this honestly

how did you make it optional?

slim panther Dec 26, 2024, 11:56 PM

#

macro

left orbit Dec 26, 2024, 11:57 PM

#

fair

fiery turtle Dec 26, 2024, 11:57 PM

#

yeah just ifdef

#

#ifndef UACPI_NATIVE_ALLOC_ZEROED
void *uacpi_kernel_alloc_zeroed(uacpi_size size)
{
    void *ptr;

    ptr = uacpi_kernel_alloc(size);
    if (uacpi_unlikely(ptr == UACPI_NULL))
        return ptr;

    uacpi_memzero(ptr, size);
    return ptr;
}
#endif

left orbit Dec 27, 2024, 12:12 AM

#

could you maybe also add a uacpi_realloc kernel api that's optional? some allocators could make use of the fact some allocations can be resized without actually resizing but idk if it's worth the hassle

frank canopy Dec 27, 2024, 12:12 AM

#

yep that might be nice if its ever used

left orbit Dec 27, 2024, 12:12 AM

#

the only place it could really be used is dynamic_array but yeah

fiery turtle Dec 27, 2024, 12:13 AM

#

Yeah it could be utilized by the dynamic array for sure

#

Wonder how many kernels have a realloc

frank canopy Dec 27, 2024, 12:13 AM

#

i do

fiery turtle Dec 27, 2024, 12:13 AM

#

(Here)

fiery turtle Dec 27, 2024, 12:13 AM

#

frank canopy i do

Nice

median crest Dec 27, 2024, 12:13 AM

#

fiery turtle ```cpp #ifndef UACPI_NATIVE_ALLOC_ZEROED void *uacpi_kernel_alloc_zeroed(uacpi_s...

so my zerod alocator will be faster

#

pweading

left orbit Dec 27, 2024, 12:13 AM

#

i will have one

fiery turtle Dec 27, 2024, 12:14 AM

#

median crest so my zerod alocator will be faster

Didnt u provide a calloc before anyway lol

median crest Dec 27, 2024, 12:14 AM

#

yeah but it just returns alloc(size * n)

fiery turtle Dec 27, 2024, 12:14 AM

#

Whats the diff?

frank canopy Dec 27, 2024, 12:15 AM

#

fiery turtle Nice

well technically zig does - zig's std provides a general-purpose allocator that just needs a backing page allocator and which has a resize function, and then the zig std provides a general realloc based on resize (resize returns true if it could resize in place)

median crest Dec 27, 2024, 12:15 AM

#

wait is that def not what I'm thinking

fiery turtle Dec 27, 2024, 12:15 AM

#

Like calloc already allocates zeroed memory

#

Its just funny api because count and size

fiery turtle Dec 27, 2024, 12:15 AM

#

frank canopy well technically zig does - zig's std provides a general-purpose allocator that ...

Oh interesting

#

Is it well suited for kernel use?

left orbit Dec 27, 2024, 12:16 AM

#

it's basically what i implemented today

#

kinda slow

median crest Dec 27, 2024, 12:16 AM

#

fiery turtle Its just funny api because count and size

oh

#

well add realloc because it would play very well with my allocator meme

fiery turtle Dec 27, 2024, 12:16 AM

#

median crest well add realloc because it would play very well with my allocator <:meme:575445...

I hear you chad

#

I guess scores might change if I add a realloc

left orbit Dec 27, 2024, 12:17 AM

#

me give good idea

#

yay

median crest Dec 27, 2024, 12:17 AM

#

I mean slab free and alloc is o(1)

fiery turtle Dec 27, 2024, 12:17 AM

#

Dynamic array resize is what might be tanking some

median crest Dec 27, 2024, 12:17 AM

#

at least on the small slabs in astral

left orbit Dec 27, 2024, 12:18 AM

#

promixma about to snap 7M

frank canopy Dec 27, 2024, 12:18 AM

#

fiery turtle Is it well suited for kernel use?

its not the absolute most high-perf thing out there but its still very good. even lets you swap out the mutex type it uses for thread safety which i do for my kernel one. im not sure all the details of what exact scheme it does other than being bucket-based. also in releasesafe and debug it can automatically detect double frees and some leaks

fiery turtle Dec 27, 2024, 12:18 AM

#

Lmao

#

Inb4 monkuous does some 4 line of code optimization for realloc that pumps out 10M/s

fiery turtle Dec 27, 2024, 12:19 AM

#

frank canopy its not the absolute most high-perf thing out there but its still very good. eve...

