uACPI - a portable and easy-to-integrate ACPI implementation | Operating System Development | Page 58

fiery turtle May 31, 2025, 10:34 PM

#

lol

slim panther May 31, 2025, 10:34 PM

#

yea i saw that

calm latch May 31, 2025, 10:34 PM

#

this is mtune znver4

📎 pmOS.iso

gentle peak May 31, 2025, 10:34 PM

#

when the clocksource is tsc or kvmclock there is no syscall involved in getting the time

fiery turtle May 31, 2025, 10:34 PM

#

what does that even look like

dusky glade May 31, 2025, 10:35 PM

#

how dare u disrespect hpet

fiery turtle May 31, 2025, 10:35 PM

#

calm latch this is mtune znver4

same score

dusky glade May 31, 2025, 10:35 PM

#

/j

calm latch May 31, 2025, 10:35 PM

#

gentle peak when the clocksource is tsc or kvmclock there is no syscall involved in getting ...

I always syscall

gentle peak May 31, 2025, 10:35 PM

#

fiery turtle what does that even look like

instead of having something like jiffies that you use when precision isn't as important as access time, whenever you want to get the time it does a full read from hw

dusky glade May 31, 2025, 10:35 PM

#

okay nyaux might also be handicapped by HPET

#

i could implement kvm clock and see how it is

slim panther May 31, 2025, 10:35 PM

#

i think nyaux is much more handicapped by its allocator

calm latch May 31, 2025, 10:35 PM

#

I use tsc

fiery turtle May 31, 2025, 10:35 PM

#

gentle peak instead of having something like jiffies that you use when precision isn't as im...

ah and jiffies are supposed to be cached?

gentle peak May 31, 2025, 10:35 PM

#

jiffies is just a counter of a periodic irq

dusky glade May 31, 2025, 10:36 PM

#

slim panther i think nyaux is much more handicapped by its allocator

how so trl

dense steppe May 31, 2025, 10:36 PM

#

yeah but that's because in reality you are compiling twenty big projects, not one.

fiery turtle May 31, 2025, 10:36 PM

#

i wonder how it updates jiffies in tickless mode

dusky glade May 31, 2025, 10:36 PM

#

im gonna implement kvm clock

#

be back in 30 minutes

frank canopy May 31, 2025, 10:36 PM

#

i should really figure out how i want to do architect timekeeping in general tbh

gentle peak May 31, 2025, 10:36 PM

#

gentle peak instead of having something like jiffies that you use when precision isn't as im...

oh and there's a bunch of places where i read the time only to read it again 4 lines later in a leaf function with no locks taken in between or anything

#

aka i don't bother reusing the last read value even if it's still pretty much valid

fiery turtle May 31, 2025, 10:37 PM

#

its probably fine to rely on time reads to be fast

#

if you target relatively modern hw

frank canopy May 31, 2025, 10:37 PM

#

hitting the wall of i want my code to pick a time source (and calibrate tsc) as early as possible for stuff like logging but also half the time source options rely on some level of device discovery

fiery turtle May 31, 2025, 10:37 PM

#

like modern software in general spams gettimeofday all the time etc

dusky glade May 31, 2025, 10:37 PM

#

i just need to figure out how to check if im running under kvm

#

id check the hypervisor bit but

#

what if im running on tcg

#

or

#

what if im running on hyperv

gentle peak May 31, 2025, 10:38 PM

#

wrote a quick test and it turns out my pit clock source does in fact work correctly

calm latch May 31, 2025, 10:38 PM

#

dusky glade i just need to figure out how to check if im running under kvm

there's a bit which tells you that you have kvm clock I think

dusky glade May 31, 2025, 10:39 PM

#

calm latch there's a bit which tells you that you have kvm clock I think

where?

fiery turtle May 31, 2025, 10:39 PM

#

how does your pit clocksource convert time to ticks

#

do u just bump a counter in an irq handler

gentle peak May 31, 2025, 10:39 PM

#

so in qemu+kvm, taking a spinlock and doing 1 isa io port read + 2 isa io port writes is faster than reading from the hpet main counter

calm latch May 31, 2025, 10:39 PM

#

https://docs.kernel.org/virt/kvm/x86/cpuid.html

gentle peak May 31, 2025, 10:39 PM

#

fiery turtle do u just bump a counter in an irq handler

when an irq happens i increment a counter, and when reading i read the current count from the pit and and add that

fiery turtle May 31, 2025, 10:40 PM

#

how tf is HPET so slow even emulated

gentle peak May 31, 2025, 10:40 PM

#

so basically reading is num_irqs * TICKS_PER_IRQ + (TICKS_PER_IRQ - remaining_ticks())

dusky glade May 31, 2025, 10:40 PM

#

calm latch https://docs.kernel.org/virt/kvm/x86/cpuid.html

what would happen if i cpuid this on real hardware KVM_FEATURE_CLOCKSOURCE

dense steppe May 31, 2025, 10:40 PM

#

gentle peak so in qemu+kvm, taking a spinlock and doing 1 isa io port read + 2 isa io port w...

damn, HPET read is just ndelay(3000)

calm latch May 31, 2025, 10:40 PM

#

fiery turtle how tf is HPET so slow even emulated

it's emulated accurately trl

gentle peak May 31, 2025, 10:40 PM

#

gentle peak so basically reading is `num_irqs * TICKS_PER_IRQ + (TICKS_PER_IRQ - remaining_t...

with some extra stuff to avoid off-by-ones, race conditions, clock going backwards if the irq went pending after taking the spinlock, etc

frank canopy May 31, 2025, 10:40 PM

#

calm latch there's a bit which tells you that you have kvm clock I think

was gonna say, systemd-detect-virt exists so there has to be a way somehow

gentle peak May 31, 2025, 10:41 PM

#

dusky glade i just need to figure out how to check if im running under kvm

cpuid 1 ecx bit 31 is set if in a hypervisor

calm latch May 31, 2025, 10:41 PM

#

dusky glade what would happen if i cpuid this on real hardware KVM_FEATURE_CLOCKSOURCE

it wouldn't be set

gentle peak May 31, 2025, 10:41 PM

#

then cpuid 0x40000000 has a kvm signature

dusky glade May 31, 2025, 10:41 PM

#

gentle peak cpuid 1 ecx bit 31 is set if in a hypervisor

makes sense

#

kvmclock looks piss easy

#

lets cook

gentle peak May 31, 2025, 10:42 PM

#

this is how i determine kvmclock presence

dusky glade May 31, 2025, 10:42 PM

#

i believe we use the same cpuid function lol? perhaps i stole it from you??

gentle peak May 31, 2025, 10:42 PM

#

i mean

slim panther May 31, 2025, 10:42 PM

#

yea no timer makes no noticeable difference at all

gentle peak May 31, 2025, 10:43 PM

#

i don't think there can be that many differences between cpuid functions

#

at the end of the day it's a wrapper around inline asm

dusky glade May 31, 2025, 10:43 PM

#

yes but whats ur license trl

#

its important still

#

i always explcity declare if code isnt written by me

#

with comments

#

or if its obvious

#

holy shit my laptop is a space heater

#

its 90C

#

???

gentle peak May 31, 2025, 10:44 PM

#

gentle peak this is how i determine kvmclock presence

anyway, features.hypervisor is the value of cpuid 1 bit 31, features.hyp_vendor is this union initialized from cpuid 0x40000000, features.cpuid_hyp is eax from cpuid 0x40000000

dusky glade May 31, 2025, 10:44 PM

#

im idling only????

#

im not even playing a game

#

fan control might be cooked

fiery turtle May 31, 2025, 10:45 PM

#

gentle peak when an irq happens i increment a counter, and when reading i read the current c...

damn your code is really educational

gentle peak May 31, 2025, 10:45 PM

#

dusky glade yes but whats ur license <:trl:829316641191886888>

MIT, but i really don't think this is enough to warrant attribution (and i also doubt you took it from me since this is relatively new code compared to nyaux)

fiery turtle May 31, 2025, 10:45 PM

#

like its a fully correct thing but at an understandable scale

gentle peak May 31, 2025, 10:45 PM

#

for some parts i'm not sure it's fully correct lmao especially the vfs

slim panther May 31, 2025, 10:46 PM

#

fiery turtle like its a fully correct thing but at an understandable scale

that's what i'm saying

gentle peak May 31, 2025, 10:46 PM

#

i'm still not convinced there aren't thousands of deadlocks waiting to happen in my vfs

fiery turtle May 31, 2025, 10:46 PM

#

what is this about btw hydrogen_ret_t irq = x86_64_isa_controller.ops->open(&x86_64_isa_controller, 0, 0, handle_pit_irq, NULL);

gentle peak May 31, 2025, 10:46 PM

#

especially in rename/link

dusky glade May 31, 2025, 10:46 PM

#

gentle peak MIT, but i really don't think this is enough to warrant attribution (and i also ...

still gonna credit you idc

#

trl

fiery turtle May 31, 2025, 10:47 PM

#

oh also just noticed u use uacpi barebones in hydrogen? I thought u were gonna roll your own parsing

dusky glade May 31, 2025, 10:47 PM

#

acer-wmi kernel module is fucking trolling me

gentle peak May 31, 2025, 10:48 PM

#

fiery turtle what is this about btw `hydrogen_ret_t irq = x86_64_isa_controller.ops->open(&x8...

x86_64_isa_controller is an instance of the irq_controller_t class
open(controller, n, flags, func, ctx) opens a handle to irq n of controller with flags (active low/high, edge/level triggered, shareable/not shareable), and sets it up to be masked and to invoke func(ctx) when triggered.

fiery turtle May 31, 2025, 10:48 PM

#

dusky glade im idling only????

u should repaste it

gentle peak May 31, 2025, 10:48 PM

#

fiery turtle oh also just noticed u use uacpi barebones in hydrogen? I thought u were gonna r...

that was for the abandoned "full microkernel" rewrite of hydrogen

dusky glade May 31, 2025, 10:48 PM

#

fiery turtle u should repaste it

its only 2 years old

#

trl

fiery turtle May 31, 2025, 10:48 PM

#

sometimes the paste is shit from the factory

#

and by sometimes i mean often

fiery turtle May 31, 2025, 10:48 PM

#

gentle peak `x86_64_isa_controller` is an instance of the `irq_controller_t` class `open(con...

ah

dusky glade May 31, 2025, 10:48 PM

#

yea maybe ill do that

#

how much is thermal paste anyway

fiery turtle May 31, 2025, 10:48 PM

#

ngl it could use some enum values

fiery turtle May 31, 2025, 10:48 PM

#

dusky glade how much is thermal paste anyway

dirt cheap

dusky glade May 31, 2025, 10:49 PM

#

holy shit amazon.ie opened???

#

LETS FUCKING GOOO

#

letsgo

fiery turtle May 31, 2025, 10:49 PM

#

gentle peak that was for the abandoned "full microkernel" rewrite of hydrogen

ah

#

makes sense

gentle peak May 31, 2025, 10:49 PM

#

fiery turtle ngl it could use some enum values

there's not really anything to turn into an enum there, the first constant is a device-and-controller-dependent irq number and the second constant is only zero because there aren't any flags to pass

#

i guess i could #define PIT_IRQ 0

#

but that's about it

fiery turtle May 31, 2025, 10:49 PM

#

no like

#

theres some implicit polarity and triggering there no?

gentle peak May 31, 2025, 10:50 PM

#

ah yeah if no polarity/triggering flags are passed it's active high/edge triggered

dusky glade May 31, 2025, 10:50 PM

#

i dont have enough money nooo

gentle peak May 31, 2025, 10:50 PM

#

i guess i could define those flags as 0

dusky glade May 31, 2025, 10:50 PM

#

my dad has all the money i have

#

in his bank

fiery turtle May 31, 2025, 10:50 PM

#

thats what i would do

fierce kiln May 31, 2025, 10:51 PM

#

i love ringing nvme doorbells

gentle peak May 31, 2025, 10:51 PM

#

yep

dusky glade May 31, 2025, 10:51 PM

#

coretemp-isa-0000
Adapter: ISA adapter
Package id 0: +77.0°C (high = +100.0°C, crit = +100.0°C)
Core 0: +70.0°C (high = +100.0°C, crit = +100.0°C)
Core 4: +72.0°C (high = +100.0°C, crit = +100.0°C)
Core 8: +70.0°C (high = +100.0°C, crit = +100.0°C)
Core 12: +72.0°C (high = +100.0°C, crit = +100.0°C)
Core 16: +72.0°C (high = +100.0°C, crit = +100.0°C)
Core 20: +74.0°C (high = +100.0°C, crit = +100.0°C)
Core 24: +75.0°C (high = +100.0°C, crit = +100.0°C)
Core 25: +75.0°C (high = +100.0°C, crit = +100.0°C)
Core 26: +75.0°C (high = +100.0°C, crit = +100.0°C)
Core 27: +75.0°C (high = +100.0°C, crit = +100.0°C)
Core 28: +72.0°C (high = +100.0°C, crit = +100.0°C)
Core 29: +72.0°C (high = +100.0°C, crit = +100.0°C)
Core 30: +72.0°C (high = +100.0°C, crit = +100.0°C)
Core 31: +73.0°C (high = +100.0°C, crit = +100.0°C)
thats rlly fucking bad

fierce kiln May 31, 2025, 10:51 PM

#

yep 👍

#

lgtm

fiery turtle May 31, 2025, 10:51 PM

#

gentle peak yep

Nice

slim panther May 31, 2025, 10:52 PM

#

whoever gave these cores their IDs needs to get fired

fiery turtle May 31, 2025, 10:52 PM

#

dusky glade coretemp-isa-0000 Adapter: ISA adapter Package id 0: +77.0°C (high = +100.0°C,...

