OBOS (not vibecoded) | Operating System Development | Page 23

flint idol Oct 27, 2024, 1:44 PM

#

but now I use DPCs

real pecan Oct 27, 2024, 1:44 PM

#

E.g. Managarm and astral just shutdown by doing uacp_kernel_schedule_work

real pecan Oct 27, 2024, 1:44 PM

#

flint idol but now I use DPCs

Dpcs work too

#

Although dpcs aren't good for this

white mulch Oct 27, 2024, 1:44 PM

#

but if the registers don't get saved then it might cause issues when you want to resume that thread after the wake?

flint idol Oct 27, 2024, 1:45 PM

#

I can just suspend in a throwaway thread

#

actually no

#

I think when I wake up from sleep

real pecan Oct 27, 2024, 1:45 PM

#

white mulch but if the registers don't get saved then it might cause issues when you want to...

Ehh idk

flint idol Oct 27, 2024, 1:45 PM

#

flint idol I can just suspend in a throwaway thread

yeah I think I should just do this

real pecan Oct 27, 2024, 1:45 PM

#

I think Linux invokes the actual sleep from an asm stub to preserve the regs?

flint idol Oct 27, 2024, 1:45 PM

#

and I enforce that thread to run on the bsp

#

then when I wake up from sleep

#

I can just exit the thread

#

to continue from somewhere else

white mulch Oct 27, 2024, 1:46 PM

#

real pecan I think Linux invokes the actual sleep from an asm stub to preserve the regs?

ig that would work too

real pecan Oct 27, 2024, 1:47 PM

#

Idk I haven't looked at it in depth

white mulch Oct 27, 2024, 1:48 PM

#

a throwaway thread was just the first idea that I had so I went with that (and its nicer with all the c++ stuff that I have)

real pecan Oct 27, 2024, 1:48 PM

#

Do it on the thread that requested shutdown, preserve regs, do actual enter_sleep

#

Also

#

Dont forget to quiesce other CPUs

#

Its probably ub not to

flint idol Oct 27, 2024, 1:49 PM

#

how'd I do that

#

just stop them

#

by teling them to stop with an IPI

real pecan Oct 27, 2024, 1:50 PM

#

Just ipi and make it cli hlt in the handler

flint idol Oct 27, 2024, 1:50 PM

#

yeah ok

real pecan Oct 27, 2024, 1:50 PM

#

U should do it on panic anyway

#

And on shutdown

flint idol Oct 27, 2024, 1:50 PM

#

I do it on panic and shutdown

real pecan Oct 27, 2024, 1:50 PM

#

Oh

flint idol Oct 27, 2024, 1:50 PM

#

and reboot

white mulch Oct 27, 2024, 1:50 PM

#

what does doing it on shutdown or reboot gain tho

#

if the cpu is going to be reset anyway

flint idol Oct 27, 2024, 1:51 PM

#

void Sys_Reboot()
{
    OBOSS_HaltCPUs();
    uacpi_reboot();
    while(1)
        asm volatile("");
}
void Sys_Shutdown()
{
    OBOS_Log("oboskrnl: Shutdown requested.\n");
    OBOSS_HaltCPUs();
    // We're at IRQL_DISPATCH which should probably be enough for prepare for sleep state.
    uacpi_prepare_for_sleep_state(UACPI_SLEEP_STATE_S5);
    UACPI_ARCH_DISABLE_INTERRUPTS();
    uacpi_enter_sleep_state(UACPI_SLEEP_STATE_S5);
    while(1)
        asm volatile("");
}```

#

the forever loops are because it might take a bit for the shutdown/reboot to actually happen

#

past whatever uacpi's timeout is

#

(I have observed that on real hw)

real pecan Oct 27, 2024, 1:52 PM

#

white mulch what does doing it on shutdown or reboot gain tho

I think some hw might just hang if u don't

flint idol Oct 27, 2024, 1:52 PM

#

interesting....

white mulch Oct 27, 2024, 1:53 PM

#

ah

white mulch Oct 27, 2024, 1:53 PM

#

flint idol (I have observed that on real hw)

more than 10 seconds? lol

flint idol Oct 27, 2024, 1:53 PM

#

my computer sometimes hangs on shutdown

real pecan Oct 27, 2024, 1:53 PM

#

And how do u even shutdown if your other CPUs still do interrupts and talk to hw lol

#

It would explode

white mulch Oct 27, 2024, 1:53 PM

#

I just do that lol

flint idol Oct 27, 2024, 1:53 PM

#

flint idol my computer sometimes hangs on shutdown

I wonder if it could have anything to do with that

real pecan Oct 27, 2024, 1:54 PM

#

Yes

flint idol Oct 27, 2024, 1:54 PM

#

it would happen on windows and linux

#

idr if obos had that problem

#

~~if not then obos > windows and linux~~

real pecan Oct 27, 2024, 1:55 PM

#

Well they probably also shutdown individual devices and run other aml

flint idol Oct 27, 2024, 1:55 PM

#

yeah

#

but anyway

#

I have made a small change to the smp trampoline

#

so that the kernel tells it where to jump to after cpu initialization

real pecan Oct 27, 2024, 1:59 PM

#

Inb4 it thinks its a resume from suspend when first booting

#

Causing obos to shit itself

flint idol Oct 27, 2024, 1:59 PM

#

lol

#

nah it worked fine

#

diff --git a/src/oboskrnl/arch/x86_64/smp.asm b/src/oboskrnl/arch/x86_64/smp.asm
index 24cb3e4..0053f09 100644
--- a/src/oboskrnl/arch/x86_64/smp.asm
+++ b/src/oboskrnl/arch/x86_64/smp.asm
@@ -12,6 +12,10 @@ global Arch_SMPTrampolineCPULocalPtr:data hidden
 
 section .pageable.data
 
+extern Arch_APEntry
+global Arch_SMPTrampolineWakeLocation
+Arch_SMPTrampolineWakeLocation: dq Arch_APEntry
+
 Arch_SMPTrampolineStart:
 bits 16
 real_mode:
@@ -84,12 +88,9 @@ reload_cs_addr: equ $-Arch_SMPTrampolineStart
 ; Load RSP.
        mov rsp, [Arch_SMPTrampolineRSP-Arch_SMPTrampolineStart]
 