Nice

median crest Dec 27, 2024, 12:19 AM

#

switching my mutex to a spinlock acquire like doubles performance so once I get a real lock I will be cooking

fiery turtle Dec 27, 2024, 12:19 AM

#

Zig is cool ill try it once its more stable

fiery turtle Dec 27, 2024, 12:19 AM

#

median crest switching my mutex to a spinlock acquire like doubles performance so once I get ...

LETSFUCKINGGOOOOOO

left orbit Dec 27, 2024, 12:19 AM

#

fiery turtle Zig is cool ill try it once its more stable

better try now before they fuck it up even more

deft canopy Dec 27, 2024, 12:20 AM

#

fiery turtle Zig is cool ill try it once its more stable

It is genuinely a fairly competent language thus far, and I have fairly high hopes

fiery turtle Dec 27, 2024, 12:20 AM

#

Lmao

left orbit Dec 27, 2024, 12:20 AM

#

im totally not salty

fiery turtle Dec 27, 2024, 12:20 AM

#

What did they do

left orbit Dec 27, 2024, 12:20 AM

#

they just do dumb things

#

change things that didn't need changing

deft canopy Dec 27, 2024, 12:20 AM

#

left orbit better try now before they fuck it up even more

Recently they disabled emitting dynamic binaries in freestanding targets, which was a little frustrating but i hope this is merely a hiccup

frank canopy Dec 27, 2024, 12:21 AM

#

fiery turtle Nice

the zig generalpurposeallocator im using is used instead of malloc on basically every zig program that doesnt link libc or have more specific needs (theres also an arena allocator and object pools in the std that have the same backing allocator setup. even a fixed buffer allocator and a fixed buffer with fallback allocator for overflowing the buffer) - zig just links ntdll or does manual syscall stuff on linux to get pages allocated as needed

left orbit Dec 27, 2024, 12:21 AM

#

var variables that aren't mutated are a compile time error, and unused variables are also errors which is definitely annoying

fiery turtle Dec 27, 2024, 12:21 AM

#

Niice

left orbit Dec 27, 2024, 12:21 AM

#

and you cannot turn that off lol

fiery turtle Dec 27, 2024, 12:21 AM

#

frank canopy the zig generalpurposeallocator im using is used instead of malloc on basically ...

Sounds really tempting

frank canopy Dec 27, 2024, 12:21 AM

#

technically zig also has heapallocator for the windows heap api since thats in ntdll but most people just use generalpurposeallocator on top of the page allocator

fiery turtle Dec 27, 2024, 12:22 AM

#

I also love zigs multiline strings, I haven't seen a language that does it better

frank canopy Dec 27, 2024, 12:22 AM

#

tbh i like the unmutated var and unused const warnings

deft canopy Dec 27, 2024, 12:22 AM

#

left orbit `var` variables that aren't mutated are a compile time error, and unused variabl...

Honestly i dont really mind this

frank canopy Dec 27, 2024, 12:22 AM

#

also comptime rules

left orbit Dec 27, 2024, 12:22 AM

#

what is cool is you can use std.fmt.comptimePrint + asm :)

#

and you can literally write code that generates inline assembly

frank canopy Dec 27, 2024, 12:22 AM

#

comptime being what if constexpr was taken to the extreme

fiery turtle Dec 27, 2024, 12:22 AM

#

Yeah it's crazy

left orbit Dec 27, 2024, 12:23 AM

#

if you ever do decide try zig out for a kernel project i am more than sure i'd contribute

fiery turtle Dec 27, 2024, 12:23 AM

#

Also im biased because zig used spaces

frank canopy Dec 27, 2024, 12:23 AM

#

left orbit and you can literally write code that generates inline assembly

thats how i capture interrupt stack frames actually. each field in the saved registers struct is named after a register and ive got comptime generating the pushes and pops for my interrupt handlers

left orbit Dec 27, 2024, 12:23 AM

#

damn that is nice :^)

frank canopy Dec 27, 2024, 12:23 AM

#

once we get @unrollArgumentTuple ill be really happy, rn ive got to implement it with a bit of a comptime mess

fiery turtle Dec 27, 2024, 12:23 AM

#

frank canopy thats how i capture interrupt stack frames actually. each field in the saved reg...

