i'm using ventoy to boot it
#uACPI - a portable and easy-to-integrate ACPI implementation
hallow marten
i cant really do anything if I dont have logs
fiery turtle
yeah no, even in memdisk it still is not booting
Enable earlycon in cmdline
Do u have a way to get serial?
fiery turtle
Idk if it has a visual earlycon
hallow marten
mhm i just tried on a laptop and I can reproduce
fiery turtle
Its kinda hard to die so early in boot tbh im surprised
hallow marten
lol ok i got it to print to earlycon but its slow as balls
fiery turtle
Which earlycon?
hallow marten
efifb
fiery turtle
Nice
hallow marten
it doesnt actually crash but for some reason busybox doesnt launch
I think it's a tty issue
it boots!
fiery ice
gg
hallow marten
I do get some ACPI errors
The firmware's AML is doing And(SomeLocalOrArg, ...) where that Local/Arg was never initialized. uACPI strictly refuses to treat an Uninitialized object as an Integer. This is a common firmware bug (vendors test against Windows, which tolerates it).
fiery ice
lol, ofc it's stupid firmare stuff
hallow marten
@fiery ice can you try now
fiery ice
i would have to unpack the iso
hallow marten
no
in grub
you select the entry, press e, then on the linux line you add earlycon=efifb keep_bootcon
hallow marten
hallow marten
where
fiery ice
hallow marten
weird
fiery turtle
Can you send an acpi dump as well so I could take a look later
fiery turtle
Why
fiery turtle
Well thats fine
hallow marten
what's the command again? acpidump
fiery turtle
`The firmware's AML is doing And(SomeLocalOrArg, ...) where that Local/Arg was n...
Tbh that usually happens with EC memes but ill take a look, maybe clanker forgot to initialize it early with uacpi
Its going to be obvious from the dump
hallow marten
ugh why are there no good simple file uploads lol
fiery turtle
Ill take a look when im home
torpid root
self-host it :^)
fiery turtle
Can u ask the clanker what this thing is about
hallow marten
yeah I think it f ound an issue
fiery turtle
What was it?
fiery turtle
Its the other way around
hallow marten
what
fiery turtle
Null is valid
fiery turtle
Ah nvm
fiery turtle
So why did linux fail to map it
hallow marten
idk what that file uploader is
hallow marten
So why did linux fail to map it
it didnt
fiery turtle
Lol sure
hallow marten
fiery ice
https://limewire.com/d/Om94J#1wWyCivY55
what is this
hallow marten
@fiery ice try this
hallow marten
According to this it did
wdym
linux failed to map it
but uacpi reported it as good
fiery turtle
Yes that's what I'm saying
hallow marten
i meant uacpi 
fiery ice
@hallow marten it booted
hallow marten
idk why linux failed to map it
hallow marten
<@379432087620550658> it booted
fuck yeah
fiery turtle
<@379432087620550658> it booted
Show dmesg
fiery ice
it's long AF
fiery turtle
idk why linux failed to map it
It has 256k max for early remap
fiery turtle
it's long AF
Well grep for acpi
fiery ice
hallow marten
idk if these erros are normal or not
fiery turtle
Anything before that?
fiery ice
yes
hallow marten
"failed to get unique processor ID" is sus
fiery ice
but i can't scroll back
fiery turtle
Yep
fiery turtle
but i can't scroll back
No less?
fiery ice
thx
fiery turtle
idk if these erros are normal or not
Probably not
fiery turtle
Irq 0 here is also sus
hallow marten
For Processor() declarations the kernel reads object.processor.proc_id, but my write_obj doesn't handle UACPI_OBJECT_PROCESSOR — it falls through to the default and leaves proc_id = 0 for every CPU, so they all collide → "Failed to get unique processor _UID (0x0)". Let me check uACPI's processor-info accessor and add the marshalling.
🤦♂️
fiery turtle
AGI
fiery ice
XDDDD
hallow marten
does poweroff -f work?
fiery turtle
I mean if there isn't a service listening to it, power button won't do anything
hallow marten
ok it fixed the unique processor thing
fiery ice
no
it's very low brightness
the second boot
before i even ran any cmd
poweroff -f works btw
fiery turtle
Crazy
fiery turtle
Yeah the power button press is probably not listened to by anyone
hallow marten
"can you go through the whole code and fix any similar gaps if any? and update the patch accordingly"
fiery turtle
Lol
Can you ask it about irq 0 as well
Thats definitely wrong
And it was on both computers
hallow marten
this kinda stuff is what LLMs are good at because it's boring work and it's purely iteration and it can work against a spec
its not implementing features
hallow marten
Can you ask it about irq 0 as well
fiery turtle
Hm ok
hallow marten
@fiery ice if u wanna try and check if there's no processor _UID thing
fiery ice
k
fiery turtle
Would need to compare against acpica to see if thats true
fiery turtle
Damn, it even handled dynamically loaded tables via aml perfectly fine
fiery turtle
Damn, it even handled dynamically loaded tables via aml perfectly fine
Thats also a code path no hobby os ever reached
full nimbus
uACPI upstreamed into linux would be huge for this shit
once things are a bit more stable il test it on my laptops
im a lil busy still rn
fiery turtle
Also it read the thermal zone temperature correctly
So yeah pretty great stuff
Ill investigate abbix warnings when im home
full nimbus
oh also one of my laptops i know has ACPI bugs logged from ACPICA
so may be intersting
torpid root
adding that to my todo list :^)
fiery ice
Do these scopeop warnings appear normally?
idk, never checked
fiery turtle
idk, never checked
Can you check later
fiery turtle
Great, same errors and irq numbers
fiery turtle
Btw @hallow marten can you ask why there isn't a single "new power resource" in the uACPI output
But other than that looks identical to me even error wise, which is impressive on hardware where you have 25k aml ops
hallow marten
Btw <@379432087620550658> can you ask why there isn't a single "new power resour...
