uACPI - a portable and easy-to-integrate ACPI implementation | Operating System Development | Page 32

gentle peak Jan 11, 2025, 1:11 PM

#

that's possible too but that's not really defer anymore imo

calm latch Jan 11, 2025, 1:12 PM

#

Defer in Go just calls a function

gentle peak Jan 11, 2025, 1:12 PM

#

still better than nothing (and arguably better than defer)

dense steppe Jan 11, 2025, 1:12 PM

#

if you want fancy features like this in C, imo it might be best to switch to C++

mortal yoke Jan 11, 2025, 1:12 PM

#

yeah there isn't much point in using c

#

c++ is supported basically everywhere where c is

dense steppe Jan 11, 2025, 1:13 PM

#

afaik Linux nowadays has some weird "scope guards" implemented using __attribute__((cleanup))--although I'm not sure how widely used they are

calm latch Jan 11, 2025, 1:13 PM

#

It's not fancy, error hanging sucks in C

#

You've got to use different atrocities all the time

calm latch Jan 11, 2025, 1:14 PM

#

dense steppe afaik Linux nowadays has some weird "scope guards" implemented using `__attribut...

Linux uses GNU features a lot

dense steppe Jan 11, 2025, 1:15 PM

#

yeah ik

#

but I think this is more of a recent thing

calm latch Jan 11, 2025, 1:15 PM

#

In the sense that it can be compiled with few compilers, and you as compiler dev would add missing function for Linux

fresh laurel Jan 11, 2025, 5:37 PM

#

hey can someone pass the github for this project?

north holly Jan 11, 2025, 5:37 PM

#

https://github.com/UltraOS/uACPI

GitHub

GitHub - UltraOS/uACPI: A portable and easy-to-integrate implementa...

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

fresh laurel Jan 11, 2025, 5:37 PM

#

i do not want to scroll on 30000 mesages

fresh laurel Jan 11, 2025, 5:37 PM

#

north holly https://github.com/UltraOS/uACPI

nice!

north holly Jan 11, 2025, 5:37 PM

#

#1217009725711847465 message

#

also take that

fresh laurel Jan 11, 2025, 5:38 PM

#

yeah im starting to look at AML and i dont want to parse it

fiery turtle Jan 11, 2025, 6:12 PM

#

uACPI had one skill issue vs ACPICA as I realized from @ marvin's AML blob

#

If there was an undefined AML reference in an IfOp or WhileOp the entire table load would get aborted

#

whereas ACPICA knows how to skip them cleanly

#

uACPI could only skip methods/devices etc

#

[uACPI][WARN] unable to lookup named object 'DTFS' within (or above) scope '\': not found
[uACPI][WARN] Op 0x00A0 ('IfOp'): skipping due to previous errors
[uACPI][WARN] unable to lookup named object 'DTFS' within (or above) scope '\': not found
[uACPI][WARN] Op 0x00A0 ('IfOp'): skipping due to previous errors
[uACPI][WARN] unable to lookup named object 'DTFS' within (or above) scope '\': not found
[uACPI][WARN] Op 0x00A0 ('IfOp'): skipping due to previous errors
[uACPI][WARN] unable to lookup named object 'DTFS' within (or above) scope '\': not found
[uACPI][WARN] Op 0x00A0 ('IfOp'): skipping due to previous errors
[uACPI][WARN] unable to lookup named object 'DTFS' within (or above) scope '\': not found
[uACPI][WARN] Op 0x00A0 ('IfOp'): skipping due to previous errors
[uACPI][WARN] unable to lookup named object 'DTFS' within (or above) scope '\': not found
[uACPI][WARN] Op 0x00A0 ('IfOp'): skipping due to previous errors
[uACPI][WARN] unable to lookup named object '\_SB_.PC02.GBES': not found
[uACPI][WARN] Op 0x00A0 ('IfOp'): skipping due to previous errors
[uACPI][WARN] unable to lookup named object '\_SB_.PC02.HDAS.VDID': not found
[uACPI][WARN] Op 0x00A0 ('IfOp'): skipping due to previous errors
[uACPI][WARN] unable to lookup named object '\_SB_.PC02.I2C1': not found
[uACPI][WARN] Op 0x0010 ('ScopeOp'): skipping due to previous errors
[uACPI][WARN] unable to lookup named object '\_SB_.PC02.I2C5': not found
[uACPI][WARN] Op 0x0010 ('ScopeOp'): skipping due to previous errors

#

Well not anymore LETSFUCKINGGOOOOOO

leaden fox Jan 11, 2025, 6:15 PM

#

fiery turtle If there was an undefined AML reference in an `IfOp` or `WhileOp` the entire tab...

so an undefined reference evals to false?

fiery turtle Jan 11, 2025, 6:15 PM

#

Previous output:

[uACPI][ERROR] unable to lookup named object 'DTFS' within (or above) scope '\': not found
[uACPI][ERROR] aborting table load due to previous error: AML referenced an undefined object
[uACPI][ERROR]     #0 in \()

fiery turtle Jan 11, 2025, 6:15 PM

#

leaden fox so an undefined reference evals to false?

neither, the entire if block is skipped

#

technically false yes

leaden fox Jan 11, 2025, 6:15 PM

#

honestly feels kinda like lua

fiery turtle Jan 11, 2025, 6:15 PM

#

this is more of a workaround for braindead firmware devs

#

that dont test their own code

fiery turtle Jan 11, 2025, 6:16 PM

#

fiery turtle uACPI had one skill issue vs ACPICA as I realized from @ marvin's AML blob

Cc @mortal yoke meme

#

new test coming soon

mortal yoke Jan 11, 2025, 6:18 PM

#

that same thing would happen in qacpi if it's at the top level

fiery turtle Jan 11, 2025, 6:18 PM

#

get to work ultrameme

#

they literally had one under a CondRefOf check, but the other SSDT decided to not do that

mortal yoke Jan 11, 2025, 6:20 PM

#

I guess I could just make the unwind function unwind individual frames if at the top level though that might create some interesting errors later on

fiery turtle Jan 11, 2025, 6:20 PM

#

mortal yoke I guess I could just make the unwind function unwind individual frames if at the...

i had to make it sort of hackish as well

#

#

#

basically it unwinds only pending ops but not the top one (which is the If/While)

#

which are skipped if the OPERAND ends up being NULL

dusky glade Jan 11, 2025, 8:22 PM

#

boo

fiery turtle Jan 11, 2025, 8:24 PM

#

?

dusky glade Jan 11, 2025, 8:25 PM

#

im just saying hi

#

plus good to see uacpi development as active as ever

fiery turtle Jan 11, 2025, 8:26 PM

#

👻

dusky glade Jan 11, 2025, 8:26 PM

#

dies of scaredness

slim panther Jan 11, 2025, 10:15 PM

#

infy was right

#

amd did favor menix

#

on my 265kf i now only get 3.2mil

#

on my 7700X i got 3.9

vagrant hull Jan 12, 2025, 11:26 AM

#

time to optimise :^)

thorny orbit Jan 12, 2025, 1:15 PM

#

how to check a speed with uacpi

#

like in this table

Screenshot_2025-01-12-16-16-53-548_com.android.chrome.jpg

calm latch Jan 12, 2025, 1:34 PM

#

it should print the speed once you load the namespace

thorny orbit Jan 12, 2025, 1:36 PM

#

calm latch it should print the speed once you load the namespace

how I should start qemu

#

with -q35 and kvm args ?

calm latch Jan 12, 2025, 1:37 PM

#

yes

dense steppe Jan 12, 2025, 1:37 PM

#

-M q35 -enable-kvm

thorny orbit Jan 12, 2025, 1:37 PM

#

ok

dense steppe Jan 12, 2025, 1:38 PM

#

is it allowed to just toss absurd amounts of memory to the VM? like qemu-system-x86_64 -m 48G?

thorny orbit Jan 12, 2025, 1:38 PM

#

dense steppe is it allowed to just toss absurd amounts of memory to the VM? like `qemu-syste...

qemu will return allocate error

dense steppe Jan 12, 2025, 1:38 PM

#

not if you have 64GiB of RAM

thorny orbit Jan 12, 2025, 1:38 PM

#

yes

dense steppe Jan 12, 2025, 1:38 PM

#

hence the question

#

is it allowed?

calm latch Jan 12, 2025, 1:39 PM

#

dense steppe is it allowed to just toss absurd amounts of memory to the VM? like `qemu-syste...

yes, but why would you want to do that

thorny orbit Jan 12, 2025, 1:39 PM

#

if you have 64G yes

dense steppe Jan 12, 2025, 1:39 PM

#

calm latch yes, but why would you want to do that

idk, just bump allocate literally everything maybe

#

although it'd probably perform worse tbh, because of cache behavior

calm latch Jan 12, 2025, 1:39 PM

#

is that allowed? Thonk

mortal yoke Jan 12, 2025, 1:40 PM

#

yeah no, I think it has to be like an actually usable one

#

that's not very usable in the long term

dense steppe Jan 12, 2025, 1:40 PM

#

it goes against the spirit of the competition

calm latch Jan 12, 2025, 1:41 PM

#

My score was with everything enabled

#

like mutexes, proper memory allocation and so on

#

and userspace since it's where uACPI ran

dense steppe Jan 12, 2025, 1:41 PM

#

I don't even have any options to disable any of that :p

thorny orbit Jan 12, 2025, 1:42 PM

#

with amd ryzen 5 3600

#

is this value is small or big

calm latch Jan 12, 2025, 1:42 PM

#

kinda on the smaller side

#

what CPU do you have?

mortal yoke Jan 12, 2025, 1:42 PM

#

ryzen 5 3600

calm latch Jan 12, 2025, 1:43 PM

#

bruh i can't read

dense steppe Jan 12, 2025, 1:45 PM

#

we should really do something where the competition benchmarks are performed on a "standardised" set of machines; you send a .iso to some people who are responsible for running the benchmark, and they run it. that way, everyone's code runs on the same machines

calm latch Jan 12, 2025, 1:47 PM

#

just ask someone with a fast CPU meme

calm latch Jan 12, 2025, 1:47 PM

#

thorny orbit is this value is small or big

but yeah i think it's kinda slow...

mortal yoke Jan 12, 2025, 1:48 PM

#

thorny orbit with amd ryzen 5 3600

can you send the iso?

thorny orbit Jan 12, 2025, 1:48 PM

#

mortal yoke can you send the iso?

yes

thorny orbit Jan 12, 2025, 1:50 PM

#

mortal yoke can you send the iso?

📎 output.iso

mortal yoke Jan 12, 2025, 1:53 PM

#

bruh

thorny orbit Jan 12, 2025, 1:53 PM

#

mortal yoke <:bruh:761542241381908491>

how much ram you give

#

to qemu

#

a

#

wait

#

why you paused

mortal yoke Jan 12, 2025, 1:54 PM

#

yeah that's what I were thinking about (it gets stuck when it tries to printf in some exception handler when locking a lock inside printf), Ill try with a higher amount

thorny orbit Jan 12, 2025, 1:55 PM

#

mortal yoke yeah that's what I were thinking about (it gets stuck when it tries to printf in...

I testing with 1GB

dense steppe Jan 12, 2025, 1:55 PM

#

I did qemu-system-x86_64 -enable-kvm -cpu host -M q35 -m 4G -no-reboot -drive file=/usr/share/edk2/x64/OVMF.4m.fd,if=pflash,index=0,read-only=on -debugcon stdio -nic none -cdrom output.iso on a 7950X3D and I get 1702 ops in 2ms (avg 834686/s)

fiery turtle Jan 12, 2025, 1:55 PM

#

dense steppe it goes against the spirit of the competition

A bump allocator actually performs quite terribly on this benchmark, because with a good allocator most are reused instantly and thus are still in cache

dense steppe Jan 12, 2025, 1:55 PM

#

thorny orbit Jan 12, 2025, 1:56 PM

#

dense steppe

what exactly average speed checking

dense steppe Jan 12, 2025, 1:56 PM

#

idk. that's what uACPI prints when I boot your ISO

north holly Jan 12, 2025, 1:56 PM

#

dense steppe

is it just me, or does that have like ~20 less ops

mortal yoke Jan 12, 2025, 1:57 PM

#

probably because of -nic none

fiery turtle Jan 12, 2025, 1:57 PM

#

^

dense steppe Jan 12, 2025, 1:57 PM

#

just sneakily memset(dsdt, 0, sizeof(dsdt)) before running the benchmark

fiery turtle Jan 12, 2025, 1:57 PM

#

thorny orbit what exactly average speed checking

Interpreter speed, aml ops per second

dense steppe Jan 12, 2025, 1:58 PM

#

I'm not going to integrate uACPI into my kernel until I actually have userspace and drivers that would be able to make use of it.

fiery turtle Jan 12, 2025, 1:58 PM

#

You could still use it in early table mode

#

That doesnt require much api

dense steppe Jan 12, 2025, 1:58 PM

#

yeah

#

I probably should do that

#

currently I have this: https://github.com/dbstream/davix/blob/master/acpi/tables.c

GitHub

davix/acpi/tables.c at master · dbstream/davix

A hobby operating system kernel targeting x86_64. Contribute to dbstream/davix development by creating an account on GitHub.

calm latch Jan 12, 2025, 1:59 PM

#

btw i think you could add some guards to turn the compilation of everything off and only leave table things in uACPI instead of making it a separate project (?)

fiery turtle Jan 12, 2025, 1:59 PM

#

calm latch btw i think you could add some guards to turn the compilation of everything off ...

