uACPI - a portable and easy-to-integrate ACPI implementation | Operating System Development | Page 11

north holly Nov 28, 2024, 9:42 PM

#

Related to mde

#

*kde

#

Uhh

#

1,726,000,000,000 ops a second

fiery turtle Nov 28, 2024, 9:47 PM

#

yeah uh

north holly Nov 28, 2024, 9:47 PM

#

Sounds wrong

fiery turtle Nov 28, 2024, 9:47 PM

#

what does your nanoseconds_since_boot return

north holly Nov 28, 2024, 9:47 PM

#

fiery turtle what does your nanoseconds_since_boot return

Hpet counter or smth similar

fiery turtle Nov 28, 2024, 9:48 PM

#

yeah idk sounds incorrect lol

north holly Nov 28, 2024, 9:48 PM

#

Although it uses an internal function to convert from hpet ticks to nanoseconds

#

Maybe that's bugged

fiery turtle Nov 28, 2024, 9:49 PM

#

sounds like it

north holly Nov 28, 2024, 9:50 PM

#

rn_image_picker_lib_temp_224f7269-afa0-4af9-b624-6f06b0e47ad7.jpg

#

Sorry for screenshotted code

fiery turtle Nov 28, 2024, 9:51 PM

#

why do this

#

instead of just return ...

north holly Nov 28, 2024, 9:51 PM

#

Because the hpet might not run on ns

#

Idk maybe it's stupid

fiery turtle Nov 28, 2024, 9:51 PM

#

monotonic means >=

#

not that it has to increase every time

north holly Nov 28, 2024, 9:52 PM

#

Silly me

#

rn_image_picker_lib_temp_ac3f40ba-587d-46d4-a686-8dba72b8acb9.jpg

fiery turtle Nov 28, 2024, 9:53 PM

#

looks complex

north holly Nov 28, 2024, 9:53 PM

#

It has the actual calculation done at the top of the function

#

Now I might've pulled that math outta my ass

hollow elm Nov 28, 2024, 9:53 PM

#

are you sure the multiplication by 1000000000 doesn't overflow

north holly Nov 28, 2024, 9:53 PM

#

hollow elm are you sure the multiplication by 1000000000 doesn't overflow

Nope

#

Uh oh

#

I break uacpi

#

With my buggy value

#

rn_image_picker_lib_temp_b3401665-15d1-4b38-861f-620595190893.jpg

fiery turtle Nov 28, 2024, 10:02 PM

#

uacpi skill issue

north holly Nov 28, 2024, 10:04 PM

#

Although I see no division there?

#

On that line

#

It's end_ts-begin_ts

fiery turtle Nov 28, 2024, 10:04 PM

#

blind opcodes_per_second /= end_ts - begin_ts;

north holly Nov 28, 2024, 10:04 PM

#

I'm blind nvm

#

I thought it was just equal

fiery turtle Nov 28, 2024, 10:05 PM

#

i would test managarm rn as well but it takes 10 years to rebuild, clone, run random commands and half of them break

#

maybe tomorrow

north holly Nov 28, 2024, 10:06 PM

#

Bruh

#

It gets 0

fiery turtle Nov 28, 2024, 10:06 PM

#

lol

north holly Nov 28, 2024, 10:06 PM

#

Maybe my math is a bit bugged lol

fiery turtle Nov 28, 2024, 10:06 PM

#

i guess obos takes less than a nanosecond to run 1700 opcodes

north holly Nov 28, 2024, 10:07 PM

#

Obos W

#

So fast it broke uacpi

median crest Nov 28, 2024, 10:07 PM

#

can you add time travel to obos

fiery turtle Nov 28, 2024, 10:07 PM

#

i should definitely add a guard against div by zero there

north holly Nov 28, 2024, 10:07 PM

#

median crest can you add time travel to obos

Already have it

fiery turtle Nov 28, 2024, 10:11 PM

#

uacpi_u64 opcodes_per_second = g_uacpi_rt_ctx.opcodes_executed;

opcodes_per_second *= 1000ull * 1000ull * 1000ull;
if (uacpi_likely(end_ts > begin_ts))
    opcodes_per_second /= end_ts - begin_ts;

ill add this obos protection

north holly Nov 28, 2024, 10:11 PM

#

Currently at two things added because of obos bug

#

That I remember

fiery turtle Nov 28, 2024, 10:11 PM

#

lol

north holly Nov 28, 2024, 10:11 PM

#

0 byte allocation protection

#

And this

fiery turtle Nov 28, 2024, 10:12 PM

#

both are just good things to protect against anyway

north holly Nov 28, 2024, 10:12 PM

#

Actually three, the debug log mode was added to uacpi for people whose kernels suddenly become cursed by uacpi

#

And I was the first kernel to get cursed using uacpi

fiery turtle Nov 28, 2024, 10:13 PM

#

indeed

#

some people literally stopped working on their kernels because uacpi uncovered a ton of bugs

#

e.g. frosty

north holly Nov 28, 2024, 10:14 PM

#

Poor frosty

left orbit Nov 28, 2024, 10:14 PM

#

i have yet to test my kernel against uacpi

#

if i had an actual allocator i would

#

but i am too lazy sadmeme

fiery turtle Nov 28, 2024, 10:14 PM

#

can still use it without an allocator for early_table_access meme

left orbit Nov 28, 2024, 10:15 PM

#

i do that!!!

#

:^)

fiery turtle Nov 28, 2024, 10:15 PM

#

oh damn lol

left orbit Nov 28, 2024, 10:15 PM

#

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

#

a little wrapper around table_find

fiery turtle Nov 28, 2024, 10:16 PM

#

nice

north holly Nov 28, 2024, 10:17 PM

#

If only I had written damn comments

#

In my code

#

Like cool I got a calculation

fiery turtle Nov 28, 2024, 10:17 PM

#

obofuscation

north holly Nov 28, 2024, 10:17 PM

#

NOW WTF DOES IT DO

#

😭

fiery turtle Nov 28, 2024, 10:18 PM

#

left orbit a little wrapper around table_find

cleanest glue implementation ive seen too

median crest Nov 28, 2024, 10:18 PM

#

fiery turtle some people literally stopped working on their kernels because uacpi uncovered a...

astral mogged uacpi by just working™️

north holly Nov 28, 2024, 10:19 PM

#

It looks like it overflows

#

And that's why

left orbit Nov 28, 2024, 10:19 PM

#

i love UACPI_STATUS_UNIMPLEMENTED

fiery turtle Nov 28, 2024, 10:19 PM

#

median crest astral mogged uacpi by just working™️

true, only found a few minor bugs while implementing it

jaunty fox Nov 28, 2024, 10:19 PM

#

left orbit <https://github.com/48cf/osdev/blob/main/kernel/src/acpi.c>

star'd

north holly Nov 28, 2024, 10:19 PM

#

median crest astral mogged uacpi by just working™️

I still got the first kernel to use uacpi

left orbit Nov 28, 2024, 10:19 PM

#

thank you 🙏

fiery turtle Nov 28, 2024, 10:19 PM

#

lol npnp

north holly Nov 28, 2024, 10:19 PM

#

And it wasn't buggy on my third kernel

#

And it just worked™️

#

So obos obv better than astral

fiery turtle Nov 28, 2024, 10:20 PM

#

not until we see ops per second

median crest Nov 28, 2024, 10:20 PM

#

iretq osdev proejct real

north holly Nov 28, 2024, 10:20 PM

#

fiery turtle obofuscation

Obosfuscation

north holly Nov 28, 2024, 10:20 PM

#

fiery turtle not until we see ops per second

It's only partially my fault it overflows

fiery turtle Nov 28, 2024, 10:20 PM

#

my bet is keyronex should outperform astral here

left orbit Nov 28, 2024, 10:20 PM

#

with my first star being fadanoid i should really get back to work 👀

fiery turtle Nov 28, 2024, 10:21 PM

#

we're about to see soon

jaunty fox Nov 28, 2024, 10:21 PM

#

fiery turtle my bet is keyronex should outperform astral here

it depends on whether you use mutexes

fiery turtle Nov 28, 2024, 10:21 PM

#

are they slow in keyronex?

median crest Nov 28, 2024, 10:21 PM

#

fiery turtle my bet is keyronex should outperform astral here

100%

north holly Nov 28, 2024, 10:21 PM

#

Surely acquiring a non acquired mutex isn't slow

fiery turtle Nov 28, 2024, 10:22 PM

#

yeah mutexes are now used a few order of mangitude more often than before

north holly Nov 28, 2024, 10:22 PM

#

Because it's not like uacpi is multithreaded during init

jaunty fox Nov 28, 2024, 10:22 PM

#

the uacpi integration is using full mutexes because 1) they're highly debuggable 2) keyronex master branch doesn't have efficient pushlocks integrated

fiery turtle Nov 28, 2024, 10:22 PM

#

so does astral i think

#

but there shouldnt be any contention so

north holly Nov 28, 2024, 10:22 PM

#

wdym full mutexes

#

Like everything is a mutex...

jaunty fox Nov 28, 2024, 10:23 PM

#

full mutexes are slow even if uncontended on keyronex because they are part of the generic waiting framework

fiery turtle Nov 28, 2024, 10:23 PM

#

we're about to see just how slow meme

jaunty fox Nov 28, 2024, 10:23 PM

#

it does a lot of work

fiery turtle Nov 28, 2024, 10:24 PM

#

btw if u want u can try rebasing to the thread-safety branch yourself, if making a separate branch is problematic right now

#

ill merge it soon but im not changing anything probably

jaunty fox Nov 28, 2024, 10:24 PM

#

fiery turtle we're about to see just how slow <:meme:575445094589661215>

one more thing, there is no dynticks at all on old keyronex and timekeeping is ticks driven so it's going to be low res

#

but i am curious what the result will be

#

i think astral will win

fiery turtle Nov 28, 2024, 10:25 PM

#

hmm well u do have a way to read the actual counter value on demand?

#

or does it read some cached counter

jaunty fox Nov 28, 2024, 10:26 PM

#

fiery turtle or does it read some cached counter

only that, you could improve resolution by editing kernel/include/kdk/amd64.h line 16 (KERN_HZ from 64 to say 250)

fiery turtle Nov 28, 2024, 10:27 PM

#

ah i see

#

yeah that res might be too slow, because with kvm it takes 4ms in astral

median crest Nov 28, 2024, 10:29 PM

#

do a quick and dirty impl of the kvm timer halfmemeright

fiery turtle Nov 28, 2024, 10:29 PM

#

true

north holly Nov 28, 2024, 10:30 PM

#

According to obos, 33 timer ticks is 0 nanoseconds

#

At a frequency of 500 hz

#

Which is obviously wrong

#

I assume some math overflows

fiery turtle Nov 28, 2024, 10:31 PM

#

astral is probably the only hobby os to use the kvm timer

north holly Nov 28, 2024, 10:31 PM

#

obos will soon™️

median crest Nov 28, 2024, 10:36 PM

#

fiery turtle astral is probably the only hobby os to use the kvm timer

all because I'm a stubborn little shit who wanted to keep using wsl2 despite hyperv not liking exposing the tsc to wsl

fiery turtle Nov 28, 2024, 10:36 PM

#

yeah and it looks very precise now

north holly Nov 28, 2024, 10:37 PM

#

I'm just gonna assume obos' timekeeping is shit

#

Because it says 2945 ops/s

fiery turtle Nov 28, 2024, 10:37 PM

#

lol

jaunty fox Nov 28, 2024, 10:38 PM

#

i tried out the thread_safety branch on the new kernel and it appears to release a mutex that wasn't acquired

fiery turtle Nov 28, 2024, 10:38 PM

#

a few orders of magnitude behind astral

fiery turtle Nov 28, 2024, 10:38 PM

#

jaunty fox i tried out the thread_safety branch on the new kernel and it appears to release...

did you make sure to change the acquire_mutex api to return a uacpi_status?

north holly Nov 28, 2024, 10:39 PM

#

Maybe I should make a dirty implementation of kvm timer

jaunty fox Nov 28, 2024, 10:39 PM

#

i did indeed, and it looks like it never does fail

fiery turtle Nov 28, 2024, 10:40 PM

#

interesting, got a backtrace where that happens?

north holly Nov 28, 2024, 10:40 PM

#

In case my timekeeping really is the problem (it isn't)

jaunty fox Nov 28, 2024, 10:40 PM

#

fiery turtle interesting, got a backtrace where that happens?