-extern Arch_APEntry
-
 ; Call AP initialization code.
        mov rdi, [Arch_SMPTrampolineCPULocalPtr-Arch_SMPTrampolineStart]
-       mov rax, Arch_APEntry
-       call rax
+       call [Arch_SMPTrampolineWakeLocation]
 Arch_SMPTrampolineEnd:
 section .text
 ; All of these CPUID bits are in CPUID.07H.0H:EBX```

#

that's all it was

#

obos_status OBOS_Suspend();

extern uint32_t OBOSS_WakeVector32;
extern uint64_t OBOSS_WakeVector64;

extern OBOS_WEAK obos_status OBOSS_PrepareWakeVector();```

#

nvm

real pecan Oct 27, 2024, 2:04 PM

#

Will you also supply a 64 bit vector

flint idol Oct 27, 2024, 2:04 PM

#

I changed that

#

obos_status OBOS_Suspend();

#define NO_WAKEVECTOR UINT32_MAX
extern uint32_t OBOSS_WakeVector;

extern OBOS_WEAK obos_status OBOSS_PrepareWakeVector();```

real pecan Oct 27, 2024, 2:04 PM

#

flint idol I changed that

Why

flint idol Oct 27, 2024, 2:04 PM

#

idk what a protecte mode entry is

#

 * 'addr64' is the protected mode entry-point address

real pecan Oct 27, 2024, 2:04 PM

#

Just have 0 be no entry

#

Why 0xfffff

flint idol Oct 27, 2024, 2:05 PM

#

real pecan Just have 0 be no entry

because my smp trampoline sits at 0x0

real pecan Oct 27, 2024, 2:05 PM

#

What the fuck

flint idol Oct 27, 2024, 2:05 PM

#

and I don't feel like changing that and seeing what happens

flint idol Oct 27, 2024, 2:05 PM

#

real pecan What the fuck

lmao

#

I have the barebones for suspend implemented

#

but no wake related things are called yet

real pecan Oct 27, 2024, 2:21 PM

#

is the wake code called

flint idol Oct 27, 2024, 2:21 PM

#

I do something slightly cursed

#

and by slightly I mean very cursed

real pecan Oct 27, 2024, 2:22 PM

#

i mean its obos so i trust u

flint idol Oct 27, 2024, 2:22 PM

#

bool OBOS_WokeFromSuspend;
static void suspend_impl()
{
    if (OBOS_WokeFromSuspend)
    {
        uacpi_prepare_for_wake_from_sleep_state(UACPI_SLEEP_STATE_S3);
        UACPI_ARCH_ENABLE_INTERRUPTS();
        uacpi_wake_from_sleep_state(UACPI_SLEEP_STATE_S3);
        OBOS_WokeFromSuspend = false;
        CoreH_ThreadReady(suspended_thread);
        Core_ExitCurrentThread();
    }
    uacpi_prepare_for_sleep_state(UACPI_SLEEP_STATE_S3);
    UACPI_ARCH_DISABLE_INTERRUPTS();
    // good night computer.
    uacpi_enter_sleep_state(UACPI_SLEEP_STATE_S3);
    OBOS_UNREACHABLE;
}```
suspend_impl is the entry of the suspend worker thread

#

it's entered at IRQL 2

#

(Dispatch)

#

so it cannot yield

#

so the idea is that

real pecan Oct 27, 2024, 2:22 PM

#

i mean this aint even cursed

flint idol Oct 27, 2024, 2:22 PM

#

flint idol so the idea is that

the arch's wake from suspend trampoline

#

will just switch to that thread

#

but the scheduler's structs still think that

#

it is at the entry

flint idol Oct 27, 2024, 2:23 PM

#

flint idol will just switch to that thread

with OBOS_WokeFromSuspend set to true

#

so it wakes from suspend

#

kinda cursed

real pecan Oct 27, 2024, 2:23 PM

#

nah not cursed

flint idol Oct 27, 2024, 2:24 PM

#

hmm

#

anyway I'm now gonna implement the arch side of suspend

#

wake.c under arch/x86_64

real pecan Oct 27, 2024, 2:24 PM

#

maybe its cursed in the sense that this is some other random thread

flint idol Oct 27, 2024, 2:25 PM

#

oh yeah I forgot to set any wake GPEs

#

I'll do that after the wake trampoline code

flint idol Oct 27, 2024, 2:55 PM

#

@real pecan the gpe-related functions in uacpi take in an index, can this be anything unique per gpe device, or do I need to find it some way

real pecan Oct 27, 2024, 2:55 PM

#

wdym

#

gpe is literally an index

flint idol Oct 27, 2024, 2:55 PM

#

UACPI_ALWAYS_ERROR_FOR_REDUCED_HARDWARE(
uacpi_status uacpi_gpe_enable_for_wake(
    uacpi_namespace_node *gpe_device, uacpi_u16 idx
))
UACPI_ALWAYS_ERROR_FOR_REDUCED_HARDWARE(
uacpi_status uacpi_gpe_disable_for_wake(
    uacpi_namespace_node *gpe_device, uacpi_u16 idx
))```

#

idx

#

what do I pass

real pecan Oct 27, 2024, 2:55 PM

#

oh just null

flint idol Oct 27, 2024, 2:56 PM

#

UACPI_ALWAYS_ERROR_FOR_REDUCED_HARDWARE(
uacpi_status uacpi_gpe_setup_for_wake(
    uacpi_namespace_node *gpe_device, uacpi_u16 idx,
    uacpi_namespace_node *wake_device
))```

flint idol Oct 27, 2024, 2:56 PM

#

flint idol ``` UACPI_ALWAYS_ERROR_FOR_REDUCED_HARDWARE( uacpi_status uacpi_gpe_setup_for_wa...

so would I first call this

real pecan Oct 27, 2024, 2:56 PM

#

either null or just the _GPE node

flint idol Oct 27, 2024, 2:56 PM

#

then uacpi_gpe_enable_for_wake

flint idol Oct 27, 2024, 2:56 PM

#

real pecan either null or just the _GPE node

if I pass null does it assume the latter?