I was never going to make a separate project

mortal yoke Jan 12, 2025, 1:59 PM

#

dense steppe I did `qemu-system-x86_64 -enable-kvm -cpu host -M q35 -m 4G -no-reboot -drive f...

I get like 100k with this for some reason

dense steppe Jan 12, 2025, 2:00 PM

#

what is your CPU?

fiery turtle Jan 12, 2025, 2:00 PM

#

I was going to introduce a bare bones mode

calm latch Jan 12, 2025, 2:00 PM

#

oh, i must have misunderstood it

mortal yoke Jan 12, 2025, 2:00 PM

#

dense steppe what is your CPU?

i5 13600k, its certainly not the cpu because with eg. astral I get 2M

dense steppe Jan 12, 2025, 2:00 PM

#

like #define UACPI_BARE_BONES and it doesn't compile any namespace management or AML interpreter?

fiery turtle Jan 12, 2025, 2:00 PM

#

dense steppe like `#define UACPI_BARE_BONES` and it doesn't compile any namespace management ...

Yup

dense steppe Jan 12, 2025, 2:02 PM

#

well tbh, if that existed, I would probably still just stub out unimplemented APIs like ```c
uacpi_status_t
uacpi_os_do_something (...)
{
return UACPI_NOT_IMPLEMENTED;
}

fiery turtle Jan 12, 2025, 2:02 PM

#

Yeah that's why its not a priority

#

Its not exactly hard to stub out everything

#

And just use early table access

dense steppe Jan 12, 2025, 2:03 PM

#

stop it I am getting ACPICA flashbacks

fiery turtle Jan 12, 2025, 2:03 PM

#

Lol

dense steppe Jan 12, 2025, 2:03 PM

#

btw how's 1.0 going?

fiery turtle Jan 12, 2025, 2:04 PM

#

Very close to being done

#

Next week perhaps

dense steppe Jan 12, 2025, 2:04 PM

#

calm latch Jan 12, 2025, 2:04 PM

#

but the interest is also in it being very small

#

(although I guess lto takes care of that)

fiery turtle Jan 12, 2025, 2:04 PM

#

Yeah its especially nice for separating out built-in kernel driver vs modules

#

Or microkernel etc

gentle peak Jan 12, 2025, 2:05 PM

#

i don't think you'd even need lto to make it small, just -fdata-sections -ffunction-sections combined with -Wl,--gc-sections should take care of that

fiery turtle Jan 12, 2025, 2:05 PM

#

Yep

dense steppe Jan 12, 2025, 2:06 PM

#

I've had cases where I have some function that calls into another function that I have no implementation for, which would generate a linker error, were it not for the fact that that function is optimized out because it is never called.

north holly Jan 12, 2025, 2:06 PM

#

I'm gonna test the latest uacpi commit (I don't want to debug stack corruption rn)

vagrant hull Jan 12, 2025, 2:06 PM

#

uACPI is gonna use semver right?

fiery turtle Jan 12, 2025, 2:07 PM

#

Yes

vagrant hull Jan 12, 2025, 2:07 PM

#

so is it gonna be 1.0 or 1.0.0?

fiery turtle Jan 12, 2025, 2:07 PM

#

The latter

vagrant hull Jan 12, 2025, 2:07 PM

#

awesome

#

good choice on both counts

fiery turtle Jan 12, 2025, 2:08 PM

#

I like semver

north holly Jan 12, 2025, 2:08 PM

#

I tested both master and address-spaces-support

#

the latter worked

#

master also worked

mortal yoke Jan 12, 2025, 2:10 PM

#

@thorny orbit your get uacpi_get_ns_since_boot is broken, why do you read the raw hpet counter at 0xF0 and subtract whatever the current unix time in seconds was at the time of boot from that?

#

that doesn't make any sense

vagrant hull Jan 12, 2025, 2:10 PM

#

🦆

thorny orbit Jan 12, 2025, 2:11 PM

#

mortal yoke <@456226577798135808> your get uacpi_get_ns_since_boot is broken, why do you rea...

boot time request giving time in seconds ???

mortal yoke Jan 12, 2025, 2:11 PM

#

just read the spec, its like the actual time date stamp in unix seconds and not how long the boot takes or anything

thorny orbit Jan 12, 2025, 2:12 PM

#

then do I need to convert cmos to unix subtract and convert result to nanoseconds ???

mortal yoke Jan 12, 2025, 2:12 PM

#

you don't need to use the boot time request at all for this

#

you need to figure out the hpet frequency by reading from the capabilities register and then use that to convert the hpet counter value you read to nanoseconds

fiery turtle Jan 12, 2025, 2:19 PM

#

north holly I tested both master and address-spaces-support

Im gonna need some extensive testing before merging that as thats basically the 1.0 pr

#

And I have some changes that are local atm that are also semi dangerous

thorny orbit Jan 12, 2025, 2:36 PM

#

mortal yoke you need to figure out the hpet frequency by reading from the capabilities regis...

i fixed this

#

shit

fiery turtle Jan 12, 2025, 2:46 PM

#

nyaux is faster

thorny orbit Jan 12, 2025, 2:46 PM

#

fiery turtle nyaux is faster

why it so slow 😭

fiery turtle Jan 12, 2025, 2:47 PM

#

for nyaux it was because allocator bug

thorny orbit Jan 12, 2025, 2:47 PM

#

lmfao

#

🙏

fiery turtle Jan 12, 2025, 2:48 PM

#

yeah im gonna add a warning like "your clock source might be unstable"

thorny orbit Jan 12, 2025, 2:48 PM

#

fiery turtle for nyaux it was because allocator bug

i have some troubles with hpet

#

it slow

north holly Jan 12, 2025, 2:48 PM

#

cope

thorny orbit Jan 12, 2025, 2:48 PM

#

but ok

#

a

#

is uacpi_kernel_get_nanoseconds_since_boot also used for average speed ?

fiery turtle Jan 12, 2025, 2:50 PM

#

how else do u think it can get elapsed time

thorny orbit Jan 12, 2025, 2:57 PM

#

do i need to implement uacpi_kernel_install_interrupt_handler to handle power button event

fiery turtle Jan 12, 2025, 2:57 PM

#

yes..

thorny orbit Jan 12, 2025, 2:57 PM

#

fiery turtle yes..

more pain

kindred beacon Jan 12, 2025, 8:40 PM

#

current managarm score:
uacpi-info: successfully loaded 1 AML blob, 1747 ops in 0ms (avg 4644312/s)

fiery turtle Jan 12, 2025, 8:44 PM

#

kindred beacon current managarm score: `uacpi-info: successfully loaded 1 AML blob, 1747 ops in...

which cpu?

kindred beacon Jan 12, 2025, 8:44 PM

#

14700K

fiery turtle Jan 12, 2025, 8:44 PM

#

LETSFUCKINGGOOOOOO 2nd place

fiery turtle Jan 12, 2025, 8:44 PM

#

kindred beacon current managarm score: `uacpi-info: successfully loaded 1 AML blob, 1747 ops in...

slim panther Jan 12, 2025, 9:14 PM

#

kms

fiery turtle Jan 12, 2025, 10:07 PM

#

slim panther kms

you mean kernel mode setting right

#

meme

slim panther Jan 12, 2025, 10:07 PM

#

right...

fiery turtle Jan 12, 2025, 10:07 PM

#

udrm got into your head

slim panther Jan 12, 2025, 10:08 PM

#

not the only thing i will put through my head

#

https://tenor.com/view/eye-of-rah-gif-10095752177411352905

Tenor

torpid root Jan 12, 2025, 10:10 PM

#

kms

flat badge Jan 13, 2025, 8:15 AM

#

It's quite nice that Managarm's score got 25% faster just by uacpi improvements 😄 Good job optimizing uacpi

#

actually more like 40% faster

fiery turtle Jan 13, 2025, 9:08 AM

#

thanks

slim panther Jan 13, 2025, 9:10 AM

#

yea this

thorny orbit Jan 13, 2025, 10:07 AM

#

thorny orbit lmfao

anyway I am fastest

fiery turtle Jan 13, 2025, 10:07 AM

#

If (PATH.THAT.DOES.NOT.EXIS.T) {
    Debug = "???"
} Else {
    Debug = "OK"
}

uACPI now skips the else block as well after an error like this

#

[uACPI][WARN] unable to lookup named object '\PATH.THAT.DOES.NOT_.EXIS.T___': not found
[uACPI][WARN] Op 0x00A0 ('IfOp'): skipping due to previous errors
[uACPI][WARN] Op 0x00A1 ('ElseOp'): skipping due to previous errors
[uACPI][INFO] successfully loaded 1 AML blob, 3 ops in 0ms (avg 10733/s)

#

ACPICA output:

Firmware Error (ACPI): Could not resolve symbol [\PATH.THAT.DOES.NOT.EXIS.T], AE_NOT_FOUND (20200925/psargs-395)
ACPI: Ignoring error and continuing table load
ACPI Error: Skipping While/If block (20200925/psloop-461)
ACPI Error: Skipping Else block (20200925/psloop-464)

fiery turtle Jan 13, 2025, 3:13 PM

#

question of the century

#

does NT skip stray ElseOp blocks without a previous If

#

have to disable secure boot to find out

loud ice Jan 13, 2025, 3:20 PM

#

you could rev acpi.sys

fiery turtle Jan 13, 2025, 3:21 PM

#

i have my trusted windbg + qemu

#

mortal yoke Jan 13, 2025, 3:22 PM

#

Ill bet it would as that's one of the simple ways to implement if

fiery turtle Jan 13, 2025, 3:22 PM

#

yet another acpica incompat? 🤯

fiery turtle Jan 13, 2025, 3:25 PM

#

mortal yoke Ill bet it would as that's one of the simple ways to implement if

it aborts table load 🤯

mortal yoke Jan 13, 2025, 3:25 PM

#

mortal yoke Ill bet it would as that's one of the simple ways to implement if

I did it like that in qacpi, basically for if when the condition is true it just falls through to the body and then if it encounters an elseop anywhere it just skips the body of it (and if the condition of the if was false it checks if there is an elseop following the if body and if there is it jumps to the else body)

mortal yoke Jan 13, 2025, 3:25 PM

#

fiery turtle it aborts table load 🤯

bruh

fiery turtle Jan 13, 2025, 3:26 PM

#

i guess ill do that too then

mortal yoke Jan 13, 2025, 3:26 PM

#

it probably doesn't matter tbh

#

unless someone is using it for some kind of weird nt detection

fiery turtle Jan 13, 2025, 3:26 PM

#

i could see it break in various ways

Else { 
  // Do bogus crap
}

#

if bios patches the blob poorly

#

u could run bogus aml

#

that nt would never touch

mortal yoke Jan 13, 2025, 3:27 PM

#

yeah but if you just skip the body

loud ice Jan 13, 2025, 3:27 PM

#

just precompile the tables into your own bytecode which doesn't suck as hard meme

fiery turtle Jan 13, 2025, 3:27 PM

#

Else {
}
// Do bogus crap

winter orbit Jan 13, 2025, 3:27 PM

#

fiery turtle that nt would never touch

oof

fiery turtle Jan 13, 2025, 3:28 PM

#

NT went out of their way to detect stray ElseOp

#

what can I say

fiery turtle Jan 13, 2025, 3:34 PM

#

mortal yoke yeah but if you just skip the body

i like this more

#

i guess ill just skip the body if its stray

mortal yoke Jan 13, 2025, 3:38 PM

#

its easy to implement at least, though if you already create a new frame for the if body in uacpi then you could also just abort if you want to match the nt behaviour

fiery turtle Jan 13, 2025, 3:39 PM

#

yea

#

its not something ive ever encountered but who knows how braindead firmware aml patching code might be

loud ice Jan 13, 2025, 3:41 PM

#

im sure its great

winter orbit Jan 13, 2025, 3:42 PM

#

loud ice im sure its great

lol

loud ice Jan 13, 2025, 3:42 PM

#

great at breaking shit at least lul

flat badge Jan 13, 2025, 3:43 PM

#

I'd suggest to add both nt and acpica compat modes for cases where they fundamentally disagree