#4  uacpi_release_aml_mutex (mutex=0xffff800000089fa0) at uacpi/source/mutex.c:248
#5  0xffffffff80027a85 in access_field_unit (field=field@entry=0xffff80000006d980, offset=<optimized out>,
    offset@entry=0, op=op@entry=UACPI_REGION_OP_WRITE, in_out=in_out@entry=0xffffc80000043c10)
    at uacpi/source/io.c:317
#6  0xffffffff800278ff in uacpi_write_field_unit (field=0xffff80000006d980, src=<optimized out>, size=<optimized out>)
    at uacpi/source/io.c:460
#7  0xffffffff800256a1 in object_assign_with_implicit_cast (dst=0xffff800000073da0, src=src@entry=0xffff80010403c9a0)
    at uacpi/source/interpreter.c:978
#8  0xffffffff800261e1 in store_to_reference (dst=0xffff800000073ce0, src=<optimized out>)
    at uacpi/source/interpreter.c:2334
#9  0xffffffff80026296 in store_to_target (dst=0xffff800000073ce0, src=src@entry=0xffff80010403c9a0)
    at uacpi/source/interpreter.c:4196
#10 0xffffffff80026e28 in handle_copy_object_or_store (ctx=<optimized out>)
    at uacpi/source/interpreter.c:4229
#11 0xffffffff800269a1 in exec_op (ctx=ctx@entry=0xffffc80000053000)
    at uacpi/source/interpreter.c:5405
#12 0xffffffff80026eed in uacpi_execute_control_method (scope=<optimized out>, method=method@entry=0xffff800000073c40,
    args=args@entry=0x0, out_obj=out_obj@entry=0x0) at uacpi/source/interpreter.c:5757
#13 0xffffffff80032e35 in uacpi_eval (parent=parent@entry=0xffff80000006dec0,
    path=path@entry=0xffffffff8003d4eb "_INI", args=args@entry=0x0, out_obj=out_obj@entry=0x0)
    at uacpi/source/uacpi.c:669
#14 0xffffffff80032e7d in ini_eval (ctx=ctx@entry=0xffffc80000043f08, node=node@entry=0xffff80000006dec0)
    at uacpi/source/uacpi.c:450
--Type <RET> for more, q to quit, c to continue without paging--
#15 0xffffffff800330b6 in do_sta_ini (opaque=opaque@entry=0xffffc80000043f08, node=node@entry=0xffff80000006dec0,
    depth=depth@entry=3) at uacpi/source/uacpi.c:525
#16 0xffffffff80029377 in uacpi_namespace_do_for_each_child (...)
    at uacpi/source/namespace.c868

#

wait sorry

#

i pasted somehing else

#

one sec

fiery turtle Nov 28, 2024, 10:40 PM

#

i wonder if c++ library implementation doesnt catch that if its a real bug

jaunty fox Nov 28, 2024, 10:41 PM

#

it should if it is so i'm suspicious

zenith crescent Nov 28, 2024, 10:41 PM

#

north holly Poor frosty

I'll return soon

jaunty fox Nov 28, 2024, 10:41 PM

#

just compacting the trace to fit discord

#

that's it up there

fiery turtle Nov 28, 2024, 10:43 PM

#

yeah no its a bug

#

what does field->lock_rule say?

#

good thing i didnt merge it

jaunty fox Nov 28, 2024, 10:43 PM

#

0

fiery turtle Nov 28, 2024, 10:43 PM

#

yup

#

need to add that if at the bottom

#

--- a/source/io.c
+++ b/source/io.c
@@ -314,7 +314,8 @@ static uacpi_status access_field_unit(
     );

 out:
-    uacpi_release_aml_mutex(g_uacpi_rt_ctx.global_lock_mutex);
+    if (field->lock_rule)
+        uacpi_release_aml_mutex(g_uacpi_rt_ctx.global_lock_mutex);
     return ret;
 }

#

try that real quick

#

how do i force the C++ standard library to abort if i release a mutex that wasnt acquired lol

jaunty fox Nov 28, 2024, 10:46 PM

#

[uacpi]: successfully loaded 1 AML blob in 6ms (2356 ops total, avg 373183/s) success

fiery turtle Nov 28, 2024, 10:46 PM

#

kinda annoying it wasnt caught on any platform

#

daamn faster than astral

jaunty fox Nov 28, 2024, 10:46 PM

#

it's under KVM i should note

#

i'll send on the iso, 1s

fiery turtle Nov 28, 2024, 10:47 PM

#

astral without kvm does 40k

jaunty fox Nov 28, 2024, 10:47 PM

#

this new branch i can't test without kvm as it's wedded to TSC deadline at the moment

fiery turtle Nov 28, 2024, 10:47 PM

#

ah

#

well still great results

#

managed to beat astral

#

i wonder if i should try to make it print in thousands instead

#

aka avg 373.1k/s

#

or just 373k/s

fiery turtle Nov 28, 2024, 10:50 PM

#

jaunty fox `[uacpi]: successfully loaded 1 AML blob in 6ms (2356 ops total, avg 373183/s)` ...

thanks for testing and finding the bug, very important

#

ill make a hash table in c++ and store acquired mutexes lmao

#

so that i catch stuff like this

jaunty fox Nov 28, 2024, 10:51 PM

#

fiery turtle thanks for testing and finding the bug, very important

no worries, pleased to be of assistance

#

📎 keyronex.iso

#

an ISO if you want to see if it's the same result on your machine

#

needs qemu-system-x86_64 -cdrom keyronex.iso -serial stdio -M q35 -enable-kvm -cpu host and will die immediately after

fiery turtle Nov 28, 2024, 10:53 PM

#

jaunty fox

bruh

jaunty fox Nov 28, 2024, 10:53 PM

#

fiery turtle bruh

https://tenor.com/view/waltergotme-gif-18867690

Tenor

fiery turtle Nov 28, 2024, 10:53 PM

#

lmao

jaunty fox Nov 28, 2024, 10:53 PM

#

wonder what the virus is

median crest Nov 28, 2024, 10:53 PM

#

keyronex so fast big OS is censoring it

fiery turtle Nov 28, 2024, 10:54 PM

#

jaunty fox wonder what the virus is

jaunty fox Nov 28, 2024, 10:54 PM

#

wtf

fiery turtle Nov 28, 2024, 10:54 PM

#

lmaooo

jaunty fox Nov 28, 2024, 10:55 PM

#

Two security experts point out to me that if code section is writeable, it is easily flagged by AV as malicious.

I check three of my programs, what they said are true, all with characteristic E0000000 (code section) - RWX.

#

surely it doesn't actually look in the ISO, find an elf (not a PE) with an RWX segment, and flag that?

fiery turtle Nov 28, 2024, 10:56 PM

#

apparently it can do that? lol

jaunty fox Nov 28, 2024, 10:57 PM

#

i have no rwx sections in either the kernel nor the virtio.kernext

#

maybe limine

fiery turtle Nov 28, 2024, 10:57 PM

#

yeah

zenith crescent Nov 28, 2024, 10:57 PM

#

limine is a PE because of UEFI

fiery turtle Nov 28, 2024, 10:57 PM

#

ohhh right

jaunty fox Nov 28, 2024, 10:57 PM

#

zenith crescent limine is a PE because of UEFI

this is what i was thinking, that might explain it

fiery turtle Nov 28, 2024, 10:57 PM

#

that must be it

hollow elm Nov 28, 2024, 10:58 PM

#

fiery turtle how do i force the C++ standard library to abort if i release a mutex that wasnt...

maybe -D_GLIBCXX_ASSERTIONS?

fiery turtle Nov 28, 2024, 10:58 PM

#

ill try that thanks

hollow elm Nov 28, 2024, 10:59 PM

#

not sure if it adds asserts for mutexes but it for example adds asserts in vector::operator[] etc

jaunty fox Nov 28, 2024, 10:59 PM

#

limine has no RWX sections in any of its BOOT***.efi either

fiery turtle Nov 28, 2024, 10:59 PM

#

ubsan literally couldnt care less about bogus mutex releases

#

The behavior is undefined if the mutex is not locked by the calling thread.

fiery turtle Nov 28, 2024, 11:00 PM

#

jaunty fox limine has no RWX sections in any of its BOOT***.efi either

very strange

#

i wonder what makes it go off then

hollow elm Nov 28, 2024, 11:02 PM

#

do you ship storahci or something in the iso?

jaunty fox Nov 28, 2024, 11:03 PM

#

hollow elm do you ship storahci or something in the iso?

no, not in this one

hollow elm Nov 28, 2024, 11:03 PM

#

hmm

jaunty fox Nov 28, 2024, 11:03 PM

#

here it is without virtio.kernext

📎 keyronex.iso

fiery turtle Nov 28, 2024, 11:03 PM

#

if i have writable code sections im automatically a trojan?

jaunty fox Nov 28, 2024, 11:03 PM

#

wait, is it without there? one sec

#

yeah, no virtio.kernext in there, maybe that helps

fiery turtle Nov 28, 2024, 11:03 PM

#

jaunty fox here it is without virtio.kernext

that one downloaded successfully

fiery turtle Nov 28, 2024, 11:04 PM

#

jaunty fox here it is without virtio.kernext

im on wsl2 unfortunately, so same problem that made mathew add kvm clock support:

!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!
Warning: No invariant TSC, problems likely!
!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!!

hollow elm Nov 28, 2024, 11:05 PM

#

hollow elm not sure if it adds asserts for mutexes but it for example adds asserts in vecto...

also looks like it doesn't add any for mutexes, since they just boil down to pthread calls

jaunty fox Nov 28, 2024, 11:05 PM

#

fiery turtle im on wsl2 unfortunately, so same problem that made mathew add kvm clock support...

kvm clock is probably quite simple, i can look into that

fiery turtle Nov 28, 2024, 11:05 PM

#

yeah its pretty simple iirc

#

like a struct in memory that contains the values to convert tsc to elapsed time

#

and you either re-read it every time, or if it has a stable kvm clock bit or w/e u just read it once

jaunty fox Nov 28, 2024, 11:09 PM

#

i had a quick skim, very reasonable

gentle peak Nov 28, 2024, 11:10 PM

#

I'd like to implement it as well, is there a spec for it or do you just have to figure it out from linux sources or something

fiery turtle Nov 28, 2024, 11:11 PM

#

jaunty fox here it is without virtio.kernext

btw is that like the virtio driver for your kernel or?

#

anything strange about that file?

jaunty fox Nov 28, 2024, 11:11 PM

#

fiery turtle btw is that like the virtio driver for your kernel or?

yeah

#

it's just an elf solib

fiery turtle Nov 28, 2024, 11:12 PM

#

interesting

jaunty fox Nov 28, 2024, 11:12 PM

#

Elf file type is DYN (Shared object file)
Entry point 0x0
There are 5 program headers, starting at offset 64

Program Headers:
  Type           Offset             VirtAddr           PhysAddr
                 FileSiz            MemSiz              Flags  Align
  LOAD           0x0000000000000000 0x0000000000000000 0x0000000000000000
                 0x0000000000003770 0x0000000000003770  R E    0x1000
  LOAD           0x0000000000003770 0x0000000000004770 0x0000000000004770
                 0x00000000000020b0 0x00000000000020b8  RW     0x1000
  DYNAMIC        0x0000000000003780 0x0000000000004780 0x0000000000004780
                 0x00000000000001a0 0x00000000000001a0  RW     0x8
  GNU_EH_FRAME   0x0000000000003220 0x0000000000003220 0x0000000000003220
                 0x00000000000000fc 0x00000000000000fc  R      0x4
  GNU_STACK      0x0000000000000000 0x0000000000000000 0x0000000000000000
                 0x0000000000000000 0x0000000000000000  RW     0x10

#

not RWX segments

fiery turtle Nov 28, 2024, 11:12 PM

#

looks fairly standard

#

must be some other entropy about it

left orbit Nov 28, 2024, 11:13 PM

#

jaunty fox

[uacpi]: successfully loaded 1 AML blob in 2ms (1705 ops total, avg 699779/s) brrrr

fiery turtle Nov 28, 2024, 11:13 PM

#

damn

zenith crescent Nov 28, 2024, 11:13 PM

#

fiery turtle must be some other entropy about it

might just be windows defender being a pain

left orbit Nov 28, 2024, 11:14 PM

#

fiery turtle im on wsl2 unfortunately, so same problem that made mathew add kvm clock support...