#

oh wait I get it now

#

static void set_wake_vectors()
{
    uacpi_gpe_setup_for_wake(nullptr, 0, nullptr);
    uacpi_gpe_enable_for_wake(nullptr, UACPI_FIXED_EVENT_POWER_BUTTON);
}```

#

I think I would do that to make the power button wake me

real pecan Oct 27, 2024, 3:00 PM

#

flint idol if I pass null does it assume the latter?

yeah

flint idol Oct 27, 2024, 3:00 PM

#

flint idol I think I would do that to make the power button wake me

and is this valid?

real pecan Oct 27, 2024, 3:00 PM

#

no

flint idol Oct 27, 2024, 3:00 PM

#

static void set_wake_vectors()
{
    uacpi_gpe_setup_for_wake(nullptr, 0, nullptr);
    uacpi_gpe_enable_for_wake(nullptr, 0);
}```

#

what about this

real pecan Oct 27, 2024, 3:00 PM

#

wtf

flint idol Oct 27, 2024, 3:01 PM

#

well you said idx can be null

#

 * NOTE: 'gpe_device' may be null for GPEs managed by \_GPE

#

and you also said that

real pecan Oct 27, 2024, 3:01 PM

#

i never said anything about the index

flint idol Oct 27, 2024, 3:01 PM

#

real pecan Oct 27, 2024, 3:02 PM

#

i didnt mean the idx

#

because that would make no sense

flint idol Oct 27, 2024, 3:02 PM

#

what do I pass to idx

#

/home/oberrow/Code/obos/src/oboskrnl/power/suspend.c:57:(.text+0x10f): undefined reference to `uacpi_gpe_setup_for_wake'```

#

uhhhh

real pecan Oct 27, 2024, 3:03 PM

#

https://elixir.bootlin.com/linux/v6.11.5/source/drivers/acpi/scan.c#L1026

scan.c - drivers/acpi/scan.c - Linux source code v6....

Elixir Cross Referencer - source file of Linux (version v6.11.5). Browsed file: /drivers/acpi/scan.c

flint idol Oct 27, 2024, 3:03 PM

#

hmm what do we have here

real pecan Oct 27, 2024, 3:03 PM

#

ur supposed to figure out the gpe index of the device u want to enable for wake

#

then set it up for wake

#

thats it pretty much

flint idol Oct 27, 2024, 3:03 PM

#

ok

flint idol Oct 27, 2024, 3:04 PM

#

flint idol ``` /home/oberrow/Code/obos/src/oboskrnl/power/suspend.c:57:(.text+0x10f): undef...

also what's up with this

real pecan Oct 27, 2024, 3:04 PM

#

lmao

flint idol Oct 27, 2024, 3:04 PM

#

/home/oberrow/x86_64-elf-tools/bin/../lib/gcc/x86_64-elf/13.2.0/../../../../x86_64-elf/bin/ld: src/oboskrnl/CMakeFiles/oboskrnl.dir/power/suspend.c.obj: in function `set_wake_vectors':
/home/oberrow/Code/obos/src/oboskrnl/power/suspend.c:57:(.text+0x10f): undefined reference to `uacpi_gpe_setup_for_wake'
/home/oberrow/x86_64-elf-tools/bin/../lib/gcc/x86_64-elf/13.2.0/../../../../x86_64-elf/bin/ld: /home/oberrow/Code/obos/src/oboskrnl/power/suspend.c:58:(.text+0x118): undefined reference to `uacpi_gpe_enable_for_wake'

real pecan Oct 27, 2024, 3:04 PM

#

i think its

#

it should be uacpi_setup_gpe_for_wake

#

the header is bogus

flint idol Oct 27, 2024, 3:04 PM

#

how do you just

#

mess that up

#

lmfao

real pecan Oct 27, 2024, 3:04 PM

#

thats why i need a test kernel like i said lmao

#

this is untested api

flint idol Oct 27, 2024, 3:05 PM

#

yeah it's uacpi_setup_gpe_for_wake

#

along with gpe_enable_disable_for_wake

real pecan Oct 27, 2024, 3:05 PM

#

nope

#

u dont need to enable anything yourself

flint idol Oct 27, 2024, 3:05 PM

#

oh

flint idol Oct 27, 2024, 3:05 PM

#

flint idol along with `gpe_enable_disable_for_wake`

nvm not this

real pecan Oct 27, 2024, 3:05 PM

#

it will be enabled automatically when u prepare to sleep

#

u just tell it u want it enabled via setup_for_wake

flint idol Oct 27, 2024, 3:07 PM

#

yeah ok

#

// thanks infy
UACPI_ALWAYS_ERROR_FOR_REDUCED_HARDWARE(
    uacpi_status uacpi_setup_gpe_for_wake(
        uacpi_namespace_node *gpe_device, uacpi_u16 idx,
        uacpi_namespace_node *wake_device
    ))
static void set_wake_vectors()
{
    // uacpi_gpe_setup_for_wake
    uacpi_setup_gpe_for_wake(/* TODO */);
}```

real pecan Oct 27, 2024, 3:07 PM

#

i have a pr open with fixes anyways ill get that in

flint idol Oct 27, 2024, 3:22 PM

#

@real pecan how do I find if a namespace node has a method in uacpi

#

linux seems to have acpi_has_method

#

I want something similar

#

if (!acpi_has_method(device->handle, "_PRW"))
    return;```

real pecan Oct 27, 2024, 3:23 PM

#

literally

#

see how its implemented

#

just node find _PRW

flint idol Oct 27, 2024, 3:24 PM

#

uacpi_namespace_node_find

#

on PRW

#

yeah

real pecan Oct 27, 2024, 3:25 PM

#

yes

flint idol Oct 27, 2024, 3:25 PM

#

the object returned isn't refcounted or anything right

real pecan Oct 27, 2024, 3:25 PM

#

the node?

#

no

flint idol Oct 27, 2024, 3:26 PM

#

ok

#

reading linux source to find out the thing I need

#

instead of pinging infy for the 11th time today

#

my brain hurts

real pecan Oct 27, 2024, 3:31 PM

#

yes

#

its generally more useful to consult the source than me since ive never even implemented suspend

flint idol Oct 27, 2024, 3:32 PM

#

it seems to do something like

package = (union acpi_object *)buffer.pointer;```

#

buffer is an acpi_buffer

#

so I tried to replicate that in uacpi by evaluating the PRW object as linux did

real pecan Oct 27, 2024, 3:33 PM

#

in uacpi its just obj->package

#

acpi_buffer is what ACPICA uses to return anything from methods

#

its just an opaque data buffer

flint idol Oct 27, 2024, 3:35 PM

#

ok if I understood this right

#

if the first element is an integer

#

that's the gpe number

#

and the wakeup device is null

#

otherwise if it is a package

#

and that package's first element isn't a local reference

#

wait nvm

#

ignore that

#

if ((element->package.count < 2) ||
            (element->package.elements[0].type !=
             ACPI_TYPE_LOCAL_REFERENCE)
            || (element->package.elements[1].type != ACPI_TYPE_INTEGER))
            goto out;```

