its very aggressive when it comes to optimization
#uACPI - a portable and easy-to-integrate ACPI implementation
left orbit
ub
or more like
stuff that you wouldnt expect from your program
also the lto is enabled by default
some might find that to be nice, but it tends to be too good
also clashes with inline assembly nuking half of the symbols from the binary :^)
but that is most likely a zig bug, hopefully fixed in the future
and of course the compiler is very pedantic, for example if a variable is never written to it will emit an ERROR that you HAVE to make it const
or if you have unused parameters/variables... but you can use _ = expr; syntax for that, basically discarding them... which turns into an error if you use them later lol
i hope you can see how that might get a bit annoying after a while
</rant> sorry for hijacking thread :p
fiery turtle
also the lto is enabled by default
Even Linux has a hard time with that
left orbit
there's a reason it's 0.13, not 1.x i guess
fiery turtle
</rant> sorry for hijacking thread :p
Np lol
fiery turtle
there's a reason it's 0.13, not 1.x i guess
Yeah
left orbit
found some funkiness with pointer const-ness in uacpi apis, zig's translate-c translates const char ** in uacpi_device_matches_pnp_id as a non-const array of [*c]const uacpi_char, which would make sense if it wasn't for the fact it tells zig that uacpi will mutate the array :p
i am going to solve this with a @constCast for now but it would be really nice if you could go over the apis that take read-only null terminated arrays, like uacpi_device_matches_pnp_id does, and make sure it is properly marked as const
fiery turtle
i am going to solve this with a `@constCast` for now but it would be really nice...
Oh yeah that sounds useful
Thanks for bringing that up
left orbit
happy to help!
c interop from zig is such a chore lol, but i "finished" implementing uacpi support in my kernel, now i discover pcie root buses and ps2 keyboard controllers by calling uacpi_device_matches_pnp_id :^)
fiery turtle
c interop from zig is such a chore lol, but i "finished" implementing uacpi supp...
Isnt find_devices easier?
But nice
I have an example of usage on the wiki
fiery turtle
c interop from zig is such a chore lol, but i "finished" implementing uacpi supp...
Also you might be interested in get_namespace_node_info instead of calling matches twice (again examples available on the wiki if you're interested)
Not that what you're doing is wrong, just some suggestions
fiery turtle
Strangest uACPI usage I've seen
This guy copied some files and split some of them into other files, rewrote some of it
hollow elm
ucapi
fiery turtle
Made his own structures for some reason
And still does the _S5 manual scan from osdev hacks
jaunty fox
Also https://github.com/MirMohammadd/AliNix/pull/235
Outrageous commit names
north holly
jaunty fox
Folder structure also dumb it's dozens of folders each with one or two one function files in most cases it's disgusting
hollow elm
Outrageous commit names
it's all like that, not just that pr :^)
fiery turtle
Outrageous commit names
Indeed
jaunty fox
This arsehole also too much of a bastard devoid of scruples to follow the SINGLE requirement of Mit licence
He stripped it our. You can't do that it's a violation
fiery turtle
True lol
hollow elm
dmca time 
fiery turtle
He didn't do it out of ill intent tho
I mean he does have the repo starred
So probably just doesn't know how these things work
vagrant hull
Outrageous commit names
lmao
what a knob
and of course he has a shitty borked BIOS bootsector "bootloader"
fiery turtle
Lol
hollow elm
https://github.com/MirMohammadd/AliNix?tab=readme-ov-file#important
all rights will be reserved!
hollow elm
Starting the Kheap...
Starting the Kheap..
Kheap is getting started...
vagrant hull
all rights will be reserved!
i WILL steal your code and i WILL reserve my rights!
north holly
worse code than @dusky glade's code
rain nimbus
This guy copied some files and split some of them into other files, rewrote some...
time for me to randomly rewrite the entirety of uacpi to make it slower
dusky glade
worse code than <@787341689333350450>'s code
bruh
fiery turtle
https://github.com/MirMohammadd/AliNix/blob/main/kernel/drivers/power/supply/acp...
Finally, the uACPI interpreter reduced from 5k to 50 loc
vagrant hull
https://github.com/MirMohammadd/AliNix/blob/main/kernel/Makefile best makefile ever
minor patch
fiery turtle
<https://github.com/MirMohammadd/AliNix/blob/main/kernel/Makefile> best makefile...
Lol
north holly
lol wtf is this kernel
fiery turtle
This is some cursed stuff
north holly
the description looks AI generated
dusky glade
fr
fiery turtle
the description looks AI generated
Yeah it does
So many words that carry 0 information
north holly
especially considering his other two repos' descriptions are gramatically incorrect
rain nimbus
the description looks AI generated
Like no joke wtf
sterile egret
This code is too shitty to be AI generated
fiery turtle
Yeah the code is hand written
rain nimbus
ByteSmart-Compiler Public
A compiler which calculates a mathematical/human readable data and convert it to bit code.
Dafuq does that even mean xD
fiery turtle
Maybe with copilot
sterile egret
arch/alpha? what the fuck
north holly
what the actual fuck is this naming ""convention""
hollow elm
This code is too shitty to be AI generated
it's just taken from a bunch of other projects
sterile egret
Yeah that^
hollow elm
oh hm, heisen kernel is apparently the old name of alinix? https://uwu.foundation/Q5byAN6.png
north holly
hollow elm
oh god no https://github.com/MirMohammadd/AliNix/tree/main/patch
From: Heisenberg <[email protected]> that explains the previous name
sterile egret
fiery turtle
https://github.com/MirMohammadd/AliNix/blob/main/driverso/fpga/altera-pr-ip-core...