it doesnt know
maybe the ISO I sent is old
fiery ice
maybe the ISO I sent is old
still nope
pallid lava
what are the chances this gets into mainline once it's done
fiery ice
maybe the ISO I sent is old
i found the exact line before it should call the new power resources
but i see nothing about them
hallow marten
what are the chances this gets into mainline once it's done
0.1
fiery ice
it just skips them
hallow marten
Maybe in 10 years
fiery ice
especially because of how acpica is shit now
pallid lava
i don't think the intel employee that gives 5 minutes of his time monthly to maintain acpica will be happy
lol
fiery turtle
it doesnt know
Also _AL0 evaluation failure
fiery ice
i don't think the intel employee that gives 5 minutes of his time monthly to mai...
well, considreing the last major regression XD
pallid lava
infy atp get hired in intel and get paid the big bucks only to maintain the acpi stuff
hallow marten
Also _AL0 evaluation failure
I got rate limited so gotta wait a bit
fiery turtle
Lol
fiery turtle
infy atp get hired in intel and get paid the big bucks only to maintain the acpi...
I dont think they care about the quality of the main repo as long as the linux fork works
Which doesn't accept slop changes instantly
full nimbus
@pallid lava what about becoming a Linux maintainer and getting money through that for the ACPI/AML subsystems
pallid lava
no one is paid by the linux foundation for being a maintainer unless you're one of torvalds' friends
you have to give up your free time or get paid by a corpo
for doing that work
hallow marten
it's either you work at a corpo that pays you like google, oracle, intel or redhat or you get paid by the foundation
but the foundation is only for core maintainers afaik
pallid lava
and those maintainers are people who linus has known for a very long time
hallow marten
yeah, well it makes sense
full nimbus
and those maintainers are people who linus has known for a very long time
Simple: just make a time machine
hallow marten
you wont start giving money to a random dude thast been working on the project for a month
pallid lava
just go to the 90s when linux was a hobby os
full nimbus
I was thinking more become linuses childhood friend
fiery turtle
it's either you work at a corpo that pays you like google, oracle, intel or redh...
yep
all the maintainers are mostly corpo backed
hallow marten
I was thinking more become linuses childhood friend
it's not about friends, it's about long-time maintainers
basically u start maintaining it now maybe in 10 years you'll get something
full nimbus
To be fair if you are already maintaining uacpi
The Linux shims may not be too much overhead to do
And keep up streamed etc
So just wait 10 years for money
hallow marten
- we coauthor the massive patch to linux 2. corpos notice it 3. we all get hired 4. profit
full nimbus
Mark me on there too
I helped
Well will help
By testing on a fuck ton of machines
pallid lava
i am helping with motivational support
full nimbus
After I eat lunch
full nimbus
i am helping with motivational support
Honestly though
This is like one of the best things you can do
And is the reason why my project is where it's at
Because I have had good friends here to help and talk about the project with me etc
hallow marten
i think if this does end up getting serious it'll probably end up as an experimental fork or something for a few years
/experimental set of patches
fiery turtle
Yep
fiery turtle
1. we coauthor the massive patch to linux 2. corpos notice it 3. we all get hire...
Good plan
hallow marten
also infy do you think the AML error might just be an acpica thing
pallid lava
also put it in hackernews
fiery turtle
also infy do you think the AML error might just be an acpica thing
Which one?
hallow marten
_AL0
full nimbus
If it did get merged it would be marked experanental
hallow marten
might just be ACPICA mis-interpreting it correctly
full nimbus
But it would be fucking huge
pallid lava
"we have replaced the CRUSTY intel acpica code from the 90s which is present on all computers"
hallow marten
and uACPI is actually correct
fiery turtle
might just be ACPICA mis-interpreting it correctly
But that error only happens on uACPI no?
full nimbus
and uACPI is actually correct
Another win for uACPI
hallow marten
whereas uacpi does the right thing and error
fiery turtle
I doubt it, especially since uacpi didnt log a single error there
hallow marten
well idk
fiery turtle
So the method definitely didnt get aborted
Like what does the code there actually check for
hallow marten
I think we'd have to have an aml dump maybe
fiery ice
I think we'd have to have an aml dump maybe
i sent one for this pc already if you want
hallow marten
it's in thermal code
struct acpi_handle_list devices = { 0 };
char method[] = "_PSL";
if (index != ACPI_THERMAL_TRIP_PASSIVE) {
method[1] = 'A';
method[2] = 'L';
method[3] = '0' + index;
}
if (!acpi_evaluate_reference(tz->device->handle, method, NULL, &devices)) {
acpi_handle_info(tz->device->handle, "%s evaluation failure\n", method);
return false;
}
imagine if uACPI had been written in rust
and risc-v native
instant 934859385938493 stars
fiery turtle
```
struct acpi_handle_list devices = { 0 };
char method[] = "_PSL";
...
Could be how it revolves that reference, maybe uacpi returns status not found there
Because that looks a bit SUS
full nimbus
Would we be posting this on hacker news?
hallow marten
perhaps
fiery ice
after we refine it some more
hallow marten
Could be how it revolves that reference, maybe uacpi returns status not found th...