#

NT support is not ubiquitous outside of consumer products

winter orbit Jan 13, 2025, 3:46 PM

#

depends on if you consider enterprise hardware consumer or not i guess

dense steppe Jan 13, 2025, 3:47 PM

#

maybe this compat could be like a quirk, so you pass acpi_compat=nt for NT compat and acpi_compat=acpica for ACPICA compat. this would be reflected as some sort of uacpi_set_compatibility_mode (UACPI_COMPAT_{NT,ACPICA})

flat badge Jan 13, 2025, 3:47 PM

#

yeah

dense steppe Jan 13, 2025, 3:47 PM

#

although having options for compatibility with both NT and ACPICA would make maintenance harder

#

more stuff to test

#

more behaviors to implement

flat badge Jan 13, 2025, 3:48 PM

#

but for example, AWS graviton (by far the most widely deployed ARM server) uses ACPI but doesn't support windows

#

etc

loud ice Jan 13, 2025, 3:48 PM

#

i suspect those kinds of things are less likely to have broken acpi

winter orbit Jan 13, 2025, 3:48 PM

#

lol

#

thats cute

flat badge Jan 13, 2025, 3:48 PM

#

I wouldn't count on that

loud ice Jan 13, 2025, 3:49 PM

#

lol

#

fair

#

theres less of them at least

flat badge Jan 13, 2025, 3:49 PM

#

AWS bare metal only supported bios boot until like 6months ago meme

loud ice Jan 13, 2025, 3:49 PM

#

you cant BUY a graviton, right?

flat badge Jan 13, 2025, 3:49 PM

#

on x86

loud ice Jan 13, 2025, 3:49 PM

#

flat badge AWS bare metal only supported bios boot until like 6months ago <:meme:5754450945...

any problem with that? meme

dense steppe Jan 13, 2025, 3:49 PM

#

IMO it is better to aim for good consumer firmware support, and hope that enterprise firmware eventually gets its stuff together, than to attempt to tailor to both

#

I think uACPI will end up better that way

flat badge Jan 13, 2025, 3:50 PM

#

IMO a switch "nt compat / acpica compat" makes a lot of sense

loud ice Jan 13, 2025, 3:50 PM

#

yeah

winter orbit Jan 13, 2025, 3:50 PM

#

Enterprise firmware is still going to be made by the "big 4" in bios dev anyway though

#

amazon didnt go with an inhouse solution for what is being brought up here either

loud ice Jan 13, 2025, 3:51 PM

#

winter orbit Enterprise firmware is still going to be made by the "big 4" in bios dev anyway ...

and yall test on linux, right?

flat badge Jan 13, 2025, 3:51 PM

#

where compat != bug compat. if the behavior is considered to be a bug in e.g. acpica, there's no need to support it

dense steppe Jan 13, 2025, 3:51 PM

#

well then hope that eventually, the "big 4" are practically required to test for uACPI

#

a man can dream

winter orbit Jan 13, 2025, 3:51 PM

#

loud ice and yall test on linux, right?

we test on the distros that pay us

flat badge Jan 13, 2025, 3:51 PM

#

sure, but amazon certainly never tested these machines with NT

#

considering that they cannot boot NT

winter orbit Jan 13, 2025, 3:52 PM

#

I don't think you're correct on that

fiery turtle Jan 13, 2025, 3:53 PM

#

there are some cases that make ACPICA compat very expensive/inpractical to support

#

e.g. they allow buffer resizing via StoreOp

#

which means uacpi must have a mutex per object now

loud ice Jan 13, 2025, 3:53 PM

#

#ifdef UACPI_ACPICA_COMPAT

#

or a GIL

fiery turtle Jan 13, 2025, 3:54 PM

#

yeah

flat badge Jan 13, 2025, 3:54 PM

#

fiery turtle which means uacpi must have a mutex per object now

not necessarily, it could use hashed locks

loud ice Jan 13, 2025, 3:54 PM

#

or a global buffer access lock lul

flat badge Jan 13, 2025, 3:54 PM

#

and have the mutexes outside of the objects

fiery turtle Jan 13, 2025, 3:54 PM

#

hashed locks?

flat badge Jan 13, 2025, 3:54 PM

#

to not pessimize the NT mode

loud ice Jan 13, 2025, 3:54 PM

#

fiery turtle hashed locks?

you have 32 global locks

fiery turtle Jan 13, 2025, 3:54 PM

#

ah lmfao

loud ice Jan 13, 2025, 3:54 PM

#

and you choose which one by doing global_locks[object addr % 32]

#

or whatever

#

its free inline

fiery turtle Jan 13, 2025, 3:54 PM

#

maybe at some point, if there's some firmware that breaks with NT compat

loud ice Jan 13, 2025, 3:55 PM

#

and still saves you from a lot of contention

dense steppe Jan 13, 2025, 3:55 PM

#

object addr % 32 would probably not be very good, as object addr is likely to be aligned

#

but the idea is good

winter orbit Jan 13, 2025, 3:55 PM

#

fiery turtle maybe at some point, if there's some firmware that breaks with NT compat

propose uACPI into linux kernel and then you wont need any compat mode other than what uACPI supports trollface

fiery turtle Jan 13, 2025, 3:56 PM

#

winter orbit propose uACPI into linux kernel and then you wont need any compat mode other tha...

considering we already have uacpi linux trl

flat badge Jan 13, 2025, 3:56 PM

#

linux ppl would 100% complain if it's not acpica compatible meme

loud ice Jan 13, 2025, 3:56 PM

#

dense steppe object addr % 32 would probably not be very good, as object addr is likely to be...

yeah ofc lmao

winter orbit Jan 13, 2025, 3:56 PM

#

linux people dont understand acpi

#

at all

fiery turtle Jan 13, 2025, 3:56 PM

#

so far they have complained about the opposite

dense steppe Jan 13, 2025, 3:56 PM

#

I think very few developers understand ACPI

#

like, I understand some parts of ACPI

#

but not the entire thing

winter orbit Jan 13, 2025, 3:57 PM

#

That's because it's written by liars and implemented by idiots

loud ice Jan 13, 2025, 3:57 PM

#

honestly i'd do hashed locks for everything

#

(except aml mutex objects ofc)

fiery turtle Jan 13, 2025, 3:57 PM

#

loud ice (except aml mutex objects ofc)

trl

dense steppe Jan 13, 2025, 3:57 PM

#

loud ice honestly i'd do hashed locks for everything

I'm doing something like this for my VFS. Each object (a "vnode") is hashed based on its parent and name. Each hash bucket has a lock and a linked list.

winter orbit Jan 13, 2025, 3:58 PM

#

fiery turtle considering we already have uacpi linux <:trl:829316641191886888>

we do

#

?

#

:p

dense steppe Jan 13, 2025, 3:58 PM

#

there's https://github.com/UltraOS/uacpi-linux I think

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.

loud ice Jan 13, 2025, 3:58 PM

#

oh god

fiery turtle Jan 13, 2025, 3:58 PM

#

lmfao

loud ice Jan 13, 2025, 3:58 PM

#

2027 is the year of the uacpi monopoly

#

trust

dense steppe Jan 13, 2025, 3:59 PM

#

make it happen

gentle peak Jan 13, 2025, 3:59 PM

#

in 2029 NT will switch to uacpi

loud ice Jan 13, 2025, 3:59 PM

#

no the monopoly is when nt switches 4head

winter orbit Jan 13, 2025, 3:59 PM

#

We'll be using uACPI at CLT this year

#

but i wont be able to show the cool things like suspend

fiery turtle Jan 13, 2025, 3:59 PM

#

CLT?

winter orbit Jan 13, 2025, 3:59 PM

#

its just going tobe powering the display laptops

#

https://chemnitzer.linux-tage.de/2025

Chemnitzer Linux-Tage 2025 · Chemnitzer Linux-Tage 2025

Die Chemnitzer Linux-Tage sind eine Veranstaltung rund um das Thema Linux und Open Source für jedermann, die Linux-Nutzer, Insider und Unternehmen zusammenbringt.

fiery turtle Jan 13, 2025, 4:00 PM

#

ah cool

winter orbit Jan 13, 2025, 4:00 PM

#

The reason im trying to speed run this before march

dense steppe Jan 13, 2025, 4:01 PM

#

In all seriousness, I think it would be useful to (once uacpi-linux atleast somewhat proves its usability) propose the idea of uACPI in upstream to the linux-acpi mailing list. Even if nothing becomes of it, it is a first step.

fiery turtle Jan 13, 2025, 4:01 PM

#

there's a problem with this

#

there are thousands of call sites directly using acpica

#

its not abstracted away like NT

#

so u cant do it gradually

loud ice Jan 13, 2025, 4:01 PM

#

so step 1 abstract it away?

winter orbit Jan 13, 2025, 4:01 PM

#

NT supermacy

dense steppe Jan 13, 2025, 4:01 PM

#

maybe that can be a first step

loud ice Jan 13, 2025, 4:02 PM

#

you can

dense steppe Jan 13, 2025, 4:02 PM

#

abstract away the old interfaces

loud ice Jan 13, 2025, 4:02 PM

#

gradually move to an abstraction

#

then swap the abstraction

fiery turtle Jan 13, 2025, 4:02 PM

#

if someone wants to do that work im totally down to help out where I can meme

flat badge Jan 13, 2025, 4:02 PM

#

that probably won't be accepted unless you demonstrate added value

winter orbit Jan 13, 2025, 4:03 PM

#

ACPI on linux has been proven to be completely broken

#

now its not broken

#

value

#

👏

loud ice Jan 13, 2025, 4:03 PM

#

lol

dense steppe Jan 13, 2025, 4:03 PM

#

well first you need uACPI

winter orbit Jan 13, 2025, 4:03 PM

#

"sorry suspend isnt something linux needs"

dense steppe Jan 13, 2025, 4:03 PM

#

so... to unbreak it, you need uACPI. to get uACPI, you need abstraction. for abstraction, you need value. for value, you need to unbreak it

fiery turtle Jan 13, 2025, 4:03 PM

#

if we're being more serious, acpica is slower and dies on every possible aml edge case, while also being subject to a lot of race conditions uacpi protects against

dense steppe Jan 13, 2025, 4:03 PM

#

:p

#

yeah that's something that linux people will definitely listen to

#

performance

flat badge Jan 13, 2025, 4:04 PM

#

that's true

winter orbit Jan 13, 2025, 4:04 PM

#

These are the same people that basically told us to eff off when we asked them to stopped declaring themselves as the latest windows versions

#

so we will see

fiery turtle Jan 13, 2025, 4:04 PM

#

i think iretq measured uacpi time to load the namespace vs acpica (on linux) and it was like 5x difference

#

acpica was like 600k ops/s and uacpi was about 3M

#

that was older uacpi too (i think)

winter orbit Jan 13, 2025, 4:06 PM

#

10m now

#

10x engineer infy

fiery turtle Jan 13, 2025, 4:06 PM

#

15M is the WR by monkuous ultrameme

dense steppe Jan 13, 2025, 4:06 PM

#

that's fast

fiery turtle Jan 13, 2025, 4:06 PM

#

with AVX2 in kernel

winter orbit Jan 13, 2025, 4:06 PM

#

Sadge I can't beat it

fiery turtle Jan 13, 2025, 4:07 PM

#

winter orbit <:Sadge:932196949652819999> I can't beat it

that was stubbed mutexes + avx2 + o3 + lto

gentle peak Jan 13, 2025, 4:07 PM

#

there was some other stuff too but i don't remember

#

that's the most important of it though

winter orbit Jan 13, 2025, 4:07 PM

#

let me compile reactos with avx2

fiery turtle Jan 13, 2025, 4:07 PM

#

monkuous' mutexes take 8ns to acquire tho

winter orbit Jan 13, 2025, 4:07 PM

#

actually i don't think i even can

fiery turtle Jan 13, 2025, 4:08 PM

#

and the allocator about 200ns per allocation avg

gentle peak Jan 13, 2025, 4:08 PM

#

about 150ns now on the version on github

#

and iirc the one in that benchmark had a slightly less capable but faster allocator that could do 100ns

mortal yoke Jan 13, 2025, 4:09 PM

#

winter orbit let me compile reactos with avx2

at least you can use sse/sse2 unlike some other kernels no?

winter orbit Jan 13, 2025, 4:09 PM

#

lol

#

i wish

mortal yoke Jan 13, 2025, 4:09 PM

#

doesn't windows save xmm0-xmm5 inside irqs and such?

calm latch Jan 13, 2025, 4:09 PM

#

flat badge linux ppl would 100% complain if it's not acpica compatible <:meme:5754450945896...

just don't tell them trl

dense steppe Jan 13, 2025, 4:09 PM

#

avx512 kernel when? meme

winter orbit Jan 13, 2025, 4:09 PM

#

mortal yoke doesn't windows save xmm0-xmm5 inside irqs and such?