Damn

left orbit Dec 27, 2024, 12:23 AM

#

i only used it for grabbing the cs offset from the iret frame to check if i need to swapgs

fiery turtle Dec 27, 2024, 12:23 AM

#

Really cool

left orbit Dec 27, 2024, 12:24 AM

#

never crossed my mind to generate the push/pops

frank canopy Dec 27, 2024, 12:24 AM

#

and it uses sizeof and comptimePrint to figure out the rsp offset of the bottom of the struct for making the pointer before calling the actual handler

left orbit Dec 27, 2024, 12:24 AM

#

just mov %rsp, %rdi tho

fiery turtle Dec 27, 2024, 12:25 AM

#

I might give it a try for a cli app to get a feel for it

north holly Dec 27, 2024, 12:25 AM

#

damn zig actually seems kinda cool

fiery turtle Dec 27, 2024, 12:25 AM

#

Ikr

frank canopy Dec 27, 2024, 12:25 AM

#

https://github.com/Khitiara/imaginarium/blob/main/src/krnl/hal/arch/idt.zig#L231-L321 this is the stuff i mentioned, im rather proud of it

north holly Dec 27, 2024, 12:25 AM

#

for a while I've thought of it as: oh some other dumb language, ignore

frank canopy Dec 27, 2024, 12:26 AM

#

i was on the verge of learning c to do osdev when i found zig and decided to give it a go instead lmao

left orbit Dec 27, 2024, 12:26 AM

#

you did not make a mistake

#

nooby made me learn osdev and zig at the same time

#

and it was so fun :^)

#

i miss old zig

fiery turtle Dec 27, 2024, 12:26 AM

#

frank canopy <https://github.com/Khitiara/imaginarium/blob/main/src/krnl/hal/arch/idt.zig#L23...

Smart

left orbit Dec 27, 2024, 12:27 AM

#

frank canopy <https://github.com/Khitiara/imaginarium/blob/main/src/krnl/hal/arch/idt.zig#L23...

do you do 32-bit x86?

frank canopy Dec 27, 2024, 12:27 AM

#

left orbit just mov %rsp, %rdi tho

my memory was bad, im using it with offsetof to get the vector for indexing the handler lol

frank canopy Dec 27, 2024, 12:27 AM

#

left orbit do you do 32-bit x86?

only long mode

left orbit Dec 27, 2024, 12:27 AM

#

%gs/%fs will always be zero

fiery turtle Dec 27, 2024, 12:28 AM

#

Zig freestanding: literal allocator
C++ freestanding: here's your stdint and fuck off, oh btw u have to implement initializer_list as well

frank canopy Dec 27, 2024, 12:28 AM

#

yeah im planning to rework this and remove those lol

left orbit Dec 27, 2024, 12:28 AM

#

sounds good

left orbit Dec 27, 2024, 12:28 AM

#

fiery turtle Zig freestanding: literal allocator C++ freestanding: here's your stdint and fuc...

not only do you get an allocator, you get like all of the standard library

frank canopy Dec 27, 2024, 12:28 AM

#

yeah

left orbit Dec 27, 2024, 12:28 AM

#

given you provide the necessary os things

north holly Dec 27, 2024, 12:29 AM

#

fiery turtle Zig freestanding: literal allocator C++ freestanding: here's your stdint and fuc...

C freestanding: here stdint, have fun

left orbit Dec 27, 2024, 12:29 AM

#

but if you provide a page allocator it's enough for like 60% of the std

left orbit Dec 27, 2024, 12:29 AM

#

north holly C freestanding: here stdint, have fun

that's C in general

frank canopy Dec 27, 2024, 12:29 AM

#

the ability to override the os bindings is a bit in flux rn sadly but not entirely

north holly Dec 27, 2024, 12:29 AM

#

left orbit that's C in general

true

#

I love C

#

(send help)

left orbit Dec 27, 2024, 12:29 AM

#

zig is C on steroids

frank canopy Dec 27, 2024, 12:30 AM

#

they removed the ability to override all the io stuff temporarily while trying to figure out a better abstraction than just letting you override the posix functions