id love debugging it but it's gonna take a lot of ping-pongs between me and diamond master 😭
hallow marten
i have other stuff to do rn but I'll clean things up and debug it later
fiery ice
i'll keep working on my scheduler rewrite in the meantime, lol
fiery turtle
Ok so for the _PLD here
Method (PR2S, 1, Serialized)
{
If (((CDID & 0xF000) == 0x8000))
{
Switch (Arg0)
{
Case (One)
{
Return (One)
}
Case (0x02)
{
Return (0x02)
}
Case (0x03)
{
Return (0x04)
}
Case (0x04)
{
Return (0x08)
}
Case (0x05)
{
Return (0x0100)
}
Case (0x06)
{
Return (0x0200)
}
Case (0x07)
{
Return (0x0400)
}
Case (0x08)
{
Return (0x0800)
}
Case (0x09)
{
Return (0x10)
}
Case (0x0A)
{
Return (0x20)
}
Case (0x0B)
{
Return (0x1000)
}
Case (0x0C)
{
Return (0x2000)
}
Case (0x0D)
{
Return (0x40)
}
Case (0x0E)
{
Return (0x80)
}
Case (0x0F)
{
Return (0x4000)
}
}
}
Else
{
Switch (Arg0)
{
Case (One)
{
Return (One)
}
Case (0x02)
{
Return (0x02)
}
Case (0x03)
{
Return (0x04)
}
Case (0x04)
{
Return (0x08)
}
Case (0x05)
{
Return (0x10)
}
Case (0x06)
{
Return (0x20)
}
Case (0x07)
{
Return (0x40)
}
Case (0x08)
{
Return (0x80)
}
Case (0x09)
{
Return (0x0100)
}
}
}
}
this falls through and returns nothing
Which means (CDID & 0xF000) == 0x8000) evaulates to false because Arg0 is always hardcoded by the caller
and CDID is a PCI_Config field
So either: ACPICA emulates return value 0 here (which I think it does in slack mode), or uACPI kernel api for pci config is wrong
ill check how NT behaves if the return value is omitted
iirc there were some quirks related to it returning the last value that was on the execution stack
fiery turtle
Okay so from the _DSM thing it looks like a uACPI problem
but its caused by a firmware bug
calm latch
and risc-v native
It is risc-v native
fiery turtle
Method (_DSM, 4, Serialized) // _DSM: Device-Specific Method
{
Switch (ToInteger (Arg0))
{
Case (ToUUID ("e5c937d0-3553-4d7a-9117-ea4d19c3434d") /* Device Labeling Interface */){ Switch (ToInteger (Arg2))
{
it does a switch on a UUID buffer but for whatever fucking reason converts it to an integer
So you end up comparing first 8 bytes of UUID == UUID
i could've sworn NT didnt support comparing integers and buffers
Ill confirm that and if it doesnt then we'll leave it at that
WAIT
its even worse
it's a 32-bit DSDT, so you're only comparing the first 4 bytes
ACPI Debug: Buffer [0x10]
0000: D0 37 C9 E5 53 35 7A 4D 91 17 EA 4D 19 C3 43 4D // .7..S5zM...M..CM
ACPI Debug: 0xE5C937D0
ACPI Debug: 0xFFFFFFFF
ACPICA thinks that Buffer of 10 bytes == 0xE5C937D0 => true
I doubt thats correct according to NT but ill test it
fiery ice
bruh
lethal trench
least insane aml behaviour
fiery turtle
now I have to remember how to work with windbg and amli! stuff 
Ah, these retards probably saw this warning:
x.dsl 96: Switch (Arg0)
Warning 3124 - ^ Switch expression is not a static Integer/Buffer/String data type, defaulting to Integer
And just blindly slapped a ToInteger
lethal trench
🥀
fiery turtle
alright lets see
hollow scaffold
fiery turtle
feeding nt custom AML and seeing what it thinks
fiery turtle
```
Method (_DSM, 4, Serialized) // _DSM: Device-Specific Method
{
Switch (...
If(LEqual(_T_0=0xe5c937d0,Buffer(0x10){
0xd0,0x37,0xc9,0xe5,0x53,0x35,0x7a,0x4d,0x91,0x17,0xea,0x4d,0x19,0xc3
0x43,0x4d})The target has requested that the debugger execute a command: !amli err 0xee 0000000000000000 0000000000000000 0000000000000000 0000000000000000;g
yeah i was right
ACPICA is bullshitting here
uacpi produces the correct result
0: kd> !amli err 0xee 0000000000000000 0000000000000000 0000000000000000 0000000000000000;g
an invalid parameter combination was passed to a function
As for this one, NT produces this return value:
ffff9a85cd0906bc: Store(Local0=<Obj=FFFF9A85CDF28238,Type=Unknown,Value=0x0,Buff=0000000000000000,Len=0>,Debug)Unknown()
=<Obj=FFFF9A85CD5FF090,Type=Unknown,Value=0x0,Buff=0000000000000000,Len=0>
Let's see if AndOp will crash on it as well (it probably will)
pallid lava
someone show this to that dude who wanted to write an aml interpreter 
fiery turtle
lol
I should maybe try to get the latest win11 image, but i doubt they patched much aml wise as to not break backwards compat
fiery turtle
Let's see if `AndOp` will crash on it as well (it probably will)
0: kd> !amli err 0xc6 0000000000000000 FFFFF8000532AFB8 0000000000000000 0000000000000000;g
ValidateArgTypes: expected Arg0 to be type Integer (Type=Unknown)
lmfao yeah and ACPICA in slack mode treats it as 0
so yeah that explains it
basically none of those errors are problems
as nt would react the same
lethal trench
huh
calm latch
how did this work in the first place?
lethal trench
I guess "it works on nt" is no longer the bar
the bar officially doesn't exist anymore
fiery turtle
how did this work in the first place?
those are misc methods
if they fail no one will notice really
and nt will still boot
i guess they dont look at logs
calm latch
but what's the point of them then?
fiery turtle
_PLD is physical location of device
aka it tells u where the usb port is physically on the computer
lethal trench
it works on nt has now become "it boots just fucking ship it please"
fiery turtle
_DSM is some device speciifc method that enables a feature
hallow marten
Is the _AL0 thing also something like that
pallid lava
al0 has to do with thermal zones so no
fiery turtle
Is the _AL0 thing also something like that
im almost sure it never got called
since uacpi didnt print an error
probably because that function produces a bogus path aka clanker probably fucked up in some way
fiery ice
ye
hallow marten
I know why _AL0 fails
fiery ice
<@573903151472836609> can you try the power stuff
it's not there
fiery turtle
I know why _AL0 fails
Why?
fiery turtle
ah yeah
hallow marten
@fiery turtle
fiery turtle
did u have clanker fix uacpi code?
hallow marten
yea
case UACPI_OBJECT_POWER_RESOURCE:
dst->power_resource = src->power_resource;
break;
this is the fix
fiery turtle
yeah thats the fix
fiery ice
<@573903151472836609> try this
yepp, it's working fine
hallow marten
yes
fiery turtle
crazy
hallow marten
well, both were fixed
fiery turtle
how did it fix al0?