depends on what's going on

#

also that's a win10 22 trick

fiery turtle Jan 13, 2025, 4:10 PM

#

Anyway it could be RFC'ed into Linux with demonstrable value (at least in like a week with 1.0), but its just way way too much work atm for me to go through, so maybe if someone with a lot of free time comes along meme

dense steppe Jan 13, 2025, 4:10 PM

#

it would be pretty cool to participate in this effort

#

but time is a finite resource

fiery turtle Jan 13, 2025, 4:11 PM

#

its mostly tedious manual work

#

an abstraction layer is more work tho

#

which idk where to begin

calm latch Jan 13, 2025, 4:11 PM

#

I'm getting done with my exams in 3 days ultrameme
(though I have very little idea of how Linux works inside)
so idk

mortal yoke Jan 13, 2025, 4:11 PM

#

winter orbit depends on what's going on

what does it depend on thonk though this isn't exactly related to uacpi lol

winter orbit Jan 13, 2025, 4:11 PM

#

I don't know the details but the guy who was implementing these things in reactos went on quite a rant

#

id hvae to go hunt for the long rant

fiery turtle Jan 13, 2025, 4:12 PM

#

calm latch I'm getting done with my exams in 3 days <:ultrameme:729026898666979428> (thoug...

tbh u dont even need to, u just do code drivers/acpi/* and go through every acpi_* call and convert to the uacpi counterpart (a lot of apis are very very similar so it should be relatively easy)

#

but anyway thats more of a thought experiment for now

dense steppe Jan 13, 2025, 4:13 PM

#

implementing the uacpi→kernel interfaces would probably be very easy on Linux

fiery turtle Jan 13, 2025, 4:13 PM

#

thats already done so yeah

#

its just stuff copypasted from acpica

#

but yeah they definitely won't accept a direct replacement, it would have to be configurable, at which point u need a unified abstraction

flat badge Jan 13, 2025, 4:15 PM

#

fwiw this is graviton OS support: https://github.com/aws/aws-graviton-getting-started/blob/main/os.md
so it's certainly not tested with NT

GitHub

aws-graviton-getting-started/os.md at main · aws/aws-graviton-getti...

Helping developers to use AWS Graviton2, Graviton3, and Graviton4 processors which power the 6th, 7th, and 8th generation of Amazon EC2 instances (C6g[d], M6g[d], R6g[d], T4g, X2gd, C6gn, I4g, Im4g...

fiery turtle Jan 13, 2025, 4:15 PM

#

which is probably also multiple years of work

flat badge Jan 13, 2025, 4:16 PM

#

would maybe be interesting to make managarm work + to test uacpi on it

fiery turtle Jan 13, 2025, 4:16 PM

#

true

flat badge Jan 13, 2025, 4:17 PM

#

and the story is probably similar on enterprise x86 servers

fiery turtle Jan 13, 2025, 4:18 PM

#

x86 server blobs i have do test for windows, at least via osi

flat badge Jan 13, 2025, 4:19 PM

#

for example on a random dell power edge that i just looked up, the OS support is:
VMware ESXi NFI (License Not Included)
SUSE Linux Enterprise, Non Factory Install, Requires License Selection
Ubuntu Server Non Factory Install, x64,Reqs Subscription Selection
Red Hat Enterprise Linux 9.4 Factory Install,x64,Requires Subscription Selection

winter orbit Jan 13, 2025, 4:20 PM

#

fiery turtle x86 server blobs i have do test for windows, at least via osi

ill get quite a few more for you

fiery turtle Jan 13, 2025, 4:21 PM

#

that would be cool

fiery turtle Jan 13, 2025, 4:21 PM

#

flat badge for example on a random dell power edge that i just looked up, the OS support is...

nice, would be curious to see how it behaves

winter orbit Jan 13, 2025, 4:22 PM

#

flat badge for example on a random dell power edge that i just looked up, the OS support is...

https://www.dell.com/support/home/en-us/drivers/supportedos/poweredge-xr11

#

¯_(ツ)_/¯

fiery turtle Jan 13, 2025, 4:23 PM

#

btw ACPICA on linux by default has the "slack" mode enabled

#

which basically violates both the spec and nt compatibility in many ways

#

Like it automatically pre-zeroes locals etc

#

whereas NT just aborts on uninitialized local use

#

and the spec obviously says they're uninitialized by default

#

ObjectType for predefined scopes is broken on ACPICA as well, whereas NT is spec compliant

#

the list is infinite

#

https://github.com/acpica/acpica/issues/920

GitHub

ObjectType for predefined scope names should be Uninitialized · Iss...

The spec: For typeless objects such as predefined scope names (in other words, _SB, _GPE, etc.), the type value 0 (Uninitialized) is returned. Code: Debug = ObjectType() Debug = ObjectType(_SB_...

#

they dont give a shit basically

gentle peak Jan 13, 2025, 4:27 PM

#

btw if anyone with a cpu faster than a 5800x wants to test, here's the latest proxima with all the (disqualifying) tricks that the WR ISO had

📎 proxima.iso

#

it's incompatible with smp mostly due to me not bothering to copy the xsave init asm to the smp entry point

#

on my cpu it gets ~12M

fiery turtle Jan 13, 2025, 4:28 PM

#

can u paste the command line also?

gentle peak Jan 13, 2025, 4:28 PM

#

qemu-system-x86_64 -accel kvm -cpu host,migratable=off -M q35,smm=off -cdrom proxima.iso

#

may have to test it a few times because sometimes the host schedules it weirdly and the performance gets multiplied by 0.3

#

and for reference here's the latest normal build of proxima (aka the codebase that's going to be on git somewhat soon, built normally for x86-64-v0) which gets 7.2-7.6M on my cpu

📎 proxima.iso

#

same command line

fiery turtle Jan 13, 2025, 4:33 PM

#

gentle peak and for reference here's the latest normal build of proxima (aka the codebase th...

7.6 on a non cheating build is insane

gentle peak Jan 13, 2025, 4:33 PM

#

yep

#

it has some local changes but none in the code paths that the benchmark touches so theoretically this should be the exact same as the latest build on github actions

flat badge Jan 13, 2025, 4:40 PM

#

winter orbit https://www.dell.com/support/home/en-us/drivers/supportedos/poweredge-xr11

yea it definitely differs on a model-by-model basis

#

windows servers are not exactly uncommon either

#

it's just that in contrast to consumer devices, not every enterprise device supports windows

#

actually i wonder how acpi behaves on linux-based consumer devices

#

do chrome books for example even have acpi?

fiery turtle Jan 13, 2025, 4:42 PM

#

they do

flat badge Jan 13, 2025, 4:43 PM

#

did you ever test on them?

fiery turtle Jan 13, 2025, 4:43 PM

#

well i did run them in userspace, if that counts

#

the pitust's blob

flat badge Jan 13, 2025, 4:43 PM

#

did it work?

fiery turtle Jan 13, 2025, 4:44 PM

#

yeah ofc, it's fairly carefully written aml too

#

it's only 3k lines of ASL

#

modern laptop x86 is like 300k loc sometimes

flat badge Jan 13, 2025, 4:45 PM

#

ah

#

that's not so surprising

fiery turtle Jan 13, 2025, 4:45 PM

#

it says COREBOOT as OEM ID so I assume it's some open source stuff customized

mortal yoke Jan 13, 2025, 4:46 PM

#

fiery turtle modern laptop x86 is like 300k loc sometimes

tbh I'd assume a lot of them are likely some generic amls just modified for the specific machine too

fiery turtle Jan 13, 2025, 4:46 PM

#

mortal yoke tbh I'd assume a lot of them are likely some generic amls just modified for the ...

perhaps

#

there's definitely a lot of dead code

winter orbit Jan 13, 2025, 4:46 PM

#

It’s the generic platform code unmodified

#

same with a lot of arm chrome books

flat badge Jan 13, 2025, 4:47 PM

#

do arm chrome books use acpi?

winter orbit Jan 13, 2025, 4:47 PM

#

Some of them

#

Mediatek ones don’t I’m pretty sure

fiery turtle Jan 13, 2025, 4:50 PM

#

[uACPI][INFO] successfully loaded 2 AML blobs, 1431 ops in 8ms (avg 172183/s)

#

chromebook AML is smaller than base QEMU

vale isle Jan 13, 2025, 5:04 PM

#

that's how they get that sweet fast boot

dense steppe Jan 13, 2025, 5:19 PM

#

gentle peak btw if anyone with a cpu faster than a 5800x wants to test, here's the latest pr...

that is very fast

#

but it's slower than the record, I think

gentle peak Jan 13, 2025, 5:19 PM

#

the pci stuff indicates that you're not using -M q35

dense steppe Jan 13, 2025, 5:19 PM

#

oh damn I forgot that

#

1sec

gentle peak Jan 13, 2025, 5:20 PM

#

also performance seems to improve with less memory

#

so don't bother passing -m <whatever>

dense steppe Jan 13, 2025, 5:21 PM

#

ok now it is more reasonable

#

though surprisingly slow, if the record is 15M

#

now I get avg 7011238/s

gentle peak Jan 13, 2025, 5:21 PM

#

what cpu do you have?

dense steppe Jan 13, 2025, 5:21 PM

#

dense steppe Jan 13, 2025, 5:22 PM

#

gentle peak what cpu do you have?

Ryzen 9 7950X3D

gentle peak Jan 13, 2025, 5:22 PM

#

hmm what

#

that exact iso on a 5800X gives 12M

dense steppe Jan 13, 2025, 5:22 PM

#

my QEMU command line was this: qemu-system-x86_64 -M q35 -enable-kvm -cpu host,migratable=off -drive file=/usr/share/edk2/x64/OVMF.4m.fd,if=pflash,index=0,read-only=on -debugcon stdio -m 112M -cdrom proxima.iso

gentle peak Jan 13, 2025, 5:22 PM

#

what's your qemu command line?

gentle peak Jan 13, 2025, 5:23 PM

#

gentle peak `qemu-system-x86_64 -accel kvm -cpu host,migratable=off -M q35,smm=off -cdrom pr...

try just using this one

#

no idea what the issue there is but the performance is abnormally slow

#

also is that just the first result or did you run it multiple times? sometimes the vm gets scheduled weirdly and performance suffers because of it

dense steppe Jan 13, 2025, 5:24 PM

#

gentle peak try just using this one

it still hovers around 7M with some reruns

gentle peak Jan 13, 2025, 5:24 PM

#

yeah i have no idea what the issue is here

dense steppe Jan 13, 2025, 5:24 PM

#

cpu skill issue

#

except not

gentle peak Jan 13, 2025, 5:25 PM

#

dense steppe my QEMU command line was this: `qemu-system-x86_64 -M q35 -enable-kvm -cpu host,...

this is the result i get with this command line

dense steppe Jan 13, 2025, 5:25 PM

#

maybe my QEMU binary is different

#

λ       qemu-system-x86_64 -version
QEMU emulator version 9.2.0
Copyright (c) 2003-2024 Fabrice Bellard and the QEMU Project developers

gentle peak Jan 13, 2025, 5:26 PM

#

i'm on 9.1.2

dense steppe Jan 13, 2025, 5:26 PM

#

so unless there's been some performance regression (highly unlikely), that's not the cause

gentle peak Jan 13, 2025, 5:27 PM

#

might take a while to check

hallow marten Jan 13, 2025, 5:27 PM

#

dense steppe so unless there's been some performance regression (highly unlikely), that's not...

are you on WSL?

dense steppe Jan 13, 2025, 5:27 PM

#

no

hallow marten Jan 13, 2025, 5:27 PM

#

weird then

dense steppe Jan 13, 2025, 5:27 PM

#

λ       uname -a
Linux beast 6.12.8-arch1-1 #1 SMP PREEMPT_DYNAMIC Thu, 02 Jan 2025 22:52:26 +0000 x86_64 GNU/Linux

dense steppe Jan 13, 2025, 5:28 PM

#

gentle peak might take a while to check

oh yeah, slow internet sucks

fiery turtle Jan 13, 2025, 5:29 PM

#

dense steppe

💀

dense steppe Jan 13, 2025, 5:29 PM

#

I have 250Mb internet, (mainly because my brother wouldn't stop screaming that his xbox was lagging trl ), but I can only really make use of 100Mb of it, because the cable that connects my computer with the router is actually bad and breaks down.

dense steppe Jan 13, 2025, 5:29 PM

#

fiery turtle 💀

fake record.

#

trl

gentle peak Jan 13, 2025, 5:30 PM

#

i wonder how much better/worse this would be with debugcon logging instead of fb

dense steppe Jan 13, 2025, 5:31 PM

#

dense steppe I have 250Mb internet, (mainly because my brother wouldn't stop screaming that h...