What the
rain nimbus
wtf
north holly
wtf
fiery turtle
Is this like a meme project or what
rain nimbus
And I thought my 2 files were a lot 
sterile egret
fiery turtle
Or is this guy mentally ill
hollow elm
Is this like a meme project or what
yeah i'm thinking this has to be bait
north holly
This has 2500 commits
vagrant hull
Is this like a meme project or what
andesaurux is still worse
fiery turtle
https://github.com/MirMohammadd/AliNix/blob/ca3a1f19bffc37d2410aa1d850c55f96d514...
Brooo
north holly
who would spend that much time just to troll
vagrant hull
andesaurux is still worse
at least this seems to somewhat build and do something
sterile egret
fractals of stupidity
fiery turtle
andesaurux is still worse
What is that?
vagrant hull
a meme kernel made by some Russian child that i have also gotten banned off github 3 times for harassment
fiery turtle
Lol
vagrant hull
""""kernel""""
it did not actually build
it was just a random mish mash of nonsense
fiery turtle
Have you actually verified that this builds 
vagrant hull
no lol
fiery turtle
Oh
north holly
also
fiery turtle
Just really weird stuff
sterile egret
north holly
over 2 months wtf
nvm "first commit" was misleading
fiery turtle
https://github.com/MirMohammadd/AliNix/blob/ca3a1f19bffc37d2410aa1d850c55f96d514...
Uhh
vagrant hull
grujb????
rain nimbus
fiery turtle
Wtf
rain nimbus
Straight up feels like that one kid who made a shader from scratch, which totally wasn't stolen from cherno O_O
north holly
I have lost too many brain cells as a result of this to continue looking at the code
fiery turtle
Has to be either a child or mentally ill person
rain nimbus
"Hey I'll give you my solution, just don't straight up copy it"
"Yeah sure I'll change some things"
Changes only the filename
vagrant hull
why the fuck does it have a Makefile.am that isn't automake and there is no autotools usage that i can tell
this is literally just random shit
fiery turtle
Im actually laughing out loud lmao
north holly
lol
fiery turtle
This shits insane
north holly
abbreiviation fail
vagrant hull
how did you even find out about this
fiery turtle
This guy starred uACPI and has a fork apparently
sterile egret
fiery turtle
I just saw it
vagrant hull
ah
fiery turtle
Lol
north holly
ok good my kernel isn't here
rain nimbus
I read ToaruOS as "A certain OS" O_O
north holly
so @rain nimbus, when you making kernel?
hollow elm
I read ToaruOS as "A certain OS" O_O
that is indeed why it's called that
sterile egret
There's so many different kernels in this repo lol
rain nimbus
so <@272414379083825154>, when you making kernel?
Well rn I'm doing some rendering stuff, wanting to make a path tracing based engine using wavelengths for more beautiful renderings 
And then once I'm done with the next two exams I'll go back to finishing HTTP1.1, HTTP2, HTTP3 and TLS1.3 implementations for my secrets manager (well more like a shared library I use in most of my C++ projects). And once I'm fully done with the secrets manager I'll go back to working on the kernel :>
Give it like 3 months, or roughly to august
Then maybe I'm done xD
rain nimbus
that is indeed why it's called that
Hmm, are they toaru fans perhaps :>
left orbit
Also you might be interested in get_namespace_node_info instead of calling match...
warning: struct demoted to opaque type - has bitfield 😭
translate-c doesn't seem to like that atm
fiery turtle
warning: struct demoted to opaque type - has bitfield 😭
Bitfields aren't a thing in zig?
left orbit
not really, at least not in the C sense
fiery turtle
Damn
left orbit
zig has it's own packed structs, which is basically a struct backed by a single integer type
but they aren't abi compatible with C
left orbit
so i'll just have to manually mask out bits from that field it seems Lol
fiery turtle
But for native C code this is more convenient
left orbit
that would be nice
yeah definitely
i wonder how rust handles that
with regards to C interop
left orbit
oh god the string length includes the null terminator
fiery turtle
Yeah
left orbit
i hope that it's consistent at least, but imo string lengths shouldn't include the null terminator
fiery turtle
So that all api that has lengths is consistent and also no empty strings
Yeah I just copied that api aspect from acpica
left orbit
a bit unrelated but what is the pci host bus "segment"? it seems that they it's not part of the address, what is it used for?
fiery turtle
Just another level
Its new in pcie
Look at osdev wiki I guess
It used to be BDF now its SBDF
left orbit
ahh PCIe stuff, my knowledge when it comes to that is minimal unfortunately
i still access the config space using port io, i should probably work on that!
fiery turtle
ECAM is better yeah
left orbit
looks like i can discover the base address for MMIO from the MCFG table
i need the segment and bus numbers for that..?
and then i can use that address to poke certain device/function's config space
fiery turtle
Yeah
left orbit
should i use MCFG to find the config space base address?
or can i find those using acpi?
fiery turtle
uacpi_table_find(MCFG)
left orbit
so MCFG is the way to go, got it
fiery turtle
Ye
left orbit
and now my PCI addresses should take the segment into consideration as well
left orbit
i feel like i can learn a lot by looking at the aml, what do you use to decompile it?
also where i can get a copy of qemu's DSDT? im assuming that's what i need
fiery turtle
i feel like i can learn a lot by looking at the aml, what do you use to decompil...
iasl <file>
left orbit
ah, iasl
fiery turtle
also where i can get a copy of qemu's DSDT? im assuming that's what i need
Unfortunately its generated at runtime soo
You have to dump it yourself
fiery turtle
Or write it to serial
left orbit
oh right, it just a table right?