hallow marten
it was another thing
fiery ice
hallow marten
reference.handle is null
this
fiery turtle
i guess the proper thing would be to review it and see if there are any places where it did weird shit thats better fixed on the uacpi side as a new type of api etc
for example the object attach instead of insane hashtable stuff that it did
i had the object attach thing as a TODO
fiery turtle
ok now i wont review it and send it directly to lkml
btw you could at least make a linux fork and add that patch on top and we could put it under the uacpi org at least
fiery turtle
maybe as a replacement for https://github.com/uACPI/uacpi-linux
since that just did uacpi side by side with acpica
fiery turtle
its basically just running uacpi code next to acpica, like namespace is initialized twice etc
it was more of a demo to see what sort of score linux would get
good find by the clanker also, such things can only be uncovered by more advanced use cases unfortunately
btw the real test that should be done is suspend/resume with linux uacpi
that exercises a crazy amount of api and is very fragile as well
fiery ice
well if you give me an iso for that i'll try
fiery turtle
is there a poweroff command that allows that?
full nimbus
I know my computer is very funky with sleep and resume.
I also just woke up from a nap so I can try and test things hopefully 😭
fiery turtle
meanwhile im installing window 2025 server to test if they changed any aml stuff
full nimbus
My condolences
calm latch
My laptop doesn't sleep with Linux
(I mean it has s2idle)
But like it wakes up immediately
fiery turtle
wtf
maybe it will fix itself with uacpi
ffffe30ee9706cd0: If(And(Local0=<Obj=FFFFE30EEA034C00,Type=Unknown,Value=0x0,Buff=0000000000000000,Len=0>,0x123,)ValidateArgTypes: expected Arg0 to be type Integer (Type=Unknown) yeah no change in that behavior between windows 10 and server 2025
ffffc50b840349d5: While(One)
ffffc50b840349d8: {
ffffc50b840349d8: Store(ToInteger(Arg0=Buffer(0x10){
0xd0,0x37,0xc9,0xe5,0x53,0x35,0x7a,0x4d,0x91,0x17,0xea,0x4d,0x19,0xc3
0x43,0x4d},)=0x4d7a3553e5c937d0,_T_0)=0x4d7a3553e5c937d0
ffffc50b840349e0: If(LEqual(_T_0=0xe5c937d0,Buffer(0x10){
0xd0,0x37,0xc9,0xe5,0x53,0x35,0x7a,0x4d,0x91,0x17,0xea,0x4d,0x19,0xc3
0x43,0x4d})an invalid parameter combination was passed to a function
same goes for the switch behavior
yeah looks like theres exactly 0 changes to the aml interpreter in 10 years
full nimbus
Common uACPI win
Common ACPICA L
fiery turtle
I did notice a small bug in uACPI so it wasnt wasted time ig
ffffc50b840349b8: Store(Concatenate("ARG0 as integer is ",ToInteger(Arg0=Buffer(0x10){
0xd0,0x37,0xc9,0xe5,0x53,0x35,0x7a,0x4d,0x91,0x17,0xea,0x4d,0x19,0xc3
0x43,0x4d},)=0x4d7a3553e5c937d0,)="ARG0 as integer is e5c937d0",Debug)String(:Str="ARG0 as integer is e5c937d0")
="ARG0 as integer is e5c937d0"
So as you can see here it says ToInteger(Arg0=Buffer(0x10){ 0xd0,0x37,0xc9,0xe5,0x53,0x35,0x7a,0x4d,0x91,0x17,0xea,0x4d,0x19,0xc3 0x43,0x4d},)=0x4d7a3553e5c937d0
Even though DSDT is 32-bits, it prints the result out as a 64-bit integer 0x4d7a3553e5c937d0
so uACPI copied that behavior also
But I see now that it prints out "ARG0 as integer is e5c937d0" so I guess internally it does truncate it somehow
Ill confirm that ToInteger(Arg0) == 0xe5c937d0 is indeed true and then fix that up then
full nimbus
I also meant to comment on this earlier
Shouldnt is say
ACPICA (Standard)
uACPI (Experimental)
It could also be named something better maybe? And what sub menu is this under?
Also if we got a new ISO hand it over
Il force myself to wakeup and do this shit
fiery ice
<@573903151472836609> try this
@full nimbus we have this rn
that should work correctly
full nimbus
Apparently systemd handles alot of the suspend shit?
Though there is a sysfs thing for it
fiery ice
there is no systemd in that iso afaik
full nimbus
Yeah
If there are patches for the current state
I could apply them and make my own kernel build
And install it on one of the laptops
To test suspend/resume
hallow marten
I also meant to comment on this earlier
Shouldnt is say
ACPICA (Standard)
uACP...
acpi
@full nimbus
fiery turtle
<@618537432002068491>
that last bit is already fixed on master
hallow marten
why not make a fork rn tho
because I'm gatekeeping
if i make a linux fork that means I have to do proper signed commits and stuff
ugh @fiery turtle why did you name the structs the same as acpica
fiery turtle
which ones and why is that a problem
full nimbus
<@618537432002068491>
just download ze kernel whatever the latest is source and apply these?
hallow marten
yea
fiery turtle
yeah so include the acpica header but dont include uacpi/acpi.h
hallow marten
some stuff includes it I think
full nimbus
like this right?
fiery turtle
some stuff includes it I think
dont think so, since i specifically tried to avoid that
hallow marten
like this right?
well, from git
hallow marten
dont think so, since i specifically tried to avoid that
event.h
full nimbus
i got the zip but yeah
hallow marten
nope it works
fiery turtle
yeah its a bug i think, ill get rid of that include
hallow marten
io.h wont work tho
fiery turtle
it will work just fine with a forward declaration
hallow marten
there's some cursed stuff in the code ngl 💀
full nimbus
alr im building it on my desktop
hallow marten
fiery turtle
are you going to boot into it on your desktop?