Luckily repasting takes 10 minutes

dusky glade May 31, 2025, 10:52 PM

#

slim panther whoever gave these cores their IDs needs to get fired

they should

slim panther May 31, 2025, 10:52 PM

#

but it's probably a big.little arch

dusky glade May 31, 2025, 10:52 PM

#

it is

slim panther May 31, 2025, 10:52 PM

#

so big cores get 4 ids?

dusky glade May 31, 2025, 10:52 PM

#

its 13th gen

#

i7-13700h lol

fierce kiln May 31, 2025, 10:54 PM

#

dusky glade they should

fibonacci core IDs when

#

2, 3, 5, 8, 13, 21, 34

dusky glade May 31, 2025, 10:54 PM

#

fiery turtle Luckily repasting takes 10 minutes

ill have to get it my dad gives me my money nooo

#

blud has my 950 euro

fiery turtle May 31, 2025, 10:55 PM

#

Arctic 4 costs like a few euro

gentle peak May 31, 2025, 10:55 PM

#

one issue i should fix in hydrogen is that rn if the only available clock sources are pit and 32-bit hpet the kernel will panic (if 32-bit hpet is going to be used as final clocksource instead of just calibration source it'll panic since i don't feel like dealing with overflow, and if the hpet is present it will unconditionally override pit (the logic here is that if any clock sources can be enumerated the pit will never be touched))

dusky glade May 31, 2025, 10:55 PM

#

i dont even got 4 euro

#

😭

#

in my bank

dense steppe May 31, 2025, 10:56 PM

#

gentle peak one issue i should fix in hydrogen is that rn if the only available clock source...

dealing with overflow is easy

gentle peak May 31, 2025, 10:56 PM

#

i know i just don't feel like it

#

32 bit hpet and no invtsc? fuck you

dusky glade May 31, 2025, 10:56 PM

#

brb reboot

torpid root May 31, 2025, 11:00 PM

#

@fiery turtle could you test this iso?

qemu-system-x86_64 -serial stdio -cpu host,migratable=off -M q35 -accel kvm -cdrom image-x86_64.iso

📎 image-x86_64.iso

#

I get very different results between runs so pick out the highest one :P

fiery turtle May 31, 2025, 11:12 PM

#

torpid root <@353593948523986945> could you test this iso? ``` qemu-system-x86_64 -serial st...

8.7M

torpid root May 31, 2025, 11:13 PM

#

what

#

seriously?

gentle peak May 31, 2025, 11:13 PM

#

huh??

fiery turtle May 31, 2025, 11:14 PM

#

yep

torpid root May 31, 2025, 11:14 PM

#

you've got to be joking

#

I didn't expect that

fiery turtle May 31, 2025, 11:14 PM

#

[00:00:00.014335] [info] uacpi: successfully loaded 1 AML blob, 1705 ops in 0ms (avg 8703286/s)

torpid root May 31, 2025, 11:14 PM

#

💀

fiery turtle May 31, 2025, 11:14 PM

#

it's consistent

#

i guess thats the top1

#

somehow

torpid root May 31, 2025, 11:14 PM

#

wait a sec, I must have something commented out or idk
that's too much

fiery turtle May 31, 2025, 11:14 PM

#

or wait proxima is 11M

torpid root May 31, 2025, 11:17 PM

#

fiery turtle `[00:00:00.014335] [info] uacpi: successfully loaded 1 AML blob, 1705 ops in 0ms...

try this one

#

📎 image-x86_64.iso

fiery turtle May 31, 2025, 11:18 PM

#

same

torpid root May 31, 2025, 11:18 PM

#

bruh

fiery turtle May 31, 2025, 11:18 PM

#

very good kernel ig

torpid root May 31, 2025, 11:19 PM

#

I doubt that but I'm still happy lol

frank canopy May 31, 2025, 11:20 PM

#

yeah may as well take it lol

dusky glade Jun 1, 2025, 1:45 AM

#

finally got it working

#

the generic interface

#

had to rewrite the fucking stall polling and that was pain because i suck at math

#

anyways time to get kvm clock

fiery turtle Jun 1, 2025, 10:51 AM

#

lol was looking through some old files and found my old mesa igpu triangle dumps

#

LD_PRELOAD="/home/infy/tests/mesa/build/src/glx/libGL.so" LIBGL_DRIVERS_PATH="/home/infy/tests/mesa/build/src/gallium/targets/dri/" INTEL_DEBUG="bat, submit" ./a.out 2> dump.txt

#

thats apparently how i tested it

torpid root Jun 1, 2025, 11:00 AM

#

when the next leaderboard update trl

fiery turtle Jun 1, 2025, 11:00 AM

#

next week i think

#

alongside 3.0

odd valve Jun 1, 2025, 11:56 AM

#

black on white??? theme?? my eyes!!!

fiery turtle Jun 1, 2025, 12:05 PM

#

yay boot allocator seems to work

#

just need to test dynamic growth and top down allocations now

kind mantle Jun 1, 2025, 12:35 PM

#

fiery turtle yay boot allocator seems to work

Project thread when? trl

fiery turtle Jun 1, 2025, 1:00 PM

#

after log ring trl

kindred beacon Jun 1, 2025, 2:32 PM

#

fiery turtle lol was looking through some old files and found my old mesa igpu triangle dumps

slim panther Jun 1, 2025, 2:47 PM

#

lol

dusky glade Jun 1, 2025, 5:11 PM

#

fiery turtle lol was looking through some old files and found my old mesa igpu triangle dumps

vulkan?? holy shit nie

dusky glade Jun 1, 2025, 5:11 PM

#

fiery turtle next week i think

trl

#

maybe nyaux will have better ops on kvmclock

median crest Jun 1, 2025, 9:41 PM

#

fiery turtle alongside 3.0

uacpi is already on 3.0 what

#

https://tenor.com/view/arriba-gif-12324188786972117417

Tenor

gentle peak Jun 1, 2025, 9:42 PM

#

semver, any breaking change no matter how small increments major

slim panther Jun 1, 2025, 9:43 PM

#

gentle peak semver, any breaking change no matter how small increments major

based

median crest Jun 1, 2025, 9:44 PM

#

one day I will go back to doing osdev and write the world's greatest operating system that has ever operating system'd

dusky glade Jun 1, 2025, 9:44 PM

#

based

#

wait

#

still

#

pick a choice

#

osdev or

#

being gay

#

😱

median crest Jun 1, 2025, 9:45 PM

#

I just wait till my boyfriend has a C class and then I make him do osdev with me its that simple

dusky glade Jun 1, 2025, 9:46 PM

#

thats crazy

median crest Jun 1, 2025, 9:47 PM

#

nah but fr one day I will actually start working on an os and try to make it actually like. not crap

#

maybe winter break

slim panther Jun 1, 2025, 10:05 PM

#

median crest one day I will go back to doing osdev and write the world's greatest operating s...

systemd reference

fiery turtle Jun 1, 2025, 10:06 PM

#

median crest nah but fr one day I will actually start working on an os and try to make it act...

Are you saying astral is bad

median crest Jun 1, 2025, 10:06 PM

#

fiery turtle Are you saying astral is bad

I mean it kinda is

fiery turtle Jun 1, 2025, 10:07 PM

#

Why

median crest Jun 1, 2025, 10:07 PM

#

bad style and questionable design

dusky glade Jun 1, 2025, 10:07 PM

#

it isnt

#

tbf

median crest Jun 1, 2025, 10:08 PM

#

I'm still thinking of rewriting some stuff tho maybe I will do that instead of doing it from scratch

#

just havent had the time

#

maybe once I finish some uni stuff I have to do I can leave astral doing a clean build

dusky glade Jun 1, 2025, 10:09 PM

#

listen mathew

#

if uve seen bad

#

look at nyaux

#

or boykisseros by ofluffydev

#

(real thing btw)

median crest Jun 1, 2025, 10:10 PM

#

dusky glade or boykisseros by ofluffydev

I should daily drive that

dusky glade Jun 1, 2025, 10:10 PM

#

median crest I should daily drive that

read the code for yourself 😭 this cannot be real

#

https://github.com/ofluffydev/boykisser-os

GitHub

GitHub - ofluffydev/boykisser-os: The Boykisser Operating System :3

The Boykisser Operating System :3. Contribute to ofluffydev/boykisser-os development by creating an account on GitHub.

fiery ice Jun 1, 2025, 10:11 PM

#

wait

#

wtf

#

lol what

dusky glade Jun 1, 2025, 10:12 PM

#

humanity is cooked 😭 🙏

fiery ice Jun 1, 2025, 10:12 PM

#

surreal current Jun 1, 2025, 10:18 PM

#

median crest maybe once I finish some uni stuff I have to do I can leave astral doing a clean...

astrole?

median crest Jun 1, 2025, 10:20 PM

#

casserole

fiery turtle Jun 1, 2025, 10:34 PM

#

median crest I'm still thinking of rewriting some stuff tho maybe I will do that instead of d...

yes

#

i want astral to be back sad

surreal current Jun 1, 2025, 10:36 PM

#

asterole

#

me too

hollow crescent Jun 2, 2025, 3:12 AM

#

median crest uacpi is already on 3.0 what

daang

#

by the time boron gets a uacpi port it will probably be 4.0

hollow crescent Jun 2, 2025, 3:14 AM

#

dusky glade being gay

you can do both at the same time

thorny orbit Jun 2, 2025, 1:15 PM

#

median crest uacpi is already on 3.0 what

wait what-

#

}

calm latch Jun 2, 2025, 1:18 PM

#

The update was breaking for me

left orbit Jun 2, 2025, 1:30 PM

#

nyaux update: uacpi: successfully loaded 1 AML blob, 1705 ops in 0ms (avg 1966580/s) on 13700k

#

@dusky glade

slim panther Jun 2, 2025, 2:53 PM

#

that's crazy

#

nyaux >1mil before gta6

dusky glade Jun 2, 2025, 3:00 PM

#

yes

fiery ice Jun 2, 2025, 3:46 PM

#

@fiery turtle i was playing around with the irq numbers and uacpi complained about this:
uACPI: unable to install SCI interrupt handler: already exists

#

which irq does uacpi use for SCI?

torpid root Jun 2, 2025, 3:50 PM

#

sci is always 9

fiery ice Jun 2, 2025, 3:50 PM

#

oh ok

torpid root Jun 2, 2025, 3:50 PM

#

uacpi doesn't change anything

fiery ice Jun 2, 2025, 3:50 PM

#

#

i was looking at this table

#

it showed them as available

torpid root Jun 2, 2025, 3:50 PM

#

that's why sci uses it

frank canopy Jun 2, 2025, 4:08 PM

#

i mean technically theres a field in i think the fadt that says what irq it is

#

but its basically always 9

#

(on imaginarium that ends up being ioapic mapped to vector 22, not sure what i have the targeting set to but it doesnt matter yet since i dont have AP startup yet)

fiery turtle Jun 2, 2025, 4:38 PM

#

fiery ice <@353593948523986945> i was playing around with the irq numbers and uacpi compla...

I mean uacpi doesn't pull that error out of its ass, you return that yourself lol

fiery ice Jun 2, 2025, 4:38 PM

#

yeah ik

fiery turtle Jun 2, 2025, 4:38 PM

#

Why do you return that?

fiery ice Jun 2, 2025, 4:39 PM

#

uacpi returned an error

#

and i just printed it

#

from namespace init

fiery turtle Jun 2, 2025, 4:40 PM

#

I mean you did return it from install irq handler

vocal geyser Jun 3, 2025, 6:13 PM

#

how i can get pci configuration when i using uacpi_find_devices

calm latch Jun 3, 2025, 6:22 PM

#

You're supposed to enumerate PCI devices yourself

vocal geyser Jun 3, 2025, 6:30 PM

#

calm latch You're supposed to enumerate PCI devices yourself

😭

fiery turtle Jun 3, 2025, 6:45 PM

#

Wdym by configuration

dense steppe Jun 3, 2025, 7:06 PM

#

vocal geyser how i can get pci configuration when i using `uacpi_find_devices`

ACPI gives you root buses, then you enumerate each root bus, discovering devices and possibly more downstream buses, which AFAIK don't have to be present in AML.

frank canopy Jun 3, 2025, 7:51 PM

#

yup

#

technically root bridges and its the downstream buses of those that you then enumerate but same idea

fiery turtle Jun 3, 2025, 8:04 PM

#

i like how everyone is super aware about aml and acpi now

#

i dont even have to answer questions trl

dense steppe Jun 3, 2025, 8:20 PM

#

no but you are expected to respond within three seconds of uACPI breaking because of bad kernel API implementations.

#

is that really worth it?

hollow elm Jun 3, 2025, 8:24 PM

#

fiery turtle i dont even have to answer questions <:trl:829316641191886888>

how do i acpi?

fiery turtle Jun 3, 2025, 8:24 PM

#

dense steppe no but you are expected to respond within three seconds of uACPI breaking becaus...

real

#

the uACPI SLA

fiery turtle Jun 3, 2025, 8:25 PM

#

hollow elm how do i acpi?

just do the thing

dense steppe Jun 3, 2025, 8:25 PM

#

I call uacpi_initialize, why doesn't my krenel work on real hardware?

fiery turtle Jun 3, 2025, 8:26 PM

#

trl

hollow elm Jun 3, 2025, 8:26 PM

#

i think usb hubs on managarm don't work because of uacpi, can you take a look

#

trl

fiery turtle Jun 3, 2025, 8:26 PM

#

trueee

dense steppe Jun 3, 2025, 8:28 PM

#

ACPI is a Real Hardware Thing so you are now cursed to be the Real Hardware Person and people will ping you about the most random of Real Hardware Quirks and expect you to know.

kind mantle Jun 3, 2025, 8:32 PM

#

dense steppe I call uacpi_initialize, why doesn't my krenel work on real hardware?