do i just feed the raw bytes of the table to iasl?
well, save it to a file then pass it to iasl
fiery turtle
Yup
left orbit
czapek@raptor-wsl:~/source/zigux$ file dsdt.bin
dsdt.bin: ACPI Machine Language file 'DSDT' BXPC 1 by BOCHS, revision 1, 8389 bytes, created by BXPC 1
czapek@raptor-wsl:~/source/zigux$ iasl dsdt.bin
Intel ACPI Component Architecture
ASL+ Optimizing Compiler/Disassembler version 20230628
Copyright (c) 2000 - 2023 Intel Corporation
File appears to be binary: found 81 non-ASCII characters, disassembling
Table [DSDT] is too long for file - needs: 0x20C5, remaining in file: 0xF3
Binary file does not contain a valid standard ACPI table
Compilation successful. 0 Errors, 0 Warnings, 0 Remarks, 0 Optimizations```hmmm
i am dumping sdt_hdr.length bytes starting at the header
fiery turtle
i am dumping `sdt_hdr.length` bytes starting at the header
Hmm
That should be correct
left orbit
libmagic says it's an AML file, so there's that
czapek@raptor-wsl:~/source/zigux$ file dsdt.bin
dsdt.bin: ACPI Machine Language file 'DSDT' BXPC 1 by BOCHS, revision 1, 8389 bytes, created by BXPC 1```Well it just reads the header
It doesn't do sanity checking of length
left orbit
well thats fair
fiery turtle
Can you show the dump code?
left orbit
const dsdt_table = findTable(uacpi.ACPI_DSDT_SIGNATURE);
if (dsdt_table != null) {
logger.info("DSDT: {any}", .{std.fmt.fmtSliceHexUpper(dsdt_table.?.getData())});
}```pub fn getData(self: *const @This()) []const u8 {
const hdr: *uacpi.acpi_sdt_hdr = @ptrCast(self.uacpi_table.unnamed_0.hdr);
const data: [*]const u8 = @ptrCast(hdr);
return data[0..hdr.length];
}```sorry for zig lol hope it's not too much of a trouble
fiery turtle
How do you get the pointer to it from uACPI?
left orbit
var table: ?*uacpi.uacpi_table = null;
uacpi.uacpi_table_find_by_signature(signature, &table)```and then i pass that table pointer around wrapped in a struct
it works for MADT
fiery turtle
Yeah but what is GetData
fiery turtle
Oh thats wrong
left orbit
it just constructs a byte slice starting at the sdt header spanning hdr.length bytes
fiery turtle
hdr is the header
left orbit
oh?
fiery turtle
U need virt_addr
fiery turtle
Hmm yeah I guess, maybe its some zig meme
left orbit
i cast it to a u8* later
fiery turtle
Strange
fiery turtle
Do you have the test runner handy?
left orbit
the uacpi one?
left orbit
i can clone and build it
left orbit
czapek@raptor-wsl:~/source/uACPI/tests/runner$ ./build/test-runner ../../../zigux/dsdt.bin
unexpected error: DSDT declares that it's bigger than ../../../zigux/dsdt.bin```i guess im not dumping enough bytes?
left orbit
so that should be fine
fiery turtle
Yeah
fiery turtle
so that should be fine
Should, yeah
left orbit
let me log what im getting
fiery turtle
What?
fiery turtle
Lol
left orbit
8389 bytes
OH
i am dumb, let me just log to e9 instead
my print has a length limit...
fiery turtle
Lol
left orbit
that was very dumb lmao
fiery turtle
Why do u even have a limit lmao
left orbit
im using zig's fmt library which needs a buffer to format stuff into
and instead of having a global buffer that's big + a lock around it i just have a buffer on stack
and it cant be too big for obvious reasons
fiery turtle
I see
left orbit
ill see if i can somehow use zig's fmt with some writer thing that will let me avoid buffering
fiery turtle
Cant u just byte bang to e9
left orbit
the dsdt? i am doing that right now
left orbit
Disassembly completed
ASL Output: dsdt.dsl - 88563 bytes```wooo
time to figure out how the pci host bridges are defined in there
ohh the bridge's _HID is PNP0A08 but the _CID is PNP0A03, i see
there doesn't seem to be a _CBA anywhere unfortunately
fiery turtle
there doesn't seem to be a _CBA anywhere unfortunately
-M q35?
fiery turtle
Interesting, someone did mention it existed on qemy
Maybe yours is old or they were wrong
Anyway its not mandatory and mcfg is used as fallback
left orbit
i am on 8.2.3
fiery turtle
9 is the current release
fiery turtle
Yeah give that a try
do —target-list=x86_64-softmmu
Or you will be compiling that shit for an hour
left orbit
yeah i was gonna do just x86_64, but qemu takes like a minute for me anyway
or maybe thats with the target list
i cant remember
fiery turtle
For all 9k files?
left orbit
yeah
fiery turtle
Are you on a NASA server
fiery turtle
Lol
left orbit
already at 3k
fiery turtle
Target list is only 2k files
left orbit
i am building master so i hope thats recent enough
well then that'd be done in 20 seconds
anyway im at 7.5k right now
left orbit
ok its done now, lets see
i could try looking into that but i dont think its a well supported zig target lol
aarch64 is bad enough
left orbit
hmm for some reason the build got stuck at 9401/9437
fiery turtle
Nah it was fine
left orbit
czapek@raptor-wsl:~/source/zigux$ diff dsdt.bin dsdt-9.bin
czapek@raptor-wsl:~/source/zigux$ diff dsdt.dsl dsdt-9.dsl
8c8
< * Disassembly of dsdt.bin
---
> * Disassembly of dsdt-9.bin
no but let me check lol
left orbit
yeah there is an MCFG table
fiery turtle
Damn
left orbit
sooo lack of _CBA is fine then?
fiery turtle
Its not mandatory yeah
left orbit
alright
fiery turtle
Mcfg is the fallback
fiery turtle
An opportunity for you to make a patch <:halfmemeright:715831702164996096>
Unironicallly
left orbit
a patch to qemu?
fiery turtle
It should be 3 lines of code in qemu
fiery turtle
Ye
left orbit
should it be available on i440fx?
fiery turtle
it's just an integer?