#

but like actually wth is this shit

#

I shouldn't really complain, 100Mb is great compared to some other internet speeds that people may have

#

but to be limited by the network cable, that's just sad

gentle peak Jan 13, 2025, 5:33 PM

#

my internet speed is limited to 29Mb due to my ISP

dense steppe Jan 13, 2025, 5:33 PM

#

😭

#

slow internet sucks

hallow marten Jan 13, 2025, 5:33 PM

#

dense steppe I shouldn't really complain, 100Mb is great compared to some other internet spee...

meanwhile I have like

#

1gb/s

#

but I dont have a cable going through my wall

#

so

#

I'm using wifi

dense steppe Jan 13, 2025, 5:33 PM

#

:p

hallow marten Jan 13, 2025, 5:34 PM

#

at like 200mbps

torpid root Jan 13, 2025, 5:35 PM

#

hallow marten 1gb/s

gigabyte per second

north holly Jan 13, 2025, 5:35 PM

#

I have fast internet

#

I forgor how fast

torpid root Jan 13, 2025, 5:35 PM

#

I technically have 70 mbps

#

I don't get more than 5-6mb/s

hallow marten Jan 13, 2025, 5:35 PM

#

torpid root gigabyte per second

yes

torpid root Jan 13, 2025, 5:36 PM

#

want to switch isps?

loud ice Jan 13, 2025, 5:36 PM

#

fiery turtle chromebook AML is smaller than base QEMU

i mean, there isnt a real chromebook for it to do stuff with so

north holly Jan 13, 2025, 5:36 PM

#

https://media.discordapp.net/attachments/350158983316766730/1295399339115679754/image.png?ex=67867aa5&is=67852925&hm=cd5948bb3d05ab3a0fff1a8bb7fdd7e23728011e5cfa41b1a2510295133372db&=&format=webp&quality=lossless

loud ice Jan 13, 2025, 5:36 PM

#

meme

dense steppe Jan 13, 2025, 5:36 PM

#

torpid root I don't get more than 5-6mb/s

that's called a scam

torpid root Jan 13, 2025, 5:36 PM

#

not much I can do about it

loud ice Jan 13, 2025, 5:37 PM

#

fiery turtle it says `COREBOOT` as OEM ID so I assume it's some open source stuff customized

its all open source afaik

north holly Jan 13, 2025, 5:37 PM

#

torpid root not much I can do about it

start a revolution

#

overthrow the government

torpid root Jan 13, 2025, 5:37 PM

#

It's in the works

dense steppe Jan 13, 2025, 5:37 PM

#

overthrow the ISP

north holly Jan 13, 2025, 5:37 PM

#

DOWN WITH THE ISP1!!

dense steppe Jan 13, 2025, 5:37 PM

#

just do some l33t h4xx0r1ng to get yourself better internet

torpid root Jan 13, 2025, 5:37 PM

#

isorrowproductions

north holly Jan 13, 2025, 5:37 PM

#

also turns out compiling gcc is quite slow on my laptop

torpid root Jan 13, 2025, 5:38 PM

#

dense steppe just do some l33t h4xx0r1ng to get yourself better internet

?

torpid root Jan 13, 2025, 5:38 PM

#

north holly also turns out compiling gcc is quite slow on my laptop

It takes more than a half an hour on my pc

north holly Jan 13, 2025, 5:38 PM

#

he means REDACTED

dense steppe Jan 13, 2025, 5:38 PM

#

hack the isp to get yourself better internet (except don't do that because it is illegal)

torpid root Jan 13, 2025, 5:38 PM

#

they even disabled access to my router

#

password changes on reboot

median crest Jan 13, 2025, 5:38 PM

#

north holly DOWN WITH THE ISP1!!

Pwnd by 0b3rr0w

torpid root Jan 13, 2025, 5:39 PM

#

and internet stops when I disable remote control thingy

dense steppe Jan 13, 2025, 5:39 PM

#

that's just horrible

#

like

#

that's a privacy issue goddamnit

torpid root Jan 13, 2025, 5:39 PM

#

perchance

north holly Jan 13, 2025, 5:39 PM

#

torpid root they even disabled access to my router

get a better ISP

torpid root Jan 13, 2025, 5:39 PM

#

there's only 2

mortal yoke Jan 13, 2025, 5:39 PM

#

do they force using a specific router?

north holly Jan 13, 2025, 5:39 PM

#

with fredom

torpid root Jan 13, 2025, 5:39 PM

#

both are the same shit

torpid root Jan 13, 2025, 5:39 PM

#

mortal yoke do they force using a specific router?

yes

mortal yoke Jan 13, 2025, 5:40 PM

#

and it doesn't work on other ones?

dense steppe Jan 13, 2025, 5:40 PM

#

~~clone the MAC address~~ (except don't do it because it might be illegal)

north holly Jan 13, 2025, 5:40 PM

#

get a better router, then make it spoof itself to impersonate

torpid root Jan 13, 2025, 5:40 PM

#

they supply their own and change it out once in a while

north holly Jan 13, 2025, 5:40 PM

#

the ISP's router

torpid root Jan 13, 2025, 5:40 PM

#

mortal yoke and it doesn't work on other ones?

I don't have many routers lying around lol

slow drift Jan 13, 2025, 5:41 PM

#

torpid root I technically have 70 mbps

laughs in Romania's internet speed

dense steppe Jan 13, 2025, 5:41 PM

#

mortal yoke and it doesn't work on other ones?

afaik one way that accounting is done (how much internet has ilobilo used this month? etc..) is with the MAC address

torpid root Jan 13, 2025, 5:41 PM

#

north holly get a better router, then make it spoof itself to impersonate

good idea if it was worth it

dense steppe Jan 13, 2025, 5:41 PM

#

dense steppe afaik one way that accounting is done (how much internet has ilobilo used this m...

except this is bad for many reasons

mortal yoke Jan 13, 2025, 5:41 PM

#

dense steppe afaik one way that accounting is done (how much internet has ilobilo used this m...

💀

torpid root Jan 13, 2025, 5:41 PM

#

dense steppe afaik one way that accounting is done (how much internet has ilobilo used this m...

why would they do that

dense steppe Jan 13, 2025, 5:41 PM

#

idk

#

but I don't think it is that common

#

it's just a way that it might be done

#

it is extremely bad because of MAC spoofing

#

for example, if the only authentication was MAC address and I knew your MAC address, I could very easily pretend to be you, and thus freeload on your internet

#

I think this might be one reason that ISPs provide modems, so that they can authenticate that whoever is using the internet is a legitimate customer and not someone who is MAC spoofing

torpid root Jan 13, 2025, 5:43 PM

#

It's unlimited

dense steppe Jan 13, 2025, 5:43 PM

#

unlimited* (until they decide "no more internet for you this month")

torpid root Jan 13, 2025, 5:44 PM

#

I have used a few hundred gibibytes and that hasn't happened yet

#

Maybe even a few thousand

north holly Jan 13, 2025, 5:44 PM

#

how does one use that much

loud ice Jan 13, 2025, 5:44 PM

#

dense steppe it is extremely bad because of MAC spoofing

i dont think mac spoofing matters

north holly Jan 13, 2025, 5:44 PM

#

(tbh idk how much I use)

mortal yoke Jan 13, 2025, 5:44 PM

#

north holly how does one use that much

downloading games

loud ice Jan 13, 2025, 5:44 PM

#

my ISP has "bridge mode" which exports raw ethernet

#

and i can set any mac i want

torpid root Jan 13, 2025, 5:44 PM

#

north holly how does one use that much

movies, or download dcs multiple times because you're a dumass

loud ice Jan 13, 2025, 5:45 PM

#

funnily enough, i can also run tcpdump on my router

dense steppe Jan 13, 2025, 5:45 PM

#

is your router connected directly to an external cable or is there some kind of modem that it goes through?

loud ice Jan 13, 2025, 5:45 PM

#

and you can see some managment traffic for other customers 💀

loud ice Jan 13, 2025, 5:45 PM

#

dense steppe is your router connected directly to an external cable or is there some kind of ...

a modem

#

because you need one

dense steppe Jan 13, 2025, 5:45 PM

#

loud ice and you can see some managment traffic for other customers 💀

that's... potentially a privacy and security issue

loud ice Jan 13, 2025, 5:45 PM

#

because the thing on the wire is not ethernet

torpid root Jan 13, 2025, 5:45 PM

#

dense steppe is your router connected directly to an external cable or is there some kind of ...

it's connected to something on the street with a fiber optic

loud ice Jan 13, 2025, 5:45 PM

#

dense steppe that's... potentially a privacy and security issue

i think its only managment traffic

dense steppe Jan 13, 2025, 5:45 PM

#

loud ice a modem

then, authentication is probably done through that

loud ice Jan 13, 2025, 5:46 PM

#

dense steppe then, authentication is probably done through that

yeah ofc

#

and if you dont have your own shit you can use it as a router as well

#

but its kinda shit anyway

dense steppe Jan 13, 2025, 5:46 PM

#

anyway why tf are we discussing networking in this channel?

gentle peak Jan 13, 2025, 5:47 PM

#

update finished, on qemu 9.2.0 i still get 12M/s

dense steppe Jan 13, 2025, 5:47 PM

#

so it's a me skill issue

torpid root Jan 13, 2025, 5:47 PM

#

dense steppe anyway why tf are we discussing networking in this channel?

because uACPI thread is basically lounge 2 now lol

dense steppe Jan 13, 2025, 5:47 PM

#

true

flat badge Jan 13, 2025, 7:17 PM

#

gentle peak update finished, on qemu 9.2.0 i still get 12M/s

Is this with the single core or smp safe version of proxima?

slim panther Jan 13, 2025, 7:18 PM

#

gentle peak update finished, on qemu 9.2.0 i still get 12M/s

bro what

gentle peak Jan 13, 2025, 7:20 PM

#

flat badge Is this with the single core or smp safe version of proxima?

it's with a bunch of unfeasible modifications: stubbed mutexes, stubbed spinlocks, simd in kernel, and initializing uacpi multiple times for better cache hit rates

#

but it's based on the smp safe version

#

(the single core version doesn't exist anymore)

#

the normal (smp-safe) version gets 7.2-7.6M

#

the context got a bit buried under all the networking talk

flat badge Jan 13, 2025, 7:22 PM

#

Does simd in the kernel actually improve perf for uacpi? I would have guessed that not a lot of acpi ops benefit from simd

gentle peak Jan 13, 2025, 7:22 PM

#

it's not a huge impact but it's definitely there

flat badge Jan 13, 2025, 7:22 PM

#

Ah, interesting

#

yeah, 12M is quite impressive 😄

gentle peak Jan 13, 2025, 7:23 PM

#

gentle peak it's not a huge impact but it's definitely there

just measured, 7.2-7.6 -> 7.5-8.0

flat badge Jan 13, 2025, 7:24 PM

#

Btw infy, is there some acpi implementation in a VM that is more involved than the qemu one? Asking because by now all (?) OSes in the top list finish init in 0ms meme

gentle peak Jan 13, 2025, 7:25 PM

#

gentle peak just measured, 7.2-7.6 -> 7.5-8.0

it's a much less significant impact if i also restrict it to x86-64-v1 (aka SSE2 only): 7.2-7.6 -> 7.4-7.7

hollow elm Jan 13, 2025, 7:25 PM

#

virtualbox isn't any more advanced ```
uacpi-info: successfully loaded 2 AML blobs, 2365 ops in 0ms (avg 3077535/s)

fiery turtle Jan 13, 2025, 7:26 PM

#

flat badge Btw infy, is there some acpi implementation in a VM that is more involved than t...

Perhaps VMware, but there's also a patch lying around by pitust to display load time in most applicable unit aka us/ns

fiery turtle Jan 13, 2025, 7:26 PM

#

hollow elm virtualbox isn't any more advanced ``` uacpi-info: successfully loaded 2 AML blo...

2 blobs and 2k ops is quite more advanced than qemu tbh

hollow elm Jan 13, 2025, 7:26 PM

#

qemu's also at ~2K ops?

fiery turtle Jan 13, 2025, 7:27 PM

#

hollow elm qemu's also at ~2K ops?

1700

hollow elm Jan 13, 2025, 7:27 PM

#

wasn't it 19xx?

flat badge Jan 13, 2025, 7:27 PM

#

fiery turtle Perhaps VMware, but there's also a patch lying around by pitust to display load ...

Yeah but it's also possible that more interesting effects happen for longer aml execution times

fiery turtle Jan 13, 2025, 7:27 PM

#

hollow elm wasn't it 19xx?

1705 for base qemu iirc

hollow elm Jan 13, 2025, 7:27 PM

#

the solution is to just have some baseline computer everyone tests boots their os on :^)

fiery turtle Jan 13, 2025, 7:27 PM

#

flat badge Yeah but it's also possible that more interesting effects happen for longer aml ...