I call uacpi_initialize, why doesn't that magically do everything a kernel is supposed to do?

#

trl

calm latch Jun 3, 2025, 9:08 PM

#

uACPI pma, vma and heap when trl

fiery turtle Jun 3, 2025, 9:09 PM

#

UACPI_USE_BUILTIN_OS

frank canopy Jun 3, 2025, 9:09 PM

#

lol

torpid root Jun 3, 2025, 9:15 PM

#

finally, os crate

median crest Jun 3, 2025, 11:31 PM

#

uos when @fiery turtle

torpid root Jun 3, 2025, 11:32 PM

#

I believe it's being worked on

fiery turtle Jun 4, 2025, 7:23 AM

#

Yep

vocal geyser Jun 4, 2025, 7:36 AM

#

when uacpi motherboard

torpid root Jun 4, 2025, 9:32 AM

#

with sane aml

fierce kiln Jun 4, 2025, 4:38 PM

#

vocal geyser 😭

it's not hard at all don't worry 👍

#

scan all buses devices and functions

calm latch Jun 4, 2025, 4:52 PM

#

https://gitlab.com/mishakov/pmos/-/blob/main/devicesd/generic/pci/pci.c?ref_type=heads#L571

left orbit Jun 4, 2025, 7:28 PM

#

static void probe_function(uint16_t seg, uint8_t bus, uint8_t slot, uint8_t func) {
  /* Match some drivers or add the device to the list */
}

static void probe_slot(uint16_t seg, uint8_t bus, uint8_t slot) {
  uint16_t vendor = pci_read16(seg, bus, slot, 0);
  if (vendor == 0xFFFF)
    return; /* No function present */

  probe_function(seg, bus, slot, 0);

  /* Check if the device has multiple functions */
  uint8_t header_type = pci_read8(seg, bus, slot, 0);
  if ((header_type & 0x80) == 0)
    return;

  for (uint8_t func = 0; func < 8; func++) {
    uint16_t vendor = pci_read16(seg, bus, slot, func);
    if (vendor == 0xFFFF)
      continue; /* No function present */

    probe_function(seg, bus, slot, func);
  }
}

static void probe_bus(uint16_t seg, uint8_t bus) {
  for (uint8_t slot = 0; slot < 32; slot++) {
    probe_slot(seg, bus, slot);
  }
}

void pci_init(void) {
  for (/* for each root bus */) {
    probe_bus(bus->seg, bus->bus_num);
  }
}

#

shrimple

frank canopy Jun 4, 2025, 7:32 PM

#

it seriously is yeah

gentle peak Jun 4, 2025, 7:43 PM

#

isn't there some stuff you have to do with PCI to PCI bridges?

left orbit Jun 4, 2025, 7:44 PM

#

yes but that's going to be part of probe_function

#

or is it

#

idk, can oyu have a pci-pci bridge at function != 0?

#

probably safe to do it in probe_function anyway

hollow elm Jun 4, 2025, 7:44 PM

#

a pci-pci bridge requires you to read the secondary bus number and scan there

kind mantle Jun 4, 2025, 7:45 PM

#

hollow elm a pci-pci bridge requires you to read the secondary bus number and scan there

Wait are you supposed to do that? BadgerOS just iterates all 256 busses lol

left orbit Jun 4, 2025, 7:46 PM

#

you can do that too, i guess

#

but that's yucky

hollow elm Jun 4, 2025, 7:47 PM

#

i mean you have to

#

if you find a bridge at 12:1a.6 to bus 13:00.0

#

https://github.com/managarm/managarm/blob/master/kernel/thor/system/pci/pci_discover.cpp#L1268-L1289

kind mantle Jun 4, 2025, 7:48 PM

#

Oh I think I may be doing PCI discovery very wrong then lol

left orbit Jun 4, 2025, 7:49 PM

#

i believe what they are saying is they scan all busses instead of actually properly following the pci-pci bridge

#

you don't have to scan all 256 busses, right?

kind mantle Jun 4, 2025, 7:49 PM

#

I group all functions into one device node

hollow elm Jun 4, 2025, 7:49 PM

#

i mean yeah brute forcing all 256 buses works too ig

hollow elm Jun 4, 2025, 7:49 PM

#

kind mantle I group all functions into one device node

that i don't think is correct though :^)

#

consider a gpu which has the actual gpu function and an intel hda function for audio

kind mantle Jun 4, 2025, 7:50 PM

#

hollow elm consider a gpu which has the actual gpu function and an intel hda function for a...

oh

left orbit Jun 4, 2025, 7:50 PM

#

is it better to scan bus 0 + follow pci-pci bridges properly? or can there be a non-zero bus that isn't discoverable from bus 0 or acpi?

hollow elm Jun 4, 2025, 7:50 PM

#

you'd use 2 different drivers for them

kind mantle Jun 4, 2025, 7:50 PM

#

yeah that's a bit of a thing isn't it

kind mantle Jun 4, 2025, 7:50 PM

#

left orbit is it better to scan bus 0 + follow pci-pci bridges properly? or can there be a ...

Well in my case the DTB says bus-range is [0,255] anyway

left orbit Jun 4, 2025, 7:50 PM

#

kind mantle I group all functions into one device node

i guess that works if you create sub-device nodes for each function

kind mantle Jun 4, 2025, 7:51 PM

#

left orbit i guess that works if you create sub-device nodes for each function

That would be incorrect interrupt-wise in my model.

left orbit Jun 4, 2025, 7:51 PM

#

so pci device at xx:xx -> (gpu at xx:xx.0, hda at xx:xx.1)

#

ah i see

kind mantle Jun 4, 2025, 7:51 PM

#

Because of automatically inherited interrupt-parents, which are implicit if not specified.

#

I guess I'll just make one node per function then.

left orbit Jun 4, 2025, 7:52 PM

#

yeah 100%

kind mantle Jun 4, 2025, 7:53 PM

#

That was also the only reason my device system supports multiple addresses so I might as well just make that exactly one again.

#

But that's a future thing. For now, I'm going to be doing some SATA funny business and hopefully query drive parameters like block size and count.

left orbit Jun 4, 2025, 7:54 PM

#

do you have actual ahci hardware you want to support?

kind mantle Jun 4, 2025, 7:55 PM

#

I'm going to target the generic AHCI controller and I have a NVME-to-sata thingimajig that doesn't require special drivers on my testbench.

left orbit Jun 4, 2025, 7:56 PM

#

i would honestly just go for nvme lol

kind mantle Jun 4, 2025, 7:56 PM

#

I don't have spare drives of that kind right now

left orbit Jun 4, 2025, 7:56 PM

#

ahci is not nice imo

#

wait...

kind mantle Jun 4, 2025, 7:56 PM

#

I'm going to support AHCI, NVMe and USB MSC

left orbit Jun 4, 2025, 7:56 PM

#

you have one for the nvme-to-sata thing though?

#

usb mass storage is honestly so nice

kind mantle Jun 4, 2025, 7:57 PM

#

left orbit usb mass storage is honestly so nice

Requires USB first though

left orbit Jun 4, 2025, 7:57 PM

#

minus it being wrapped scsi

#

xhci is also a nice spec to work with

#

at least my driver worked on most hardware it tested it on.

flat badge Jun 4, 2025, 8:11 PM

#

left orbit is it better to scan bus 0 + follow pci-pci bridges properly? or can there be a ...

following the topology is definitely the right way

kind mantle Jun 4, 2025, 8:11 PM

#

Damn the physical memory allocation for an AHCI driver is going to be a bit of a PITA.

#

Just look at all this indirection

flat badge Jun 4, 2025, 8:14 PM

#

flat badge following the topology is definitely the right way

there are no busses that are unreachable from the roots (that'd be a bug in the acpi tables or the DT)

kind mantle Jun 4, 2025, 8:21 PM

#

How do y'all implement the memory allocation for all this? Do you have static command tables to be filled out and then dispatched, or do you allocate them as needed, use them, and release the memory later?

dense steppe Jun 4, 2025, 8:31 PM

#

probably the sane thing is to allocate the command ring buffer at setup time

#

and the response ringbuffer too

frank canopy Jun 4, 2025, 8:43 PM

#

left orbit is it better to scan bus 0 + follow pci-pci bridges properly? or can there be a ...

i think i read somewhere that reading or interacting with buses that dont exist can cause problems. reading the relevant byte to determine if a device exists is fine as long as the bus is real though. (also a root bridge can list a _BBN acpi method that gets you its root bus number but its almost always 0)

#

i could be wrong tho

left orbit Jun 4, 2025, 8:44 PM

#

frank canopy i think i read somewhere that reading or interacting with buses that dont exist ...

that's what my pseudocode basically does, assuming you get the segment and bus number by evaluating _SEG and _BBN

#

i didn't know about the non-existing bus thing tho

frank canopy Jun 4, 2025, 8:44 PM

#

i dont remember where i saw it and it might be something else im confusing tbf

dense steppe Jun 4, 2025, 8:45 PM

#

people have said it in here before

flat badge Jun 4, 2025, 8:51 PM

#

i'm not sure if reading their config space can be problematic but you can definitely run into devices that are behind bridges with mmio/pio decoding disabled etc

#

so if you do naive brute force, you'd think that these devices are usable while they are actually not usable until you reprogram their bridges (or until ACPI does that through AML, not sure if that's a thing though)

calm latch Jun 4, 2025, 8:52 PM

#

Nah

calm latch Jun 4, 2025, 8:52 PM

#

kind mantle Just look at all this indirection

To this

#

I allocate two memory regions per controller I think

#

Like fixed size command list I think?

dense steppe Jun 4, 2025, 8:54 PM

#

dense steppe people have said it in here before

the situation probably is something like this: on most hardware it is fine and just returns 0xffffffff, but on some hardware it hangs or raises MCEs

#

goddamn family interrupting me for half an hour when I'm in the middle of writing multiple messages

#

they raised NMI

hollow elm Jun 4, 2025, 8:58 PM

#

dense steppe the situation probably is something like this: on most hardware it is fine and j...

on the pi4 instead of yielding 0xffffffff it hangs the core for like 8s and then raises an SError :^)

#

although this happens for all nonexistent devices, so in addition to not doing a brute force scan you also need a different optimization

calm latch Jun 4, 2025, 8:59 PM

#

calm latch https://gitlab.com/mishakov/pmos/-/blob/main/devicesd/generic/pci/pci.c?ref_type...

I think this is infy approved?

dense steppe Jun 4, 2025, 8:59 PM

#

hollow elm although this happens for all nonexistent devices, so in addition to not doing a...

oh wtf

hollow elm Jun 4, 2025, 8:59 PM

#

    uint32_t nSlots = 32;

    // A PCIe downstream port has only one device (slot 0) attached.
    // In theory, this is only an optimization, in practice however omitting
    // this causes a SError on the BCM2711 when trying to access the vendor ID
    // of a non-existant device.
    if (bridge && bridge->isPcie && bridge->isDownstreamPort)
        nSlots = 1;

    for(uint32_t slot = 0; slot < nSlots; slot++)
        checkPciDevice(bus, slot, enumerateDownstream);

calm latch Jun 4, 2025, 8:59 PM

#

(except the raspberry pi thing)

calm latch Jun 4, 2025, 9:00 PM

#

dense steppe oh wtf

Because Broadcom's PCI controller is funny, and they do weirdness with ACPI

hollow elm Jun 4, 2025, 9:00 PM

#

this is not even acpi-related, it's just a bug (?) in their pcie rc

flat badge Jun 4, 2025, 9:00 PM

#

there is no proper ACPI for the pi4

hollow elm Jun 4, 2025, 9:00 PM

#

they fixed it for the pi5

flat badge Jun 4, 2025, 9:00 PM

#

the community EDK2 fork is not compliant

flat badge Jun 4, 2025, 9:01 PM

#

hollow elm ```cpp uint32_t nSlots = 32; // A PCIe downstream port has only one dev...

did you get this optimization from linux?

gentle peak Jun 4, 2025, 9:01 PM

#

most x86 acpi isn't spec compliant either, sounds like it fits right in

calm latch Jun 4, 2025, 9:01 PM

#

hollow elm this is not even acpi-related, it's just a bug (?) in their pcie rc

Not really a bug trl

calm latch Jun 4, 2025, 9:01 PM

#

hollow elm they fixed it for the pi5

Wasn't it even worse?

hollow elm Jun 4, 2025, 9:01 PM

#

flat badge did you get this optimization from linux?

yes, and conversely if you turn it off on linux (with pci=pcie_scan_all) then you get a kernel panic on linux too

flat badge Jun 4, 2025, 9:02 PM

#

ah

flat badge Jun 4, 2025, 9:02 PM

#

gentle peak most x86 acpi isn't spec compliant either, sounds like it fits right in

yeah but the pi community fork's ACPI doesn't even expose the pcie controller

gentle peak Jun 4, 2025, 9:03 PM

#

that's because of the weird config access mechanism right?

hollow elm Jun 4, 2025, 9:03 PM

#

yeah

#

the pi4/5 pcie rc exposes a single 4K config space window you slide around between functions

#

instead of ecam

dense steppe Jun 4, 2025, 9:04 PM

#

why do they do this??

hollow elm Jun 4, 2025, 9:04 PM

#

(and for 00:00.0 you use the pcie rc's mmio directly)

dense steppe Jun 4, 2025, 9:05 PM

#

hollow elm (and for 00:00.0 you use the pcie rc's mmio directly)

of course there are damn exceptions, why wouldn't there be!

gentle peak Jun 4, 2025, 9:05 PM

#

dense steppe why do they do this??