Aml objects are sort of typeless in bytecode perspective , or rather methods are transparent
There isnt a method call operator
So its automatically invoked
So method or named object is same thing
left orbit
lol it doesnt work
i think im regarded or smth
if (mcfg_valid) {
printf("adding _CBA %zx\n", mcfg.base);
aml_append(dev, aml_name_decl("_CBA", aml_int(mcfg.base)));
}```the printf gets printed but the name doesn't appear in the decompiled aml
nvm im blind
Name (_CBA, 0xB0000000) // _CBA: Configuration Base Address
seems to have worked?
left orbit
time to become a qemu contributor
left orbit
hmm there is already a method called acpi_dsdt_add_gpex which adds a pci bus device to the aml, and that adds does add a _CBA thing
left orbit
it's only called for loongarch
left orbit
i did try without smm and there wasn't a _CBA anyway
fiery turtle
Yeah ok
left orbit
nvm the acpi_dsdt_add_gpex is called from multiple places, including i386 from a file called acpi-microvm.c
also from whatever the hell hw/pci-host is
ah that's where the method is defined lmao nvm
fiery turtle
microvm is amazon firecracker
fiery turtle
also from whatever the hell hw/pci-host is
Why the hell is it not called for you
left orbit
good question! thanks qemu
fiery turtle
Lol
left orbit
i think its because it checks if its q35 or i440fx?
fiery turtle
Dont u specify q35
left orbit
yeah i do
fiery turtle
Huh
left orbit
but q35 builds its own dsdt
fiery turtle
Strange
left orbit
honestly i'm going to leave that for someone who actually knows what they are doing, especially since it looks like i need to send a patch to a mailing list (!!!) and i've never done that before lol
left orbit
yeah it is trivial, but like you said maybe it's a bug or a regression and in that case the patch is pretty useless
fiery turtle
I guess
left orbit
#define ACPI_MADT_POLARITY_MASK 0b11
#define ACPI_MADT_POLARITY_CONFORMING 0b00
#define ACPI_MADT_POLARITY_ACTIVE_HIGH 0b01
#define ACPI_MADT_POLARITY_ACTIVE_LOW 0b11```
how does that work? why are both mask and active_low 0b11?spec says
Polarity of the APIC I/O input signals:
00 Conforms to the specifications of the bus (for example,
EISA is active-low for level-triggered interrupts).
01 Active high
10 Reserved
11 Active low
fiery turtle
```c
#define ACPI_MADT_POLARITY_MASK 0b11
#define ACPI_MADT_POLARITY_CONFORMING ...
mask is the mask of the bit field
its not some other value lol
fiery turtle
value & MASK == ACITIVE_LOW etc
fiery turtle
The only "feature" uacpi is currently missing is basically thread safety
and its so fkn boring
but i guess i have to bite the bullet and just do it
namespace & table should be easy to protect with just a mutex
interpreter is more complicated
particularly the only edge cases where something does a CopyObject
Which either needs an RCU-like object replacement thing or some other bs
Anyway, made an issue https://github.com/UltraOS/uACPI/issues/74
fiery turtle
@hollow elm btw I finally cleaned up & merged While timeouts
any fuzzer timeout is now considered a bug
fiery turtle
also since the test runner with --enumerate-namespace now runs _PRS and _CRS we might uncover new conversion bugs with fuzzer especially against real hw blobs
if fuzzer generates bogus resource lengths or offsets
i do have validation there but i doubt its perfect
i was too lazy to write out a dumper for each resource, but at least it prints out the length
wtf is this
tired crater
Because you plug it in and play
dusky glade
real
fiery turtle
why is it called that lmao
No idea, some legacy standard
vast kestrel
In general the idea is that you just plug the hardware and the os can automatically detect it
At least afaik
Like pcie is plug and play since you as the user don't need to configure anything be it in the hardware or the software
unlike legacy hardware which is just assumed to exist or you need to properly configure the ids/addresses for devices
fiery turtle
soo im working on mutexes and im a bit unsure
this_id = uacpi_kernel_get_thread_id();
if (mutex->owner == this_id) {
mutex->depth++;
return UACPI_TRUE;
}
since mutexes are recursive i have to do this
but owner is 0 by default
and technically thread_id of 0 is also valid
so there must be some sentinel value that corresponds to no owner
i could just say that 0 is that value, or maybe make it customizable by the client
or maybe thats bad api in general
like i have no idea
rustic compass
why do you give a mutex a owner?
fiery turtle
since mutexes are recursive i have to do this
.
ok ACPICA just does this #define ACPI_MUTEX_NOT_ACQUIRED ((ACPI_THREAD_ID) -1)
i think ill just do this as well but make it customizable as well
loud ice
```c
this_id = uacpi_kernel_get_thread_id();
if (mutex->owner == this_id) {
...