-cpu host,+invtsc

fiery turtle Nov 28, 2024, 11:14 PM

#

left orbit `-cpu host,+invtsc`

same warning and it dies on PF lol

jaunty fox Nov 28, 2024, 11:14 PM

#

left orbit `[uacpi]: successfully loaded 1 AML blob in 2ms (1705 ops total, avg 699779/s)` ...

czapeczka setting the new standard

jaunty fox Nov 28, 2024, 11:14 PM

#

fiery turtle same warning and it dies on PF lol

the death is expected

#

i fucked something up

fiery turtle Nov 28, 2024, 11:15 PM

#

ah

left orbit Nov 28, 2024, 11:15 PM

#

global_constructors_init: running

#

is there some c++?

jaunty fox Nov 28, 2024, 11:15 PM

#

left orbit is there some c++?

objc

left orbit Nov 28, 2024, 11:15 PM

#

ah, fair enough

fiery turtle Nov 28, 2024, 11:15 PM

#

so 700k is the current record

#

i wonder how managarm fares here, will take a look later maybe

left orbit Nov 28, 2024, 11:16 PM

#

i can consistently get 740k ops/s with that keyronex image

fiery turtle Nov 28, 2024, 11:16 PM

#

damn

left orbit Nov 28, 2024, 11:16 PM

#

but im wondering why there's only 1705 ops

fiery turtle Nov 28, 2024, 11:17 PM

#

its 1705 in my qemu as well

jaunty fox Nov 28, 2024, 11:17 PM

#

oh i see the fault cause now

fiery turtle Nov 28, 2024, 11:17 PM

#

fadanoid must have a newer version

jaunty fox Nov 28, 2024, 11:17 PM

#

uacpi_namespace_node_get_object got privatised

fiery turtle Nov 28, 2024, 11:17 PM

#

jaunty fox uacpi_namespace_node_get_object got privatised

indeed

left orbit Nov 28, 2024, 11:17 PM

#

im on 9.1.1 so it's a pretty new version but i dont follow qemu closely so idk what's the latest

fiery turtle Nov 28, 2024, 11:17 PM

#

see the new eval heleprs and object api instead

fiery turtle Nov 28, 2024, 11:18 PM

#

left orbit im on 9.1.1 so it's a pretty new version but i dont follow qemu closely so idk w...

8.1.1 here

#

could also be a command line difference

left orbit Nov 28, 2024, 11:18 PM

#

i used the one fadanoid sent

fiery turtle Nov 28, 2024, 11:18 PM

#

ah

left orbit Nov 28, 2024, 11:19 PM

#

except for the +invtsc on cpu :^)

#

but i dont think that would change the aml

fiery turtle Nov 28, 2024, 11:19 PM

#

fadanoid what's your current qemu version

jaunty fox Nov 28, 2024, 11:19 PM

#

fiery turtle fadanoid what's your current qemu version

9.1.0

fiery turtle Nov 28, 2024, 11:19 PM

#

interesting

left orbit Nov 28, 2024, 11:19 PM

#

only one patch behind me

fiery turtle Nov 28, 2024, 11:20 PM

#

jaunty fox uacpi_namespace_node_get_object got privatised

btw, what were u using it for?

left orbit Nov 28, 2024, 11:20 PM

#

very interesting

jaunty fox Nov 28, 2024, 11:20 PM

#

fiery turtle btw, what were u using it for?

oh, just to check if the object at some node was a device or not, now i do uacp_namespace_node_type()

fiery turtle Nov 28, 2024, 11:21 PM

#

jaunty fox oh, just to check if the object at some node was a device or not, now i do uacp_...

is that during namespace iteration? If so, that now supports predefined object filters

jaunty fox Nov 28, 2024, 11:21 PM

#

and i did that when building the ACPI plane (the creation of device nodes according to the acpi layout, i discussed that some time ago)

jaunty fox Nov 28, 2024, 11:21 PM

#

fiery turtle is that during namespace iteration? If so, that now supports predefined object f...

it is, that would be even better

fiery turtle Nov 28, 2024, 11:21 PM

#

UACPI_OBJECT_DEVICE_BIT in for_each_child

jaunty fox Nov 28, 2024, 11:22 PM

#

 * allowed to reach, where 1 is direct children of 'parent', 2 is children of
 * first-level children etc.```
-1 for any?

fiery turtle Nov 28, 2024, 11:22 PM

#

yeah, or UACPI_MAX_DEPTH_ANY

#

should add that to the comment, thanks for pointing out

jaunty fox Nov 28, 2024, 11:28 PM

#

ascending_callback nullable as well which is worth noting probably

fiery turtle Nov 28, 2024, 11:29 PM

#

true

jaunty fox Nov 28, 2024, 11:30 PM

#

i wonder now how much faster it would be using pushlocks instead of the full mutex

fiery turtle Nov 28, 2024, 11:30 PM

#

keyronex pushing the 1M ops per second ultrameme

frank canopy Nov 29, 2024, 12:54 AM

#

well im officially biting the bullet and adding acpi to my os (at least partially; i need sync primitives at some point too before fulling committing) and thank goodness for this library making that an even remotely viable thing to do lol

left orbit Nov 29, 2024, 1:09 AM

#

welcome to the uacpi gang

frank canopy Nov 29, 2024, 1:10 AM

#

immediately running into the issue of not having stdbool trying to include it because im using zig lmao

#

but that should be solvable

#

(also out of curiosity, anyone got a sample of using uacpi with zig successfully? at least as far as build integration goes)

left orbit Nov 29, 2024, 1:16 AM

#

https://github.com/48cf/zigux

#

quite an old version of uacpi but it should still be the same-ish

#

as far as integration goes

#

ofc update the list of sources for starters

frank canopy Nov 29, 2024, 1:20 AM

#

left orbit <https://github.com/48cf/zigux>

oh dang dont think i can steal from that in that it pulls in libc which i have no intention of doing lol

#

my current build issue is that i dont have a single libc header lol

left orbit Nov 29, 2024, 1:22 AM

#

i dont pull in the libc

#

just some freestanding headers

#

wait lemme check

#

ok yeah my bad

#

you can use uhhh

#

https://github.com/osdev0/freestnd-c-hdrs-0bsd

#

or https://github.com/osdev0/freestnd-c-hdrs

#

but those are gpl(3?)

frank canopy Nov 29, 2024, 1:23 AM

#

it seems the only thing uacpi pulls in from libc is in the platform/types.h

#

which can be overridden

left orbit Nov 29, 2024, 1:24 AM

#

not too sure, haven't went down that road

#

but i don't think it's too much pain

#

to do it manually

frank canopy Nov 29, 2024, 1:47 AM

#

ok got it to only complain about the uacpi_kernel_* things being not defined so thats a good start

fiery turtle Nov 29, 2024, 8:11 AM

#

frank canopy which can be overridden

True

#

And welcome

fiery turtle Nov 29, 2024, 9:14 AM

#

[uACPI][INFO] successfully loaded 23 AML blobs in 1874ms (15804 ops total, 8K ops/s)

Do u guys like this output more than

[uACPI][INFO] successfully loaded 23 AML blobs in 1874ms (15804 ops total, avg 8194/s)

#

#polls message

#

vote ig

hasty plinth Nov 29, 2024, 9:24 AM

#

what about 8.1k ops/s?

fiery turtle Nov 29, 2024, 9:24 AM

#

doable, just more code to write ig lol

hasty plinth Nov 29, 2024, 9:24 AM

#

lol, I mean not that the precision of this really matters

fiery turtle Nov 29, 2024, 9:24 AM

#

yea

dusky glade Nov 29, 2024, 3:46 PM

#

@fiery turtle is there any big uacpi kernel api changes i should know about

north holly Nov 29, 2024, 3:46 PM

#

Yes

dusky glade Nov 29, 2024, 3:46 PM

#

i felt like u said there'd be one

#

oo[

north holly Nov 29, 2024, 3:46 PM

#

It has been reworked

#

From the ground up

dusky glade Nov 29, 2024, 3:46 PM

#

when is this getting pushed

north holly Nov 29, 2024, 3:46 PM

#

Jk

dusky glade Nov 29, 2024, 3:46 PM

#

nooo

north holly Nov 29, 2024, 3:46 PM

#

There are like 3 changws

#

*changes

dusky glade Nov 29, 2024, 3:46 PM

#

okay

north holly Nov 29, 2024, 3:47 PM

#

All of which you probably don't have implemented anyway

dusky glade Nov 29, 2024, 3:47 PM

#

okok

torpid root Nov 29, 2024, 4:06 PM

#

2 functions iirc

#

mutex lock and ns since boot

fiery turtle Nov 29, 2024, 4:35 PM

#

yup

north holly Nov 29, 2024, 4:37 PM

#

torpid root mutex lock and ns since boot

So basically nothing he has implemented

#

Yeah

frank canopy Nov 29, 2024, 8:35 PM

#

reading the comments on kernel_api.h, "event" is basically a semaphore yeah?

left orbit Nov 29, 2024, 8:36 PM

#

iirc yeah

fiery turtle Nov 29, 2024, 8:36 PM

#

yup

#

similar to eventfd on linux

frank canopy Nov 29, 2024, 8:36 PM

#

ok cool

#

i saw event and was thinking like AutoResetEvent or ManualResetEvent at first

fiery turtle Nov 29, 2024, 8:43 PM

#

fiery turtle https://discord.com/channels/440442961147199490/1227738482152181761/131198569959...

ok hear me out
[uACPI][INFO] successfully loaded 23 AML blobs, 15804 ops in 1453ms (10870 avg/s)

#

i think this might be better tbh

torpid root Nov 29, 2024, 8:48 PM

#

looks nice

frank canopy Nov 29, 2024, 8:49 PM

#

id put avg before the number at the end still personally

#

but does look nice

fiery turtle Nov 29, 2024, 8:50 PM

#

u would do (avg 10870/s)?

frank canopy Nov 29, 2024, 8:50 PM

#

yeah

fiery turtle Nov 29, 2024, 8:50 PM

#

hmm idk

frank canopy Nov 29, 2024, 8:50 PM

#

idk why but it looks better to me shrug

fiery turtle Nov 29, 2024, 8:50 PM

#

yeah im bad at these things so

fiery turtle Nov 29, 2024, 8:53 PM

#

frank canopy idk why but it looks better to me *shrug*

[uACPI][INFO] successfully loaded 23 AML blobs, 15804 ops in 1163ms (avg 13580 ops/s) what do u think of this one

#

or maybe thats too verbose

#

yeah i dont like that it says ops twice

#

[uACPI][INFO] successfully loaded 23 AML blobs, 15804 ops in 1152ms (avg 13708/s) fuck it ig this one it is

torpid root Nov 29, 2024, 8:57 PM

#

there's always something to not like, but that one looks good

fiery turtle Nov 29, 2024, 8:57 PM

#

i can always change it later i guess

#

uacpi log messages are not part of the stable api meme

fiery turtle Nov 29, 2024, 9:28 PM

#

stuff left to do for 1.0

#

some of those are skippable too tbh

#

like racey tests or BAREBONES_MODE

#

as far as thread safety is concerned most scary stuff is done too

#

finally

torpid root Nov 29, 2024, 9:43 PM

#

based

deft canopy Nov 29, 2024, 9:50 PM

#

fiery turtle as far as thread safety is concerned most scary stuff is done too

awesome!!!!

fiery turtle Nov 29, 2024, 9:54 PM

#

thanks!

frank canopy Nov 29, 2024, 10:42 PM

#

\o/

#

eyy renamed the unimplemented get_ticks to the ns since boot name and my thing still compiles

fiery turtle Nov 29, 2024, 10:47 PM

#

is your kernel available somewhere?

frank canopy Nov 29, 2024, 11:09 PM

#

https://github.com/Khitiara/imaginarium i havent pushed any uacpi things yet tho

left orbit Nov 29, 2024, 11:11 PM

#

b.verbose_llvm_ir = "agony.ir";
b.verbose_llvm_bc = "agony.bc";```

#

the zig experience

frank canopy Nov 29, 2024, 11:11 PM

#

lol

fiery turtle Nov 29, 2024, 11:12 PM

#

lol

fiery turtle Nov 29, 2024, 11:13 PM

#

frank canopy <https://github.com/Khitiara/imaginarium> i havent pushed any uacpi things yet t...