hallow marten
this is probably not that bad
fiery turtle
this is probably not that bad
why is pld stuff done on uacpi side?
full nimbus
Sometimes it takes 30 seconds for one monitor to come up
And somtimes on a fresh boot my mouse dosnt work
💀
fiery turtle
does acpica give helpers for that or smth
hallow marten
this is in acpi_decode_pld_buffer
full nimbus
And somtimes on a fresh boot my mouse dosnt work
Imagine if uACPI fixes this
It’s like not that annoying
But like every few weeks
I turn on my pc
And I have to reboot it
fiery turtle
this is in acpi_decode_pld_buffer
ah ok thats not that bad for sure, ill just add that helper to uacpi natively
fiery turtle
why is it so complex
resources are really complex
its what aml returns via _CRS etc
the resource code converts a compressed varaible length insane aml repr into usable C structs
I just hard to rebuild qemu on windows with an insane MSYS2 setup just to switch the dsdt revision to 2
but it helped me confirm my research
hallow marten
the code isnt actually that bad
the resource code is insane
but I think that's ACPI's fault
fiery turtle
yeah thats true
fiery turtle
nice
hallow marten
this should work 🤞
did it work
full nimbus
did it work
still building driver modules
🥀
i copied my main kernels config
i stg if nvme or ext4 is a moduke
fiery turtle
the header changes unfortunately break backwards compat i.e. may accidentally break compilation for some people so its time for uacpi 6.0 soon
i have a few other things befor then tho
including PLD helpers now ig
full nimbus
full nimbus
i alr make cleaned because im making a more light config that has nvme built in
hallow marten
where u have acpi_install_gpe_handler
fiery turtle
@hallow marten its probably because clanker didnt add EXPORT_SYMBOL_GPL to the implementations
so modpost doesnt see them
full nimbus
make a patch that includes uacpi sources maybe
hallow marten
<@379432087620550658> its probably because clanker didnt add EXPORT_SYMBOL_GPL t...
ahh
fuck
can you try it evalyn
or i can ask the clanka
fiery turtle
or i guess just EXPORT_SYMBOL since its not GPL
full nimbus
ask the clanker
hallow marten
or i guess just EXPORT_SYMBOL since its not GPL
it is
i have decided
it is now gpl
fiery turtle
gg
hallow marten
full nimbus
okay this config has nvme/ext4/efifb/amdgpu/uacpi as builtins and not modules
lets try this
Okay it's doing something
It logged a bunch of ACPI errors
And now it's booting
I think
Also it didn't use limine and its by default using the kernel from my main bootloader
So
💀
And it's stuck
@fiery turtle
fiery turtle
The pld stuff is harmless
Its stuck for some other reason
Im assuming because of the warnings below
full nimbus
this is where it died and i had to force shit it
fiery turtle
Huh, why wasn't that in the logs on the screen?
fiery turtle
acpidump > dunp.txt
full nimbus
fiery turtle
Why does it keep spamming that method
07 20:09:20 EvalynPC kernel: ACPI: requested invalid table index 28 (28 tables installed) also lol
Clanker tried to out of bounds it or smth?
Or maybe it doesn't know about the count getter
full nimbus
acpidump > dunp.txt
where do i get this package
fiery turtle
Thanks
full nimbus
this is my board
Or should I make a PR with it or whatever else later
For ze GitHub points
fiery turtle
this is where it died and i had to force shit it
Why is the thin you posted different from this log?
fiery turtle
For ze GitHub points <:PSP1G_pspTrollar:1434611394782691429>
Ofc
full nimbus
Why is the thin you posted different from this log?
I had two boots
I got the one that I let run the longest
leme get the otherrr
fiery turtle
Yeah but that one looked like it made it to Userspace just fine?
full nimbus
full nimbus
Yeah but that one looked like it made it to Userspace just fine?
Maybe something else was dying?
Unrelated to ACPI
I couldn't open any other virtual terminals
They just were blank
full nimbus
Yes
full nimbus
Lol
fiery turtle
Unrelated to ACPI
Yeah definitely
Its probably because you changed the config and it didnt like that somehow
full nimbus
@hallow marten can you make an ISO that has a way to put the thing to sleep?
fiery turtle
Does poweroff really not have that option
full nimbus
Let me check but I can't remember
nope
Options:
--help Show this help
--halt Halt the machine
-p --poweroff Switch off the machine
--reboot Reboot the machine
-f --force Force immediate halt/power-off/reboot
-w --wtmp-only Don't halt/power-off/reboot, just write wtmp record
-d --no-wtmp Don't write wtmp record
--no-wall Don't send wall message before halt/power-off/reboot
Unless halt is sleep?
hallow marten
what does halt do
full nimbus
Idfk
full nimbus
We need a full distro <:trl:829316641191886888>
or just somthing super minimal that glues on just enough ad-hoc support
fiery turtle
Ye
hollow elm
i think for sleep you can just write some value to some sysfs file
echo deep > /sys/power/mem_sleep; echo mem > /sys/power/state
full nimbus
Test it on the uACPI ISO
fiery ice
yes
fiery ice
Test it on the uACPI ISO
it went to sleep
full nimbus
Can it wake up?
full nimbus
That's probably not good
@fiery turtle
We may need a distro with systemd
That can have better logging for this shit
hallow marten
someone should rewrite acpica and call it acpicb
full nimbus
someone should rewrite acpica and call it acpicb
Doesn't ACPICA have a bunch of core architectural problems that make it harder to properly be NT compatible etc?
full nimbus
someone should still rewrite acpica and call it acpicb
Rewrite it in the B programming language too
lethal trench
acpica-rs when?
flat badge
Rewrite uacpi in rust 
fiery turtle
That's probably not good
Waking up is done by firmware, aka you should at least hear the fan spinning etc, if it doesn't get further its a kernel bug
If it doesn't wake up at all maybe there are no wake sources or smth
Basically it probably never even reached the kernel trampoline
Idk if clanker did GPE programming mapping correctly
Since thats how you set wake sources
full nimbus
Maybe it could have been that?