flint idol Oct 27, 2024, 3:37 PM

#

flint idol otherwise if it is a package

otherwise if the first element of this package is also a package

real pecan Oct 27, 2024, 3:37 PM

#

why cant u just read the acpi spec

flint idol Oct 27, 2024, 3:37 PM

#

real pecan why cant u just read the acpi spec

too much work\

real pecan Oct 27, 2024, 3:37 PM

#

https://uefi.org/specs/ACPI/6.5/07_Power_and_Performance_Mgmt.html#prw-power-resources-for-wake

flint idol Oct 27, 2024, 3:38 PM

#

oh wow that doesn't give me brain damage

#

from reading linux src

real pecan Oct 27, 2024, 3:38 PM

#

in uacpi ACPI_TYPE_LOCAL_REFERENCE would just be a string that u would uacpi_namespace_node_resolve_from_aml_namepath

flint idol Oct 27, 2024, 3:48 PM

#

if (pkg->objects[0]->type == UACPI_OBJECT_INTEGER)
{
    gpe_idx = pkg->objects[0]->integer;
    gpe_dev = nullptr;
} 
else if (pkg->objects[0]->type == UACPI_OBJECT_PACKAGE)
{
    uacpi_package* pkg2 = pkg->objects[0]->package;
    gpe_dev = ;
    gpe_idx = pkg2->objects[1]->integer;
}```
in the 2nd branch, I must set gpe_dev, which is a namespace node, to the pkg2->objects[0]

#

but idk how because that is a uacpi_object

real pecan Oct 27, 2024, 3:49 PM

#

what

flint idol Oct 27, 2024, 3:49 PM

#

well

#

real pecan Oct 27, 2024, 3:49 PM

#

u mean this case DeviceName // Reference?

flint idol Oct 27, 2024, 3:49 PM

#

yes

real pecan Oct 27, 2024, 3:49 PM

#

did u just ignore what i sent above

flint idol Oct 27, 2024, 3:50 PM

#

uacpi_namespace_node_resolve_from_aml_namepath

real pecan Oct 27, 2024, 3:50 PM

#

yes

#

package[0] would be a string object

#

that u resolve against that device

flint idol Oct 27, 2024, 3:50 PM

#

oh

#

thanks infy

real pecan Oct 27, 2024, 3:50 PM

#

np

flint idol Oct 27, 2024, 3:51 PM

#

and I assume a string behind the scenes is a uacpi_buffer

real pecan Oct 27, 2024, 3:51 PM

#

yes

#

and that has a .text

flint idol Oct 27, 2024, 3:52 PM

#

if (pkg->objects[0]->type == UACPI_OBJECT_INTEGER)
{
    gpe_idx = pkg->objects[0]->integer;
    gpe_dev = nullptr;
}
else if (pkg->objects[0]->type == UACPI_OBJECT_PACKAGE)
{
    uacpi_package* pkg2 = pkg->objects[0]->package;
    gpe_dev = uacpi_namespace_node_resolve_from_aml_namepath(node, pkg2->objects[0]->buffer->text);
    gpe_idx = pkg2->objects[1]->integer;
}
else
{
    OBOS_UNREACHABLE;
}```

#

before that I also make sure that device can wake me in S3

real pecan Oct 27, 2024, 3:52 PM

#

besically yes if u dont do any error checking at all

flint idol Oct 27, 2024, 3:52 PM

#

using the 2nd object of the first package

#

who needed error checking anyway

real pecan Oct 27, 2024, 3:53 PM

#

this is going to be the new interface btw #1217009725711847465 message

#

you wont be able to access shit like that anymore

flint idol Oct 27, 2024, 3:53 PM

#

well that is in the future

real pecan Oct 27, 2024, 3:53 PM

#

yes

flint idol Oct 27, 2024, 3:54 PM

#

rn I need something down

real pecan Oct 27, 2024, 3:54 PM

#

btw

#

the second branch is only relevant for non x86

#

or well

#

probably some very very niche x86

flint idol Oct 27, 2024, 3:55 PM

#

well I do eventually plan on aarch64

#

or riscv

#

oh cool it's suspended rn

real pecan Oct 27, 2024, 3:56 PM

#

did u suspend qemu

flint idol Oct 27, 2024, 3:56 PM

#

and it triple faults

flint idol Oct 27, 2024, 3:56 PM

#

real pecan did u suspend qemu

yes

flint idol Oct 27, 2024, 3:56 PM

#

flint idol and it triple faults

on wake

#

lmao

real pecan Oct 27, 2024, 3:56 PM

#

lol

#

did it have any wake devices

flint idol Oct 27, 2024, 3:56 PM

#

probably

real pecan Oct 27, 2024, 3:56 PM

#

how did u wake it up

flint idol Oct 27, 2024, 3:56 PM

#

sendkey a

real pecan Oct 27, 2024, 3:56 PM

#

interesting

flint idol Oct 27, 2024, 3:56 PM

#

in qemu

#

weird

#

I added a log to see if it finds any wake gpes

#

and I never get one

real pecan Oct 27, 2024, 3:57 PM

#

lol

flint idol Oct 27, 2024, 3:57 PM

#

yet sendkey triple faults it

real pecan Oct 27, 2024, 3:57 PM

#

how do u uh

#

even look for it

flint idol Oct 27, 2024, 3:58 PM

#

namespace iterate

#

then I make sure I see a device

#

otherwise I go on

white mulch Oct 27, 2024, 3:58 PM

#

does any of the wake code ever get executed?

flint idol Oct 27, 2024, 3:59 PM

#

good question

white mulch Oct 27, 2024, 3:59 PM

#

also did you put the wake vector at a non zero address? because the facs doesn't allow zero as the wake address because it means invalid

flint idol Oct 27, 2024, 3:59 PM

#

white mulch also did you put the wake vector at a non zero address? because the facs doesn't...

uhh no lol

#

my trampoline is at phys 0

white mulch Oct 27, 2024, 3:59 PM

#

well yeah that's probably why lol

real pecan Oct 27, 2024, 4:00 PM

#

i mean

#

what does -d int say

flint idol Oct 27, 2024, 4:00 PM

#

good question

#

nothing

#

*of interest

#

the last exception is normal

real pecan Oct 27, 2024, 4:03 PM

#

normal exception?