are u going to make a progress thread?

frank canopy Nov 29, 2024, 11:13 PM

#

probably at some point

#

ive mostly been doing stuff in the zig server's osdev channel

fiery turtle Nov 29, 2024, 11:13 PM

#

ah

left orbit Nov 29, 2024, 11:14 PM

#

i like zig but i still think this server might be objectively better meme

frank canopy Nov 29, 2024, 11:15 PM

#

alright pushed my stuff, uacpi shim impl and also my uacpi zig externs in https://github.com/Khitiara/imaginarium/tree/main/src/krnl/hal/acpi/uacpi and the necessary c headers in https://github.com/Khitiara/imaginarium/tree/main/include if youre interested in how i integrated uacpi ig

#

this thread is like my first time doing stuff in this server lol

left orbit Nov 29, 2024, 11:15 PM

#

anyway you should definitely create a progress thread, it's nice to show off some cool stuff and maybe rant when things don't work

frank canopy Nov 29, 2024, 11:15 PM

#

ye

#

i hadnt worked on it in over a month until like yesterday lol

fiery turtle Nov 29, 2024, 11:16 PM

#

oh interesting

#

u made actual bindings?

frank canopy Nov 29, 2024, 11:16 PM

#

as i need em yeah

fiery turtle Nov 29, 2024, 11:17 PM

#

could be cool to eventually make it a separate library like the rust bindings

left orbit Nov 29, 2024, 11:17 PM

#

i was gonna do that at some point

#

but... C took over my kernel

#

rest in peace zigux

#

welcome yet unnamed project i randomly abandon for weeks at a time and suddenly figure out the bug that's been haunting me since day 0

frank canopy Nov 29, 2024, 11:18 PM

#

mood

#

but yeah could be neat to extract zig bindings to a separate library

#

i do use existing types from my kernel for some of the uacpi types in my bindings though

fiery turtle Nov 29, 2024, 11:19 PM

#

lol

frank canopy Nov 29, 2024, 11:19 PM

#

like i already had a struct for the sdt headers so i just use that directly

#

one nice thing about using bindings like this is the pointer types on the zig-side uacpi_kernel_xxx functions dont have to all be ?*anyopaque (which is what uacpi_handle translates to), i can just directly strong-type them

fiery turtle Nov 29, 2024, 11:20 PM

#

btw what is this about

pub export fn _memmove(dest: [*]u8, src: [*]u8, size: usize) callconv(.C) [*]u8 {
    @memcpy(dest, src[0..size]);
    @memset(src[0..size], 0);
    return dest;
}

#

looks very very wrong lol

frank canopy Nov 29, 2024, 11:21 PM

#

copied from https://github.com/kora-org/ydin/blob/main/src/uacpi/uacpi_libc.zig

#

just to initially get working

fiery turtle Nov 29, 2024, 11:22 PM

#

memmove doesnt zero the src lol

#

it just handles the edge case where src and dst overlap is all

#

which memcpy doesnt handle

frank canopy Nov 29, 2024, 11:22 PM

#

ahh

fiery turtle Nov 29, 2024, 11:22 PM

#

https://en.cppreference.com/w/cpp/string/byte/memmove

cpp/string/byte/memmove

frank canopy Nov 29, 2024, 11:24 PM

#

pub export fn _memmove(dest: [*]u8, src: [*]u8, size: usize) callconv(.C) [*]u8 {
    if(dest > src) {
        std.mem.copyBackwards(u8, dest[0..size], src[0..size]);
    } else {
        std.mem.copyForwards(u8, dest[0..size], src[0..size]);
    }
    return dest;
}

then this is a valid impl given that contract

fiery turtle Nov 29, 2024, 11:24 PM

#

think so

#

looks like u even have your own helpers for the RSDP and stuff

#

fyi uacpi has that functionality as well

frank canopy Nov 29, 2024, 11:34 PM

#

which helpers?

#

cause all my old acpi code is in there for the various fixed tables

#

not removed yet

fiery turtle Nov 29, 2024, 11:35 PM

#

For finding the tables

#

uacpi_table_find etc

frank canopy Nov 29, 2024, 11:35 PM

#

wait you can use that to find the rsdp?

#

assumed it was for the others mostly

#

if so i might reimplement in terms of that

#

also not sure why im bothering with the uacpi_libc stuff tbh. only major difference over the naive impl would be vectorization and i have that turned off in my kernel so i dont have to save the vector registers on interrupts unless i do usermode context switching when i eventually add that

left orbit Nov 29, 2024, 11:36 PM

#

idk what infy is talking about but afaik uacpi requires you to pass in a physical rsdp address?

#

he may be referring to that you don't need your own acpi table parsing code

#

and you can use uacpi_table_find instead

frank canopy Nov 29, 2024, 11:37 PM

#

yeah that part is leftover code from before i decided to bring in uacpi

#

my plan is to use the installation handler callback actually

left orbit Nov 29, 2024, 11:38 PM

#

left orbit idk what infy is talking about but afaik uacpi requires you to pass in a physica...

btw sorry infy didn't mean for that to sound disrespectful

#

man i love overthinking :^)

fiery turtle Nov 29, 2024, 11:39 PM

#

Yeah no youre right

#

I was talking about xsdt walking code mostly

frank canopy Nov 29, 2024, 11:39 PM

#

yeah thats now officially leftover legacy code since im pulling in uacpi now

#

predates that decision

fiery turtle Nov 29, 2024, 11:39 PM

#

I see

frank canopy Nov 29, 2024, 11:40 PM

#

had all my table parsing working perfectly but was getting annoyed by not being able to enumerate serial ports (on the isa bus) without parsing aml while being too intimidated by aml to try and write the parsing

#

and then someone mentioned uacpi lol

fiery turtle Nov 29, 2024, 11:42 PM

#

frank canopy my plan is to use the installation handler callback actually

Idk what your plan for that is exactly, but note that the virtual mapping passed to that callback is only alive for the duration of the callback

frank canopy Nov 29, 2024, 11:43 PM

#

fiery turtle Idk what your plan for that is exactly, but note that the virtual mapping passed...

i do a static mapping of all phys mem in the upper half so all basic phys->virtual mappings are permanent in my os

fiery turtle Nov 29, 2024, 11:43 PM

#

I see, then it would work ofc, but still kinda hacky ig since that callback is meant to be more of a filtering mechanism

frank canopy Nov 29, 2024, 11:44 PM

#

fair ig

#

though i dont think i keep pointers to any of the tables generally

#

im extracting info into my own variables

fiery turtle Nov 29, 2024, 11:45 PM

#

Should be fine yeah

frank canopy Nov 29, 2024, 11:46 PM

#

oh yeah other thing i was going to ask about is can we get a function to get a namespace_node's parent? afaict from reading headers the only public function for that is to go through the relative-path option passing "^"

fiery turtle Nov 29, 2024, 11:46 PM

#

Its a function that exists now lol

#

Grep again

frank canopy Nov 29, 2024, 11:46 PM

#

oh neat

#

was reading before it got added then

fiery turtle Nov 29, 2024, 11:46 PM

#

Yeah

#

Whats your use case for it?

frank canopy Nov 29, 2024, 11:48 PM

#

the ascending/descending might cover it tbh

fiery turtle Nov 29, 2024, 11:48 PM

#

Now im curious

frank canopy Nov 29, 2024, 11:48 PM

#

my kernel device object keeps track of its parent so i build a true tree

fiery turtle Nov 29, 2024, 11:48 PM

#

Ah

frank canopy Nov 29, 2024, 11:49 PM

#

so i need to either do a bunch of depth-1 passes (which might fail given scopes i think) or track parents somehow

#

im not 100% sure how scopes and stuff work when trying to iterate devices since im just getting started rn

fiery turtle Nov 29, 2024, 11:50 PM

#

Scopes just allow to change the currently active scope, relative named objects created inside would reside inside that namespace instead of the current parents namespace

frank canopy Nov 29, 2024, 11:51 PM

#

so if im iterating then theyll show up where the scope wants to put them then? dont need to worry about that then really ig

fiery turtle Nov 29, 2024, 11:52 PM

#

Yeah after the namespace is created the Scope() stuff is sort of abstracted away in that for you it will be as if it was natively created under that namespace

frank canopy Nov 29, 2024, 11:53 PM

#

cool

fiery turtle Nov 29, 2024, 11:53 PM

#

That is if I understood your question correctly

frank canopy Nov 29, 2024, 11:53 PM

#

yep

fiery turtle Nov 29, 2024, 11:54 PM

#

Cool cool

frank canopy Nov 30, 2024, 1:38 AM

#

welp

#

i messed up but thats for a separate thread ig

hasty plinth Nov 30, 2024, 6:08 AM

#

fiery turtle stuff left to do for 1.0

congrats on the merge, and just wanted to say I'm excited for 1.0 🎉

fiery turtle Nov 30, 2024, 9:12 AM

#

hasty plinth congrats on the merge, and just wanted to say I'm excited for 1.0 🎉

Thank you ❤️

fiery turtle Nov 30, 2024, 9:13 AM

#

frank canopy i messed up but thats for a separate thread ig

What happened

fiery turtle Nov 30, 2024, 12:03 PM

#

a ton of different address spaces require special handling

#

this is from ACPICA

#

ill have to implement all of that before 1.0 unfortunately

frank canopy Nov 30, 2024, 4:00 PM

#

fiery turtle What happened

officially now kernel too big for my bootloader lol

left orbit Nov 30, 2024, 4:10 PM

#

limine time trl

vagrant hull Nov 30, 2024, 4:11 PM

#

limine

frank canopy Nov 30, 2024, 4:25 PM

#

anyway made a thread in progress report where details can go lol

deft canopy Nov 30, 2024, 4:28 PM

#

frank canopy officially now kernel too big for my bootloader lol

"Too big" in what way?

#

Well, it might be better to go to your thread

frank canopy Nov 30, 2024, 4:29 PM

#

deft canopy Well, it might be better to go to your thread

yeah lol

vast kestrel Nov 30, 2024, 6:10 PM

#

fiery turtle a ton of different address spaces require special handling

Have you ever seen anything using these stuff?

frank canopy Nov 30, 2024, 6:27 PM

#

only ones id think would be useful are memory, io, pci conf, and pci bar

kindred beacon Nov 30, 2024, 6:29 PM

#

vast kestrel Have you ever seen anything using these stuff?

I would assume that on some systems like embedded or laptops it might be useful to have devices downstream of an I2C/SPI/whatever controller be enumerable via ACPI

fiery turtle Nov 30, 2024, 6:40 PM

#

vast kestrel Have you ever seen anything using these stuff?

Well hobby osses ofc no, but Linux and stuff yes

#

And I think this is a thing on x86 as well on some hw

#

Or even on a lot of hw

vast kestrel Nov 30, 2024, 6:40 PM

#

Well, I only ever looked at server hardware that has the most generic and plain acpi tables :^)

fiery turtle Nov 30, 2024, 6:41 PM

#

Laptops have this serial and i2c crap

vast kestrel Nov 30, 2024, 6:41 PM

#

Damn

#

Makes sense

#

Cause EC

fiery turtle Nov 30, 2024, 6:41 PM

#

frank canopy only ones id think would be useful are memory, io, pci conf, and pci bar

Pci bar isnt used by any existing hardware

#

Strangely enough

fiery turtle Nov 30, 2024, 6:41 PM

#

vast kestrel Cause EC

Yeah

fiery turtle Nov 30, 2024, 6:41 PM

#

fiery turtle Pci bar isnt used by any existing hardware

Its not even supported by Linux

gentle peak Nov 30, 2024, 6:42 PM

#

is it supported by NT?

fiery turtle Nov 30, 2024, 6:42 PM

#

Probably not otherwise Linux would support it too

gentle peak Nov 30, 2024, 6:42 PM

#

then that's probably also the reason no hw uses it

fiery turtle Nov 30, 2024, 6:43 PM

#

Some stuff they added to the acpi standard just didnt stick

frank canopy Nov 30, 2024, 6:55 PM

#

fiery turtle Pci bar isnt used by any existing hardware

huh, interesting

frank canopy Nov 30, 2024, 9:43 PM

#

happy to report that ive got as far as uacpi_initialize working now in my kernel \o/

#

and using find_table etc for the fixed table init stuff

fiery turtle Nov 30, 2024, 9:45 PM

#

cool cool

#

now load dat namespace

frank canopy Nov 30, 2024, 9:46 PM

#

thats up next!