I'm guessing it's a remnant of when they had a 32 bit physical address space since proper ecam has huge mmio regions

dense steppe Jun 4, 2025, 9:05 PM

#

that seems likely

hollow elm Jun 4, 2025, 9:05 PM

#

yeah their pcie ip is descended from their mips socs

calm latch Jun 4, 2025, 9:06 PM

#

didn't raspberry pi 4 also have some weirdness with 8gb versions or something?

hollow elm Jun 4, 2025, 9:07 PM

#

pcie dma is limited to the first 3GB of ram iirc?

dense steppe Jun 4, 2025, 9:07 PM

#

because 32-bit?

hollow elm Jun 4, 2025, 9:08 PM

#

no it was a bug in the pcie controller integration with the memory controller in early pi4 chips

flat badge Jun 4, 2025, 9:08 PM

#

the pi4 also has 3 (?) possible physical memory layouts

#

because of backwards compat shenanigans

#

it's lovely

hollow elm Jun 4, 2025, 9:09 PM

#

yeah there are 3 phys address spaces, the legacy 32-bit addr one which some dma devices see, the full 35-bit addr bus one, and a low-peripheral 35-bit addr bus which moves a bunch of peripherals under 4G

dense steppe Jun 4, 2025, 9:10 PM

#

this is actually so cursed wtf

hollow elm Jun 4, 2025, 9:10 PM

#

iirc the gpu is always using the normal 35-bit one, and the cpu can be toggled by applying a dtb overlay in config.txt

#

and as i mentioned some peripheral dma (for stuff like i2c etc ig) uses the legacy 32-bit phys addresses

#

but yeah you can clearly tell that the broadcom socs have a long heritage :^)

#

all the way to the mips ones :^)

dense steppe Jun 4, 2025, 9:13 PM

#

the x86 of chipsets.

calm latch Jun 4, 2025, 9:21 PM

#

How did it work on raspberry pi 3? thonk

hollow elm Jun 4, 2025, 9:23 PM

#

pi3 has no pcie

calm latch Jun 5, 2025, 9:22 AM

#

fiery turtle Jun 5, 2025, 9:29 AM

#

impossible

hollow scaffold Jun 5, 2025, 10:56 AM

#

calm latch

No way! ||Isn't testing usually done using google test or some other programs that feed program test binaries and check outputs xd||

#

Things may appear working but it doesn't mean it will always work

calm latch Jun 5, 2025, 10:56 AM

#

huh?

#

You try on physical hardware

hollow scaffold Jun 5, 2025, 10:57 AM

#

And that decides if program works?

hallow marten Jun 5, 2025, 12:30 PM

#

hollow scaffold And that decides if program works?

Yes, uacpi already has a test suite but it cannot simulate hardware

hollow scaffold Jun 5, 2025, 12:39 PM

#

hallow marten Yes, uacpi already has a test suite but it cannot simulate hardware

Aw that's sad

torpid root Jun 5, 2025, 12:57 PM

#

uacpi hardware simulator

left orbit Jun 5, 2025, 3:43 PM

#

i believe they are testing and expecting known behavior

#

oh sorry discord didn't scroll all the way down trl

fiery turtle Jun 5, 2025, 3:51 PM

#

thats always so annoying

rustic compass Jun 5, 2025, 4:19 PM

#

flat badge i'm not sure if reading their config space can be problematic but you can defini...

The pcie spec mandates that all reads from non existing device has to return something like all 1s and writes to be discarded

#

And whatever pi4,5 do is not spec compliant

flat badge Jun 5, 2025, 4:22 PM

#

rustic compass The pcie spec mandates that all reads from non existing device has to return som...

Where does it say that?

#

Also, that doesn't contradict what i said

rustic compass Jun 5, 2025, 4:23 PM

#

Wasn't intended to

vast kestrel Jun 6, 2025, 11:33 AM

#

https://enaix.github.io/2025/06/03/acpi-conspiracy.html

#

the complaining at the start about the OS requiring the smbus hc driver is a bit stupid and just means he doesn't understand acpi, but the second part is funny nevertheless

fiery turtle Jun 6, 2025, 2:08 PM

#

Yeah SMBus is a mess

calm latch Jun 6, 2025, 2:11 PM

#

I wanted to implement it

fiery turtle Jun 6, 2025, 9:42 PM

#

okay spent some time studying how linux handles pgd/p4d/pud/pmd/pte and page faults, I think i get it enough to implement something like that

#

once i have that ill start building the sparse struct page array

#

ill try to use less stupid names for page table level abstractions

vast kestrel Jun 6, 2025, 9:50 PM

#

infy speed running re-writing linux from scratch

fiery turtle Jun 6, 2025, 9:50 PM

#

more like slow running

vast kestrel Jun 6, 2025, 9:51 PM

#

tbf you have like 20 years of work to catch up :^)

fiery turtle Jun 6, 2025, 9:51 PM

#

but yeah i guess this is a mini-linux project tbh

#

i wanna research all subsystems in depth

slim panther Jun 6, 2025, 9:51 PM

#

fiery turtle ill try to use less stupid names for page table level abstractions

what's wrong with just having a loop?

fiery turtle Jun 6, 2025, 9:52 PM

#

reusing the same code on all arches and on all runtime configurations

#

an arch just needs to provide some basic defines and helpers

slim panther Jun 6, 2025, 9:52 PM

#

yea i have a generic paging impl with a loop

fiery turtle Jun 6, 2025, 9:53 PM

#

plus very easy to make that stuff handle huge pages etc

slim panther Jun 6, 2025, 9:53 PM

#

that can do an arbirary amount of levels (as long as the arch supports it)

fiery turtle Jun 6, 2025, 9:53 PM

#

well how do u handle 5 vs 4 levels

#

in linux level 4 is just folded transparently if the cpu doesnt support it

slim panther Jun 6, 2025, 9:54 PM

#

if there's 5 levels then i start at 5, otherwise at 4 :P

fiery turtle Jun 6, 2025, 9:54 PM

#

lol fair

#

i mean a loop should work just fine probably as wel

slim panther Jun 6, 2025, 9:55 PM

#

there are obviously some arch dependent bits to determine the flags for each level

fiery turtle Jun 6, 2025, 9:55 PM

#

do u do stuff like arch_flags_for_level(i)?

slim panther Jun 6, 2025, 9:56 PM

#

i do this

fiery turtle Jun 6, 2025, 9:56 PM

#

like this linux stuff is basically a loop just unrolled and optimized at compile time

fiery turtle Jun 6, 2025, 9:56 PM

#

slim panther i do this

makes sense

slim panther Jun 6, 2025, 9:57 PM

#

you're going to have a blast with DRM actually

#

linux drm is one of the best documented subsystems IME

fiery turtle Jun 6, 2025, 9:58 PM

#

yeah im kinda ridiculously far from that atm, but ill have to devise some shitty impl to have a dumbfb over the vesa/gop thing

slim panther Jun 6, 2025, 9:58 PM

#

just do linux fbdev

#

it's the simplest thing you can do

fiery turtle Jun 6, 2025, 9:58 PM

#

fbdev works over a drm dumbfb i think

#

like internally

slim panther Jun 6, 2025, 9:59 PM

#

not quite

#

there's helper functions that can translate fbdev stuff into drm

#

that's what i used to translate the s3fb driver to drm

#

but you can still have a raw fbdev backed driver

fiery turtle Jun 6, 2025, 10:00 PM

#

ah ok ill have to look into that

#

i do want to do the kernel text mode thing

#

so i can at least run shit without drm

torpid root Jun 6, 2025, 10:00 PM

#

kernel text mode?

fiery turtle Jun 6, 2025, 10:01 PM

#

yeah like /dev/ttyX

slim panther Jun 6, 2025, 10:01 PM

#

just kernel vt i guess

torpid root Jun 6, 2025, 10:01 PM

#

ah

#

I thought you meant something else

fiery turtle Jun 6, 2025, 10:01 PM

#

the thing u get when u press ctrl+alt+a number or wahtever

slim panther Jun 6, 2025, 10:01 PM

#

yea kernel vt

torpid root Jun 6, 2025, 10:01 PM

#

tty

fiery turtle Jun 6, 2025, 10:02 PM

#

that

#

i think thats the default mode and then a drm master can switch it into graphical

slim panther Jun 6, 2025, 10:03 PM

#

roughly the kernel is the drm master, but it's like a weak ref because the second a user process requests to be the master it hands it over

fiery turtle Jun 6, 2025, 10:03 PM

#

makes sense

#

ill have to make a uflanterm trl

slim panther Jun 6, 2025, 10:03 PM

#

multiple people already started doing that

fiery turtle Jun 6, 2025, 10:04 PM

#

really?

slim panther Jun 6, 2025, 10:04 PM

#

yea

torpid root Jun 6, 2025, 10:04 PM

#

would it possible to do a gui in vga text mode, if you loaded a 8x16 font that had 128 characters with each having a single pixel occupied at different position
then you could draw basically anything

slim panther Jun 6, 2025, 10:04 PM

#

(including me)

slim panther Jun 6, 2025, 10:04 PM

#

torpid root would it possible to do a gui in vga text mode, if you loaded a 8x16 font that h...

no

torpid root Jun 6, 2025, 10:04 PM

#

why not?

fiery turtle Jun 6, 2025, 10:04 PM

#

slim panther (including me)

is it public

slim panther Jun 6, 2025, 10:04 PM

#

no

#

it doesn't work lmao

fiery turtle Jun 6, 2025, 10:04 PM

#

are u doing it because NIH

#

or rust

#

or

slim panther Jun 6, 2025, 10:05 PM

#

because i wanted to do some drm userland programming

#

i've done very little drm programming

fiery turtle Jun 6, 2025, 10:06 PM

#

ah u mean for userland

#

are u going to have a kernel vt?

slim panther Jun 6, 2025, 10:07 PM

#

i have a flanterm buffer but i do want a simpledrm fb

frank canopy Jun 6, 2025, 10:35 PM

#

fiery turtle once i have that ill start building the sparse struct page array

how do you plan to handle the sparse array accesses? assuming by sparse you mean what I think you do, how do you control what is in the array and check for if a page is in the array?

fiery turtle Jun 6, 2025, 10:39 PM

#

frank canopy how do you plan to handle the sparse array accesses? assuming by sparse you mean...

im not sure what u mean exactly, by sparse i mean that the struct page array has a fixed base, but only present physical memory will have an associated struct page mapped at the corresponding offset

#

some fake struct pages for the buddy allocator max order alignment

tired temple Jun 6, 2025, 10:53 PM

#

PFN database?

fiery turtle Jun 6, 2025, 10:55 PM

#

i think thats what they call it in NT land yeah

slim panther Jun 6, 2025, 11:14 PM

#

frank canopy how do you plan to handle the sparse array accesses? assuming by sparse you mean...

not usable memory = indices aren't mapped

#

easy as

frank canopy Jun 6, 2025, 11:15 PM

#

slim panther not usable memory = indices aren't mapped

yes but how do you actually guard against that, and also how do you handle when a not usable page has a struct page in the same page as a usable page

dense steppe Jun 6, 2025, 11:15 PM

#

frank canopy how do you plan to handle the sparse array accesses? assuming by sparse you mean...

you pretty much never need to test if there is a page at phys addr X

slim panther Jun 6, 2025, 11:15 PM

#

frank canopy yes but how do you actually guard against that, and also how do you handle when ...

then you map it

dense steppe Jun 6, 2025, 11:16 PM

#

if you got phys addr X from someone, that means there is a page there

#

(or it is device memory and you're not going to use struct pages for it anyway*)

frank canopy Jun 6, 2025, 11:17 PM

#

dense steppe (or it is device memory and you're not going to use struct pages for it anyway*)

my map for mmio stuff is the same function as to make an temporary extra kernel-mode mapping of existing memory so i need to do that check in there for page refcounting

dense steppe Jun 6, 2025, 11:18 PM

#

ah

#

so you need some kind of pfn_is_ram() or something?