flint idol Oct 27, 2024, 4:04 PM

#

one that is expected

#

it was a page fault because the kernel needed to page something in

real pecan Oct 27, 2024, 4:04 PM

#

so uh

#

how do u even know it triple faults

flint idol Oct 27, 2024, 4:04 PM

#

it reboots

#

so I assumed triple fault

white mulch Oct 27, 2024, 4:05 PM

#

void VISIBLE32FLAT
handle_resume32(int status)
{
    ASSERT32FLAT();
    dprintf(1, "In 32bit resume\n");

    if (status == 0xfe)
        s3_resume();

    // Must be a soft reboot - invoke a hard reboot.
    tryReboot();
}
``` s3_resume returns early if the facs waking vector is zero and then it gets to tryReboot (this is on seabios)

real pecan Oct 27, 2024, 4:05 PM

#

oh cool

flint idol Oct 27, 2024, 4:05 PM

#

soo ummm

#

I need a non-zero vector

real pecan Oct 27, 2024, 4:05 PM

#

placing anything at 0 is a terrible idea ngl

flint idol Oct 27, 2024, 4:06 PM

#

now since it'd be too much work to make my smp trampoline be relocatable

#

I will just have some temp page

#

that jumps to address zero

#

hacky, but whatever

real pecan Oct 27, 2024, 4:07 PM

#

bruh

flint idol Oct 27, 2024, 4:07 PM

#

the smp trampoline assumes addresses off zero in many places

white mulch Oct 27, 2024, 4:08 PM

#

its not that hard to make a relocatable trampoline, you can use cs relative addressing and put the boot info at the end of the trampoline or whatever

flint idol Oct 27, 2024, 4:09 PM

#

doesn't change the fact that

#

I don't wanna

#

ok now it triple faults

#

which is better

real pecan Oct 27, 2024, 4:17 PM

#

lmao

flint idol Oct 27, 2024, 4:18 PM

#

it's a GPF

#

nvm

#

it's a bad instruction

flint idol Oct 27, 2024, 4:54 PM

#

werid stuff

#

I think the bios uses phys. zero as a scratch space

#

because when I am actually in the trampoline it's corrupted

#

but before I go to sleep it's not

real pecan Oct 27, 2024, 4:54 PM

#

phys zero is bios ivt

flint idol Oct 27, 2024, 4:55 PM

#

oh yeahhhhhhh

#

since I still don't feel like making my smp trampoline relocatable

#

what I'll do now is

#

keep a copy of the trampoline after the reset vector

#

and copy it to phys. 0

real pecan Oct 27, 2024, 4:56 PM

#

this guy will do anything but move his goddamn trampoline

white mulch Oct 27, 2024, 4:56 PM

#

at this point you would have already made it relocatable lol

flint idol Oct 27, 2024, 4:59 PM

#

Arch_ACPIWakeTrampoline:
bits 16
    call $
    pop ax
    mov es, ax
    mov si, es:.data-Arch_ACPIWakeTrampoline
    mov di, 0
    mov cx, Arch_SMPTrampolineEnd-Arch_SMPTrampolineStart
    rep movsb
    jmp 0x00:0x0000
Arch_ACPIWakeTrampoline_data:
times Arch_SMPTrampolineEnd-Arch_SMPTrampolineStart db 0```

#

bam

white mulch Oct 27, 2024, 5:00 PM

#

cursed

flint idol Oct 27, 2024, 5:00 PM

#

then I just copy the smp trampoline into Arch_ACPIWakeTrampoline_data

#

and I'm good

flint idol Oct 27, 2024, 5:00 PM

#

white mulch cursed

Yes.

torpid wigeon Oct 27, 2024, 5:01 PM

#

flint idol ```x86asm Arch_ACPIWakeTrampoline: bits 16 call $ pop ax mov es, ax ...

🤢

flint idol Oct 27, 2024, 5:01 PM

#

this code is cursed

#

but I will not look past at it past today

#

ok now it crashes in a different way

#

instead of it triple faulting because of an invalid opcode

#

nvm

#

bro wtf is up with the cpu

#

how is it at

#

long mode

#

where is the gdt????

#

GDT=     0000000000000000 00000000```

real pecan Oct 27, 2024, 5:14 PM

#

did u expect registers to survive

flint idol Oct 27, 2024, 5:15 PM

#

no

#

which is why I load a gdt

#

the first thing in the smp trampoline

thick jolt Oct 27, 2024, 5:15 PM

#

hi infy hi oberrow

#

my favourite peopple

flint idol Oct 27, 2024, 5:15 PM

#

hi nyauxmaster

thick jolt Oct 27, 2024, 5:15 PM

#

hi

flint idol Oct 27, 2024, 5:15 PM

#

is your day good

thick jolt Oct 27, 2024, 5:16 PM

#

yes

#

😎

flint idol Oct 27, 2024, 5:16 PM

#

good to hear

#

nyaux is returning to greatness it seems

thick jolt Oct 27, 2024, 5:16 PM

#

yes indeed

#

im eta'ing scheduling in maybe 3-5 days depending on what memes ill encounter :)

flint idol Oct 27, 2024, 5:17 PM

#

W

thick jolt Oct 27, 2024, 5:17 PM

#

indeed

flint idol Oct 27, 2024, 5:17 PM

#

I'm tryna setup bochs

thick jolt Oct 27, 2024, 5:17 PM

#

ooo

#

does obos run in bochs

flint idol Oct 27, 2024, 5:17 PM

#

I'm tryna get it running

#

so idk

real pecan Oct 27, 2024, 5:17 PM

#

yeah just boot it in bochs

#

bochs will tell u what u did wrong

thick jolt Oct 27, 2024, 5:18 PM

#

exactly

flint idol Oct 27, 2024, 5:22 PM

#

bruh how do I use bochs

#

Message: ROM: couldn't open ROM image file '/usr/share/bochs/BIOS-bochs-latest'.

real pecan Oct 27, 2024, 5:23 PM

#

u need a config

flint idol Oct 27, 2024, 5:23 PM

#

yuh huh

thick jolt Oct 27, 2024, 5:24 PM

#

good luck on this oberrow!

#

ima take a nap

flint idol Oct 27, 2024, 5:24 PM

#

thanks

thick jolt Oct 27, 2024, 5:24 PM

#

np

flint idol Oct 27, 2024, 5:24 PM

#

gn

thick jolt Oct 27, 2024, 5:24 PM

#

gn

real pecan Oct 27, 2024, 5:24 PM

#

nyaux nap%

flint idol Oct 27, 2024, 5:27 PM

#

wtf

#

why don't I have any BIOS rom

#

but I do have a vga bios

#

it is in the bochsbios package

#

life without apt-file

#

would be bad

short mortar Oct 27, 2024, 6:02 PM