that does not look correct
fiery turtle
that does not look correct
how?
loud ice
how?
that's not how you usually do concurrent algorithms
read: this is definitely broken
fiery turtle
what algorithms
jaunty fox
well, if the mutex is currently held by yourself, then it ought not to be pulled out from under you
fiery turtle
yeah i dont understand
looks correct to me
other than missing relaxed atomic load
if owner is larger than the word size it might potentially be write in progress
but other than that i think its ok
its always one of two values: sentiel no owner placeholder, a valid thread id
well also depth overflow, i should check for that
rain nimbus
I mean it's not very atomic that's for sure...
fiery turtle
does that matter if owner is less than or equal to word size
but the proper way is to define atomic helpers 
does msvc even have them
InterlockedOrNoFence
lmao
fiery turtle
its only used for userspace tests so i dont care to be honest
but odd how?
ACPICA:
/* Support for multiple acquires by the owning thread */
if (ObjDesc->Mutex.ThreadId == ThreadId)
{
/*
* The mutex is already owned by this thread, just increment the
* acquisition depth
*/
ObjDesc->Mutex.AcquisitionDepth++;
return_ACPI_STATUS (AE_OK);
}
0 fucks given
overflow, atomicity
tbh if your thread_id is larger than word size just take the L
fiery turtle
nvm
rain nimbus
but odd how?
They're atomic operations yet they're not called atomic
They used the Interlocked word instead xD
fiery turtle
welll who cares its just a name\
jaunty fox
True, just an odd choice imo
the meaning is quite clear and the usage is traditional
rain nimbus
tbh if your thread_id is larger than word size just take the L
UUIDs for thread_id would be funny
fiery turtle
bruh
rain nimbus
the meaning is quite clear and the usage is traditional
It's just not a commonly used word for the concept of atomicity, that's all xD
except for with msvc I suppose
jaunty fox
i don't know of any kernel where there isn't at least part of the thread structure that's always resident and i can't immediately think of any where the thread structure is relocated, so the pointer to it is a reasonable thread id
rain nimbus
I don't think anyone has thread_id's larger than 32 bits...
Tbf even 16 bits is enough
until you're like windows and spawn a separate thread for every window regardless and you spawn like 100k windows...
fiery turtle
yeah i think a pointer or a 32-bit value is reasonable
jaunty fox
i will say i was actually a bit surprised
"atomic read" does return significantly more results than does "interlocked read" searching published articles indexed on google scholar prior to 1986
fiery turtle
interlocked is more of a legacy term I guess
jaunty fox
enjoy this use of it by the people who invented Mach then
rain nimbus
amd doesn't use the term in their amd64 docs
jaunty fox
NT does have spiritual lineage to Mach so it's not overly surprising
vast kestrel
```c
this_id = uacpi_kernel_get_thread_id();
if (mutex->owner == this_id) {
...
Shouldn't this be fully offloaded to the implementer?
Also because of single copy atomicity on any sane architecture it's actually completely fine (obviously as long as the size and alignment allow for it )
It will never have a case where you see your own thread id when the lock is taken by someone else because you had to reset it yourself and a thread is in sync with itself (alpha doesn't exist and can't hurt me)
Obviously assuming that the compiler will do it nicely is not the correct thing to do so using a relaxed atomic is the correct thing to do
But Microsoft be Microsoft, and Microsoft loves doing this kind of shit all over
kindred beacon
yeah offloading to the OS seems like a good choice
fiery turtle
Shouldn't this be fully offloaded to the implementer?
Its more convenient this way, because mutexes are sort of intertwined with AML and this way its easier for me to deal with recursive acquires
Not to mention there's e.g. a global lock that has special acquire semantics
But offloading would work with a few extra helpers on top I guess
vast kestrel
I still think the recursive part should be handled by the kernel, it makes sure that if anything strange needs to happen it does
And the global lock works very different to an os lock
Your wakeup is using an interrupt
And tbh that requires very specific semantics from the OS
fiery turtle
That uses a combination of events and mutexes
vast kestrel
Your primitives should be cond var + mutex
fiery turtle
Yeah thats basically what it is
I could even force the os to implement RCU
I just think less is better
loud ice
Also because of single copy atomicity on any sane architecture it's actually com...
no it's not, because the compiler is allowed to break that
vast kestrel
Where rcu doesn't make much sense
vast kestrel
no it's not, because the compiler is allowed to break that
Yeah that's why I said the nice thing would be to use relaxed atomic, but if you look at Microsoft's code they do that all the time
fiery turtle
Yup
If I do a relaxed load I need to force the os to use a specific type for thread id
Of force it to specify an atomic helper to use to load it
loud ice
you could just
fiery turtle
Its just all super annoying
fiery turtle
Which stack?
loud ice
the stack of mutexes you have acquired
fiery turtle
That works well for managed mutexes in aml
loud ice
the one you maintain already
fiery turtle
But not the mutexes that could be taken by me or the os
loud ice
you could just ban those uses from being recursive?
fiery turtle
Technically there's still one mutex that is shared by aml and user
So it must be recursive
Or maybe it could be only recursive in AML side idk
fiery turtle
you could just ban those uses from being recursive?