full nimbus
The firmware for wakeup though basically puts you in a similar state after a SIPI though right? And you basically do the same as an AP trampoline but maybe a little different for setup ig?
fiery turtle
Yep
rustic compass
@fiery turtle for the FACS can i just read the ptr in FADT or do i have to use the same table api as for other tables?
fiery turtle
same api, why do u need it tho?
rustic compass
does the facs doesnt contain anything of interest for the kernel?
fiery turtle
not really, and its managed by uacpi anyway
rustic compass
then i wont add it now, maybe later if someone wants to be able to dump the tables via uacpi-rs
rustic compass
i implement them in uacpi-rs as getter
fiery turtle
I see
rustic compass
so that the user doesnt have to implement a single table in their kernel
flat badge
/var/lib/managarm-buildenv/build/system-root/usr/share/uacpi/source/stdlib.c:606:25: error: 'signed char' to 'uacpi_u64' (aka 'unsigned long long') conversion; consider casting to 'unsigned char' first. [bugprone-signed-char-misuse,-warnings-as-errors]
606 | value = (signed char)uacpi_va_arg(vlist, int);
fiery turtle
yeah but what am i supposed to do with it
it's intended since it's a formatting function
flat badge
clang-tidy wants you to do (unsigned char)(signed char)...
fiery turtle
is that better?
flat badge
it doesn't sign extend to 0xffffffffffffffxx
fiery turtle
if (fm->is_signed) {
uacpi_i64 as_ll = value;
if (as_ll < 0) {
value = -as_ll;
negative = UACPI_TRUE;
}
}
It does this later
if I dont sign extend this logic will blow up
flat badge
I think in that case it wants you to cast to uint64_t directly (?)
I will silence this by ignoring uacpi source i guess but it's still something that you want to fix upstream
fiery turtle
how do I run this check?
it's funny that it got triggered by signed char but not signed short for example
fiery turtle
ill take a look, why not fix it if its cheap ig
fiery turtle
> /var/lib/managarm-buildenv/build/system-root/usr/share/uacpi/source/stdlib.c:6...
will be fixed in https://github.com/uACPI/uACPI/pull/155
i found a few important memory leaks/potential crashes on OOM as well
might roll 6.0.0 very soon
fiery ice
lol, the quickest release ever
fiery turtle
public header changes 😭
fiery turtle
Also updated the wiki a bit
if someone could write an example of how to use the barebones/early table stuff thatd be nice
steel inlet
might roll 6.0.0 very soon
5.0.1
fiery turtle
thats not how semver semvers
steel inlet
tbh i have no idea what new features there are
fiery turtle
i wish changelog existed that i posted and that is available on github releases
steel inlet
lmao
calm latch
if someone could write an example of how to use the barebones/early table stuff ...
How many people even use it? 
fiery turtle
idk but the page has 30k views
rustic compass
if someone could write an example of how to use the barebones/early table stuff ...
I could write one in rust
calm latch
I mean my kernel uses it...
pallid lava
i don't
calm latch
Or the barebones mode
pallid lava
i have allocators and scheduling before acpi init
i just have my own functions for iterating over the madt and retrieving other tables from the rsdt
pallid lava
then after scheduling is done i initialize acpi and then pci
fiery turtle
u just reimplemented the early tables mode yourself lol
pallid lava
yeah basically
calm latch
I set up uACPI barebones mode after allocation and before scheduling
Though I guess I can swap those
fiery turtle
i set it up after the boot allocator & early MMIO mapping helpers but thats pretty much it, nothing else is available at that point
calm latch
Since my kernel bootstraps itself by finding the largest memory region, and bump allocating from it, so uACPI should work
pallid lava
i set it up after the boot allocator & early MMIO mapping helpers but thats pret...
boot allocator boot allocator
calm latch
Though I guess I can swap those
Like when decide to support NUMA or whatever
pallid lava
fiery turtle
yeah if im going to support numa id have to redo the order a bit
early mmio would have to be first
calm latch
I have no early mmio mappings
pallid lava
imagine having early stuff
fiery turtle
how do u map acpi stuff?
pallid lava
lmao
calm latch
how do u map acpi stuff?
As normal
But like not through a special mechanism
I set up vmem before pfndb && palloc
fiery turtle
so u have an early mapping thing?
calm latch
so u have an early mapping thing?
It is a mapping thing
fiery turtle
yeah but how do u allocate the chunk of virtual memory where to map it
calm latch
I don't use HHDM
calm latch
yeah but how do u allocate the chunk of virtual memory where to map it
Though vmem
pallid lava
based tbh
fiery turtle
so like u initialize all these allocators before acpi right
calm latch
Yes
fiery turtle
i see
i dont do that so i can get a clock source early
so my logs have timestamps
- early hypervisor detection
which can skip tsc calibration which speeds up boot by 50ms
calm latch
I map the kernel, then temporary mapper, then switch to kernel page tables, then initialize vmem, then pfndb and page frame allocator after that, then uACPI
Then scheduling, interrupts, SMP, userspace
(in that order)
fiery turtle
I see
calm latch
Idr the order of the first two
fiery turtle
early mmio i just have a fixed bitmap of slots, for the late mmio it just reuses vmalloc for allocating virtual memory + applies arch caching attributes etc
calm latch
Like my kernel allocation functions just bump from the temporary region, then it gets reserved in my pma
fiery turtle
the pain for supporting multiboot
calm latch
So vmem's slab just allocates from that and I map stuff through the normal mechanism
calm latch
the pain for supporting multiboot
The pain is filtering the memory map
The rest of the code is the same
I guess mapping hhdm would be painful
But I don't do it
fiery turtle
i mean i dont have a single point in my kernel where i dont have access to every single ram page via the hhdm
that simplifies every piece of init code a lot
calm latch
I just have helpers for it
Which use bootloaders' hhdm/direct map which I set up with multiboot, then I switch to temporary mapper thing
On la64 I use the canonical gap hhdm
fiery turtle
did u check if linux uses it btw
calm latch
I think someone said it does
But idk if it only sets it up or actually uses it
Idk why limine doesn't use it though
fiery turtle
wow
static inline void __iomem *ioremap_prot(phys_addr_t offset, unsigned long size,
pgprot_t prot)
{
if (offset > TO_PHYS_MASK)
return NULL;
switch (pgprot_val(prot) & _CACHE_MASK) {
case _CACHE_CC:
return (void __iomem *)(unsigned long)(CACHE_BASE + offset);
case _CACHE_SUC:
return (void __iomem *)(unsigned long)(UNCACHE_BASE + offset);
case _CACHE_WUC:
return (void __iomem *)(unsigned long)(WRITECOMBINE_BASE + offset);
default:
return NULL;
}
}
#define ioremap(offset, size) \
ioremap_prot((offset), (size), PAGE_KERNEL_SUC)
This is crazy
calm latch
Probably because of kaslr?