#

Bochs build in Debian'/Ubuntu's repos is ... weird

real pecan Oct 27, 2024, 6:05 PM

#

U mean GNU/Linux/Debian/Ubuntu?

short mortar Oct 27, 2024, 6:14 PM

#

Lol yes

#

Is there bochs for Debian GNU/Hurd? thonk

flint idol Oct 27, 2024, 6:29 PM

#

idk what I did wrong

#

but I don't even get to my kernel in bochs

real pecan Oct 27, 2024, 6:35 PM

#

bochs debugger time

short mortar Oct 27, 2024, 6:37 PM

#

I just compiled mine from source on Ubuntu

real pecan Oct 27, 2024, 6:39 PM

#

as you should tbh

short mortar Oct 27, 2024, 7:14 PM

#

Fedora's repos were really nice

flint idol Oct 27, 2024, 7:54 PM

#

flint idol Oct 27, 2024, 8:14 PM

#

uacpi_namespace_node* prw = uacpi_namespace_node_find(node, "_PRW");
if (prw)
    goto end;

#

lmao\

#

I was wondering where all the devices

#

that could wake the system were

#

why does qemu have no devices with PRW

#

and why can't bochs get past "Booting from hard disk"

#

without triple faulting

flint idol Oct 27, 2024, 8:24 PM

#

flint idol instead of it triple faulting because of an invalid opcode

on kvm it just dies

#

KVM internal error. Suberror: 1```

white mulch Oct 27, 2024, 8:31 PM

#

flint idol why does qemu have no devices with PRW

yeah looks like it unconditionally causes a wakeup on ps2 keyboard and mouse press

real pecan Oct 27, 2024, 8:33 PM

#

obos try not to triple fault challenge

flint idol Oct 27, 2024, 8:34 PM

#

ok it stopped triple faulting

#

I think

#

now it's doing something different

#

unless

#

I think it's just executing zeroes

#

or the ivt

flint idol Oct 27, 2024, 9:06 PM

#

@real pecan I think I accidentally found a workaround to not being able to set the FACS reset vector to zero

#

if you set it to 0x800

#

it seems to actually throw me at address 0x0

flint idol Oct 27, 2024, 9:06 PM

#

flint idol if you set it to 0x800

or really any non-page aligned address

#

wait wtf

real pecan Oct 27, 2024, 9:20 PM

#

implementation details of a specific bios

flint idol Oct 27, 2024, 9:20 PM

#

it's always going to 0x0

flint idol Oct 27, 2024, 9:20 PM

#

real pecan implementation details of a specific bios

well yeah

#

I wasn't trying to do that

#

I found out by accident

real pecan Oct 27, 2024, 9:21 PM

#

flint idol it's always going to 0x0

wtf

flint idol Oct 27, 2024, 9:21 PM

#

at least if gdb isn't lying to me

#

I have two bps

#

one at 0x0

#

and one at the reset vector

#

the latter never gets hit

real pecan Oct 27, 2024, 9:21 PM

#

i mean print the value in facs

flint idol Oct 27, 2024, 9:22 PM

#

yeah

#

that's what I'm doing

real pecan Oct 27, 2024, 9:23 PM

#

p g_uacpi_rt_ctx.facs->whatever

flint idol Oct 27, 2024, 9:23 PM

#

I printed it in uacpi_set_waking_vector

#

and it's what I expected

real pecan Oct 27, 2024, 9:23 PM

#

print in gdb as well

flint idol Oct 27, 2024, 9:23 PM

#

yeah I am

#

it's 0x1000

real pecan Oct 27, 2024, 9:24 PM

#

u could of course

#

compile the bios

#

and add printfs there

flint idol Oct 27, 2024, 9:32 PM

#

Bah

flint idol Oct 27, 2024, 10:45 PM

#

real pecan compile the bios

my compiled seabios no works :(

#

weird

#

it doesn't like hyper

#

@real pecan helppppppppp

#

seabios no working

real pecan Oct 27, 2024, 10:52 PM

#

idk figure it out

flint idol Oct 27, 2024, 10:53 PM

#

nooo

#

now it's not even building

#

on the 1.12-stable branch

#

finally

#

it booted

#

In resume (status=254)
In 32bit resume
table(50434146)=0x3ffe21f3 (via rsdt)
Found 4 cpu(s) max supported 4 cpu(s)
PCIe: using q35 mmconfig at 0xb0000000
Running option rom at c000:0003
Start SeaVGABIOS (version 1.16.3-debian-1.16.3-2)
VGABUILD: gcc: (Debian 13.2.0-7) 13.2.0 binutils: (GNU Binutils for Debian) 2.41
enter vga_post:
   a=00000000  b=0000ffff  c=00000000  d=0000ffff ds=0000 es=f000 ss=0000
  si=00000000 di=000060c0 bp=00000000 sp=00000f36 cs=f000 ip=d004  f=0000
Jump to resume vector (1000)

real pecan Oct 27, 2024, 11:16 PM

#

There u go

flint idol Oct 27, 2024, 11:16 PM

#

well 0x1000 is the vector I wanted....

real pecan Oct 27, 2024, 11:16 PM

#

Does it not actually jump there

flint idol Oct 27, 2024, 11:17 PM

#

let me add an artifical breakpoint

#

it does

real pecan Oct 27, 2024, 11:18 PM

#

Sooo how are u not seeing it via grub

flint idol Oct 27, 2024, 11:18 PM

#

grub?

real pecan Oct 27, 2024, 11:18 PM

#

Lmao

#

Sorry

#

Gdb

flint idol Oct 27, 2024, 11:18 PM

#

idk, it's quite weird

real pecan Oct 27, 2024, 11:18 PM

#

Probably some skill issue

flint idol Oct 27, 2024, 11:18 PM

#

bruh

#

I put b *0x1000

real pecan Oct 27, 2024, 11:20 PM

#

Maybe its waiting on virtual 0x1000

flint idol Oct 27, 2024, 11:20 PM

#

perhaps

#

yay it triple faults now

#

or maybe not yay

#

probaby yay!

real pecan Oct 27, 2024, 11:20 PM

#

Wdym yay

flint idol Oct 27, 2024, 11:21 PM

#

I now know I'm not jumping to arbitrary code

real pecan Oct 27, 2024, 11:22 PM

#

Big

flint idol Oct 27, 2024, 11:22 PM

#

now I just need to find out the reason for the triple fault

#

which I will do after a tiny bit

#

it's a gpe

#

bruh

#

 53176: v=0d e=0012 i=0 cpl=0 IP=0008:00000000000fd089 pc=00000000000fd089 SP=0010:0000000000000000 env->regs[R_EAX]=0000000000000000

#

that ain't my code

torpid wigeon Oct 27, 2024, 11:25 PM