#

by the nature of page tables, it is actually possible to implement this with another "virtual" array, where pfn_is_page is just

extern uint64_t *pfn_bitmap;
static inline bool pfn_is_ram(pfn_t pfn)
{
        return pfn_bitmap[pfn/64] & (pfn%64);
}

#

this can be done by aliasing portions of the array with identical contents

#

and this aliasing can possibly even be done at higher page table levels too (so e.g. pml4[foo] = pml4[bar] = a pml3, with all entries pointing to the same pml2, with all entries pointing to the same pml1, with all entries pointing to a zero page

#

IMO this is a little too much fun to have with page tables

dense steppe Jun 6, 2025, 11:25 PM

#

frank canopy my map for mmio stuff is the same function as to make an temporary extra kernel-...

why does the temporary extra kernel-mode mapping need refcounting on the pages?

frank canopy Jun 6, 2025, 11:27 PM

#

dense steppe why does the temporary extra kernel-mode mapping need refcounting on the pages?

to prevent them getting paged out

frank canopy Jun 6, 2025, 11:28 PM

#

dense steppe by the nature of page tables, it is actually possible to implement this with ano...

ok yeah using a bitmap is what im doing, was just curious if there was a different method i was missing

dense steppe Jun 6, 2025, 11:29 PM

#

frank canopy to prevent them getting paged out

but presumably the caller holds a reference to the page already

#

and will unmap before dropping that reference

#

or, by refcount do you mean something like mlock?

frank canopy Jun 6, 2025, 11:30 PM

#

dense steppe but presumably the caller holds a reference to the page already

the main case im running into is io buffers from usermode

#

so a usermode thing passes a usermode buffer to the kernel io stuff

dense steppe Jun 6, 2025, 11:31 PM

#

and the kernel wants to access that buffer from another context e.g. DPC or some other callback somewhere (when it might be running in a different address space)?

frank canopy Jun 6, 2025, 11:32 PM

#

dense steppe and the kernel wants to access that buffer from another context e.g. DPC or some...

yep, so the problem case is:

usermode gives buffer to kernel io stuff
user process isnt running on processor that runs the io callback
user process phys page gets swapped out to disk

#

could have the refcount not part of the map function but i prefer it in there so i cant forget

dense steppe Jun 6, 2025, 11:33 PM

#

yup, makes sense

frank canopy Jun 6, 2025, 11:34 PM

#

i ended up using a bitmap after looking into how reactos does it since my kernel is very NT-inspired so react is a good reference for some general stuff like that

dense steppe Jun 6, 2025, 11:34 PM

#

frank canopy could have the refcount not part of the map function but i prefer it in there so...

or you could have a

page *get_user_page(uintptr_t addr)
{
    // increase mlock_count
}
void put_user_page(page *pg)
{
    // decrease mlock_count
}

frank canopy Jun 6, 2025, 11:35 PM

#

yeah ig

#

and i still might end up changing this idk

#

hell i dont even have usermode yet so it doesnt matter lmao

dense steppe Jun 6, 2025, 11:37 PM

#

lol

#

well

#

it's good to plan ahead

#

anyway so before you've mapped a page into the kernel address space, do you hold some lock or something which prevents it from being swapped concurrently?

dense steppe Jun 6, 2025, 11:40 PM

#

frank canopy yep, so the problem case is: 1. usermode gives buffer to kernel io stuff 1. user...

also there's a tricky case to be aware of: another thread can be performing a simultaneous mmap(MAP_FIXED) that is replacing the portion of the address space you are doing IO on. this means that you can pin the page, dispatch IO, some other thread unmaps the page, IO callback writes to the page, kaboom unless you are careful

#

IMO: for synchronous IO it is best to have a copy_to_user that runs in the context of the calling process.

#

and then the page fault handler can fault in pages or make copy_to_user return EFAULT

#

asynchronous IO is tricky, but since you're doing NT-like you may want to be aware of the existence of something called the APC (iirc) which runs in the context of the enqueueing process (iirc).

#

someone will probably dispatch a serial killer to me for butchering NT terminology lmao

frank canopy Jun 6, 2025, 11:47 PM

#

dense steppe asynchronous IO is tricky, but since you're doing NT-like you may want to be awa...

context of a thread even but yeah

frank canopy Jun 6, 2025, 11:48 PM

#

dense steppe also there's a tricky case to be aware of: another thread can be performing a si...

the whole thing is that theres a separate kernel mode temp mapping so changing out the virtual mapping of the buffer page doesnt break anything

frank canopy Jun 6, 2025, 11:48 PM

#

dense steppe anyway so before you've mapped a page into the kernel address space, do you hold...

yeah theres a one-bit spinlock in the page struct

#

anyway ig if this is going to turn into rambling about my kernel it should be done in #1312232715193683988 instead lol

dense steppe Jun 7, 2025, 12:02 AM

#

well that's not going to happen since I'm a human being who needs sleep meme

fiery turtle Jun 7, 2025, 12:02 AM

#

refuse sleep

#

program 24 7 instead trl

fiery turtle Jun 7, 2025, 11:17 AM

#

What's strange is Linuxes PAE support seems to be compile time

#

Like it cant dynamically disable it if the CPU doesnt have it

#

Like it can for 5 level paging for example

gentle peak Jun 7, 2025, 11:21 AM

#

pae is much more invasive than adding an extra level

#

the new level is a different size than other levels, it behaves differently, and most importantly the page table entry size (and number of entries per table) changes

fiery turtle Jun 7, 2025, 11:22 AM

#

i think the most important one is the entry size change, their model cant easily adapt to that

gentle peak Jun 7, 2025, 11:22 AM

#

yeah

#

both that and the number of entries per table (and thus the vaddr index encoding change) are really annoying to deal with at runtime

#

especially if your model hasn't been made with it in mind

fiery turtle Jun 7, 2025, 11:23 AM

#

number of entries they can handle dynamically, they pretend the p4d has 1 entry if 5 lvl paging is not supported

gentle peak Jun 7, 2025, 11:23 AM

#

is that not just a special case for folding?

fiery turtle Jun 7, 2025, 11:24 AM

#

well im not sure what u mean, but folding works by just unconditionally returning true to pgd_present() and reusing the pgd offset in p4d_offset

#

im not sure exactly what code relies on number of entries per thing

gentle peak Jun 7, 2025, 11:25 AM

#

the problem with number of entries per table actually changing, and not just pretending X amount of top-level tables are 1 entry, is that it completely changes the layout of the vaddr bitfield

#

with pae you go from 10:10:12 to 2:9:9:12

fiery turtle Jun 7, 2025, 11:26 AM

#

technically all that is hidden behind thing_offset, but some code might be using publically available "constants", not sure

gentle peak Jun 7, 2025, 11:26 AM

#

fair enough

fiery turtle Jun 7, 2025, 11:27 AM

#

i think some code in ASAN or hugetlbfs are places i've seen referencing these

#

but not sure how much of a problem this is

#

arch/x86/boot/cpu.c-75- if (err_flags) {
arch/x86/boot/cpu.c:76:         puts("This kernel requires the following features "
arch/x86/boot/cpu.c-77-              "not present on the CPU:\n");

Yeah it just dies on boot if u dont have PAE

#

i guess pae is old enough so whatever

#

this is also interesting

/*
 * Checks whether a page fault on the given mm is still reliable.
 * This is no longer true if the oom reaper started to reap the
 * address space which is reflected by MMF_UNSTABLE flag set in
 * the mm. At that moment any !shared mapping would lose the content
 * and could cause a memory corruption (zero pages instead of the
 * original content).
 *
 * User should call this before establishing a page table entry for
 * a !shared mapping and under the proper page table lock.
 *
 * Return 0 when the PF is safe VM_FAULT_SIGBUS otherwise.
 */
static inline vm_fault_t check_stable_address_space(struct mm_struct *mm)
{
    if (unlikely(test_bit(MMF_UNSTABLE, &mm->flags)))
        return VM_FAULT_SIGBUS;
    return 0;
}

#

while a process is being reaped by the oom killer its considered to have an unstable address space

#

there are some ifs in the pf handler to not cause recursion i guess

gentle peak Jun 7, 2025, 11:51 AM

#

does the oom reaper free the mm before terminating the threads or something?

fiery turtle Jun 7, 2025, 11:51 AM

#

gentle peak does the oom reaper free the mm before terminating the threads or something?

yeah

#

if the process being killed is the one that caused the OOM it gets to satisfy the allocation that caused the OOM by killing itself

#

so __alloc_pages spins until the OOM killer reclaims enough memory to satisfy the allocation

#

which could be in the context of the process being killed

gentle peak Jun 7, 2025, 11:53 AM

#

ah makes sense ig

fiery turtle Jun 7, 2025, 12:14 PM

#

im confused about this math lol

static inline int page_level_shift(enum pg_level level)
{
    return (PAGE_SHIFT - PTE_SHIFT) + level * PTE_SHIFT;
}

#

why the hell does it subtract pte shift from page shift

#

shouldn't it be just PAGE_SHIFT + level * PTE_SHIFT?

dense steppe Jun 7, 2025, 12:15 PM

#

maybe it's one-indexed?

fiery turtle Jun 7, 2025, 12:15 PM

#

so it basically does 3 + level * 9

gentle peak Jun 7, 2025, 12:15 PM

#

then the user of this api probably operates on VFNs, not vaddrs

#

actually

#

hm

fiery turtle Jun 7, 2025, 12:16 PM

#

hmm not sure

#

https://elixir.bootlin.com/linux/v6.15/source/arch/x86/include/asm/pgtable.h#L1550

#

this code here

gentle peak Jun 7, 2025, 12:17 PM

#

yeah nevermind my guess for what PTE_SHIFT was was incorrect

fiery turtle Jun 7, 2025, 12:17 PM

#

it's basically the number of bits covered by one level

#

in this case the bottom most table

gentle peak Jun 7, 2025, 12:18 PM

#

yeah i was guessing it was the number of bits in a hardware page (to support software page size > hardware page size)

fiery turtle Jun 7, 2025, 12:18 PM

#

im gonna assume this function is just bullshit

#

because it doesnt math

#

Author: Dave Hansen <[email protected]>
Date:   Tue Jan 22 13:24:31 2013 -0800

    x86, mm: Pagetable level size/shift/mask helpers

    I plan to use lookup_address() to walk the kernel pagetables
    in a later patch.  It returns a "pte" and the level in the
    pagetables where the "pte" was found.  The level is just an
    enum and needs to be converted to a useful value in order to
    do address calculations with it.  These helpers will be used
    in at least two places.

    This also gives the anonymous enum a real name so that no one
    gets confused about what they should be passing in to these
    helpers.

    "PTE_SHIFT" was chosen for naming consistency with the other
    pagetable levels (PGD/PUD/PMD_SHIFT).

    Cc: H. Peter Anvin <[email protected]>
    Signed-off-by: Dave Hansen <[email protected]>
    Link: http://lkml.kernel.org/r/[email protected]
    Signed-off-by: H. Peter Anvin <[email protected]>

gentle peak Jun 7, 2025, 12:20 PM

#

ah i figured it out

#

pg_level is 1-indexed

#

a value of 0 is PG_LEVEL_NONE

fiery turtle Jun 7, 2025, 12:21 PM

#

yeah

gentle peak Jun 7, 2025, 12:21 PM

#

PG_LEVEL_4K is 1, so (12 - 9) + 1 * 9 = 3 + 9 = 12

fiery turtle Jun 7, 2025, 12:21 PM

#

what is that number tho?

gentle peak Jun 7, 2025, 12:21 PM

#

it's the number of bits to shift the vaddr by before masking to get the index for the given level

fiery turtle Jun 7, 2025, 12:22 PM

#

jesus lol

gentle peak Jun 7, 2025, 12:22 PM

#

page_level_size is then the size of 1 entry in the given level

fiery turtle Jun 7, 2025, 12:22 PM

#

least convoluted way to write this

gentle peak Jun 7, 2025, 12:22 PM

#

yeah i can't help but feel that PAGE_SHIFT + (level - 1) * PTE_SHIFT would've been clearer lol

fiery turtle Jun 7, 2025, 12:22 PM

#

yeah like wtf

#

that results in the same number right?

gentle peak Jun 7, 2025, 12:23 PM

#

yes

fiery turtle Jun 7, 2025, 12:23 PM

#

some peoples brains are just built different idk lmao

fiery turtle Jun 7, 2025, 7:38 PM

#

for whatever reason i spent a few hours studying how rseq(2) works

#

which is fine ig because its one of the first syscalls used by glibc

#

also me when set_robust_list

#

Also if anyone has updated scores they want to submit for the uacpi leaderboard now is a good time

frank canopy Jun 7, 2025, 7:44 PM

#

i think you already have my latest

fiery turtle Jun 7, 2025, 7:46 PM

#

nice

#

i have too much temporary shit in my head, I need to implement it quickly before i forget everything lmao

frank canopy Jun 7, 2025, 7:48 PM

#

(not on the readme mind you, but you tested mine at some point past)

fiery turtle Jun 7, 2025, 7:49 PM

#

can u send a link if u have it?

frank canopy Jun 7, 2025, 7:49 PM

#

#1217009725711847465 message found the message

fiery turtle Jun 7, 2025, 7:49 PM

#

ah

dusky glade Jun 7, 2025, 8:08 PM

#

infy cooking with ukernel

fiery turtle Jun 7, 2025, 8:17 PM

#

Nah

#

Absolutely not

slim panther Jun 7, 2025, 8:18 PM

#

libk

#

literally

torpid root Jun 7, 2025, 8:19 PM

#

new mod colour

vast kestrel Jun 7, 2025, 8:21 PM

#

fiery turtle for whatever reason i spent a few hours studying how rseq(2) works

Rseq is cool

flat badge Jun 7, 2025, 8:32 PM

#

rseq is about the least important syscall halfmemeleft

fiery turtle Jun 7, 2025, 8:34 PM

#

it's quite invasive which is interesting

#

like various code paths must check for rseq_cs presence explicitly every time

#

in the core scheduling and signal delivery etc

#

but its pretty simple anyway

#

also i learned about mm_cid, which is a relatively new thing

#

aka concurrency id

#

basically it allows u to cheaply index per-cpu data in userspace using the cid instead of the cpu id

#

e.g. if u have 256 cpus and e.g. 4 threads in your process and your thread 4 is scheduled on cpu 255 the cpu_id will be 255 but cid will be only 3

#

basically allowing you to have smaller arrays

fiery turtle Jun 7, 2025, 9:57 PM

#

damn didnt expect this much macro nesting to work tbh

#

but this means I can do this trl

fiery turtle Jun 7, 2025, 11:05 PM

#

i think I have most x86-specific page table code done, one thing left is huge page support and implementing ptX_populate()/prot bits

dense steppe Jun 7, 2025, 11:07 PM

#

no one needs huge pages anyways, outside of mapping the HHDM, right?

fiery turtle Jun 7, 2025, 11:07 PM

#

basically yeah, atm im only gonna use them for HHDM

dense steppe Jun 7, 2025, 11:08 PM

#

dense steppe no one needs huge pages anyways, outside of mapping the HHDM, right?