#

just happy thata ive got uacpi initialized and caught back up to the same place i was at before pulling in the dependency

#

(and got to throw out a bunch of code that i didnt really like anyway)

fiery turtle Nov 30, 2024, 9:48 PM

#

nice

fiery turtle Dec 1, 2024, 2:34 PM

#

uacpi is now at 140 stars, never expected to get this much
Thanks to everyone who starred!

median crest Dec 1, 2024, 2:41 PM

#

hell yeah

vagrant hull Dec 1, 2024, 2:57 PM

#

me not remembering if i starred it

#

let me check

#

oh yeah i already did, nice

fiery turtle Dec 1, 2024, 2:59 PM

#

Based

frank canopy Dec 3, 2024, 12:33 AM

#

kinda wild having it just work™️ when adding the namespace load and namespace init calls, thanks uacpi

#

there is some unimplemented errors because i return the unimplemented status from my pci read/write shims

#

but hey working is working and uacpi continues to work great

frank canopy Dec 3, 2024, 2:21 AM

#

for uacpi_kernel_install_interrupt_handler, do i need to apply redirects etc from the madt myself when using uacpi?

north holly Dec 3, 2024, 2:36 AM

#

frank canopy for `uacpi_kernel_install_interrupt_handler`, do i need to apply redirects etc f...

Yes

frank canopy Dec 3, 2024, 2:39 AM

#

cool

#

ill have to implement ioapic and irq redirect stuff then because id rather it not be overlapping with the exception vectors (even if 0x09 is deprecated)

fiery turtle Dec 3, 2024, 5:32 AM

#

frank canopy ill have to implement ioapic and irq redirect stuff then because id rather it no...

well if using pic u can just rebase both to start at 32

fiery turtle Dec 3, 2024, 5:32 AM

#

frank canopy kinda wild having it just work™️ when adding the namespace load and namespace in...

nice

frank canopy Dec 3, 2024, 5:33 AM

#

fiery turtle well if using pic u can just rebase both to start at 32

thats true ig

#

though i dont have proper code for either pic or io/apic yet so might as well go with ioapic/apic stuff for interrupts

hasty plinth Dec 3, 2024, 5:47 AM

#

I don't imagine it's what you meant, but don't rebase both pics to 32 - put one at 40.

fiery turtle Dec 3, 2024, 5:48 AM

#

yeah pic0 at 32, pic1 at 32 +however many pins (idr)

frank canopy Dec 3, 2024, 5:48 AM

#

it is 8 per pic yeah

#

iirc

torpid root Dec 3, 2024, 4:25 PM

#

interrupts are being disabled when initialising uacpi. this seems to be related to my mutex disabling and re-enabling them

#

I just save the status on lock and restore on unlock

#

and even if I just sti without checking the previous value, interrupts are still disabled in the end

#

which leads me to believe that uacpi might be forgetting to release a lock

#

hmmmmm maybe not

#

// ints on
lock1->acquire()
// ints off, lock1 knows that they were enabled before
lock2->acquire()
// ints off, lock2 doesn't know they were enabled
lock1->release()
// ints on, enabled by lock1
lock2->release()
// ints off, disabled by lock2

#

@fiery turtle something like this might be happening

#

one solution is to never disable them in my release function, only enable

#

I'll see if it works

#

yeah that's a fix but I don't really like it

#

or per cpu global state

#

idk

frank canopy Dec 3, 2024, 4:40 PM

#

torpid root interrupts are being disabled when initialising uacpi. this seems to be related ...

do you use the same lock for both spinlock and mutex? i was under the impression that mutexes are generally higher level and dont leave interrupts disabled

jaunty fox Dec 3, 2024, 4:40 PM

#

torpid root or per cpu global state

let the lock acquisition function return a token and let the lock release function take one

#

then it's uACPI's responsibility to call providing the appropriate tokens

frank canopy Dec 3, 2024, 4:41 PM

#

the uacpi mutex methods dont pass around tokens iirc

#

only the spinlock ones

#

might be misremembering

torpid root Dec 3, 2024, 4:41 PM

#

jaunty fox let the lock acquisition function return a token and let the lock release functi...

but what information would be stored in that token? interrupts status?

#

I don't think that would solve the problem

jaunty fox Dec 3, 2024, 4:42 PM

#

torpid root but what information would be stored in that token? interrupts status?

or IPL or similar

jaunty fox Dec 3, 2024, 4:42 PM

#

frank canopy the uacpi mutex methods dont pass around tokens iirc

oh, the mutexes

torpid root Dec 3, 2024, 4:42 PM

#

frank canopy do you use the same lock for both spinlock and mutex? i was under the impression...

I don't have a scheduler yet, so my mutexes are spinlocks

frank canopy Dec 3, 2024, 4:43 PM

#

ah ok

#

uacpi might just be holding onto the mutexes tbh

#

depending on what its doing exactly

#

do you have SMP set up yet? if not just do what i do and make mutexes a no-op until youve got scheduler and smp up tbh

#

or what i did anyway, ive got scheduler set up now

torpid root Dec 3, 2024, 4:47 PM

#

frank canopy do you have SMP set up yet? if not just do what i do and make mutexes a no-op un...

I do

fiery turtle Dec 3, 2024, 4:52 PM

#

torpid root <@353593948523986945> something like this might be happening

Store previous int state in your spinlock/mutex struct

torpid root Dec 3, 2024, 4:52 PM

#

I do that

fiery turtle Dec 3, 2024, 4:52 PM

#

Oh nvm

#

Why not implement like an actual mutex? They aren't supposed to disable interrupts because you might call into something that blocks while holding one

#

Your current implementation might sleep with interrupts disabled which will cause a deadlock

torpid root Dec 3, 2024, 4:55 PM

#

yes I'm doing that right now

fiery turtle Dec 3, 2024, 5:00 PM

#

Or don't disable interrupts at least

#

Because that's certainly not intended

torpid root Dec 3, 2024, 5:01 PM

#

yep until I have motivation to write a scheduler it does that

frank canopy Dec 3, 2024, 6:04 PM

#

is there a way to get the rsdt's oem id from uacpi?

#

i think thats the only thing my acpi code returned that im not getting back out yet from my uacpi code

deft canopy Dec 3, 2024, 6:39 PM

#

frank canopy is there a way to get the rsdt's oem id from uacpi?

I don't think there'd be a way to get it directly from uACPI, but you can probably bring in uACPI's ACPI header and use that to help parsing the RSDP->RSDT/XSDT->OEMID

fiery turtle Dec 3, 2024, 7:10 PM

#

frank canopy is there a way to get the rsdt's oem id from uacpi?

it doesnt actually store those tables, maybe it could tho

frank canopy Dec 3, 2024, 7:25 PM

#

fiery turtle it doesnt actually store those tables, maybe it could tho

honestly even just a sort of callback or out pointer for the oem id would be nice, afaik thats basically the only thing of note in the rsdt/xsdt other than the rest of the tables that uacpi already stores

fiery turtle Dec 3, 2024, 7:27 PM

#

frank canopy honestly even just a sort of callback or out pointer for the oem id would be nic...

every other table has oem id as well

#

and iirc they're the same for all tables so u can use any other

#

not just the RSDT

#

well at least on qemu for sure

frank canopy Dec 3, 2024, 7:28 PM

#

fiery turtle and iirc they're the same for all tables so u can use any other

i did a dump of tables from my laptop the other day and got a couple of different ones iirc

#

on qemu they would all be the same yeah

fiery turtle Dec 3, 2024, 7:28 PM

#

on real hw it makes sense yeah

#

some tables come from the chipset, some come from hardware e.g. the gpu or device-specific power state information etc etc

frank canopy Dec 3, 2024, 7:29 PM

#

yep

#

im not even really doing anything with the oemid either i was just printing it to my logs lol

fiery turtle Dec 3, 2024, 7:29 PM

#

btw

#

i was going to add that in to this pr i merged last week but

#

i have something like that in the works

#

[uACPI][INFO] RSDP 0x00007FFC133EBF60 36     v2   (      )
[uACPI][INFO] XSDT 0x0000604000001090 44     v1   (uTEST          )
[uACPI][INFO] DSDT 0x000060D0000002B0 130    v1   (uTEST  RUNRIDTB)
[uACPI][INFO] FACP 0x00006120000004C0 276    v6   (               )

#

would something like this work if uacpi dumped it instead of your kernel?

frank canopy Dec 3, 2024, 7:31 PM

#

yeah absolutely

#

the only other reason id want it is to try and detect running in a hypervisor/qemu and that i can do by checking any of the oemids

fiery turtle Dec 3, 2024, 7:31 PM

#

[uACPI][INFO] RSDP 0x00007FFF237AACB0 36     v2   (      )
[uACPI][INFO] XSDT 0x0000611000000CC0 220    v2   (ALASKA         )
[uACPI][INFO] DSDT 0x00007F8A9B571800 577646 v2   (ALASKA A M I   )
[uACPI][INFO] FACP 0x00006120000004C0 276    v6   (               )
[uACPI][INFO] SSDT 0x0000619000000A80 916    v2   (PmRef  Cpu0Cst )
[uACPI][INFO] SSDT 0x000061B000000080 1531   v2   (PmRef  Cpu0Hwp )
[uACPI][INFO] SSDT 0x000061F000000080 3238   v2   (ALASKA PtidDevc)
[uACPI][INFO] SSDT 0x0000622000000100 5247   v2   (ALASKA I2Pm_Rvp)
[uACPI][INFO] SSDT 0x000062D000000400 39780  v2   (INTEL  TcssSsdt)
[uACPI][INFO] SSDT 0x0000622000001900 4937   v2   (PmRef  ApPsd   )
[uACPI][INFO] SSDT 0x0000626000000100 10266  v2   (SocGpe SocGpe  )
[uACPI][INFO] SSDT 0x0000619000000F80 908    v2   (PmaxDv Pmax_Dev)
[uACPI][INFO] SSDT 0x0000624000002100 7087   v2   (PmRef  ApIst   )
[uACPI][INFO] SSDT 0x0000621000000100 4444   v2   (ALASKA UsbCTabl)
[uACPI][INFO] SSDT 0x000062A000000200 21338  v2   (DptfTb DptfTabl)
[uACPI][INFO] SSDT 0x0000614000000040 427    v2   (PmRef  Cpu0Psd )
[uACPI][INFO] SSDT 0x000062B000000200 24098  v2   (ALASKA TbtTypeC)
[uACPI][INFO] SSDT 0x000061C000000080 1838   v2   (PmRef  Cpu0Ist )
[uACPI][INFO] SSDT 0x0000628000000100 14492  v2   (ALASKA Ther_Rvp)
[uACPI][INFO] SSDT 0x000062B000007200 27658  v2   (ALASKA MtlS_APA)
[uACPI][INFO] SSDT 0x0000621000001500 4027   v2   (PmRef  ApCst   )
[uACPI][INFO] SSDT 0x0000626000003100 10451  v2   (SocCmn SocCmn  )
[uACPI][INFO] SSDT 0x0000625000034900 9342   v2   (INTEL  IgfxSsdt)
[uACPI][INFO] SSDT 0x0000622000003100 5664   v2   (PmRef  ApHwp   )
[uACPI][INFO] SSDT 0x000062500003C100 9828   v2   (INTEL  xh_mts2r)
[uACPI][INFO] SSDT 0x000061F000000E80 3331   v2   (INTEL  PDatTabl)

Examples with some real hw dumps

frank canopy Dec 3, 2024, 7:31 PM

#

neat

fiery turtle Dec 3, 2024, 7:31 PM

#

frank canopy the only other reason id want it is to try and detect running in a hypervisor/qe...

fair

frank canopy Dec 3, 2024, 7:32 PM

#

since "BOCHS " isnt showing up on real hardware anywhere most likely lol

fiery turtle Dec 3, 2024, 7:34 PM

#

yeah

#

but also its a hardcoded string in qemu, so u can always kinda replace it and work around the check

frank canopy Dec 3, 2024, 7:35 PM

#

true

gentle peak Dec 3, 2024, 7:41 PM

#

frank canopy the only other reason id want it is to try and detect running in a hypervisor/qe...