Yeah, for that I guess we'd have to have THE osdev computer to test against lol

flat badge Jan 13, 2025, 7:28 PM

#

yeah that'd be annoying

fiery turtle Jan 13, 2025, 7:30 PM

#

How is davids cpu so slow tho💀

gentle peak Jan 13, 2025, 7:32 PM

#

hmm

#

does 86box have anything that supports acpi?

#

or some other "real" emulator

hollow elm Jan 13, 2025, 7:32 PM

#

86box doesn't have anything 64-bit iirc?

gentle peak Jan 13, 2025, 7:32 PM

#

didn't think of that yeah

hollow elm Jan 13, 2025, 7:33 PM

#

it goes up to P3 iirc

#

and they just don't think it makes sense to do anything more modern than that

fiery turtle Jan 13, 2025, 7:34 PM

#

You can add tons of pcie root ports and stuff to qemu

#

That will add extra aml objects

#

Same goes for CPUs, hotpluggable ram slots

gentle peak Jan 13, 2025, 7:35 PM

#

with -smp 4 proxima gets ~6.3M/s

#

i wonder if that'd improve if i moved smp init to after acpi init

#

yes it does

#

i now have the same performance regardless of cpu count

#

interesting

#

(up to nproc/2 that is, because once i go above that one of the uninitialized aps gets scheduled onto the same physical core as the bsp)

flat badge Jan 13, 2025, 7:42 PM

#

APs (or qemu vcpus) that sleep shouldn't really impact perf though

gentle peak Jan 13, 2025, 7:43 PM

#

i'm using limine for smp init so uninitialized aps are in a spin loop waiting for a field to be set

#

they're not really sleeping

flat badge Jan 13, 2025, 7:43 PM

#

ah yeah, that obviously has an impact

#

limine could use mwait :^) @vagrant hull

gentle peak Jan 13, 2025, 7:45 PM

#

fiery turtle Perhaps VMware, but there's also a patch lying around by pitust to display load ...

i was curious, so uh...

#

this is the normal proxima image

#

x86-64-v0, no tricks

flat badge Jan 13, 2025, 7:46 PM

#

but larger aml blob

#

this probably makes sense

loud ice Jan 13, 2025, 7:47 PM

#

less time loss on all the one-time setup

flat badge Jan 13, 2025, 7:47 PM

#

since uacpi will be in L1 cache already for a greater fraction of the running time etc

gentle peak Jan 13, 2025, 7:48 PM

#

can't really measure the one-time setup cost on qemu but if i simulate the cache part by initializing uacpi multiple times it only gets up to 8.0M

dense steppe Jan 13, 2025, 7:49 PM

#

fiery turtle How is davids cpu so slow tho💀

no idea mate. it's a good CPU, so I'm surprised the 5800X (or was it 5700X?) performs better

gentle peak Jan 13, 2025, 7:49 PM

#

5800x yeah

dense steppe Jan 13, 2025, 7:50 PM

#

probably tuning or interference (but lots of cores on the 7950X3D, so that is unlikely)

gentle peak Jan 13, 2025, 7:51 PM

#

i've done like 0 tuning on this system

dense steppe Jan 13, 2025, 7:51 PM

#

maybe the 5800X has better single-thread performance
but probably no

hallow marten Jan 13, 2025, 7:51 PM

#

your cpu is just bad

#

give it to me

dense steppe Jan 13, 2025, 7:51 PM

#

yeah sure meme

gentle peak Jan 13, 2025, 7:51 PM

#

there's only one performance-related thing that isn't default on my system and that's ram speed (it's set to the rated 3200MHz instead of the default 2133MHz)

dense steppe Jan 13, 2025, 7:52 PM

#

well my ram is like 5600MHz iirc

#

so that's not it

hallow marten Jan 13, 2025, 7:52 PM

#

let's start a laptop company and make firmware with huge aml blobs

#

the official OSDEV™️ laptop

dense steppe Jan 13, 2025, 7:52 PM

#

yeah

#

one target system to rule them all

gentle peak Jan 13, 2025, 7:53 PM

#

dense steppe maybe the 5800X has better single-thread performance but probably no

the 7950X3D scores better than the 5800X on Geekbench 6 single-thread by 731 points

median crest Jan 13, 2025, 7:54 PM

#

gentle peak it's a much less significant impact if i also restrict it to x86-64-v1 (aka SSE2...

Is there a point to like femtobenchmarking uacpi like this

gentle peak Jan 13, 2025, 7:55 PM

#

it's fun

median crest Jan 13, 2025, 7:58 PM

#

Fair

fiery turtle Jan 13, 2025, 8:03 PM

#

gentle peak i was curious, so uh...

6k ops 💀

#

This is like close to a lot of real desktops

vagrant hull Jan 13, 2025, 8:04 PM

#

flat badge limine could use mwait :^) <@379163868364275712>

isn't this like the 10th time at least that either you or someone else has brought this up?

#

i will look into doing that

dense steppe Jan 13, 2025, 8:20 PM

#

dense steppe well my ram is like 5600MHz iirc

just rebooted and checked. it is 5200MHz. I thought that maybe it was misconfigured somehow and that causes the slower-than-expected benchmark results.

hasty plinth Jan 13, 2025, 11:08 PM

#

dense steppe probably tuning or interference (but lots of cores on the 7950X3D, so that is un...

Isn't the extra cache only available to half of the cores on the 950x3d chips? You can get wonky performance characteristics depending on which cores you run on.

dense steppe Jan 13, 2025, 11:08 PM

#

true

#

although I haven't noticed any wonky performance characteristics

#

maybe I should try running it with CPU pinning

#

(however one would set that up)

hasty plinth Jan 13, 2025, 11:10 PM

#

Disable some cores in firmware halfmemeleft

hasty plinth Jan 13, 2025, 11:11 PM

#

dense steppe although I haven't noticed any wonky performance characteristics

I don't know if that's the issue, just something that came to mind

dense steppe Jan 13, 2025, 11:11 PM

#

hasty plinth Isn't the extra cache only available to half of the cores on the 950x3d chips? Y...

afaik there are two 'core complexes' (abbreviated as CCX iirc), each with 8 physical cores giving 16 threads per CCX, for 32 in total. L3 cache is per CCX, and only one CCX has the extra cache.

hasty plinth Jan 13, 2025, 11:11 PM

#

Yeah that was my understanding

dense steppe Jan 13, 2025, 11:14 PM

#

hasty plinth Disable some cores in firmware <:halfmemeleft:715831625233203250>

inb4 accidentally disables the 3D V-cache™️ cores

#

no but in all seriousness, I don't think it matters. like at all.

hasty plinth Jan 13, 2025, 11:16 PM

#

Fair enough

calm latch Jan 13, 2025, 11:56 PM

#

I think 3D cache cores run on a slower clock though

#

(next step: hobby OSes casually getting heterogenous multi-core support because of uACPI benchmark trl )

gentle peak Jan 14, 2025, 12:00 AM

#

wouldn't matter for uacpi benchmark since it's done in qemu and when in a vm it's all transparent to the guest

calm latch Jan 14, 2025, 12:00 AM

#

Didn't Windows have some drama with its scheduler?

dense steppe Jan 14, 2025, 12:02 AM

#

calm latch I think 3D cache cores run on a slower clock though

wtf... I've never heard of such a thing

#

why would they do that?

calm latch Jan 14, 2025, 12:03 AM

#

Because Zen 4 cache impedes cooling

#

So the cores have to run slower

#

I think it was changed with Zen 5, where they flipped the chips

#

So you have 8 cores with fast clock and 8 cores with a lot of cache

#

And they were doing some weirdness, like disabling half the cores in games

#

Because Windows' scheduler was dumb

#

I think it was improved in 24h2?

dense steppe Jan 14, 2025, 12:07 AM

#

I hardly ever reach top clock speeds anyways, unless I am compiling something :^)

calm latch Jan 14, 2025, 12:07 AM

#

calm latch (next step: hobby OSes casually getting heterogenous multi-core support because ...

There's no excuse now that both Intel and AMD use it trl

calm latch Jan 14, 2025, 12:08 AM

#

dense steppe I hardly ever reach top clock speeds anyways, unless I am compiling something :^...

But idk Windows might be doing something dumb vs. Linux, and having 2 heterogenous CCDs might be hurting the performance a lot

dense steppe Jan 14, 2025, 12:08 AM

#

idk

#

I have a dual-boot setup (although it's been a while since I last booted windows)

#

I've never noticed any issues

calm latch Jan 14, 2025, 12:09 AM

#

Like my PC ramps the fans up when I open vscode, even though it's mostly single-threaded, because it boosts one core

dense steppe Jan 14, 2025, 12:09 AM

#

but again, workload

#

any gaming I do is usually limited by my GPU as well :p

fiery turtle Jan 14, 2025, 9:58 AM

#

uACPI previously had slightly hacky if (ctx->skip_else) in order to detect whether it should skip an arbitrary ElseOpblock.
In order to make IfOp skipping during table loads and stuff more nice (+ skipping stray ElseOp) i microcoded both to no longer depend on global ctx state:

UACPI_OP(                                                        
    IfOp, 0xA0,                                                  
    {                                                            
        UACPI_PARSE_OP_TRACKED_PKGLEN,                           
        UACPI_PARSE_OP_OPERAND,                                  
        UACPI_PARSE_OP_IF_NULL, 1, 4,                            
            UACPI_PARSE_OP_EMIT_SKIP_WARN,                       
            UACPI_PARSE_OP_SWITCH_TO_NEXT_IF_EQUALS, 0xA1,       
            UACPI_PARSE_OP_END,                                  
        UACPI_PARSE_OP_IF_FALSE_OBJECT, 1, 3,                    
            UACPI_PARSE_OP_SWITCH_TO_NEXT_IF_EQUALS, 0xA1,       
            UACPI_PARSE_OP_END,                                  
        UACPI_PARSE_OP_INVOKE_HANDLER,                           
    }                                                            
)                                                                
UACPI_OP(                                                        
   ElseOp, 0xA1,                                                 
   {                                                             
       UACPI_PARSE_OP_TRACKED_PKGLEN,                            
       UACPI_PARSE_OP_IF_SWITCHED_FROM, 0xA0, 0x00, 3,          
            UACPI_PARSE_OP_SKIP_WITH_WARN_IF_NULL, 1,            
            UACPI_PARSE_OP_INVOKE_HANDLER,                       
   }                                                             
)

#

instead IfOp now just skips directly to ElseOp if it exists after the If block, and Else on its own skips itself by default unless it was invoked via If

kindred beacon Jan 14, 2025, 11:52 AM

#

AML microcode???????

mortal yoke Jan 14, 2025, 12:00 PM

#

its actually a nice way to parse aml, I also stole that idea from uacpi to my aml interpreter

slim panther Jan 14, 2025, 12:01 PM

#

fiery turtle uACPI previously had slightly hacky `if (ctx->skip_else)` in order to detect whe...

c preprocessor my beloved

mortal yoke Jan 14, 2025, 12:01 PM

#

res[LoadTableOp] = {7, {
    Op::TermArg,
    Op::TermArg,
    Op::TermArg,
    Op::TermArg,
    Op::TermArg,
    Op::TermArg,
    Op::CallHandler
}, OpHandler::LoadTable};
res[BankFieldOp] = {8, {
    Op::PkgLength,
    Op::NameString,
    Op::NameString,
    Op::TermArg,
    Op::Byte,
    Op::StartFieldList,
    Op::FieldList,
    Op::CallHandler
}, OpHandler::BankField};

fiery turtle Jan 14, 2025, 1:24 PM

#

mortal yoke ```cpp res[LoadTableOp] = {7, { Op::TermArg, Op::TermArg, Op::TermAr...

yours are kinda more high level i guess, because for uACPI FieldList is implemented in microcode too

#

mortal yoke Jan 14, 2025, 1:27 PM

#

yeah I don't really have any kind of control flow or anything

fiery turtle Jan 14, 2025, 1:28 PM

#

non turing complete microcode TheVoices

#

how many lines is your interpreter btw?

loud ice Jan 14, 2025, 1:32 PM

#

fiery turtle non turing complete microcode <a:TheVoices:1306604167682592839>

lol

#

why

fiery turtle Jan 14, 2025, 1:32 PM

#

idk

#

ray tracing or x86 emulator

loud ice Jan 14, 2025, 1:32 PM

#

i wonder what's the performance penalty for this microcode

loud ice Jan 14, 2025, 1:32 PM

#

fiery turtle ray tracing or x86 emulator

x86 emulator

#

clearly

mortal yoke Jan 14, 2025, 1:32 PM

#

the fields are slightly weird with how I did them, basically the StartFieldList pushes a fieldlist object to the object stack that contains the frame that the fields are within and then the FieldList op parses one field from the frame at a time until the frame is consumed (and sets the op block context's processed flag to false which makes it return keep processing that same FieldList op instead of continuing to verify it and get the next op, though in case of connectfield it actually parses the buffer/object in the middle on a new ctx before returning back to parsing the fields)

mortal yoke Jan 14, 2025, 1:33 PM

#

fiery turtle how many lines is your interpreter btw?