fiery turtle
on other arches ioremap is really complex
pallid lava
holy epic
fiery turtle
on other arches its vmalloc + remap
calm latch
on other arches ioremap is really complex
Tbh it sounds like a Linux skill issue
calm latch
on other arches its vmalloc + remap
But this isn't complex
fiery turtle
fiery turtle
accessing everything thats not ram
it's also like tlb sync, traffic for invlidations etc
and LA64 is like, nah just grab an hhdm offset
crazy arch
fiery turtle
on x86 its ub to map the same physical page with different caching attributes
and it will blow up
so yes everything is stopping you
calm latch
on x86 its ub to map the same physical page with different caching attributes
What
Wdym
Wtf
fiery turtle
linux has a really complex interval tree on x86 thats used to make sure that never happens
e.g. if some ram is mapped as both WB but something wants to map it as WC
that will cause an MCE
calm latch
Bruh
I don't have checks for that for userspace
I mean my kernel is safe because microkernel
fiery turtle
well u should if u allow mmap to specify cache attributes
calm latch
I'll check it out later (when I get an actual MCE because of that )
fiery turtle
if it detects that RAM is remapped with a different caching attribute it splits the large HHDM pages and also remaps them with that
fiery turtle
and this https://elixir.bootlin.com/linux/v7.0.11/source/arch/x86/mm/pat/set_mem...
(thats this code)
fiery turtle
From doing what
pallid lava
yeah, i was going to ask something similar
calm latch
From mapping ram with other permissions
pallid lava
how common is it to map the same physical addresses with different caching attributes?
calm latch
Like if you're explicitly mapping something as WT or UC check that it's not ram
Since you would only do that for mmio or framebuffer anyway
Though WT and UC still needs to be checked for collisions
fiery turtle
how common is it to map the same physical addresses with different caching attri...
u can grep the code for set_memory_*
calm latch
But like in reality if somebody tries to map that it's a bug
fiery turtle
most hits are DRM
But also some control structures, e.g. virtio queues are usually allocated as uncached dma memory in linux (idk if it does that on x86)
probably only for non coherent PCI
calm latch
Also, this doesn't apply to different ASIDs/address spaces, right?
fiery turtle
so maybe thats fine
fiery turtle
Also, this doesn't apply to different ASIDs/address spaces, right?
cache works on physical pages not on virtual ones
so it doesnt matter which asid it came from
calm latch
Because then some process can trash other's mappings
fiery turtle
if it tries to remap ram as UC?
calm latch
Yeah
fiery turtle
well on linux u need to be admin for that
and with admin u can crash the system by a lot of means
its intended
calm latch
Like map /dev/memory or whatever as UC
fiery turtle
yeah sure
calm latch
Idk, this sounds like a big headache
That's solvable
You can't reject userspace
The issue is with framebuffers
Do GPU memory locations change on mode change?
fiery turtle
the framebuffer is usually mapped at an offset in a GPU BAR
so its not physical memory
calm latch
Or how does the firmare communicate the framebuffer's location?
pallid lava
in the gpu itself you can change the offset in the vram where the framebuffer is located
fiery turtle
firmware or software
calm latch
the framebuffer is usually mapped at an offset in a GPU BAR
So it's reserved in memory map as reserved
calm latch
Like where do limine/hyper get framebuffer memory region information from?
fiery turtle
well igpu stolen memory might be in the memory map but thats another unrelated story
fiery turtle
Like where do limine/hyper get framebuffer memory region information from?
its allocated by firmware
gentle peak
Like where do limine/hyper get framebuffer memory region information from?
vesa/gop
gentle peak
which is implemented by the GPU's vbios
pallid lava
aa long as it's not marked as usable
calm latch
Like what I don't get it how do I track the framebuffer
fiery turtle
on igpu u have a bar that is a view into the global igpu page tables, u map a stolen region of igpu memory in the gpu page tables and set a video mode
for example
calm latch
For UC/WT
fiery turtle
wdym by track
calm latch
Like if you connect a display, you get a second framebuffer
fiery turtle
linux tracks all mmio resources in a resource rb tree
it has one for memory and io ports
you dont magically get it, you allocate the framebuffer backing, set it on the gpu, and get a cursor into it
calm latch
But I'm a microkernel
fiery turtle
you can have device mappings requests go through a server
calm latch
How does Managarm deal with it?
fiery turtle
that validates the caching attributes etc
and then actually requests the kernel to map it
calm latch
I've been validating it in kernel
Yeah, I can do that through capabilities
Like right now anyone can map any memory, but I can just restrict it to my acpi/dt server and make it give out the capabilities to stuff
fiery turtle
to be clear there are MTRRs right, so the double remap is only a problem if you override the MTRR settings
calm latch
(I wanted to do it that way anyway)
fiery turtle
which happens with WC etc
calm latch
Yeah, the framebuffer
calm latch
or if u remap ram as UC
I can just reject that
fiery turtle
yeah
calm latch
Though I'd have to reject UC as well
fiery turtle
well u have to have that for non-x86 platforms but thats not a problem there i suppose
calm latch
Which you won't get anyway unless you're the ACPI server, which deduces the memory from memory map anyway
calm latch
well u have to have that for non-x86 platforms but thats not a problem there i s...