#

i stole it

flint idol Oct 27, 2024, 11:25 PM

#

yeah it's because of my smp trampoline isn't it

#

indirectly

#

time to give in and write a relocatable smp trampoline

real pecan Oct 27, 2024, 11:52 PM

#

flint idol indirectly

Wdym

flint idol Oct 27, 2024, 11:54 PM

#

real pecan Wdym

basically

#

I was triple faulting because I was getting an irq in an AP when it was it was still in bios code

#

and the IVT was overwritten

real pecan Oct 27, 2024, 11:55 PM

#

Ah

#

Oberrow finds out that fucking with bios structures is not good

weak kestrel Oct 27, 2024, 11:57 PM

#

https://tenor.com/view/fafo-findout-fuckaround-fuckaroundandfindout-you-gif-26826252

Tenor

flint idol Oct 27, 2024, 11:58 PM

#

lol

flint idol Oct 28, 2024, 1:14 AM

#

flint idol time to give in and write a relocatable smp trampoline

actually

#

I'll just assume it's at 0x1000 instead

#

@real pecan

#

IT WORKS

#

defgdsfudrhesfw\

#

I woke up from S3

#

static void on_wake()
{
    for (volatile bool b = true; b; );
}```

#

I was able to get the state to the bare minimum to get something to print

#

gdb is being stoobid

flint idol Oct 28, 2024, 2:08 AM

#

idk why but my tss is broken

#

fixed

#

idk why but I must reinitialize the tss' segment on wake

flint idol Oct 28, 2024, 2:40 AM

#

I do now properly restart the cpu

#

but it's still not in a preferable state to yield to another thread

#

I am now gonna wake the other CPUs

#

bruh

#

nooo

#

I triple fault

#

check_exception old: 0xffffffff new 0xd
 56719: v=0d e=0012 i=0 cpl=0 IP=0008:00000000000fd089 pc=00000000000fd089 SP=0010:0000000000000000 env->regs[R_EAX]=0000000000000000```

#

fixed

#

I now restart all CPUs

#

what's next is restarting the LAPIC timer

#

then I can finally switch back to whatever that thread was doing behind

#

check_exception old: 0xffffffff new 0xd
 44433: v=0d e=0012 i=0 cpl=0 IP=0008:00000000000fd089 pc=00000000000fd089 SP=0010:0000000000000000 env->regs[R_EAX]=0000000000000000```

#

sigh

#

I tried restarting the lapic timer

#

for this to happen again

flint idol Oct 28, 2024, 3:22 AM

#

after syncing all CPUs to initialize their timers at the same time

#

it no longer triple faults

#

but now it crashes (best word to describe what just happened) when trying to switch to a new thread

#

@real pecan your wake from sleep state might be broken

#

I'll see once I actually fix this code

real pecan Oct 28, 2024, 6:29 AM

#

flint idol <@353593948523986945> your wake from sleep state might be broken

x to doubt

real pecan Oct 28, 2024, 6:30 AM

#

flint idol IT WORKS

LETSFUCKINGGOOOOOO

flint idol Oct 28, 2024, 10:48 AM

#

real pecan x to doubt

I will debug it after school

real pecan Oct 28, 2024, 10:54 AM

#

debug it at school

flint idol Oct 28, 2024, 11:11 AM

#

real pecan debug it at school

Not even a terrible idea

#

I just can't get caught

real pecan Oct 28, 2024, 11:11 AM

#

suspend school pc

flint idol Oct 28, 2024, 11:11 AM

#

Also a good idea

real pecan Oct 28, 2024, 11:12 AM

#

lol

flint idol Oct 28, 2024, 11:15 AM

#

So what I will do is bring debian livecd

#

Or just a debian installer

#

So that I can dd the iso

#

And I'll just debug at school

#

I need to push my code first thiugh

#

*though

real pecan Oct 28, 2024, 11:16 AM

#

win32 imager or fuse

flint idol Oct 28, 2024, 11:16 AM

#

Those need admin access, no?

real pecan Oct 28, 2024, 11:16 AM

#

not sure

flint idol Oct 28, 2024, 12:20 PM

#

Forgot to commit n'stuff

flint idol Oct 28, 2024, 8:01 PM

#

time to debug

#

what I know now is that:

it hangs after all CPUs get a halt NMI

real pecan Oct 28, 2024, 8:03 PM

#

halt all cpus
everything hangs

flint idol Oct 28, 2024, 8:04 PM

#

no but it's unintentional

#

idk why it's happening

#

hm

#

it panics

#

but where are all the log sources?

#

oh

#

wait

#

it hangs while waiting for all CPUs to stop

#

which I think I know why

#

yeah

#

I also noticed that there is a slight (big) bug

#

the other CPUs restart fine

#

but they jump to some previous point of the thread they were running

#

since the thread contexts are not saved

#

so I'll need to do that

real pecan Oct 28, 2024, 8:08 PM

#

btw u can cherry-pick the fix for the event header if u want

#

its not in master yet

flint idol Oct 28, 2024, 8:08 PM

#

nah I'll just wait for the PR to be merged

white mulch Oct 28, 2024, 8:14 PM

#

flint idol since the thread contexts are not saved

I were thinking about this, one way could be that you save the rsp after the irq frame + current thread to somewhere within eg. the cpu struct and then when the ap wakes up after you have done all the lapic init and whatever you would set rsp to the saved rsp + pop regs + possibly swapgs + iretq (after first setting your current thread to the old thread)

#

and wait for all the cpus to have done that before actually proceeding with the sleep ofc

flint idol Oct 28, 2024, 8:18 PM

#

@real pecan

#

I fixed the bug

#

I can now restart the scheduler

#

damn it

#

it doesn't work well when it tries to run the uacpi code (but that's because of me)

real pecan Oct 28, 2024, 8:19 PM

#

why is that

flint idol Oct 28, 2024, 8:20 PM

#

some sorta irql bug

#

brb though

#

wait