But thats a good idea
loud ice
Technically there's still one mutex that is shared by aml and user
which one?
fiery turtle
_GL
loud ice
oh global lock?
fiery turtle
Yeah
loud ice
yeah just make it exclusive on the kernel side only
fiery turtle
yeah just make it exclusive on the kernel side only
yeah i think this should work actually
i dont even need an owner pointer in this case
and a thread id helper
the only problem is i have to iterate the entire array so it could be more expensive
but aml usually just acquires one mutex
if any at all
loud ice
so yeah
its not that bad
you can do another thing tho
instead of a thread id
you can use the stack pointer at interpreter entry
as a "thread id"
loud ice
true
fiery turtle
but yeah i think this should be fine
and i can use an atomic load helper for it
since its a pointer
rustic compass
can i run uapci in userspace?
fiery turtle
can i run uapci in userspace?
cd tests && ./run_tests.py or compile runner manually with cmake
rustic compass
so i could run uacpi as a userspace driver in my microkernel?
fiery turtle
absolutely
your kernel must provide capabilities to map memory or io ports if you're a microkernel anyway
left orbit
is the irq passed to uacpi_kernel_install_interrupt_handler an ISA irq? or is it just a vector that uacpi wants it to be registered at?
left orbit
so i would route it through ioapic, right?
taking ISOs into account and all that
if there happens to be one for that irq of course
fiery turtle
yeah
left orbit
thanks!
fiery turtle
np!
left orbit
also one more question, what is the return value of the interrupt handler meant to be?
oh, whether it was handled or not
i guess i can just ignore it if i dont plan on sharing that vector then
fiery turtle
yeah its just handled/unhandled
left orbit
sorry for stupid questions but what is this sci interrupt used for?
is it just a way for firmware/acpi to callback into the operating system?
when like the power button is pressed or whatever
fiery turtle
its called the system control interrupt
and its a replacement of the SMM interrupt
its used by the firmware to make the acpi subsystem do something
left orbit
ah interesting
fiery turtle
sometimes its just something simple like power button press
sometimes more involved like a general purpose event to execute arbitrary AML
left orbit
qemu doesnt generate an sci interrupt for me when i press "power off" in the menu
unfortunate
fiery turtle
it does!
left orbit
then i guess i broke something :p
left orbit
i call these functions in this order:
uacpi_initialize
uacpi_install_notify_handler
uacpi_namespace_load
uacpi_namespace_initialize
uacpi_set_interrupt_model
i will, one sec
ah i need a fixed event handler
i was expecting an sci at the very least though
fiery turtle
yeah, the status bit for the power button must be set
otherwise a SCI is not generated
its a mask register
left orbit
ahhh that makes sense
let me do that real quick!
woooo
thanks :^)
hmm, repeat power button presses are not generating sci interrupts though
is this some flag i have to clear?
oh wait my kernel seems to have hung
fiery turtle
is this some flag i have to clear?
its cleared automatically
left orbit
yeah then i think something went horribly wrong because all of a sudden everything died lol
fiery turtle
qemu might spam it forever, but note that real hardware generally doesnt
there's a timeout before it sends the event again
it waits for the os to enter some sleep state for a bit
left orbit
wait, does uacpi_enter_sleep_state(UACPI_SLEEP_STATE_S5); perform the actual shutdown?
lol cool
so simple
left orbit
fair
fiery turtle
did you figure out why it hangs? 
left orbit
nope, but i dont really care since i'm not gonna be scheduling after that anymore 
fiery turtle
lol
fiery turtle
yeah but like it says there calling these from an interrupt context is really fragile
especially for real hw
it might call like Sleep() or try to grab a mutex
torpid ferry
MS even has annotations to prevent you from calling stuff at the wrong irql
fiery turtle
linux also with __might_sleep or whatever
torpid ferry
ms sal is really cool
left orbit
yeah yeah my stuff is shit though
torpid ferry
also has in/out params and length
fiery turtle
yeah
left orbit
i love the annotations on windows apis
but when i accidentally go to definition of some function
i tend to reconsider my life choices
i wonder who at microsoft willingly decided to sit down and annotate all the parameters
hollow elm
there's probably a policy that you have to annotate them from the start nowadays
left orbit
yeah, qookie and nooby helped me out a lot :^)
jaunty fox
MS even has annotations to prevent you from calling stuff at the wrong irql
i have entertained the thought of trying to do something like that, maybe with, what do you call it
libclang
fiery turtle
Okay I think i implemented global lock stuff, now i just need to hook it up to the interpreter and that's it
that will mark the end of missing ACPI-specific features
everything else is done
jaunty fox
everything else is done
remarkable work
what will you work on next? ultraOS?
https://github.com/UltraOS/UltraLegacy/tree/master/Kernel looks like it was already a reasonable amount of work done on it
fiery turtle
what will you work on next? ultraOS?
There's still this unfortunately https://github.com/UltraOS/uACPI/issues/74
fiery turtle
remarkable work
Thank you
fiery turtle
https://github.com/UltraOS/UltraLegacy/tree/master/Kernel looks like it was alre...
This one is archived, its in c++ and kind of what I used to learn the general idea, im not going to work on it anymore
fiery turtle
what will you work on next? ultraOS?