5681

fiery turtle Jan 14, 2025, 1:33 PM

#

damn

#

uACPI's was smaller before I added esoteric field stuff

#

rn its at 6000

#

at least my local copy

mortal yoke Jan 14, 2025, 1:37 PM

#

to be fair things like this add a lot of lines

fiery turtle Jan 14, 2025, 1:38 PM

#

yeah i have stuff like this too

#

although this would be handled in microcode for me

mortal yoke Jan 14, 2025, 1:39 PM

#

reminds me that I should really refactor that ObjectRef to instead default construct as empty, its kinda weird to use if it by allocates by default and then you have to check if it failed (and if you want to construct an empty non-allocated one then you have to use ::empty()) KEKW

#

its just another huge painful thing to do because it has billions of uses

fiery turtle Jan 14, 2025, 1:40 PM

#

lol

frank canopy Jan 14, 2025, 2:06 PM

#

sometimes I wonder how a fully zig acpi library would be and then I'm like heck no I don't actually want to have to make that to find out

#

zig has good enough c interop that uacpi is painless enough that it wouldn't be worth

fiery turtle Jan 14, 2025, 2:08 PM

#

lol

#

i mean u can give it a try for some tiny aml subset

#

like math stuff

#

AML is simple: 0 + 1 is encoded as AddOp 0x00 0x01 trl

#

trust

#

the rest is this ez too

mortal yoke Jan 14, 2025, 2:10 PM

#

I mean most of it is not that bad

fiery turtle Jan 14, 2025, 2:10 PM

#

yeah

#

AddOp AddOp AddOp ... trl

frank canopy Jan 14, 2025, 2:11 PM

#

lol

#

I was actually in the middle of starting one when I found out about uacpi

#

might revisit at some point

fiery turtle Jan 14, 2025, 2:12 PM

#

its a fun exercise for the most part

#

dont do references tho 💀

frank canopy Jan 14, 2025, 2:13 PM

#

lol

mortal yoke Jan 14, 2025, 2:13 PM

#

yeah the semantics are something I don't really understand to this day lol

#

and specifically the nt compatible ones

#

I just looked at uacpi and twiddled with it enough so it passes the tests

fiery turtle Jan 14, 2025, 2:14 PM

#

I did understand it entirely while implementing it but I no longer remember anything becaues its very counterintuitive

frank canopy Jan 14, 2025, 2:14 PM

#

the actual spec being a partial lie because of nt compat is what disillusioned me from the whole doing it in zig project

fiery turtle Jan 14, 2025, 2:14 PM

#

yeah acpica violates it quite a lot too

#

so the truth is somewhere in the subset of both

mortal yoke Jan 14, 2025, 2:16 PM

#

tbh playing with nt giving it different kinds of cases could help understanding them but idk how exactly do I even pass aml blobs to nt lol

fiery turtle Jan 14, 2025, 2:16 PM

#

i had to do sooo much of that

fiery turtle Jan 14, 2025, 2:18 PM

#

mortal yoke tbh playing with nt giving it different kinds of cases could help understanding ...

install com0com (virtual com driver)
install comByTCP
start comByTCP at COM3
run windbg -k com:port=com4,baud=115200
qemu-system-x86_64 -M q35 -hda windows.qcow2 -serial tcp::4445,server,nowait -m 4G -smp 8 -acpitable file="..\..\uACPI\tests\reftest.aml" -accel whpx -L "C:\Program Files\qemu\share"
connect to the debugger with windbg
!amli set traceon dbgbrkon errbrkon spewon;g
wait for Acpi.sys to get to your AML

frank canopy Jan 14, 2025, 2:19 PM

#

wait can't you run named pipes from qemu?

#

or is the tcp to escape wsl with it

fiery turtle Jan 14, 2025, 2:19 PM

#

idk if u can do that on windows

frank canopy Jan 14, 2025, 2:19 PM

#

I'm pretty sure you can

fiery turtle Jan 14, 2025, 2:19 PM

#

but anyway this works for me after 20 hours of setting it up

#

so i aint touching it

frank canopy Jan 14, 2025, 2:19 PM

#

I went looking for that at some point

#

and that's 100% fair

mortal yoke Jan 14, 2025, 2:20 PM

#

fiery turtle - install com0com (virtual com driver) - install comByTCP - start comByTCP at CO...

thanks, maybe Ill try it with something

fiery turtle Jan 14, 2025, 2:20 PM

#

i've tried just TCP and online guides suggested running two vms and stuff

#

i couldnt get anything to work but this

#

and it was very very painful to do

frank canopy Jan 14, 2025, 2:21 PM

#

oof

fiery turtle Jan 14, 2025, 2:21 PM

#

this requires disabling secure boot

#

because com0com isnt signed for secure boot

#

but it works on normal windows tho

frank canopy Jan 14, 2025, 2:21 PM

#

my instinct would def be named pipes but something working is better than trying other things

#

aint broke dont fix and all that

fiery turtle Jan 14, 2025, 2:22 PM

#

if you could show me how to do it with pipes id give it a try too

#

im not sure if windbg has a mode like that tho

gentle peak Jan 14, 2025, 2:24 PM

#

iirc on windows you have to use a path like \\.\pipe\<name>

frank canopy Jan 14, 2025, 2:24 PM

#

yep

winter orbit Jan 14, 2025, 2:25 PM

#

frank canopy my instinct would def be named pipes but something working is better than trying...

QEMU crashes on windows when you conntect windbg via pipes

gentle peak Jan 14, 2025, 2:25 PM

#

but I don't know the specifics of how to set it up with qemu

frank canopy Jan 14, 2025, 2:25 PM

#

gentle peak iirc on windows you have to use a path like `\\.\pipe\<name>`

that exact path actually

frank canopy Jan 14, 2025, 2:25 PM

#

winter orbit QEMU crashes on windows when you conntect windbg via pipes

when you use the qemu support?

#

or when you pass pipe through to a guest serial

#

because the latter is whats used here

#

qemu's chardev in theory does named pipes

winter orbit Jan 14, 2025, 2:25 PM

#

this is a very old issue

#

https://github.com/DarkFire01/qemu/tree/QemuThatDoesntSuckOnWindows

GitHub

GitHub - DarkFire01/qemu at QemuThatDoesntSuckOnWindows

Official QEMU mirror. Please see http://wiki.qemu.org/Contribute/SubmitAPatch for how to submit changes to QEMU. Pull Requests are ignored. Please only use release tarballs from the QEMU website. -...

fiery turtle Jan 14, 2025, 2:26 PM

#

winter orbit https://github.com/DarkFire01/qemu/tree/QemuThatDoesntSuckOnWindows

why not upstream it?

winter orbit Jan 14, 2025, 2:26 PM

#

https://github.com/DarkFire01/qemu/commit/8bf156bc9c0922bf960c5de60ca1f7bcb031336f

GitHub

fixes QEMU serial debugging on Windows author:gaasedelen · DarkFire...

#

because look at this

#

it was a quick work around i adapted

fiery turtle Jan 14, 2025, 2:26 PM

#

💀

frank canopy Jan 14, 2025, 2:27 PM

#

again is this for the qemu built in debug or for the named pipe serial/chardev?

#

is there a difference?

winter orbit Jan 14, 2025, 2:27 PM

#

this is for pipe serial/chardev

frank canopy Jan 14, 2025, 2:27 PM

#

RIP

winter orbit Jan 14, 2025, 2:27 PM

#

The issue is how Windbg handles its pipe communication

#

qemu just explodes

#

works with putty though

frank canopy Jan 14, 2025, 2:27 PM

#

ah that sounds about right then lmao

gentle peak Jan 14, 2025, 2:27 PM

#

wouldn't you be able to work around the issue by using -serial stdio and redirecting stdin and stdout to the pipe

winter orbit Jan 14, 2025, 2:28 PM

#

it'll still be too much unless you force windbg output to be sub 4000~

#

iirc

frank canopy Jan 14, 2025, 2:28 PM

#

damn

gentle peak Jan 14, 2025, 2:29 PM

#

that's annoying

frank canopy Jan 14, 2025, 2:29 PM

#

yeah that really sucks

winter orbit Jan 14, 2025, 2:29 PM

#

The proper solution would be rewriting that file a bit to use OS specific events

#

I;m lazy though

#

Ah anyway;
if you grab that hack and recompile qemu itll work fine but I feel like that's a pain and what @fiery turtle does now is far better and more stable

frank canopy Jan 14, 2025, 2:31 PM

#

yeah if you can just make something other than qemu handle the serial port then its fine

#

oh god i went to read the source for combytcp and that was a fuckin mistake

#

why is this visual basic written like its 3-5 years older than it is

#

(well ig it probably actually is tbh)

#

but dear lord

#

anyway i could probably make a tcp-named pipe adapter in less than an hour to save having to do the com0com bit

#

might even do that after breakfast, take a refresher from all this osdev with some nice high-level c#

mortal yoke Jan 14, 2025, 2:53 PM

#

I think I managed to get windbg working on linux with two vm's, though its painfully slow (and prints like 3 asl lines in a second in windbg) ```
host (used to run windbg connected to com1)
qemu-system-x86_64 -m 6G -M q35 -cpu qemu64 -smp 4 -drive file=win10-main.qcow2,media=disk,if=none,id=ide-disk -device piix4-ide,id=ide -device ide-hd,drive=ide-disk,bus=ide.0 -enable-kvm -serial tcp::4445,server,nowait -name WinMain -vga virtio
debuggee (configured with bcdedit /debug on; bcdedit /dbgsettings serial debugport:1 baudrate:115200):
qemu-system-x86_64 -m 6G -M q35 -cpu qemu64 -smp 4 -drive file=win10-debug.qcow2,media=disk,if=none,id=ide-disk -device piix4-ide,id=ide -device ide-hd,drive=ide-disk,bus=ide.0 -enable-kvm -serial tcp:127.0.0.1:4445 -name WinDebug -acpitable file=~/Projects/uACPI/tests/bin/to-integer.aml

fiery turtle Jan 14, 2025, 2:55 PM

#

mortal yoke I think I managed to get windbg working on linux with two vm's, though its painf...

nice, its painfully slow for me as well

mortal yoke Jan 14, 2025, 2:56 PM

#

is it also supposed to have some extra stuff printed? because I don't think a lot of the stuff its printing was within the to integer table that I tried to give it

#

like pci stuff and whatever

fiery turtle Jan 14, 2025, 2:56 PM

#

mortal yoke like pci stuff and whatever

i mean it first loads qemus dsdt

#

and only then this thing

mortal yoke Jan 14, 2025, 2:56 PM

#

ah that explains it

#

its going to take ages to get there tho lol

fiery turtle Jan 14, 2025, 2:57 PM

#

maybe its even slower for u than it is for me

#

mine takes maybe a minute to get to my tables

#

acpi.sys is slow af i guess lol

frank canopy Jan 14, 2025, 2:59 PM

#

the dbg might slow it down too

mortal yoke Jan 14, 2025, 3:00 PM

#

yeah all the printing and decompilation probably does

#

especially as with two vm's as it has to first go to the host from the first qemu then from there to the second qemu over tcp and from there qemu feeds it to the guest

fiery turtle Jan 14, 2025, 3:02 PM

#

yeah i mean makes sense

frank canopy Jan 14, 2025, 5:24 PM

#

well it took slightly longer than an hour to make a named-pipe/tcp bridge because i kept getting interrupted and im too lazy to test it in general but upsides it might actually work on non-windows ig lmao

fiery turtle Jan 14, 2025, 5:25 PM

#

we need to make like a guide for both approaches because its very easy to forget, and your option might be better because secure boot can still be on

frank canopy Jan 14, 2025, 5:27 PM

#

https://github.com/Khitiara/PipeByTcp/releases/tag/v0.1 well if anyone wants it heres a download and you can read the source yourself (it also prints out usage)

#

passing --pipe:xyz means giving windbg com:pipe,port=\\.\pipe\xyz

fiery turtle Jan 14, 2025, 5:27 PM

#

Wtf you actually made one from scratch?

frank canopy Jan 14, 2025, 5:28 PM

#

and it can act as either tcp host or client

#

yeah

fiery turtle Jan 14, 2025, 5:28 PM

#

Damn

frank canopy Jan 14, 2025, 5:28 PM

#

took around an hour ish

fiery turtle Jan 14, 2025, 5:28 PM

#

Well done

frank canopy Jan 14, 2025, 5:28 PM

#

ive been using c#/dotnet for over half my life i can almost do it in my sleep now lmao