for this you can use cpuid, bit 31 of either ecx or edx in leaf 1 is 0 on real hw and 1 in hypervisors

frank canopy Dec 3, 2024, 7:42 PM

#

oh id forgotten that leaf, had been checking leaf 0x8000000A eax bit 8

gentle peak Dec 3, 2024, 7:43 PM

#

just checked it's ecx

#

but yeah pretty much all hypervisors set that bit to 1

#

it's also how to check that the cpuid leaves at 0x40000000 exist

frank canopy Dec 3, 2024, 7:46 PM

#

yeah that bit is working

#

whereas the 8000000A bit should also work for some hypervisors but isnt getting set

#

of curiosity, since i couldnt find any info on it with a quick search, is there any guidance for what interrupt priority level the SCI etc should be set at? ditto for remapping isa irqs, any general guidance?

gentle peak Dec 3, 2024, 8:01 PM

#

ISA interrupt priorities are generally indicated by their IRQ numbers (higher number is lower priority) since that's how the PIC works

#

but I usually just mostly ignore interrupt priorities since it becomes a mess to manage very quickly

fiery turtle Dec 3, 2024, 9:00 PM

#

frank canopy of curiosity, since i couldnt find any info on it with a quick search, is there ...

lowest possible

frank canopy Dec 3, 2024, 9:01 PM

#

cool that makes things easier lol

deft canopy Dec 3, 2024, 11:11 PM

#

frank canopy since "BOCHS " isnt showing up on real hardware anywhere most likely lol

little did you know, in 10 years when I eventually become dictator of IMD (Intel and AMD merged in the future, forming Intel Micro Devices) I'm gonna make them put BOCHS on the CPUs

fiery turtle Dec 6, 2024, 11:07 AM

#

wake up babe new uacpi changes are about to drop

#

fixed bogus error handling in namespace_node_resolve
uacpi now dumps all newly installed table headers, example:

[uACPI][INFO] RSDP 0x00007FFE882BFC80 00000024 v02 (uOEMID)
[uACPI][INFO] XSDT 0x0000604000001090 0000002C v02 (uTEST  uTESTTBL)
[uACPI][INFO] DSDT 0x0000614000000040 00000199 v02 (uTEST  TESTTABL)
[uACPI][INFO] FACP 0x00006120000004C0 00000114 v06 (uOEMID uTESTTBL)
[uACPI][INFO] FACS 0x0000606000000320 00000040
[uACPI][INFO] SSDT 0x0000000000000000 00000055 v01 (uTEST  RUNRIDTB)

a few misc thigns

fiery turtle Dec 6, 2024, 11:42 AM

#

looks pretty nice in Astral i think

median crest Dec 6, 2024, 3:33 PM

#

Epic

frank canopy Dec 6, 2024, 4:07 PM

#

nice

fiery turtle Dec 6, 2024, 4:31 PM

#

Thanks!

frank canopy Dec 6, 2024, 9:05 PM

#

im getting some invalid opcode exceptions in the FACS header logging for some reason now

#

on an actual UD1 evidently

fiery turtle Dec 6, 2024, 9:26 PM

#

unlucky

#

this works in astral soo idk

fiery turtle Dec 6, 2024, 9:27 PM

#

frank canopy im getting some invalid opcode exceptions in the FACS header logging for some re...

65 should be unreachable there

#

is your memcmp implemented correctly?

frank canopy Dec 6, 2024, 9:27 PM

#

heres the jump

#

and im using your stdlib.c one

#

idk why that test is there, theres none at any of the other log lines

fiery turtle Dec 6, 2024, 9:29 PM

#

so does my userspace test runner tbh

#

add a printf there and see why its not entering the first if statement

frank canopy Dec 6, 2024, 9:30 PM

#

the if signatures match one? that should be succeeding, its on the FACS and already fell through that memcmp

#

based on this objdump anyway

fiery turtle Dec 6, 2024, 9:30 PM

#

what if u add a printf instead of conjecture 😛

frank canopy Dec 6, 2024, 9:31 PM

#

on it

fiery turtle Dec 6, 2024, 9:31 PM

#

well there's a return statement on line 46 soo like

#

idk how it's not reached in this case

frank canopy Dec 6, 2024, 9:33 PM

#

#

added some printfs

fiery turtle Dec 6, 2024, 9:34 PM

#

frank canopy heres the jump

i think its maybe checking the pointer alignment

#

since FACS is not packed

frank canopy Dec 6, 2024, 9:34 PM

#

in that case, what are your compiler flags on your working use?

fiery turtle Dec 6, 2024, 9:34 PM

#

the default ones

#

well

frank canopy Dec 6, 2024, 9:34 PM

#

this might be zig being more strict when it invokes the c compiler than default

fiery turtle Dec 6, 2024, 9:35 PM

#

can u print the address of FACS?

#

like the pointer

frank canopy Dec 6, 2024, 9:35 PM

#

on it

#

0x000000007FFE0000 is the phys address

fiery turtle Dec 6, 2024, 9:36 PM

#

not that

#

the facs variable in that if statement

frank canopy Dec 6, 2024, 9:36 PM

#

ohh

fiery turtle Dec 6, 2024, 9:37 PM

#

i think i know why

#

the header is created on line 613 as a stack variable

#

since it's PACKED it has an alignment on 1

frank canopy Dec 6, 2024, 9:37 PM

#

ah yep 0xFFFFFFFF80A3FB2C

fiery turtle Dec 6, 2024, 9:37 PM

#

i should use a memcpy there

#

1 sec lemme give u a patch to try

frank canopy Dec 6, 2024, 9:38 PM

#

and zig defaults to strict compiler flags when compiling C code so the unaligned access causes the UD test to be generated instead that makes sense

fiery turtle Dec 6, 2024, 9:39 PM

#

good thing u use zig then

#

--- a/source/tables.c
+++ b/source/tables.c
@@ -37,11 +37,9 @@ static void dump_table_header(
     struct acpi_sdt_hdr *sdt = hdr;

     if (uacpi_signatures_match(hdr, ACPI_FACS_SIGNATURE)) {
-        struct acpi_facs *facs = hdr;
-
         uacpi_info(
             "FACS 0x%016"UACPI_PRIX64" %08X\n", UACPI_FMT64(phys_addr),
-            facs->length
+            sdt->length
         );
         return;
     }

#

i think even this works lol

frank canopy Dec 6, 2024, 9:39 PM

#

giving it a try

#

yup that worked

fiery turtle Dec 6, 2024, 9:40 PM

#

yeah the length is at the same offset in both structs but one is packed

#

should never have used this struct at all lol

#

thanks for finding the bug

#

or reporting it i guess

frank canopy Dec 6, 2024, 9:41 PM

#

nw lol, always good to get bugs found

#

now to go find the off by one error in my own code lmao

fiery turtle Dec 6, 2024, 9:42 PM

#

lol gl

#

pushed the fix

#

i wonder what flags its using to catch misaligned accesses

frank canopy Dec 6, 2024, 9:43 PM

#

turns out there was just a literal constant 1 labeled offset that i had due to misreading something so that was an easy off by one to find lol, though thats more for my own progress thread ig lol

fiery turtle Dec 6, 2024, 9:43 PM

#

i compile with both asan and ubsan

#

none of them cared in any of the test cases

#

and it runs like thousands of tests on all platforms for each commit

fiery turtle Dec 6, 2024, 9:44 PM

#

frank canopy turns out there was just a literal constant 1 labeled offset that i had due to m...

what was it for?

frank canopy Dec 6, 2024, 9:46 PM

#

fiery turtle what was it for?

im automatically generating the bit mask type for object_type_bits since zig has bitpacked structs built in but i made a mental off by one error and thought the integer flag was bit 0 etc and programmed that in

fiery turtle Dec 6, 2024, 9:46 PM

#

ah

frank canopy Dec 6, 2024, 9:52 PM

#

-fsanitize=undefined -fsanitize-trap=undefined is what im finding in the clang invocation from zig

#

might actually be something with my target features tbh

fiery turtle Dec 6, 2024, 9:53 PM

#

-fsanitize=address,undefined -g3 -Wall -Wextra -Werror

#

this is what my tests do

#

maybe i need trap as well idk

frank canopy Dec 6, 2024, 9:55 PM

#

idk

#

C:\Users\robot\.zvm\master\zig.exe clang Z:\imaginarium\uACPI\source\opregion.c --no-default-config -fno-caret-diagnostics -mcmodel=kernel -fsanitize=undefined -fsanitize-trap=undefined -fno-sanitize=function -mred-zone -fno-omit-frame-pointer -fno-stack-protector -O2 -D_FORTIFY_SOURCE=2 -Werror=date-time -funwind-tables -fno-PIC -gdwarf-4 -gdwarf32 -MD -MV -MF Z:\imaginarium\.zig-cache\tmp\d44124af22e359ab-opregion.o.d -ffreestanding -I Z:\imaginarium\include -I Z:\imaginarium\uACPI\include -ffreestanding -nostdlib -DUACPI_OVERRIDE_LIBC -DUACPI_SIZED_FREES -DUACPI_OVERRIDE_TYPES -DUACPI_OVERRIDE_ARCH_HELPERS -DUACPI_DEFAULT_LOG_LEVEL=UACPI_LOG_INFO -c -o Z:\imaginarium\.zig-cache\tmp\d44124af22e359ab-opregion.o --serialize-diagnostics Z:\imaginarium\.zig-cache\tmp\d44124af22e359ab-opregion.o.diag heres the full invocation without the target feature stuff, maybe you can see something i dont

fiery turtle Dec 6, 2024, 9:57 PM

#

hmm

left orbit Dec 6, 2024, 9:57 PM

#

is there a list of all configuration macros i can use? i.e sized frees

fiery turtle Dec 6, 2024, 9:57 PM

#

config.h

left orbit Dec 6, 2024, 9:58 PM

#

ah thanks!

fiery turtle Dec 6, 2024, 9:58 PM

#

np!

frank canopy Dec 9, 2024, 9:27 PM

#

one thing that might be nice to have would be docs on what errors each uacpi function can return (if thats even possible to determine)

#

would be useful in my zig bindings to try and make the error returns as narrowly typed as possible

fiery turtle Dec 9, 2024, 9:28 PM

#

definitely possible for at least some of them

#

a lot of them branch into the interpreter and from there its impossible to predict

frank canopy Dec 9, 2024, 9:28 PM

#

yeah

#

but for the ones that would be possible it might be nice

fiery turtle Dec 9, 2024, 9:28 PM

#

true

frank canopy Dec 9, 2024, 9:29 PM

#

not a huge deal since for the ones that are possible to predict i can usually go through and try to figure it out as i use the functions

fiery turtle Dec 9, 2024, 9:38 PM

#

yeah docs overall are kinda lacking right now so

#

something to improve in the future

fiery turtle Dec 10, 2024, 6:29 PM

#

Unrelated, but my ACPICA fix got upstreamed into the Linux kernel as well, which is nice

hollow elm Dec 10, 2024, 11:35 PM

#

what were the api changes for the new uacpi again?