(note: I am aware of the usecases of hugepages, this is just me memeing, no need to lecture me about database uses for it and such)

fiery turtle Jun 7, 2025, 11:08 PM

#

im definitely not gonna support THP or hugetlbfs any time soon

#

its too much adhoc code everywhere

dense steppe Jun 7, 2025, 11:09 PM

#

THP is a mess

fiery turtle Jun 7, 2025, 11:11 PM

#

yep

flat badge Jun 8, 2025, 6:23 AM

#

fiery turtle but this means I can do this <:trl:829316641191886888>

you know, there are languages with proper support for things like that

#

meme

vast kestrel Jun 8, 2025, 8:26 AM

#

Never heard of em

lofty dragon Jun 8, 2025, 9:02 AM

#

@fiery turtle shame that uacpi doesn't have madt helpers

#

the madt is annoying to parse so those would've been nice

#

i mean it's not a big big deal

#

just something that'd be nice

fiery turtle Jun 8, 2025, 9:25 AM

#

lofty dragon <@353593948523986945> shame that uacpi doesn't have madt helpers

U mean like an iterator for it? It was suggested earlier by someone else as well

fiery turtle Jun 8, 2025, 9:25 AM

#

flat badge you know, there are languages with proper support for things like that

trl

dense steppe Jun 8, 2025, 9:39 AM

#

it's really just two functions that would need to be implemented:

acpi_entry_hdr *uacpi_subtable_first(acpi_sdt_hdr *hdr, size_t hdr_len)
{
    uintptr_t end = (uintptr_t) hdr + hdr->length;
    uintptr_t addr = (uintptr_t) hdr + hdr_len;

    return (addr < end) ? (acpi_entry_hdr *) addr : nullptr;
}

acpi_entry_hdr *uacpi_subtable_next (acpi_sdt_hdr *hdr, acpi_entry_hdr *entry)
{
    uintptr_t end = (uintptr_t) hdr + hdr->length;
    uintptr_t addr = (uintptr_t) entry + entry->length;

    return (addr < end) ? (acpi_entry_hdr *) addr : nullptr;
}

this should give a nice iterator interface to any such table with subtables in that same format

#

acpi_entry_hdr *uacpi_madt_first(acpi_madt *madt)
{
    return uacpi_subtable_first(&madt->hdr, sizeof(*madt));
}

acpi_entry_hdr *uacpi_madt_next(acpi_madt *madt, acpi_entry_hdr *entry)
{
    return uacpi_subtable_next(&madt->hdr, entry);
}

fiery turtle Jun 8, 2025, 9:48 AM

#

Yeah something like that ig

#

I might have both a macro version and a function version

#

Similar to resource iteration

lofty dragon Jun 8, 2025, 11:57 AM

#

fiery turtle U mean like an iterator for it? It was suggested earlier by someone else as well

yeah, like a function to parse it handily or so

fiery turtle Jun 8, 2025, 11:59 AM

#

yeah

#

ill add that

north hull Jun 10, 2025, 8:39 AM

#

do I need to call any of the init before using uacpi to discover tables

fiery turtle Jun 10, 2025, 8:40 AM

#

just uacpi_setup_early_table_access

north hull Jun 10, 2025, 8:40 AM

#

ah

#

thanks

#

so should I just pass like a page of memory to the early access

#

realistically I need access to maybe three tables lol

fiery turtle Jun 10, 2025, 8:43 AM

#

it doesnt matter how many tables u want to access, the array will store all existing ones so yeah

#

it has a very large doc string above the function btw

north hull Jun 10, 2025, 8:43 AM

#

ahh

#

yeah I just misread it

#

page should suffice tho?

fiery turtle Jun 10, 2025, 8:43 AM

#

yeah

#

or if u have heap available at this stage u can just uacpi_initialize

#

and not worry about static buffers

north hull Jun 10, 2025, 8:43 AM

#

oh

#

alright then ill do that

#

mb

fiery turtle Jun 10, 2025, 8:44 AM

#

yeah

fiery turtle Jun 10, 2025, 8:47 PM

#

https://github.com/acpica/acpica/issues/1027

GitHub

Calling method with wrong number of arguments leads to unbootable L...

A recent Dell firmware update has caused various Dell laptops, and other systems, to become unbootable on several Linux distributions, including Fedora Linux. This is a particularly severe problem ...

#

wait how tf is this still not handled by acpica lmao

kind mantle Jun 10, 2025, 8:53 PM

#

bruh

#

how

#

-# Rust would've prevented this trl

fiery turtle Jun 10, 2025, 8:54 PM

#

rewrite acpica in rust trl

dense steppe Jun 10, 2025, 9:03 PM

#

I believe the real problem is that ACPICA is architectured in a way where it blindly trusts a machine's AML. This is not from a security perspective (i.e. prevent malicious AML from pwning you and stealing all of your cookies) but rather a fault-tolerance one, AKA firmware devs should be able to ship sloppy firmware and its effect on us should be limited in scope.

kind mantle Jun 10, 2025, 9:04 PM

#

Fortunately, uACPI is good at that

#

But of course, making Linux use uACPI is probably never gonna happen

fiery turtle Jun 10, 2025, 9:28 PM

#

kind mantle But of course, making Linux use uACPI is probably never gonna happen

true

sterile egret Jun 11, 2025, 1:03 AM

#

fiery turtle wait how tf is this still not handled by acpica lmao

Someone should reply with "I think this is another example of the need to move to a modern ACPI implementation, one that luckily enough already exists: <uACPI link here>"

#

meme

flat badge Jun 11, 2025, 5:43 AM

#

well, somebody would need to put in the work to replace all the acpica calls by uacpi

rustic compass Jun 11, 2025, 6:33 AM

#

We need a crowdstrike style Blackout for Linux systems caused by a firmware update, that might help get uacpi support meme

calm latch Jun 11, 2025, 6:52 AM

#

What about BSDs? thonk trl

fiery turtle Jun 11, 2025, 7:35 AM

#

depending on their level of abstraction for the acpi driver

vast kestrel Jun 11, 2025, 7:36 AM

#

My guess is zero :^)

flat badge Jun 11, 2025, 7:56 AM

#

i mentioned this before but if you want to get uacpi into linux, you need a acpica API compat

#

otherwise it's not going to happen

rustic compass Jun 11, 2025, 7:58 AM

#

Or someone with suspiciously insane amounts of free time and autistic level focus rewriting all call sites to uacpi

fiery turtle Jun 11, 2025, 8:04 AM

#

flat badge i mentioned this before but if you want to get uacpi into linux, you need a acpi...

yeah, which is gonna be extremely annoying because they have global variables as public api, they have unsafe practices like allocating aml objects on the stack as part of user api, no opaque objects etc

#

but if anyone wants to do it unironically ill gladly merge an optional compat layer

woeful rock Jun 11, 2025, 8:22 AM

#

wait, I thought the only way to parse aml was to lookup the function to see how many argument it expected... been a while, I'm probably misremembering.

fiery turtle Jun 11, 2025, 8:24 AM

#

it's possible that the caller frame just ends at two arguments so theres nothing to parse after, so acpica just continues the call

#

they didnt attach a dump so its impossible to say

#

i wanted to make a test case for this but u have to use a hex editor unfortunately

#

ah nvm managed to make it work

#

yeah lol acpica just jumps to the function anyway

#

Method (TEST, 3) {
    Printf("Arg0 %o Arg1 %o Arg2 %o", Arg0, Arg1, Arg2)
}

TEST(123, 321)

#

Table [DSDT: Local   ] (id 01) -    0 Objects with   0 Devices,   0 Regions,    0 Methods (0/0/0 Serial/Non/Cvt)
ACPI Error: Uninitialized Arg[2] at node 0x5977751aed78 (20230628/dsmthdat-471)
ACPI Error: AE_AML_UNINITIALIZED_ARG, While resolving operands for [Concatenate] (20230628/dswexec-505)
Failed at AML Offset 00001, Opcode 0073: Concatenate (-Return Value- (), Arg2)
No Local Variables are initialized for Method [TEST]

Initialized Arguments for Method [TEST]:  (3 arguments defined for method invocation)
    Arg0:   0x5977751aebc0 <Obj>             Integer 000000000000007B
    Arg1:   0x5977751ae910 <Obj>             Integer 0000000000000141

ACPI Error: Aborting method \TEST due to previous error (AE_AML_UNINITIALIZED_ARG) (20230628/psparse-580)
ACPI Error: Aborting method \ due to previous error (AE_AML_UNINITIALIZED_ARG) (20230628/psparse-580)
ACPI Error: Invalid zero thread count in method (20230628/dsmethod-869)
ACPI Error: Invalid OwnerId: 0x000 (20230628/utownerid-210)
ACPI Error: AE_AML_UNINITIALIZED_ARG, (SSDT:RUNRIDTB) while loading table (20230628/tbxfload-244)

#

idk if it actually corrupted some memory with this or not

#

uacpi just skips this call, it's never invoked because it doesnt parse enough arguments

#

[uACPI][TRACE] load of 'DSDT' (OEM ID 'uTEST' OEM Table ID 'RUNRIDTB')
[uACPI][DEBUG] BEGIN OP 'MethodOp' (0x0014)
[uACPI][DEBUG]     pOP: TRACKED_PKGLEN (0x0F)
[uACPI][DEBUG]     pOP: CREATE_NAMESTRING_OR_NULL_IF_LOAD (0x11)
[uACPI][DEBUG]     pOP: LOAD_IMM (0x1F)
[uACPI][DEBUG]     pOP: SKIP_WITH_WARN_IF_NULL (0x01)
[uACPI][DEBUG] END OP 'MethodOp' (0x0014)
[uACPI][DEBUG]     pOP: RECORD_AML_PC (0x1B)
[uACPI][DEBUG]     pOP: OBJECT_ALLOC_TYPED (0x1A)
[uACPI][DEBUG]     pOP: INVOKE_HANDLER (0x15)
[uACPI][DEBUG]     pOP: INSTALL_NAMESPACE_NODE (0x25)
[uACPI][DEBUG]     pOP: <END-OF-OP> (0x00)
[uACPI][DEBUG] END OP 'MethodOp' (0x0014)
[uACPI][DEBUG] BEGIN OP 'UACPI_InternalOpUnresolvedNameString_T' (0x0054)
[uACPI][DEBUG]     pOP: AML_PC_DECREMENT (0x2C)
[uACPI][DEBUG]     pOP: EXISTING_NAMESTRING_OR_NULL (0x13)
[uACPI][DEBUG]     pOP: CONVERT_NAMESTRING (0x31)
[uACPI][DEBUG]     pOP: LOAD_INLINE_IMM (0x1D)
[uACPI][DEBUG]     pOP: IF_NOT_NULL (0x35)
[uACPI][DEBUG]     pOP: TERM_ARG_UNWRAP_INTERNAL (0x07)
[uACPI][DEBUG] BEGIN OP 'BytePrefix' (0x000A)
[uACPI][DEBUG]     pOP: LOAD_IMM_AS_OBJECT (0x20)
[uACPI][DEBUG]     pOP: OBJECT_TRANSFER_TO_PREV (0x26)
[uACPI][DEBUG]     pOP: <END-OF-OP> (0x00)
[uACPI][DEBUG] END OP 'BytePrefix' (0x000A)
[uACPI][DEBUG] RESUME OP 'InternalOpMethodCall3Args' (0x00FA)
[uACPI][DEBUG]     pOP: OBJECT_CONVERT_TO_SHALLOW_COPY (0x18)
[uACPI][DEBUG]     pOP: IMM_DECREMENT (0x2D)
[uACPI][DEBUG]     pOP: JMP (0x3C)
[uACPI][DEBUG]     pOP: IF_NOT_NULL (0x35)
[uACPI][DEBUG]     pOP: TERM_ARG_UNWRAP_INTERNAL (0x07)
[uACPI][DEBUG] BEGIN OP 'WordPrefix' (0x000B)
[uACPI][DEBUG]     pOP: LOAD_IMM_AS_OBJECT (0x20)
[uACPI][DEBUG]     pOP: OBJECT_TRANSFER_TO_PREV (0x26)
[uACPI][DEBUG]     pOP: <END-OF-OP> (0x00)
[uACPI][DEBUG] END OP 'WordPrefix' (0x000B)
[uACPI][DEBUG] RESUME OP 'InternalOpMethodCall3Args' (0x00FA)
[uACPI][DEBUG]     pOP: OBJECT_CONVERT_TO_SHALLOW_COPY (0x18)
[uACPI][DEBUG]     pOP: IMM_DECREMENT (0x2D)
[uACPI][DEBUG]     pOP: JMP (0x3C)
[uACPI][DEBUG]     pOP: IF_NOT_NULL (0x35)
[uACPI][DEBUG]     pOP: TERM_ARG_UNWRAP_INTERNAL (0x07)
[uACPI][INFO] successfully loaded 1 AML blob, 4 ops in 2ms (avg 1887/s)

woeful rock Jun 11, 2025, 8:41 AM

#

What if there is a termarg after the the bad call to test ? Like just a math op or a literal...

fiery turtle Jun 11, 2025, 8:42 AM

#

in the byte stream it would be parsed as an argument to the function invocation, simply because there's no way to tell if its a standalone op or an argument

#

well up to the number of arguments the function has

#

the rest of standalone ops would be skipped

#

Like if u have TEST(123, 321, 123, 321, 123 , 321)

#

it would parse the first 3

#

the rest would be skipped since it's basically an rvalue that just dies without side effects

woeful rock Jun 11, 2025, 8:44 AM

#

Aml really bad. Maybe not much you can do to mitigate a corrupt stream...

fiery turtle Jun 11, 2025, 8:44 AM

#

well best u can do is just make it so u dont care if its a corrupted stream

#

but yeah it was a really stupid decision to not have a MethodCall opcode

#

By doing this they also left it to the implementation to decide when a named method should be invoked and when it simply produces a reference to the method

#

so whether u invoke a method is context dependant

#

which is insane

#

for example Package { METHOD } here u just store a ref, but Add (123, METHOD) here u call the method

#

but both just have the method name there, no other opcodes

woeful rock Jun 11, 2025, 8:49 AM

#

Yeah, I remember having headaches dealing with refs

fiery turtle Jun 11, 2025, 8:49 AM

#

yeah especially since refs work entirely different on acpica vs nt

woeful rock Jun 11, 2025, 8:51 AM

#

That bug report above is also bogus... an external doesn't specify the number of args. (I think, right ?)

surreal current Jun 11, 2025, 8:51 AM

#

NT!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!! >:(((((((((((((((((((((( NT!!!!!!!!!! >:((((

#

D:<<<< NT!!!

fiery turtle Jun 11, 2025, 8:51 AM

#

woeful rock That bug report above is also bogus... an external doesn't specify the number of...

external is generated by the iasl diasssembler, it doesnt even exist in the original dump

#

ExternalOp is always a no-op anyway

woeful rock Jun 11, 2025, 8:51 AM

#

Exactly

lofty dragon Jun 11, 2025, 3:26 PM

#

@fiery turtle to disengage uACPI wholly what does one need to do? is calling uacpi_state_reset() enough?

fiery turtle Jun 11, 2025, 3:39 PM

#

Yes

lofty dragon Jun 11, 2025, 3:46 PM

#

fiery turtle Yes

asking because even after doing that you need to long press the power button to turn the system off

vast kestrel Jun 11, 2025, 4:36 PM

#

Tbf I am not sure how many bioses properly support disabling acpi mode

torpid root Jun 11, 2025, 4:44 PM

#

leaderboard :(

fiery turtle Jun 11, 2025, 5:39 PM

#

lofty dragon asking because even after doing that you need to long press the power button to ...