fiery turtle Dec 27, 2024, 12:30 AM

#

I also like that its not forcing borrow checking or other stuff down your throat

frank canopy Dec 27, 2024, 12:30 AM

#

yeah thats why i cant handle rust lol

north holly Dec 27, 2024, 12:30 AM

#

rust is brainrot

frank canopy Dec 27, 2024, 12:30 AM

#

borrow checker sounds cool but i cannot wrap my head around what it wants

fiery turtle Dec 27, 2024, 12:30 AM

#

Apparently once you figure it out completely its good

#

But you need a PhD in language theory

frank canopy Dec 27, 2024, 12:31 AM

#

one thing that i do thats kinda fucked up tbh but im still doing it anyway is that zig's std (and compiler_rt aka builtins) are, like everything else in zig, included as source instead of linking, so i literally have a fork of the zig std for my os that overrides stuff lmaoo

left orbit Dec 27, 2024, 12:32 AM

#

ultra zig rewrite when 🙏

deft canopy Dec 27, 2024, 12:32 AM

#

fiery turtle Apparently once you figure it out completely its good

I think it could be pretty good for some userspace stuff but ive kinda given up on the idea of rust in the kernel

fiery turtle Dec 27, 2024, 12:32 AM

#

I've heard it doesn't work well for long lived apps, but good for data processing like compilers etc

north holly Dec 27, 2024, 12:32 AM

#

idk but I think kernels live for a long time

#

just a theory

left orbit Dec 27, 2024, 12:33 AM

#

i feel like kernel stuff is inherently incompatible with the idea of rust

fiery turtle Dec 27, 2024, 12:33 AM

#

Yeah perhaps

median crest Dec 27, 2024, 12:33 AM

#

north holly idk but I think kernels live for a long time

you dont restart yours every 10 seconds?

left orbit Dec 27, 2024, 12:33 AM

#

sure you can write safe abstractions over kernel concepts

north holly Dec 27, 2024, 12:33 AM

#

nyaux is a perfect example of a kernel that doesn't live long

left orbit Dec 27, 2024, 12:33 AM

#

but it all needs to be unsafe lol

frank canopy Dec 27, 2024, 12:33 AM

#

rn just to fix a bug that the pr hasnt been merged yet where the cryptographic rng doesnt use overridden getrandom if you tell it to always use getrandom instead of just for seed and then a prng, which bug means i need language threadlocals to work which isnt happening

fiery turtle Dec 27, 2024, 12:33 AM

#

left orbit but it all needs to be unsafe lol

Yeah have u seen Linux rust drivers

#

Its insane

north holly Dec 27, 2024, 12:34 AM

#

nyauxmaster: ```diff
+1
-1

(10 hours debugging)

frank canopy Dec 27, 2024, 12:34 AM

#

anyway back to trying to figure out unit tests for an avl tree because i decided to put vmm in my what to do next poll and suddenly i need a self-balancing tree

fiery turtle Dec 27, 2024, 12:35 AM

#

OMG my brain couldn't handle rbtrees, like it did work in the end but I still barely understand it

north holly Dec 27, 2024, 12:35 AM

#

I just copied bsd's sys/tree.h

#

ezpz

fiery turtle Dec 27, 2024, 12:35 AM

#

Especially removing nodes

#

U have to recursively walk up and reconfigure it if there's a double black node

frank canopy Dec 27, 2024, 12:36 AM

#

i found a c avl tree lib to copy but i also have to port it to zig in the process lol

#

zig has a treap in the std but its not intrusive so not really suited to my kernel work

fiery turtle Dec 27, 2024, 12:36 AM

#

Yeah

#

Intrusive containers ftw

frank canopy Dec 27, 2024, 12:36 AM

#

side note, @fieldParentPtr is goated in zig

fiery turtle Dec 27, 2024, 12:36 AM

#

How hard is it to make a linked list in zig

fiery turtle Dec 27, 2024, 12:37 AM

#

frank canopy side note, `@fieldParentPtr` is goated in zig

container_of?

frank canopy Dec 27, 2024, 12:37 AM

#

fiery turtle container_of?

yeah

fiery turtle Dec 27, 2024, 12:37 AM

#

Yeah container of is a based idea

frank canopy Dec 27, 2024, 12:37 AM

#

takes a comptime string field name

fiery turtle Dec 27, 2024, 12:37 AM

#

Does zig have reflection btw

left orbit Dec 27, 2024, 12:37 AM

#

yeah

#

@typeInfo

fiery turtle Dec 27, 2024, 12:37 AM

#

Big

left orbit Dec 27, 2024, 12:37 AM

#

and @Type to create a type from type infoi

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

what?

what