#

i want to see if i can quickly update managarm to latest uacpi

fiery turtle Dec 10, 2024, 11:36 PM

#

get_ticks -> nanoseconds_since_boot

#

and mutex acquires returning a status

#

just those two

hollow elm Dec 10, 2024, 11:36 PM

#

also a change to the enumeration api i see

#

UACPI_RESOURCE_ITERATION_CONTINUE no longer exists?

fiery turtle Dec 10, 2024, 11:37 PM

#

those got unified into one

#

its just UACPI_ITERATION_DECISION_CONTINUE

hollow elm Dec 10, 2024, 11:37 PM

#

ah

fiery turtle Dec 10, 2024, 11:37 PM

#

for all iteration callbacks

hollow elm Dec 10, 2024, 11:42 PM

#

hm, uacpi_package is also no longer a thing

#

and uacpi_namespace_node_find

#

er no, the latter is just missing an arg

#

ah it has an out ptr now

fiery turtle Dec 10, 2024, 11:43 PM

#

yeah

fiery turtle Dec 10, 2024, 11:43 PM

#

hollow elm hm, `uacpi_package` is also no longer a thing

uacpi_object_get_package

#

object is now opaque

hollow elm Dec 10, 2024, 11:44 PM

#

hollow elm ah it has an out ptr now

can the out ptr just be null if i only want to check if it exists?

fiery turtle Dec 10, 2024, 11:45 PM

#

hollow elm can the out ptr just be null if i only want to check if it exists?

ofc

hollow elm Dec 10, 2024, 11:53 PM

#

to free the object arrays i just uacpi_free them right? i don't see a special dtor function in the header

fiery turtle Dec 10, 2024, 11:53 PM

#

u dont free them

#

u just uacpi_object_unref the original object

hollow elm Dec 10, 2024, 11:54 PM

#

ah it gives you a pointer to memory managed by the object

#

got it

fiery turtle Dec 10, 2024, 11:54 PM

#

yup

hollow elm Dec 10, 2024, 11:55 PM

#

ah wait, it's just an out ptr to a normal struct, not an out ptr to a ptr to a struct

#

i'm blind

fiery turtle Dec 10, 2024, 11:55 PM

#

lol

#

curious how managarm is going to hold up

#

in terms of ops/s

hollow elm Dec 11, 2024, 12:00 AM

#

i guess we'll see shortly

#

assuming i didn't mess anything up so far

fiery turtle Dec 11, 2024, 12:01 AM

#

api breakage was mostly cosmetic so i doubt it

#

no ideological changes

hollow elm Dec 11, 2024, 12:02 AM

#

what's the replacement for uacpi_namespace_node_get_object?

fiery turtle Dec 11, 2024, 12:02 AM

#

what was it used for

#

to get the type?

hollow elm Dec 11, 2024, 12:02 AM

#

yeah

fiery turtle Dec 11, 2024, 12:03 AM

#

uacpi_namespace_node_type/uacpi_namespace_node_is

hollow elm Dec 11, 2024, 12:03 AM

#

ah nice

fiery turtle Dec 11, 2024, 12:03 AM

#

where did it get the said object from btw?

#

or rather, the node

hollow elm Dec 11, 2024, 12:06 AM

#

uacpi_namespace_for_each_node_depth_first, that was gonna be my next question :^)

fiery turtle Dec 11, 2024, 12:06 AM

#

yeah there's now a for_each_child

hollow elm Dec 11, 2024, 12:06 AM

#

we're looking for a Device with a specific _ADR

fiery turtle Dec 11, 2024, 12:06 AM

#

which supports filtering by type

#

so u can just use that with UACPI_OBJECT_DEVICE_BIT

hollow elm Dec 11, 2024, 12:06 AM

#

ah nice

fiery turtle Dec 11, 2024, 12:07 AM

#

didnt realize managarm used all that api tbh lol

hollow elm Dec 11, 2024, 12:07 AM

#

this is in the pci code specifically

fiery turtle Dec 11, 2024, 12:07 AM

#

makes sense

hollow elm Dec 11, 2024, 12:07 AM

#

the package stuff was in the battery code

#

for _BIF and _BST

fiery turtle Dec 11, 2024, 12:08 AM

#

makes sense as wel

#

the current ops/s record is keyronex at about 850k/s peak i think

#

and it has slow mutex acquires too

hollow elm Dec 11, 2024, 12:09 AM

#

okay now one last thing, uacpi_init_params is gone

#

how do i feed it the rsdp now

fiery turtle Dec 11, 2024, 12:09 AM

#

uacpi_kernel_get_rsdp

hollow elm Dec 11, 2024, 12:09 AM

#

ah

fiery turtle Dec 11, 2024, 12:09 AM

#

that one is an old change

#

and just pass 0 to uacpi_initialize

hollow elm Dec 11, 2024, 12:13 AM

#

very minor typo: c // Returns the PHYSICAL address of the RSDP structure via *out_rsdp_address. uacpi_status uacpi_kernel_get_rsdp(uacpi_phys_addr *out_rdsp_address); the argument name contain be rsdp not rdsp

fiery turtle Dec 11, 2024, 12:13 AM

#

thanks for spotting

#

i like how its correct even in the comment

hollow elm Dec 11, 2024, 12:15 AM

#

hm ```
uacpi-info: successfully loaded 1 AML blob, 589 ops in 0ms (avg 806251/s)

#

this is with KASAN enabled though

fiery turtle Dec 11, 2024, 12:15 AM

#

huh

#

589 ops?

#

whats ur QEMU version

#

pretty much everyone whos posted here had ~1.7k

hollow elm Dec 11, 2024, 12:15 AM

#

this should be 9.something, but without -M q35

fiery turtle Dec 11, 2024, 12:16 AM

#

ah maybe thats why

#

can u try q35 + no asan?

hollow elm Dec 11, 2024, 12:16 AM

#

sure

fiery turtle Dec 11, 2024, 12:16 AM

#

800k with asan is kinda impressive

hollow elm Dec 11, 2024, 12:18 AM

#

uacpi-info: successfully loaded 1 AML blob, 1747 ops in 1ms (avg 1569521/s)

fiery turtle Dec 11, 2024, 12:18 AM

#

what the hell

#

phd code actually paid off

hollow elm Dec 11, 2024, 12:18 AM

#

the power of building with -O2

fiery turtle Dec 11, 2024, 12:19 AM

#

@jaunty fox you build keyronex with O2 as well right?

#

thats kinda insane damn

hollow elm Dec 11, 2024, 12:19 AM

#

i like the new list of tables that's printed

fiery turtle Dec 11, 2024, 12:19 AM

#

thanks!

#

sorta suggested by fadanoid as well

hollow elm Dec 11, 2024, 12:20 AM

#

overall diff is pretty small ```
8 files changed, 71 insertions(+), 61 deletions(-)

fiery turtle Dec 11, 2024, 12:20 AM

#

nice! i could take a look at the PR if u post it

fiery turtle Dec 11, 2024, 12:20 AM

#

hollow elm ``` uacpi-info: successfully loaded 1 AML blob, 1747 ops in 1ms (avg 1569521/s) ...

#

managarm is the new official record holder at 1.5M ops/s

#

nyaux is the anti record holder at 150 ops/s

#

like 4 orders of magnitude behind?

hollow elm Dec 11, 2024, 12:22 AM

#

sounds about right lol

fiery turtle Dec 11, 2024, 12:22 AM

#

lmaoo

#

i wonder what would be ops/s if i did this in linux

#

probably not that much higher

hollow elm Dec 11, 2024, 12:28 AM

#

fiery turtle nice! i could take a look at the PR if u post it

https://github.com/managarm/managarm/pull/785

fiery turtle Dec 11, 2024, 12:31 AM

#

Very clean, LGTM

fiery turtle Dec 11, 2024, 12:32 AM

#

hollow elm <https://github.com/managarm/managarm/pull/785>

There's also uacpi_eval simple helpers, that dont take in an arg

#

If u want to use that

hollow elm Dec 11, 2024, 12:32 AM

#

ci failed because the regex that was supposed to detect Depends on managarm/bootstrap_managarm#420 doesn't work

#

afaics

#

or i'm stupid

fiery turtle Dec 11, 2024, 12:33 AM

#

I don't have access to the ci output I think

fiery turtle Dec 11, 2024, 12:33 AM

#

fiery turtle There's also uacpi_eval simple helpers, that dont take in an arg

This basically just saves an extra nullptr param

#

Aka eval simple package

hollow elm Dec 11, 2024, 12:34 AM

#

oh it fails because i'm stupid and it's bootstrap-managarm not bootstrap_managarm

fiery turtle Dec 11, 2024, 12:34 AM

#

Lol

hollow elm Dec 11, 2024, 12:34 AM

#

nice it automatically retriggered the checks after changing the pr message

fiery turtle Dec 11, 2024, 12:35 AM

#

Nice

hollow elm Dec 11, 2024, 12:35 AM

#

ah that's why ```yml
on:
push:
pull_request:
types: [opened, edited, synchronize, reopened]

fiery turtle Dec 11, 2024, 12:35 AM

#

Makes sense

hollow elm Dec 11, 2024, 12:35 AM

#

fiery turtle Aka eval simple package

good to know, lemme change it to use that

fiery turtle Dec 11, 2024, 12:36 AM

#

We have more cases that could technically be simplified like that in managarm iirc

#

Like evaluation of ADR and stuff

hollow elm Dec 11, 2024, 12:36 AM

#

true

#

_ADR could be simple_integer ig?

fiery turtle Dec 11, 2024, 12:36 AM

#

Yup

#

Its just nicer to look at when there are no nullptrs anywhere I guess

#

Cleaner intent too

hollow elm Dec 11, 2024, 12:38 AM

#

yeah

#

okay pushed that and it looks like ci is gonna pass

#

i think i'll go to sleep now

fiery turtle Dec 11, 2024, 12:44 AM

#

Based

#

Looks cool

#

Gn qookie

hollow elm Dec 11, 2024, 12:44 AM

#

gn

frank canopy Dec 11, 2024, 1:00 AM

#

now yall made me curious and
INFO (uacpi): successfully loaded 1 AML blob, 1705 ops in 0ms (avg 1705000000000/s)

#

with zig compiling uacpi using -Doptimize=ReleaseSafe

#

though i dont have a timer thing yet

#

so this is probably quite wrong

frank canopy Dec 11, 2024, 1:25 AM

#

ok yeah that was just my lack of timer, here's the correct speed INFO (uacpi): successfully loaded 1 AML blob, 1723 ops in 7ms (avg 240107/s)

#

so fairly middling numbers

#

with ReleaseFast im up to 949526/s

fiery turtle Dec 11, 2024, 8:01 AM

#

frank canopy with ReleaseFast im up to 949526/s

not bad as well

mortal yoke Dec 11, 2024, 8:50 AM

#

I also had to test uacpi on my kernel KEKW (ignore the page fault, its 99% because there is code that expects qacpi to be initialized)

#

this is on kvm btw, on tcg its like 48700/s in 35ms

torpid root Dec 11, 2024, 9:08 AM

#

on kvm I get only 170k

#

also how do I get that table printout?

#

I just cloned the repo but it's not showing

torpid root Dec 11, 2024, 9:30 AM

#

torpid root on kvm I get only 170k

what affects speed the most?

torpid root Dec 11, 2024, 10:04 AM

#

not mapping memory as uncachable increased it to 250k

fiery turtle Dec 11, 2024, 10:25 AM

#

mortal yoke I also had to test uacpi on my kernel <:KEKW:913708372329644042> (ignore the pag...

not bad

fiery turtle Dec 11, 2024, 10:25 AM

#

torpid root I just cloned the repo but it's not showing

u sure u cloned the latest?

torpid root Dec 11, 2024, 10:25 AM

#

it should have

fiery turtle Dec 11, 2024, 10:25 AM

#

torpid root what affects speed the most?

allocator performance

#

also libc

torpid root Dec 11, 2024, 10:26 AM

#

sadge

fiery turtle Dec 11, 2024, 10:27 AM

#

torpid root it should have

well check your tables.c, does it have a dump_table_header?

#

or just tell me what git log shows as latest

torpid root Dec 11, 2024, 10:28 AM

#

fiery turtle well check your tables.c, does it have a dump_table_header?

yes

fiery turtle Dec 11, 2024, 10:28 AM

#

what does your uacpi log look like

torpid root Dec 11, 2024, 10:28 AM

#

is this it?

#

FACS

fiery turtle Dec 11, 2024, 10:29 AM

#

yes but it swallowed every other print lol

#

are u using a custom printf?

torpid root Dec 11, 2024, 10:29 AM

#

hmm

#

fmtlib

fiery turtle Dec 11, 2024, 10:29 AM

#

maybe something about the precision formatting in other printouts tripping it

torpid root Dec 11, 2024, 10:31 AM

#

oh for printf I have eyalroz/printf

fiery turtle Dec 11, 2024, 10:32 AM

#

add a printf into your log callback to print unformatted meme

#

to see if its ever called

torpid root Dec 11, 2024, 10:33 AM

#

only FACS

#

nothing else

fiery turtle Dec 11, 2024, 10:33 AM

#

can u add printf to dump_table_header?

#

maybe something in your code discarding those early logs?

torpid root Dec 11, 2024, 10:33 AM

#

that's difficult with my build system

fiery turtle Dec 11, 2024, 10:33 AM

#

are u using early table access?

torpid root Dec 11, 2024, 10:33 AM

#

yes

fiery turtle Dec 11, 2024, 10:34 AM

#

fiery turtle maybe something in your code discarding those early logs?

could be this

jaunty fox Dec 11, 2024, 10:34 AM

#

fiery turtle <@391791686881247234> you build keyronex with O2 as well right?

most probably it was built with -Og

fiery turtle Dec 11, 2024, 10:35 AM

#

oh damn, u might get crazier numbers at O2 then

torpid root Dec 11, 2024, 10:35 AM

#

fiery turtle maybe something in your code discarding those early logs?