Either that or igpu 3d driver
jaunty fox
Either that or igpu 3d driver
you could link up with leo of LiL fame
fiery turtle
Lil is focused on modesetting but yeah
jaunty fox
Lil is focused on modesetting but yeah
as i understand it that's just the beginning
tired crater
Either that or igpu 3d driver
Cool project idea
kindred beacon
Lil is focused on modesetting but yeah
not necessarily, I'd like to get 3d accel working there too
right now, however, yeah
modesetting comes first 
sterile egret
seems like the kind of thing where collaboration would be the best way forward
Mostly I want to encourage this so that there's a good option for hobby os 3d accel, because I sure as hell can't write a modern GPU driver
fiery turtle
yeah I can definitely see value in that, but lil is C++ and just sort of different from how I would approach this
will see if i even get anywhere anyway
i might make a library similar to uacpi or i might make it coupled to my kernel
fiery turtle
@mortal yoke btw if u still work on your acpi interpreter, i now have --large for tests, which runs 354 real hardware blobs against the interpreter
mortal yoke
nice, maybe I should try running those
fiery turtle
yeah definitely
some of them will run forever for you unless you have while timeouts implemented
mortal yoke
do they use loadtable tho? because I still don't have that
fiery turtle
nah
mortal yoke
speaking of which I also lack ToString lol
fiery turtle
LoadTable is only used if you're _OSC'ing advanced capabilities
fiery turtle
speaking of which I also lack ToString lol
should be fairly rare
mortal yoke
yeah I haven't seen it being used in any of the real hw/virtual machines I have tried it on
fiery turtle
❯ grep -rn 'ToString' ./ --exclude \*.bin | wc -l
1169
well
looks like its somewhat common
but probably not during init
fiery turtle
okay looks like global lock stuff passes the tests
im not yet 100% confident about this pr and interface, but will see
ill definitely test this in managarm before merging
left orbit
uacpi works on my old ass dell optiplex together with the power button, very nice :^)
fiery turtle
Pog
hollow crescent
@fiery turtle does the U in uACPI stand for micro or Ultra?
fiery turtle
found some funkiness with pointer const-ness in uacpi apis, zig's translate-c tr...
i'm going to address this in the latest PR
as well as bitfields in the public API
since obviously it makes interop kind of annoying
left orbit
based
mortal yoke
yeah definitely
I tried it and this is the result 
fiery turtle
lol
mortal yoke
now I just need to go through the test outputs and look what exactly they are all about
fiery turtle
hey you got two of them passing 
mortal yoke
actually I realized that I didn't update the test runner's function that gets object type enum from an object variant, that seems to be why a lot of them failed
so Ill run it again and see what the actual number of failures is
fiery turtle
try it
mortal yoke
fiery turtle
not bad
mortal yoke
looks like 60 of them are either timeouts, bufferacc being unsupported or using connectfield which I apparently implemented wrong (and idk how should I actually implement it in a good way because the data is stuff that would need to be parsed using the main interpreter so I couldn't just parse the fields in a loop like I do now) and the other 2 were a mistake in the amount of args/locals as I had the counts the wrong way around
well I guess I could like have a new frame type for the field list and then have a separate opcode block table + handlers for that or smth, anyway I probably shouldn't be posting about this in here as this is for uACPI
fiery turtle
Lol
left orbit
oh man 
left orbit
i wonder what the heck is my firmware doing that it wants events for
mortal yoke
@fiery turtle btw there are a few usages of PRIu64 in source/utilities.c that should probably be UACPI_PRIu64
fiery turtle
<@353593948523986945> btw there are a few usages of PRIu64 in source/utilities.c...
ffs
fiery turtle
i wonder what the heck is my firmware doing that it wants events for
i doubt it will actually be used
just make create return something like 0x1
and then panic if its used
fiery turtle
<@353593948523986945> btw there are a few usages of PRIu64 in source/utilities.c...
thanks
i fixed that and nyaux warning in the PR
im not merging that today but anyway
left orbit
wtaf
fiery turtle
lol
left orbit
i wonder whats so special about this dsdt that makes uacpi fail
it doesnt seem to be on an error path though, maybe its just some zig stupidity
fiery turtle
i think its making your allocator fail
left orbit
do you know of any tool for windows that can dump the dsdt?
fiery turtle
acpidump -b
left orbit
on windows?
fiery turtle
yes
left orbit
fiery turtle
yup
left orbit
thanks
dusky glade
i think its making your allocator fail
u always love blaming everyones allocators lol
left orbit
it is my allocator tho
fiery turtle
lol because its almost always the case
fiery turtle
but ofc lets see what the test runner has to say
fiery turtle
how is it allocating an event in free() code lmao
left orbit
[uACPI][INFO] namespace initialization done: 295 devices, 0 thermal zones, 0 processors
[uACPI][TRACE] _STA calls: 105 (1 errors), _INI calls: 30 (0 errors)```yeah looks like it works
left orbit
how is it allocating an event in free() code lmao
it's not, it's just a log from earlier i believe
fiery turtle
the error is probably related to an undefined reference (because u didnt load SSDTs obviously)
but other than that looks like it works
fiery turtle
u cant, it only accepts one blob at a time
fiery turtle
nah
left orbit
time to look into the god damned allocator
fiery turtle
what was the _STA related error?
left orbit
[uACPI][ERROR] unable to lookup named object '\_SB_.PC00.LPCB.H_EC._STA.^^^GFX0.CLID': not found
[uACPI][ERROR] aborting method invocation due to previous error: AML referenced an undefined object
[uACPI][ERROR] #0 in \_SB_.PC00.LPCB.H_EC._STA()
[uACPI][WARN] aborted execution of '\_SB_.PC00.LPCB.H_EC._STA' due to an error: AML referenced an undefined object```anyway yeah do that
left orbit
assert(old_mem.len != 0);
thats the assert it fails on
apparently uacpi passes in a zero size
i guess ill just @min(size, 8) on both alloc and free
fiery turtle
hmm i dont think thats a code path that exists
maybe assert that size != 0 during allocation
// 2. Release the object array
uacpi_free(pkg->objects, sizeof(*pkg->objects) * pkg->count);
technically it could that the package is empty ofc
left orbit
how unlikely is that?