#

and as i said untested

fiery turtle Jan 14, 2025, 5:28 PM

#

Need to test if it works ig

fiery turtle Jan 14, 2025, 5:28 PM

#

frank canopy ive been using c#/dotnet for over half my life i can almost do it in my sleep no...

chad

frank canopy Jan 14, 2025, 5:29 PM

#

the idea is qemu still does tcp and you just use this and a named pipe on windbg instead of com ports and com0com/combytcp

fiery turtle Jan 14, 2025, 5:29 PM

#

Big

#

Lol

#

Ill definitely try it

frank canopy Jan 14, 2025, 5:29 PM

#

thank goodness for dotnet having "copytoasync" for streams lol, def made this easier

#

can just tell the tcp socket and named pipe to copy to each other and be set

#

oh wait fuck im stupid hang on

#

rn its copying the same direction twice instead of both directions

#

#

there we go https://github.com/Khitiara/PipeByTcp/releases/tag/v0.2

#

i dont have a windows virtual machine or windbg set up rn or id try it directly lol

#

it even handles ctrl-c to exit gracefully (lets ctrl-break through to just have the os kill it though)

#

(and a second ctrl-c will get passed through to the os to kill it too)

fiery turtle Jan 14, 2025, 5:39 PM

#

Nice

slim panther Jan 14, 2025, 7:49 PM

#

frank canopy

c# ❤️

frank canopy Jan 14, 2025, 7:49 PM

#

its a good language for high-level things

#

idk if yall know cosmos but thats how i got into osdev even

#

jankest thing ever though

#

much prefer doing it all manually now

#

anyway please do let me know if the PipeByTcp thing works out if any of yall try it, hopefully it does so figuring out how winnt does acpi is a bit easier

vast kestrel Jan 14, 2025, 7:59 PM

#

Come and help with tomatos then Smuggy

frank canopy Jan 14, 2025, 8:10 PM

#

ive done low-ish level interop in c# before even recently i have no more interest in using it in osdev lol

fiery turtle Jan 14, 2025, 10:35 PM

#

FYI: I've made a separate org for uACPI and transferred uACPI and all related repos there: https://github.com/uACPI/uACPI

GitHub

GitHub - uACPI/uACPI: A portable and easy-to-integrate implementati...

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

#

only chosen ones will remember it was born as an acpi library for the ultra os

dense steppe Jan 14, 2025, 10:38 PM

#

great, now prune all references to UltraOS from history and you are done

fiery turtle Jan 14, 2025, 10:38 PM

#

i dont think there were any meme

rustic compass Jan 14, 2025, 10:38 PM

#

There are some

fiery turtle Jan 14, 2025, 10:38 PM

#

really?

rustic compass Jan 14, 2025, 10:39 PM

#

In uacpi-rs

#

Submodule

fiery turtle Jan 14, 2025, 10:39 PM

#

F

rustic compass Jan 14, 2025, 10:39 PM

#

Maybe we should move that there too

fiery turtle Jan 14, 2025, 10:39 PM

#

yeah im not sure how long that redirect is going to work for

rustic compass Jan 14, 2025, 10:40 PM

#

Also I mean that we could move uacpi-rs to uacpi/

fiery turtle Jan 14, 2025, 10:40 PM

#

ohhh

#

that would be cool definitely

rustic compass Jan 14, 2025, 10:41 PM

#

So that that rs version is by definition the only true api wrapper meme

fiery turtle Jan 14, 2025, 10:41 PM

#

sure

#

I can even link it in the readme once its complete

rustic compass Jan 14, 2025, 10:42 PM

#

once

fiery turtle Jan 14, 2025, 10:42 PM

#

how much is there left ultrameme

rustic compass Jan 14, 2025, 10:42 PM

#

blob_sweats

fiery turtle Jan 14, 2025, 10:42 PM

#

lol

rustic compass Jan 14, 2025, 10:42 PM

#

Types.h mostly done

fiery turtle Jan 14, 2025, 10:43 PM

#

thats probably the biggest

#

well resources.h also but thats just structs

rustic compass Jan 14, 2025, 10:43 PM

#

Most other stuff is done fast, but the types things was annoying sometimes

fiery turtle Jan 14, 2025, 10:44 PM

#

makes sense

rustic compass Jan 14, 2025, 10:44 PM

#

Did a few type safety crimes

fiery turtle Jan 14, 2025, 10:44 PM

#

lol

#

like which

rustic compass Jan 14, 2025, 10:45 PM

#

transmute rust slice to uacpi slices

#

transmute on generic structs

fiery turtle Jan 14, 2025, 10:45 PM

#

ah

#

do I need to add u as a member so u can transfer?

rustic compass Jan 14, 2025, 10:46 PM

#

Probably never moved a repository around

#

Can try it in 12h need some sleep

fiery turtle Jan 14, 2025, 10:46 PM

#

settings, scroll down, red button called move or whatever

#

yeah sure no rush

rustic compass Jan 14, 2025, 10:49 PM

#

Also I will update to your newest commit to see if they break win clang trl

vagrant hull Jan 14, 2025, 10:55 PM

#

fiery turtle yeah im not sure how long that redirect is going to work for

i'm pretty sure the answer is "undefinitely"

#

unless you make another repo with the same name as the old one

#

then it's gone

fiery turtle Jan 14, 2025, 10:56 PM

#

ah cool

fiery turtle Jan 14, 2025, 10:56 PM

#

rustic compass Also I will update to your newest commit to see if they break win clang<:trl:829...

lol

slim panther Jan 14, 2025, 11:01 PM

#

@fiery turtle the acpi memes continue

#

iASL Warning: There were 155 external control methods found during disassembly, but only 667 were resolved (4294966784 unresolved).

fiery turtle Jan 14, 2025, 11:07 PM

#

slim panther `iASL Warning: There were 155 external control methods found during disassembly,...

iasl bug i guess

#

Least retarded uACPI bug

#

(this is in my WIP PR dw)

dense steppe Jan 14, 2025, 11:10 PM

#

yeah, doing it on one line is unspecified because the order of evaluation is not defined

fiery turtle Jan 14, 2025, 11:10 PM

#

yeah idk what i was thinking

#

thanks to asan for breaking this order

#

because it was working without it

slim panther Jan 14, 2025, 11:10 PM

#

fiery turtle iasl bug i guess

ooooor buggy bios -> buggy decomp

fiery turtle Jan 14, 2025, 11:10 PM

#

slim panther ooooor buggy bios -> buggy decomp

nahh

#

no broken aml should break the disassembler

dense steppe Jan 14, 2025, 11:10 PM

#

it's tricky to know when it is and isn't allowed to depend on evaluation order

#

I tend to only depend on evaluation order in cases like ```c
if (obj && obj->somebool) {
// ...
}

slim panther Jan 14, 2025, 11:12 PM

#

dense steppe I tend to only depend on evaluation order in cases like ```c if (obj && obj->som...

well that eval order is defined i think?

fiery turtle Jan 14, 2025, 11:12 PM

#

re-read their msg lol

dense steppe Jan 14, 2025, 11:13 PM

#

slim panther well that eval order is defined i think?

yeah that case is defined

#

and if your compiler says otherwise, go shopping for a new compiler

slim panther Jan 14, 2025, 11:13 PM

#

true

#

can i load patched dsdts outside of the initrd

#

i kinda don't want to have to regenerate my initrd on every update

fiery turtle Jan 14, 2025, 11:24 PM

#

No idea

dense steppe Jan 14, 2025, 11:26 PM

#

slim panther can i load patched dsdts outside of the initrd

like, on linux? unlikely, given that the kernel's view of the ACPI tables is populated prior to any userspace execution.

#

so you don't have filesystems yet

north holly Jan 14, 2025, 11:35 PM

#

fiery turtle FYI: I've made a separate org for uACPI and transferred uACPI and all related re...

zamn
https://github.com/OBOS-dev/obos/commit/93d8b69975ae8762714eab08347606eaeab58161

GitHub

Update uacpi dependency · OBOS-dev/obos@93d8b69

fiery turtle Jan 14, 2025, 11:35 PM

#

north holly zamn https://github.com/OBOS-dev/obos/commit/93d8b69975ae8762714eab08347606eaeab...

Why did you comment it out instead of replacing lmao

north holly Jan 14, 2025, 11:36 PM

#

as a tribute

#

to the old uacpi link

fiery turtle Jan 14, 2025, 11:36 PM

#

Lol

north holly Jan 14, 2025, 11:36 PM

#

anyway gtg make me NIC driver support receiving packets

fiery turtle Jan 14, 2025, 11:37 PM

#

Letsgo

#

What are the next plans after that

north holly Jan 14, 2025, 11:37 PM

#

network stack

#

should be fun

fiery turtle Jan 14, 2025, 11:38 PM

#

Yeah no

north holly Jan 14, 2025, 11:38 PM

#

I'm sure it'll be fine trl

#

if I end up not being able to understand anything, I will use lwIP

fiery turtle Jan 14, 2025, 11:39 PM

#

Full TCP impl is like 10k loc probably

north holly Jan 14, 2025, 11:39 PM

#

flobsh

#

nvm

#

yeah lwIP's TCP implementation is 6k+ loc

fiery turtle Jan 14, 2025, 11:43 PM

#

north holly yeah lwIP's TCP implementation is 6k+ loc

Does it require much to port?

north holly Jan 14, 2025, 11:43 PM

#

I don't believe it's too hard to port

#

https://github.com/lwip-tcpip/lwip/blob/master/doc/NO_SYS_SampleCode.c

GitHub

lwip/doc/NO_SYS_SampleCode.c at master · lwip-tcpip/lwip

lwIP mirror from http://git.savannah.gnu.org/cgit/lwip.git - lwip-tcpip/lwip

fiery turtle Jan 14, 2025, 11:45 PM

#

north holly https://github.com/lwip-tcpip/lwip/blob/master/doc/NO_SYS_SampleCode.c

Yeah idk which parts here are lwip

north holly Jan 14, 2025, 11:45 PM

#

neither do I

calm latch Jan 14, 2025, 11:46 PM

#

dense steppe it's tricky to know when it is and isn't allowed to depend on evaluation order

|| and && do, | and & don't

dense steppe Jan 14, 2025, 11:46 PM

#

calm latch `||` and `&&` do, `|` and `&` don't

yeah, I've done a fair bit of C and C++

#

but in some cases,
it is better to be safe than to be sorry

vagrant hull Jan 14, 2025, 11:48 PM

#

in LICENSE

fiery turtle Jan 14, 2025, 11:50 PM

#

Right

#

Thx

vagrant hull Jan 14, 2025, 11:51 PM

#

np

frank canopy Jan 15, 2025, 12:06 AM

#

well looks like i might actually be able to test the PipeByTcp thing myself at some point given i need to install windbg to figure out this dang bugcheck my laptop keeps doing lmao

#

probably gonna be too lazy still though

median crest Jan 15, 2025, 3:28 AM

#

fiery turtle Does it require much to port?

Boooring

#

Write your own tcp stack

flat badge Jan 15, 2025, 10:13 AM

#

A simple but compliant tcp implementation is not that big

#

It becomes more messy if you want to support difficult congestion control algorithms and all the TCP options etc

loud ice Jan 15, 2025, 10:16 AM

#

you only really need one congestion control algorithm

flat badge Jan 15, 2025, 10:17 AM

#

Right, but a generic tcp implementation will probably give you more than one to chose from and hence it'll be larger than a simple implementation

loud ice Jan 15, 2025, 10:17 AM

#

yeah

flat badge Jan 15, 2025, 10:18 AM

#

One thing that's also more complicated than a basic implementation is a proper SACK implementation

loud ice Jan 15, 2025, 10:18 AM

#

true

#

hmm

vast kestrel Jan 15, 2025, 10:18 AM

#

Just make a stop and wait tcp

#

Window is optional

loud ice Jan 15, 2025, 10:18 AM

#

how hard is SACK, really

#

bad for performance

#

i will have CUBIC in my bootloader.

#

or maybe not idk

#

probably will tho

loud ice Jan 15, 2025, 10:20 AM

#

loud ice i will have CUBIC in my bootloader.

~~by stealing a rust ip library which does it all already~~

flat badge Jan 15, 2025, 10:21 AM

#

vast kestrel Window is optional

That's not true, is it?

loud ice Jan 15, 2025, 10:22 AM

#

flat badge That's not true, is it?

it is

#

when you want to read more bytes, you wait for a tcp packet that contains the next frame

#

and ack it immediatly

#

if you want to write more bytes, you send the frame every 1ms until you get an ACK back

#

congestion control is boring anyway meme

flat badge Jan 15, 2025, 10:23 AM

#

okay, but that is extremely stupid and relies on retransmission

vast kestrel Jan 15, 2025, 10:24 AM

#

Yeah lol

#

Like you can do it, it's extremely simple, but any PL or reorder destroys the perf

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