I don't think so

#

I put everything in a vector and printed it later too

#

it just isn't called

fiery turtle Dec 11, 2024, 10:36 AM

#

torpid root I don't think so

FACS is mapped and printed out only after uacpi_initialize

#

the rest are printed out in early table access init

#

in your case

torpid root Dec 11, 2024, 10:37 AM

#

well nothing is printed when calling uacpi_setup_early_table_access

fiery turtle Dec 11, 2024, 10:37 AM

#

i have a guess actually

#

try doing uacpi_context_set_log_level before setting up early table access

#

it's probably still zeroed at that point

#

so all logs are discarded

jaunty fox Dec 11, 2024, 10:39 AM

#

on my system:

[uacpi]: successfully loaded 1 AML blob, 2356 ops in 4ms (avg 478537/s)

#

with -O2

torpid root Dec 11, 2024, 10:39 AM

#

fiery turtle try doing `uacpi_context_set_log_level` before setting up early table access

fiery turtle Dec 11, 2024, 10:39 AM

#

torpid root

mhm makes sense now

torpid root Dec 11, 2024, 10:39 AM

#

is this a uacpi bug?

#

shocked_bao

fiery turtle Dec 11, 2024, 10:39 AM

#

well

#

kinda

#

i should probably force the default log level in early table init

fiery turtle Dec 11, 2024, 10:40 AM

#

jaunty fox on my system: > [uacpi]: successfully loaded 1 AML blob, 2356 ops in 4ms (avg 47...

i think iretq was getting 850k even with Og on his?

#

This might yield numbers close to managarm on a fast system

jaunty fox Dec 11, 2024, 10:41 AM

#

here is iso for anyone who would like to try (i can also try managarm if someone has an ISO)

📎 barebones.iso

fiery turtle Dec 11, 2024, 10:41 AM

#

managarm gets built nightly i think

fiery turtle Dec 11, 2024, 10:41 AM

#

jaunty fox here is iso for anyone who would like to try (i can also try managarm if someone...

this still needs invariant tsc right?

jaunty fox Dec 11, 2024, 10:42 AM

#

fiery turtle this still needs invariant tsc right?

yeah, i need to integrate that kvm clock everyone talks of

fiery turtle Dec 11, 2024, 10:42 AM

#

yee

fiery turtle Dec 11, 2024, 10:42 AM

#

fiery turtle managarm gets built nightly i think

although the PR is still not merged, qookie needs to press the button

jaunty fox Dec 11, 2024, 10:43 AM

#

older keyronex i was trying hard to leave the way open for things like high-precision, dynamic tick on tricky clocks, but i now think this is not overly important as on modern power-efficient systems there should be an invariant tsc, while on older servers and such, neither dynamic tick nor high resolution times are desirable

fiery turtle Dec 11, 2024, 10:44 AM

#

thats probably true

jaunty fox Dec 11, 2024, 10:44 AM

#

so they will probably fall back to a fully tickful approach. which is fine and data from linux suggests that's what works best for server systems

fiery turtle Dec 11, 2024, 10:45 AM

#

makes sense

torpid root Dec 11, 2024, 10:45 AM

#

jaunty fox here is iso for anyone who would like to try (i can also try managarm if someone...

mortal yoke Dec 11, 2024, 10:45 AM

#

fiery turtle Dec 11, 2024, 10:46 AM

#

very close to 1M basically

torpid root Dec 11, 2024, 10:46 AM

#

also

jaunty fox Dec 11, 2024, 10:46 AM

#

torpid root also

that's to be expected

fiery turtle Dec 11, 2024, 10:46 AM

#

no invariant tsc

torpid root Dec 11, 2024, 10:47 AM

#

-cpu max,+invtsc -M q35

fiery turtle Dec 11, 2024, 10:47 AM

#

i have tried this here, didnt work

#

still no support for invtsc

torpid root Dec 11, 2024, 10:47 AM

#

did you enable kvm?

fiery turtle Dec 11, 2024, 10:48 AM

#

torpid root -cpu max,+invtsc -M q35

does it print a warning about no invtsc support in keyronex log?

torpid root Dec 11, 2024, 10:48 AM

#

no, it did without it

#

I need a faster allocator

fiery turtle Dec 11, 2024, 10:48 AM

#

torpid root did you enable kvm?

yes but im using wsl so that might have something to do with that

jaunty fox Dec 11, 2024, 10:48 AM

#

it will go on with a warning about no invtsc if there's no invtsc, but if there's no TSC-deadline, it dies

torpid root Dec 11, 2024, 10:48 AM

#

fiery turtle yes but im using wsl so that might have something to do with that

you can enable it in wsl2

fiery turtle Dec 11, 2024, 10:48 AM

#

wsl2 is not propagated invtsc i think

#

from the hyperv thing

jaunty fox Dec 11, 2024, 10:49 AM

#

torpid root I need a faster allocator

this is probably the area where most of the gains can be made

fiery turtle Dec 11, 2024, 10:49 AM

#

fast memcpy + fast allocator is what makes uacpi fast i bet

jaunty fox Dec 11, 2024, 10:49 AM

#

i will like to see how uacpi's performance changes if/when i add magazines or a similar layer

torpid root Dec 11, 2024, 10:49 AM

#

fiery turtle fast memcpy + fast allocator is what makes uacpi fast i bet

I believe my memcpy is quite fast

#

well "my"

fiery turtle Dec 11, 2024, 10:49 AM

#

jaunty fox i will like to see how uacpi's performance changes if/when i add magazines or a ...

should definitely go up

fiery turtle Dec 11, 2024, 10:50 AM

#

torpid root I believe my memcpy is quite fast

is it rep movsb?

torpid root Dec 11, 2024, 10:50 AM

#

yes

fiery turtle Dec 11, 2024, 10:50 AM

#

should be fast yeah

jaunty fox Dec 11, 2024, 10:50 AM

#

cache colouring another fun idea

fiery turtle Dec 11, 2024, 10:50 AM

#

then it's your allocator probably

#

are your mutex acquires fast if the mutex is not contended?

torpid root Dec 11, 2024, 10:51 AM

#

it's a variation of the ticket lock from frigg

#

iirc

fiery turtle Dec 11, 2024, 10:51 AM

#

should be fast then

#

well managarm is using those as well right

#

and it gets 1.5M so

torpid root Dec 11, 2024, 10:52 AM

#

managarm uses the allocator from frigg right?

fiery turtle Dec 11, 2024, 10:53 AM

#

think so

torpid root Dec 11, 2024, 10:53 AM

#

I'll figure out how to use that when I have some free time

fiery turtle Dec 11, 2024, 10:54 AM

#

well at least you're still on the same order of magnitude and not 150/s LULW

hollow elm Dec 11, 2024, 10:55 AM

#

ill try keyronex on my machine to see how it compares to managarm on the same cpu

fiery turtle Dec 11, 2024, 10:55 AM

#

nice

median crest Dec 11, 2024, 10:58 AM

#

fiery turtle wsl2 is not propagated invtsc i think

Yeah whole reason I did go for kvm clock

torpid root Dec 11, 2024, 11:09 AM

#

fiery turtle well at least you're still on the same order of magnitude and not 150/s <:LULW:5...

I guess I'll use frigg allocator

fiery turtle Dec 11, 2024, 11:11 AM

#

2x perf?

torpid root Dec 11, 2024, 11:11 AM

#

yes

fiery turtle Dec 11, 2024, 11:11 AM

#

not bad

torpid root Dec 11, 2024, 11:12 AM

#

513k in release mode

dusky glade Dec 11, 2024, 11:14 AM

#

torpid root I guess I'll use frigg allocator

at least ur not nyaux

#

💀

#

wanna know how much ops nyaux can perform per second

#

151

#

1838 ops in 12114ms

torpid root Dec 11, 2024, 11:16 AM

#

kekw~3

dusky glade Dec 11, 2024, 11:24 AM

#

(on tcg)

#

idk about real life hardware

#

never tested it on real hardware yet

torpid root Dec 11, 2024, 11:26 AM

#

me neither

#

I don't have a terminal yet

#

I froggot

#

not using ovmf slows it down by almost 200k

#

I still have some optimising to do I guess

#

I'll switch from bitmap to buddy for my physical memory allocator

#

I should make a progress report thread instead of cluttering this one lol

fiery turtle Dec 11, 2024, 11:51 AM

#

torpid root not using ovmf slows it down by almost 200k

hmm why does that even matter

#

what changes with ovmf?

torpid root Dec 11, 2024, 11:51 AM

#

I don't know really

#

maybe memory layout

fiery turtle Dec 11, 2024, 11:52 AM

#

strange

torpid root Dec 11, 2024, 11:58 AM

#

oh I didn't change anything, ran it again and slowdown is only 50k now

#

I love when stuff like this happen

fiery turtle Dec 11, 2024, 12:17 PM

#

lol

hollow elm Dec 11, 2024, 2:07 PM

#

jaunty fox here is iso for anyone who would like to try (i can also try managarm if someone...

on my laptop [uacpi]: successfully loaded 1 AML blob, 1726 ops in 3ms (avg 551790/s) with ```
qemu-system-x86_64 -M q35 -enable-kvm -cpu host -smp 4 -m 2G -cdrom barebones.iso -serial stdio

#

which is approximately the same as the managarm qemu args

fiery turtle Dec 11, 2024, 2:08 PM

#

wtf is managarm doing that its so fast

jaunty fox Dec 11, 2024, 2:08 PM

#

hollow elm on my laptop ``` [uacpi]: successfully loaded 1 AML blob, 1726 ops in 3ms (avg 5...

while managarm gets about 3x that?

hollow elm Dec 11, 2024, 2:08 PM

#

apparently yeah

jaunty fox Dec 11, 2024, 2:09 PM

#

striking, i will investigate this

#

how annoyed would you be if i gave you more ISOs to try later?

hollow elm Dec 11, 2024, 2:09 PM

#

not a problem

fiery turtle Dec 11, 2024, 2:09 PM

#

managarm doesnt even optimize mappings to use HHDM i think

hollow elm Dec 11, 2024, 2:09 PM

#

inb4 managarm just has broken time keeping

fiery turtle Dec 11, 2024, 2:10 PM

#

lol

torpid root Dec 11, 2024, 2:10 PM

#

lmao

fiery turtle Dec 11, 2024, 2:10 PM

#

try adding a sleep(1) in uacpi

#

or a 1000 i guess

hollow elm Dec 11, 2024, 2:10 PM

#

fiery turtle managarm doesnt even optimize mappings to use HHDM i think

well thor only has memory areas covered by the pmm mapped into the hhdm

fiery turtle Dec 11, 2024, 2:11 PM

#

yeye

#

can try adding a sleep in uacpi_namespace_load, right after it reads the starting counter

hollow elm Dec 11, 2024, 2:11 PM

#

sure

fiery turtle Dec 11, 2024, 2:12 PM

#

and if the numbers add up at the end, its probably correct

jaunty fox Dec 11, 2024, 2:12 PM

#

by my tests, the cost of mutex acquisition seems to be minimal

#

that leaves the primary suspect the allocator

fiery turtle Dec 11, 2024, 2:12 PM

#

yeah the frigg allocator doubled ilobilo's ops/s

#

could be that

hollow elm Dec 11, 2024, 2:13 PM

#

uacpi-info: successfully loaded 1 AML blob, 1747 ops in 1002ms (avg 1743/s)

fiery turtle Dec 11, 2024, 2:14 PM

#

looks fairly accurate then

hollow elm Dec 11, 2024, 2:14 PM

#

feels like it's about a second too

fiery turtle Dec 11, 2024, 2:14 PM

#

who said microkernels were slow meme

hollow elm Dec 11, 2024, 2:14 PM

#

just need to have a phd :^)

fiery turtle Dec 11, 2024, 2:15 PM

#

phd memcpy and allocator paying off

jaunty fox Dec 11, 2024, 2:16 PM

#

merely removing the printf in uacpi_kernel_map helps me significantly

#

so here's one that doesn't

📎 barebones.iso

#

it also nicely panics immediately after namespace init

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