left orbit
okay so the @max(...) is already done by the allocator
const new_aligned_size = @max(len, @as(usize, 1) << @as(Allocator.Log2Align, @intCast(log2_ptr_align)));
fiery turtle
uacpi_bool uacpi_package_fill(uacpi_package *pkg, uacpi_size num_elements)
{
uacpi_size i;
pkg->objects = uacpi_kernel_calloc(num_elements, sizeof(uacpi_handle));
if its indeed an empty package you will see this
fiery turtle
send
left orbit
can i just send it here?
fiery turtle
yea
left orbit
is that a package of size 0?
fiery turtle
yup
left orbit
oh i didnt know the size was right there lmao
i was looking for one with empty {}
sooooo my aml is kinda weird
but thats okay
im just going to align size up to 8 :)
fiery turtle
yeah i mean, alloc(0), free(0), should work i think?
left orbit
well not in zig, here we usually dont allocate 0 bytes for anything lol
fiery turtle
im just going to align size up to 8 :)
yup
left orbit
yea true, i just never hit that so i didnt think of it
now with that accounted for its time for another bare metal run
fiery turtle
thats absolutely giant dsdt tho
left orbit
this time i enabled memory space access and bus mastering in my xhci driver
fiery turtle
its 120k lines of asl
left orbit
so hopefully? it gives some sign of life
left orbit
its 120k lines of asl
yeah i thought it was a bit of a chonker too
fiery turtle
ohh i see
left orbit
this machine i had some issues with because it has too much memory
fiery turtle
how modern is this hw?
left orbit
i was getting a page fault inside a page fault and ended up smashing the stack
z790 board
last intel gen
left orbit
if you want to look at them here u go
fiery turtle
thanks
fiery turtle
i could add it to the large tests that uacpi runs
left orbit
as long as that's legal, sure
i mean anyone can dump them i guess
but since they are binaries idk what rules apply to redistributing them
fiery turtle
I mean, if this is legal https://github.com/linuxhw/ACPI
left orbit
right, fair enough
fiery turtle
I have a fork of this with some fixes
left orbit
im glad i can contribute at least in that way
fiery turtle
can u do acpidump > res.txt and send that so its more of a standard format
fiery turtle
im glad i can contribute at least in that way
yeah thats definitely appreciated
that dump lacks latest hardware
i mean the repo
fiery turtle
lol
left orbit
oh right i forgot, the allocator interface accepts "log 2 of align" and i was passing the align itself, i fixed that for alloc but forgot to update for free
so instead of 8 i should be passing 3 (1 << 3 == 8)
ofc the god damn scheduler dies, ill tackle that another time, im glad i got uacpi to work
left orbit
yep, just dies when doing some xhci stuff, presumably because of the scheduler or whatever else
left orbit
that didnt work but im not sure if it wasnt working or because it was dead
i can try without spawning any thread to rule that out
fiery turtle
can u do `acpidump > res.txt` and send that so its more of a standard format
can u do this btw
all other dumps in that repo are that format so
left orbit
fiery turtle
Control Method Sleep Button (V1) : 1
Control Method Power Button (V1) : 0
yeah should be working as a fixed event
left orbit
yup works!
fiery turtle
Nice!
hollow elm
lol because its almost always the case
that was also the case with lai lol
people would complain about heap corruption or whatever
but in the end it was their allocator shitting itself
fiery turtle
always blame the tools
if anyone else wants to volunteer to be a part of uACPI test suite, im open to acpidump > res.txt submissions
currently have 355 blobs
fiery turtle
Thanks
north holly
wait
it has some MSDM
table
which apparently contains some windows licensing info
fiery turtle
I doubt that matters but u can remove it if u want
fiery turtle
which apparently contains some windows licensing info
Thats probably encoded to only work on your motherboard
I would hope
fiery turtle
Yeah not that simple lmao
fiery turtle
Yeah doubt
mortal yoke
at least you can usually transfer retail windows licenses between pc's (at least when you have signed in with an ms acc)
hollow elm
it's an oem key so it's probably not usable outside the acpi table?
left orbit
i was just kidding anyway, but people do make it seem that easy
fiery turtle
Huh
fiery turtle
people say this is enough to get the key: `strings /sys/firmware/acpi/tables/MSD...
Yeah but can you use it?
left orbit
no idea, should i try?
dusky glade
🐈
mortal yoke
at least you can usually transfer retail windows licenses between pc's (at least...
(because I am not going to pay 100€ to ms when I can just transfer it from an old prebuilt pc that cost less than that)
fiery turtle
no idea, should i try? <:tf:968899548258914394>
Lol
left orbit
(because I am not going to pay 100€ to ms when I can just transfer it from an ol...
||massgrave.dev||
fiery turtle
Anyway oberrow delete it I guess and reupload without that table
north holly
Don't forget to send us the output
mortal yoke
||~~massgrave.dev~~||
I mean yeah but I like to do it legally
north holly
for educational purposes
mortal yoke
or at least what I consider legal enough
north holly
.
left orbit
Yep u can have it
zenith crescent
just realised this has the MSDM table
north holly
we have the windows installation keys for a couple hundred computers
fiery turtle
Probably all 500 public dumps have it
fiery turtle
Its ascii
zenith crescent
there we go:
fiery turtle
Thanks! Were now at 356
fiery turtle
What
dusky glade
what are we doing here exactly
fiery turtle
Im expanding the test suite
north holly
now how do we tell ms to get their shit together and stop putting activation keys as plain text in acpi tables
dusky glade
now how do we tell ms to get their shit together and stop putting activation key...
bru
activation keys in plain text
in acpi tables
💀