It's an arch-independent code
fiery turtle
on non x86 u have tons of cases for mapping ram as UC
calm latch
In my os
calm latch
I think I handle that already
fiery turtle
for non coherent pci basically
calm latch
I allow userspace to request DMA memory
Actually, I don't
I mean I can add it easily
fiery turtle
for aarch64 yeah
calm latch
I think I already map that as UC
Anyway
I'll come back to this in a few weeks when I have time to work on my os
calm latch
I'll just track it though the memory objects
Is this also UB on arm?
Like mapping the same memory with different caching stuff
fiery turtle
i dont think so because arm doesnt have any code for bookkeeping of that sort
calm latch
Cool
So like all of this in not a problem then
I can just have a MAP_DMA or whatever thing, which will map as normal ram on x86 and as UC on ARM
When I port to ARM (if I port to ARM )
fiery turtle
iirc linux has different allocators for this
there's basically dma_alloc_coherent which you would use to e.g. allocate an nvme queue, so the entire mapping would be UC (or WB for coherent PCI). The second one is dma_map_* thats for mapping arbitrary physical memory as DMA (that also transparently handles IOMMU)
fiery turtle
Is this also UB on arm?
ive just research it a bit more
on arm all it does is evict that physical memory from the data cache to the PoC
then remap it with vmalloc outside of hhdm
hhdm remains mapped as wb-cached
its assumed safe because a speculative access will just read the cache line but not dirty it
U can read b2.11 in ARMARM
it describes the specific side effects of mismatched attributes
whereas on x86 its just UB
fiery turtle
U can read `b2.11` in ARMARM
it might require a PhD to understand the specific implications tho
cc @ korona
full nimbus
im actually in the need to make a somewhat proper (albiet very minimal) distro for an i686 laptop to make it into a dummy terminal and have somthing that dosnt need a persistent disk. (and possibly for recovery drives to boot into and have a shell + utils)
if @hallow marten finishes up thoes patches a bit more il have that by default build with uACPI + some other crap we need to test it as an "offical" uACPI linux testing distribution
cc: @fiery turtle
full nimbus
Im not sure how good its acpi is tho if its i686
it will also have an x86-64 build
hallow marten
im actually in the need to make a somewhat proper (albiet very minimal) distro f...
What do I need to finish tho
fiery turtle
well ideally we debug the suspend thing
full nimbus
General cleanup of anything and if there are any bugs found when testing it
I'm not sure the complete state of it etc
And I don't wana push fully clanker patches for the kernel when I set this up
hallow marten
well ideally we debug the suspend thing
Does it not work?
fiery turtle
it doesnt resume so im assuming the way the clanker translated wake gpe programming is wrong
or other related stuff
but also we only tested on one laptop and idk if that even has s3 or if that uses s2idle or smth
gentle peak
where's the iso? I can test it on some of my systems, they're pretty straight forward acpi wise
hollow elm
what kernel version is the patch for? :^)
fiery turtle
wow nice
gentle peak
i'll try on my desktop as well but i need to recompile the kernel for that (need several modules incl. amdgpu)
kindred beacon
wait did I miss linux uACPI patches???
full nimbus
wait did I miss linux uACPI patches???
And hopefully a uACPI Linux distro
fiery turtle
wait did I miss linux uACPI patches???
abbix had clanker port linux to uacpi via a compat interface
kindred beacon
holy shit, uACPI everywhere
full nimbus
@hallow marten it would be nice if you could get your patches to run for kernel v6.18.35
gentle peak
i'll try on my desktop as well but i need to recompile the kernel for that (need...
btw what kernel config do i use? just defconfig?
fiery turtle
✅ Windows
✅ ReactOS
✅ Haiku
✅ Linux
✅ Hobby OSes
so only *BSDs remain?
MacOS
kindred beacon
apple silicon doesn't have ACPI tho?
fiery turtle
well legacy macos
steel inlet
what if i do a hackintosh
vocal geyser
uacpi rules world
fiery turtle
to me thats really cool and impressive to see, because these gpe and suspend related apis are really advanced/complex logic that no hobby os uses
and to see it work on real hw is cool
also the fact that clanker managed to port it correctly
hallow marten
<@379432087620550658> it would be nice if you could get your patches to run for ...
Probably would work anyway
fiery turtle
i dont think there have been many changes to any acpica apis between 6 and 7
so it should work
hallow marten
Yeah
pallid lava
wait did I miss linux uACPI patches???
no, those are ultra logs
fiery turtle
close enough
hollow elm
well legacy macos
they dropped it in the latest macos i thibnk :^)
er, legacy hw i mean
fiery ice
Does it not work?
no, it sleeps and never wakes up
gentle peak
i'll try on my desktop as well but i need to recompile the kernel for that (need...
ok on my desktop it's broken (can't tell where beyond it being after amdgpu enters standby and before the system enters sleep (the power led and fans are still on))
fiery turtle
do u have a com port on it?
fiery turtle
maybe serial over usb could work?
gentle peak
what does that entail
fiery turtle
iirc its like a usb cable that you insert, alternatively theres also pci serial cards but that costs some money ofc
basically theres probably no easy way to debug this
gentle peak
well this is after the drivers are all shut down so I doubt anything relying on the USB controller would work
fiery turtle
there has to be some sort of a command line/config option that linux uses to debug suspend issues right
like in general
like early con but the other way around
gentle peak
nope
actually i do have a COM header on the motherboard
but like I said no equipment to use it
fiery ice
and usb to serial should work there
fiery turtle
yeah youd have to buy a special cable for it
fiery ice
or even a cheap arduino / rpi
gentle peak
yeah youd have to buy a special cable for it
I might do that anyway, it seems like a good thing to have
fiery turtle
yeah that sounds pretty useful for debugging in general
sharp lichen
or even a cheap arduino / rpi
you probably need some level shifters too then
fiery ice
not really