Csmwrap doesn't even enter acpi mode to begin with

#

Iirc

#

So that shouldn't matter

fiery turtle Jun 11, 2025, 5:42 PM

#

torpid root leaderboard :(

Lol

#

Ill get on it soon lol

#

Need to draft 3.0

median crest Jun 11, 2025, 6:16 PM

#

multiply astral score by 10 billion

fiery turtle Jun 11, 2025, 6:33 PM

#

real

fierce kiln Jun 11, 2025, 6:37 PM

#

Astral ops/s = NaNeInf

fiery turtle Jun 11, 2025, 6:39 PM

#

balatro ref

fiery turtle Jun 12, 2025, 11:26 AM

#

alright uh

#

time to release uacpi 3.0 and do leaderboard updates

#

i also have 10 billion ultra changes not commited

torpid root Jun 12, 2025, 11:30 AM

#

fiery turtle i also have 10 billion ultra changes not commited

how are ultra changes different from normal ones? trl

calm latch Jun 12, 2025, 11:32 AM

#

uACPI jit when? trl

fiery turtle Jun 12, 2025, 11:40 AM

#

torpid root how are ultra changes different from normal ones? <:trl:829316641191886888>

lol

#

#

added some comments to resources.h

#

since its a very very common thing to fuck up

#

I kinda want to try making a speed benchmark comparison for ACPICA vs uacpi to add to the readme

#

like im almost 100% sure uacpi is at least 2-3 times faster, but i cant prove it without a lot of effort

#

both uacpi's test runner and acpica's acpiexec have a ton of extra crap that they do as part of init just to test that stuff works

#

so id have to modify both

#

ah fuck it ill give it a try anyway

#

ok i have acpiexec built with -flto and -O3 and with some extra init stripped out

#

ill make one for uacpi as well

fiery turtle Jun 12, 2025, 12:37 PM

#

infy@DESKTOP-IGT40G0:/mnt/d/uACPI/tests$ /mnt/e/Projects/acpica/generate/unix/bin/acpiexec -to 100 x.aml

Intel ACPI Component Architecture
AML Execution/Debug Utility version 20250404
Copyright (c) 2000 - 2025 Intel Corporation

Automatic loop timeout after 100 seconds

Input file x.aml, Length 0x62 (98) bytes

ACPI: RSDP 0x00006111189D3A80 000024 (v02 Intel )
ACPI: XSDT 0x000061112F6FB4E0 00003C (v00 Intel  AcpiExec 00001001 INTL 20250404)
ACPI: FACP 0x00006111189D3B40 000114 (v05 Intel  AcpiExec 00001001 INTL 20250404)
ACPI: DSDT 0x00006111189D2100 000024 (v02 Intel  Local    00000001 INTL 20090730)
ACPI: FACS 0x00006111189D3A40 000040
ACPI: SSDT 0x000061112F6FC4F0 000062 (v02 uTEST  RUNRIDTB F0F0F0F0 INTL 20230628)

ACPI table initialization:
Table [DSDT: Local   ] (id 01) -    0 Objects with   0 Devices,   0 Regions,    0 Methods (0/0/0 Serial/Non/Cvt)
ACPI Debug:  "10,000,000 additions: 00005AF31112D680"
Table [SSDT: RUNRIDTB] (id 02) -    1 Objects with   0 Devices,   0 Regions,    1 Methods (0/1/0 Serial/Non/Cvt)
ACPI: 2 ACPI AML tables successfully acquired and loaded

Final data object initialization: Namespace contains 10 (0xA) objects
elapsed nanoseconds 19443959481 (19443 ms)
Initializing General Purpose Events (GPEs):
    Initialized GPE 00 to 7F [_GPE] 16 regs on interrupt 0x0 (SCI)
    Initialized GPE 80 to FF [_GPE] 16 regs on interrupt 0x0 (SCI)
Initializing Device/Processor/Thermal objects and executing _INI/_STA methods:
    Executed 0 _INI methods requiring 0 _STA executions (examined 2 objects)
- q
ACPI: No outstanding allocations
infy@DESKTOP-IGT40G0:/mnt/d/uACPI/tests$ runner/build-linux-64bits/test-runner x.aml --while-loop-timeout 100
[uACPI][INFO] starting uACPI, version 2.1.1
[uACPI][INFO] RSDP 0x00007FFE288BF200 00000024 v02 (uOEMID)
[uACPI][INFO] XSDT 0x000058F0A492D5C0 0000002C v02 ( uTEST uTESTTBL)
[uACPI][INFO] DSDT 0x000058F0A492E5D0 00000062 v02 ( uTEST RUNRIDTB)
[uACPI][INFO] FACP 0x000058F0A492D600 00000114 v06 (uOEMID uTESTTBL)
[uACPI][INFO] FACS 0x000058F0A492D720 00000040
[uACPI][TRACE] load of 'DSDT' (OEM ID 'uTEST' OEM Table ID 'RUNRIDTB')
[uACPI][TRACE] [AML DEBUG] String => "10,000,000 additions: 5af31112d680"
[uACPI][INFO] elapsed nanoseconds 6275607129 (6275 ms)
[uACPI][INFO] successfully loaded 1 AML blob, 110000015 ops in 6275ms (avg 17528186/s)
[uACPI][TRACE] initialized GPE block _GPE[0->128], 0 AML handlers (IRQ 0)
[uACPI][TRACE] initialized GPE block _GPE[128->256], 0 AML handlers (IRQ 0)
[uACPI][TRACE] fixed event 0 enabled successfully
[uACPI][TRACE] activated all 'EmbeddedControl' opregions controlled by '\', 0 _REG() calls (0 errors)
[uACPI][TRACE] activated all 'PCI_Config' opregions controlled by '\', 0 _REG() calls (0 errors)
[uACPI][INFO] namespace initialization done in 0ms: 0 devices, 0 thermal zones
[uACPI][TRACE] _STA calls: 0 (0 errors), _INI calls: 0 (0 errors)
[uACPI][TRACE] fixed event 0 disabled successfully

#

uACPI is 3 times faster LULW

#

this is just 10M additions

#

  Method (ADDM, 1, NotSerialized) {
      Local0 = 0

      While (Arg0) {
          Local0 += Arg0 + Arg0
          Arg0--
      }

      Return (Local0)
  }

  Printf("10,000,000 additions: %o", ADDM(10000000))

dusky glade Jun 12, 2025, 12:38 PM

#

les fuckin go

fiery turtle Jun 12, 2025, 12:39 PM

#

[uACPI][INFO] elapsed nanoseconds 6275607129 (6275 ms) vs elapsed nanoseconds 19443959481 (19443 ms)

dusky glade Jun 12, 2025, 12:39 PM

#

aml has a printof method??

fiery turtle Jun 12, 2025, 12:39 PM

#

its a macro that makes a series of ConcatOp

dusky glade Jun 12, 2025, 12:39 PM

#

ah okay i see

fiery turtle Jun 12, 2025, 12:40 PM

#

this is a completely fair test, both runners built with O3 and i put the exact same function to calculate time in the exact same places

dusky glade Jun 12, 2025, 12:40 PM

#

when uacpi 3.0 gets commited, ill see the ops for it on nyaux so u can update it

fiery turtle Jun 12, 2025, 12:40 PM

#

ah actually i can use the Timer operator

fiery turtle Jun 12, 2025, 12:40 PM

#

dusky glade when uacpi 3.0 gets commited, ill see the ops for it on nyaux so u can update it

it's already commited

#

just not formally

dusky glade Jun 12, 2025, 12:40 PM

#

wait really?

fiery turtle Jun 12, 2025, 12:40 PM

#

yes

dusky glade Jun 12, 2025, 12:41 PM

#

epic

fiery turtle Jun 12, 2025, 12:41 PM

#

nothing actually changed lol

#

just some minor things that caused backward compat breakage

dusky glade Jun 12, 2025, 12:41 PM

#

ah okay

gentle peak Jun 12, 2025, 12:42 PM

#

does acpica have sized frees?

fiery turtle Jun 12, 2025, 12:42 PM

#

dont think so

left orbit Jun 12, 2025, 12:42 PM

#

probably not

gentle peak Jun 12, 2025, 12:42 PM

#

fuck

fiery turtle Jun 12, 2025, 12:42 PM

#

why?

left orbit Jun 12, 2025, 12:42 PM

#

acpica is a dumpster fire

fiery turtle Jun 12, 2025, 12:43 PM

#

tbf i didnt expect a 3x difference on a simple math test lmao

#

i thought it would mostly be cpu bound

#

as in not interpreter logic

gentle peak Jun 12, 2025, 12:43 PM

#

fiery turtle why?

trying to implement it it in proxima to see an actual OS comparison with uacpi

fiery turtle Jun 12, 2025, 12:43 PM

#

ah damn

left orbit Jun 12, 2025, 12:43 PM

#

all that this really tests is the decode loop

fiery turtle Jun 12, 2025, 12:43 PM

#

yeah that would be cool if u did that

left orbit Jun 12, 2025, 12:43 PM

#

but you can only make it faster than that

#

considering add is the simplest op there is

fiery turtle Jun 12, 2025, 12:44 PM

#

yeah i guess simple math test basically tests how fast the interpreter loop is

#

but i think thats kinda what matters no

gentle peak Jun 12, 2025, 12:44 PM

#

guess i'll just have to do something like this

fiery turtle Jun 12, 2025, 12:44 PM

#

lmao

left orbit Jun 12, 2025, 12:44 PM

#

exactly

#

so while this is only 3x faster, in reality it's probably even faster

#

depending on how shit the implementation is

#

which i dont have high hopes for

fiery turtle Jun 12, 2025, 12:44 PM

#

ill try other test cases as well

left orbit Jun 12, 2025, 12:44 PM

#

but they also had like

#

numerous years

#

to optimize it

fiery turtle Jun 12, 2025, 12:45 PM

#

rn i want to convert it to use the TimerOp from AML so that the test is entirely self contained

fiery turtle Jun 12, 2025, 12:45 PM

#

left orbit numerous years

yeah thats kinda crazy

#

its 25 years old

left orbit Jun 12, 2025, 12:45 PM

#

holy shit

fiery turtle Jun 12, 2025, 12:45 PM

#

and still cant handle incorrect number of function arguments

#

Method (ADDM, 1, NotSerialized) {
    Local0 = 0

    While (Arg0) {
        Local0 += Arg0 + Arg0
        Arg0--
    }

    Return (Local0)
}

Method (TOMS, 2, NotSerialized) {
    Return ((Arg1 - Arg0) / 10000)
}

Local0 = Timer
Local1 = ADDM(10000000)
if (Local1 != 0x5AF31112D680) {
    Printf("Unexpected test result %o", Local1)
    Return (1)
}
Local2 = Timer

Printf("10,000,000 additions took %o ms",
       ToDecimalString(TOMS(Local0, Local2)))

There you go

dusky glade Jun 12, 2025, 12:50 PM

#

left orbit probably not

yooo iretq morning

hallow marten Jun 12, 2025, 12:50 PM

#

fiery turtle yeah i guess simple math test basically tests how fast the interpreter loop is

you should add libgccjit

#

to uacpi

#

unironically though would be neat for userspace meme

left orbit Jun 12, 2025, 12:51 PM

#

dusky glade yooo iretq morning

i did not sleep

#

"morning"

fiery turtle Jun 12, 2025, 12:51 PM

#

[uACPI][TRACE] [AML DEBUG] String => "10,000,000 additions took 6444 ms" vs ACPI Debug: "10,000,000 additions took 17094 ms"

#

there u go

#

super fair self contained test

dusky glade Jun 12, 2025, 12:51 PM

#

gentle peak guess i'll just have to do something like this

bro is also pro gamer, bro also uses sized frees

fiery turtle Jun 12, 2025, 12:51 PM

#

same cpu, same compiler options

dusky glade Jun 12, 2025, 12:51 PM

#

left orbit i did not sleep

that is fair

#

i did not sleep either

#

until like 2 hours ago

fiery turtle Jun 12, 2025, 12:52 PM

#

hallow marten unironically though would be neat for userspace <:meme:575445094589661215>

yeah but i wanna get out of uacpi hell

#

ive been working on it for two years

hallow marten Jun 12, 2025, 12:52 PM

#

fair

fiery turtle Jun 12, 2025, 12:52 PM

#

i kinda wanna work on my os

#

Any other interesting test case ideas?

hallow marten Jun 12, 2025, 12:52 PM

#

watch me add libgccjit to uacpi on my os only so it obliterates everyone on the leaderboard

left orbit Jun 12, 2025, 12:52 PM

#

aml jit sounds like a project at least as big as the interpreter itself

dusky glade Jun 12, 2025, 12:53 PM

#

real

fiery turtle Jun 12, 2025, 12:53 PM

#

Yeah its a huge project

left orbit Jun 12, 2025, 12:53 PM

#

would it even be feasible with the amount of weird shit aml does implicitly?

#

i mean, there's javascript jit

#

so why not lmao

fiery turtle Jun 12, 2025, 12:53 PM

#

well depends on how u do it, if u do a pass over the aml and convert it to jit calling into native interpreter handlers that would probably already be an order of magnitude speed up

#

and its not that hard

dusky glade Jun 12, 2025, 12:54 PM

#

famous last words

fiery turtle Jun 12, 2025, 12:54 PM

#

especially for math test like this we could be looking at 100x even

hallow marten Jun 12, 2025, 12:54 PM

#

fiery turtle well depends on how u do it, if u do a pass over the aml and convert it to jit c...

huh?

#

what I was thinking of was more like

fiery turtle Jun 12, 2025, 12:54 PM

#

just aml to machine code converter

hallow marten Jun 12, 2025, 12:55 PM

#

instead of interpreting it all you JIT it

hallow marten Jun 12, 2025, 12:55 PM

#

fiery turtle just aml to machine code converter

ahh

#

yeah

fiery turtle Jun 12, 2025, 12:55 PM

#

so e.g. AddOp would be call uacpi_handle_add

hallow marten Jun 12, 2025, 12:55 PM

#

that'd be doable

left orbit Jun 12, 2025, 12:55 PM

#

fiery turtle so e.g. `AddOp` would be `call uacpi_handle_add`

would you do it as a stack machine and have handlers operate on that or do proper regalloc? meme

fiery turtle Jun 12, 2025, 12:56 PM

#

depending on how much time you have on your hands trl

hallow marten Jun 12, 2025, 12:56 PM

#

just generate x86 directly

left orbit Jun 12, 2025, 12:57 PM

#

last time i checked x86 did not contain the capability to operate on arbitrary data types other than numbers

hallow marten Jun 12, 2025, 12:57 PM

#

https://gcc.gnu.org/onlinedocs/gcc-14.3.0/jit/cp/intro/tutorial04.html idk much about AML, but that doesnt seem so bad to do

left orbit Jun 12, 2025, 12:57 PM

#

so you can't really "generate x86" without a bunch of runtime support

fiery turtle Jun 12, 2025, 12:58 PM

#

well yeah first step of jitting something is always just transforming stuff into calls into C

#

next step is hand optimizing each thing so it can be handled inline

#

etc

gentle peak Jun 12, 2025, 12:58 PM

#

why do the acpica docs say AcpiOsReadPciConfiguration can have a Width parameter of 64??

#

wtf is a 64 bit pci read

fiery turtle Jun 12, 2025, 12:58 PM

#

who knows lol

left orbit Jun 12, 2025, 12:58 PM

#

lmao

fiery turtle Jun 12, 2025, 12:59 PM

#

its also strange it gives u the number of bits

gentle peak Jun 12, 2025, 12:59 PM

#

i'm just returning AE_BAD_PARAMETER for that

fiery turtle Jun 12, 2025, 12:59 PM

#

like can u do a 13 bit read

left orbit Jun 12, 2025, 12:59 PM

#

i sure hope you can

gentle peak Jun 12, 2025, 1:00 PM

#

fiery turtle like can u do a 13 bit read

is this like "the parameter format allows passing it" or "this will be passed and you have to deal with it"

fiery turtle Jun 12, 2025, 1:00 PM

#

i can add this math test case results as well as proxima's result if u get that done, would be cool to have both

fiery turtle Jun 12, 2025, 1:00 PM

#

gentle peak is this like "the parameter format allows passing it" or "this will be passed an...

i hope it aligns them up to next byte boundary

#

but idk lol

fiery turtle Jun 12, 2025, 1:01 PM

#

gentle peak why do the acpica docs say AcpiOsReadPciConfiguration can have a Width parameter...

https://elixir.bootlin.com/linux/v6.15.1/source/drivers/acpi/osl.c#L816

#

well according to this

#

it's always aligned

gentle peak Jun 12, 2025, 1:02 PM

#

alright good

#

why does readmemory take a physical address? this is painful

fiery turtle Jun 12, 2025, 1:04 PM

#

because trl

fiery turtle Jun 12, 2025, 1:05 PM

#

hallow marten https://gcc.gnu.org/onlinedocs/gcc-14.3.0/jit/cp/intro/tutorial04.html idk much ...

yeah doesnt look that bad

gentle peak Jun 12, 2025, 1:06 PM

#

how fun

fiery turtle Jun 12, 2025, 1:11 PM

#

Lol

#

wait I was testing uacpi with sized frees with every allocation recorded in a hash table

#

that might've slowed it down a bit trl

hallow marten Jun 12, 2025, 1:15 PM

#

fiery turtle yeah doesnt look that bad

I will unironically try it once I get it running in userspace

#

just to beat you all

fiery turtle Jun 12, 2025, 1:16 PM

#

that would be cool

hallow marten Jun 12, 2025, 1:16 PM

#

I'm in userspace so I can abuse a lot more stuff

fiery turtle Jun 12, 2025, 1:16 PM

#

fiery turtle wait I was testing uacpi with sized frees with every allocation recorded in a h...

yeah, that shaves off another second [uACPI][TRACE] [AML DEBUG] String => "10,000,000 additions took 5124 ms"

hallow marten Jun 12, 2025, 1:16 PM

#

like simd and I can port mimalloc

fiery turtle Jun 12, 2025, 1:16 PM

#

yeah true

hallow marten Jun 12, 2025, 1:17 PM

#

this will be the theoretical limit of uacpi speeds

fiery turtle Jun 12, 2025, 1:17 PM

#

yep

#

I can test on my cpu as well since i have avx512 it should be faster

#

[uACPI][INFO] successfully loaded 1 AML blob, 110000046 ops in 5125ms (avg 21463362/s)
So i guess ATM the maximum speeed uACPI can run at is 21M ops/s

#

i wonder what happens if i tune for my cpu

#

@gentle peak what were the compiler options u used for tuning

#

(for my cpu)

gentle peak Jun 12, 2025, 1:19 PM

#

-march=znver5 -mtune=znver5 i think

fiery turtle Jun 12, 2025, 1:19 PM

#

thanks

hallow marten Jun 12, 2025, 1:19 PM

#

whats the record again

fiery turtle Jun 12, 2025, 1:19 PM

#

ah fuck my clang is too old did you mean ‘znver1’?

fiery turtle Jun 12, 2025, 1:20 PM

#

hallow marten whats the record again

11M

#

iirc

#

on qemu's blob

hallow marten Jun 12, 2025, 1:20 PM

#

ok I can easily double that

#

meme

fiery turtle Jun 12, 2025, 1:20 PM

#

probably

hallow marten Jun 12, 2025, 1:20 PM

#

fiery turtle on qemu's blob

did you try uacpi linux on it

#

like

#

userspace uacpi

#

on it

fiery turtle Jun 12, 2025, 1:20 PM

#

yeah monkuous did

#

idr the score there

gentle peak Jun 12, 2025, 1:20 PM

#

i'm getting ACPI Error: AE_NO_ACPI_TABLES, While loading namespace from ACPI tables (20250404/tbxfload-110) but it isn't even calling AcpiOsGetRootPointer ???

fiery turtle Jun 12, 2025, 1:21 PM

#

lmfao

gentle peak Jun 12, 2025, 1:21 PM

#

hallow marten did you try uacpi linux on it

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

GitHub

GitHub - monkuous/uacpi-linux

Contribute to monkuous/uacpi-linux development by creating an account on GitHub.

hallow marten Jun 12, 2025, 1:21 PM

#

no I meant like

gentle peak Jun 12, 2025, 1:21 PM

#

iirc the score on first run (with cold caches and memory not being paged) is ~7M

hallow marten Jun 12, 2025, 1:21 PM

#

infy is currently running uacpi vs acpica

gentle peak Jun 12, 2025, 1:21 PM

#

hallow marten no I meant like

this repo is uacpi running in userspace linux

hallow marten Jun 12, 2025, 1:21 PM

#

thats not in a VM

#

ahh

#

but it can run the qemu blob?

fiery turtle Jun 12, 2025, 1:22 PM

#

it would be pointless to try qemu's blob in my current set up since it does IO accesses etc, so I would be testing the speed of the test harness

#

which is kinda useless

gentle peak Jun 12, 2025, 1:22 PM

#

hallow marten but it can run the qemu blob?

yes

#

up until the point where it tries to install an irq handler anyway since that's impossible in pure userspace linux

#

but that's enough to get the score

hallow marten Jun 12, 2025, 1:23 PM

#

do you have a test iso already

#

with a kernel compiled

gentle peak Jun 12, 2025, 1:24 PM

#

i did at one point but this was last touched 4 months ago

fiery turtle Jun 12, 2025, 1:29 PM

#

so with CPU tuning uacpi gets about 23M/s and ACPICA gets 6.5M/s

hallow marten Jun 12, 2025, 1:30 PM

#

uACPI on windows when

#

oh I think someone did it already tho

#

the reactOS person

fiery turtle Jun 12, 2025, 1:31 PM

#

yeah it literally exists lol

#

https://www.youtube.com/shorts/gUEzkglsXn4

YouTube

The_DarkFire_

ReactOS ACPI / uACPI windows driver wakes up windows 10 for the fir...

▶ Play video

hallow marten Jun 12, 2025, 1:32 PM

#

creaking noise on laptop, hello kitty lamp, insane cheering

#

this is truly a 10/10 video

fiery turtle Jun 12, 2025, 1:33 PM

#

lol true

#

So basically 3.5x diff vs acpica is what i can get on a stable basis

#

ill use that for the readme

#

and im using the best result i managed to get on acpica

#

it's znver4 because my clang is too old

flat badge Jun 12, 2025, 1:37 PM

#

fiery turtle tbf i didnt expect a 3x difference on a simple math test lmao

I'd expect that the difference is higher on a math test than on real world AML code

fiery turtle Jun 12, 2025, 1:37 PM

#

yeah ofc, this is purely an interpreter benchmark

flat badge Jun 12, 2025, 1:37 PM

#

since real world AML code will spent a large fraction of the time talking to the HW

fiery turtle Jun 12, 2025, 1:37 PM

#

yes

#

monkuous is currently implementing acpica in their os

#

we'll see what that generates

#

especially interesting to see real hw diff

dusky glade Jun 12, 2025, 1:38 PM

#

gentle peak how fun

yo at least u have signals 😭

fiery turtle Jun 12, 2025, 1:38 PM

#

but ill mention in the readme that this speed is interepreter only and doesnt matter that much irl

fiery turtle Jun 12, 2025, 1:39 PM

#

dusky glade yo at least u have signals 😭

its literally like

#

write shit into stack frame

#

or kill if no handler

dusky glade Jun 12, 2025, 1:39 PM

#

interesting

#

ive heard its like really annoying to get working

left orbit Jun 12, 2025, 1:39 PM

#

signals can get messy tho

dusky glade Jun 12, 2025, 1:39 PM

#

from some ppl

fiery turtle Jun 12, 2025, 1:40 PM

#

well u have to implement sigreturn as well etc

left orbit Jun 12, 2025, 1:40 PM

#

there is a reason they are a universally hated posix interface

#

:^)

fiery turtle Jun 12, 2025, 1:40 PM

#

to pop the fake frame

#

and also handle syscall interruption

left orbit Jun 12, 2025, 1:40 PM

#

maybe the way they interact with syscalls

#

yeah that

dusky glade Jun 12, 2025, 1:40 PM

#

fun

left orbit Jun 12, 2025, 1:40 PM

#

yay interruptible syscalls :^)

#

nyaux is gonna have so much fun with that

dusky glade Jun 12, 2025, 1:41 PM

#

nooo

#

oh no

#

that is fuckin pain

flat badge Jun 12, 2025, 1:41 PM

#

signals are way worse on the userspace side than on the kernel side

fiery turtle Jun 12, 2025, 1:41 PM

#

well u have checkpoints at which syscalls check if current thread has a signal pending

#

its not that bad

dusky glade Jun 12, 2025, 1:41 PM

#

you see

fiery turtle Jun 12, 2025, 1:41 PM

#

and if it does u just return -EAGAIN

dusky glade Jun 12, 2025, 1:41 PM

#

the thing is i designed nyaux's syscall shit without it being interruptable in mind

#

so

#

fucking pain is near

fiery turtle Jun 12, 2025, 1:42 PM

#

u dont have to interrupt them at all

#

if u dont want to

flat badge Jun 12, 2025, 1:42 PM

#

fiery turtle and if it does u just `return -EAGAIN`

EINTR

dusky glade Jun 12, 2025, 1:42 PM

#

but u still need to handle signal shit for syscalls right?

fiery turtle Jun 12, 2025, 1:42 PM

#

its not required to make stuff work

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