#

it might actually have ran the uacpi code

#

@real pecan your uacpi code actually seemed to have not crashed

real pecan Oct 28, 2024, 8:21 PM

#

who would have thought meme

white mulch Oct 28, 2024, 8:21 PM

#

also on an unrelated note would be funny if the acpi global lock would be contended and the wake aml would try to acquire it when the io-apic is reset (so no sci's) meme

real pecan Oct 28, 2024, 8:21 PM

#

probably would crash windows as well

#

so they wouldnt do that

flint idol Oct 28, 2024, 9:00 PM

#

bro fuck this shit

#

HOWWWWWWWWWWWw

#

fixed

#

@real pecan something interesting is happening

#

after wake from suspend, the framebuffer stops working

white mulch Oct 28, 2024, 9:09 PM

#

I also observed that though I didn't look into it yet because all of the other annoying bugs that I have

flint idol Oct 28, 2024, 9:09 PM

#

anyway, first log message is before the suspend

#

and the rest are after I wake from suspend

white mulch Oct 28, 2024, 9:09 PM

#

probably the vga card is also reset or something

flint idol Oct 28, 2024, 9:10 PM

#

that's gonna be hard to tackle...

white mulch Oct 28, 2024, 9:10 PM

#

well on qemu it isn't that bad but on real hw however...

#

if that happens on real hw that is

flint idol Oct 28, 2024, 9:11 PM

#

yeah S3 leaves like everything dead

real pecan Oct 28, 2024, 9:11 PM

#

flint idol after wake from suspend, the framebuffer stops working

yeah that makes sense

#

ur supposed to reinit everything

flint idol Oct 28, 2024, 9:12 PM

#

when uGPU so I can reinit the GPU

real pecan Oct 28, 2024, 9:13 PM

#

i think u dont even need a gpu driver

#

to post the gpu

flint idol Oct 28, 2024, 9:14 PM

#

well then what do I do?

real pecan Oct 28, 2024, 9:14 PM

#

linux/seabios code

flint idol Oct 28, 2024, 9:14 PM

#

I don't feel like brain damage

real pecan Oct 28, 2024, 9:14 PM

#

well on real hw _WAK might post the gpu

#

qemu doesnt have that

white mulch Oct 28, 2024, 9:14 PM

#

qemu has https://osdev.wiki/wiki/Bochs_VBE_Extensions tho and I have a feeling that configuring it + writing to the enable register might make it work again

#

though it would make sense to use some more portable mechanism if there is one

real pecan Oct 28, 2024, 9:15 PM

#

anyway yeah idk

flint idol Oct 28, 2024, 9:15 PM

#

I can just put a warning before suspend

real pecan Oct 28, 2024, 9:15 PM

#

its uncharted territory for me

flint idol Oct 28, 2024, 9:15 PM

#

OBOS_Warning("Note: Framebuffer might die\n");```

real pecan Oct 28, 2024, 9:16 PM

#

u should try on real hw

flint idol Oct 28, 2024, 9:16 PM

#

good idea

real pecan Oct 28, 2024, 9:16 PM

#

althought very large chance u will see a blank screen

flint idol Oct 28, 2024, 9:16 PM

#

I'll make it shutdown after wake from suspend

real pecan Oct 28, 2024, 9:16 PM

#

lol

flint idol Oct 28, 2024, 9:16 PM

#

so I can know if it worked

real pecan Oct 28, 2024, 9:16 PM

#

thats my stupid pc when it wakes for no reason

flint idol Oct 28, 2024, 9:17 PM

#

well does your PC use obos?

real pecan Oct 28, 2024, 9:17 PM

#

worse

#

its nt

flint idol Oct 28, 2024, 9:17 PM

#

no way

#

LET THE SHITTY NT KERNEL DIE!!!!

#

REPLACE IT ALL WITH NYAYX~!!!!!!!!!!!

#

ummm

#

shutdown broke

real pecan Oct 28, 2024, 9:17 PM

#

f

torpid wigeon Oct 28, 2024, 9:18 PM

#

real pecan u should try on real hw

just write a dri driver

flint idol Oct 28, 2024, 9:18 PM

#

real pecan f

this time it might be you

real pecan Oct 28, 2024, 9:18 PM

#

??

flint idol Oct 28, 2024, 9:18 PM

#

well

real pecan Oct 28, 2024, 9:18 PM

#

literally nothing changed

flint idol Oct 28, 2024, 9:18 PM

#

let me log stuff from wake from suspend

torpid wigeon Oct 28, 2024, 9:18 PM

#

oberrow try not to deflect blame challenge

flint idol Oct 28, 2024, 9:19 PM

#

well it gets to uacpi code

real pecan Oct 28, 2024, 9:19 PM

#

oberrow try not to blame uacpi for every bug in his kernel insane challenge

flint idol Oct 28, 2024, 9:19 PM

#

it's always uacpi

#

/j

torpid wigeon Oct 28, 2024, 9:19 PM

#

i dont use uacpi yet and im still having kernel bugs

#

must be infys fault

real pecan Oct 28, 2024, 9:19 PM

#

it is

white mulch Oct 28, 2024, 9:20 PM

#

meanwhile I only have myself to blame

real pecan Oct 28, 2024, 9:21 PM

#

techinically u copied some stuff from uacpi right

#

so u can blame it as well

flint idol Oct 28, 2024, 9:21 PM

#

uhh wtf

#

why is my user program getting SIGTRAP

torpid wigeon Oct 28, 2024, 9:21 PM

#

if you steal all your code from other projects you can blame the people that wrote it galaxybrain

white mulch Oct 28, 2024, 9:22 PM

#

real pecan techinically u copied some stuff from uacpi right

well yeah some of the sleep/wake/whatever stuff + looking how reference semantics work + gas read/write helpers, ig I could troll though its more likely that its my own fault than the uacpi code's fault

real pecan Oct 28, 2024, 9:23 PM

#

lmao

flint idol Oct 28, 2024, 9:23 PM

#

kinda sounds like the entirety of uacpi

torpid wigeon Oct 28, 2024, 9:23 PM

#

copying inline asm

real pecan Oct 28, 2024, 9:23 PM

#

well his stuff is c++

flint idol Oct 28, 2024, 9:24 PM

#

idk why but when running my kernel

torpid wigeon Oct 28, 2024, 9:24 PM

#

https://tenor.com/view/gnu-efi-uefi-osdev-programming-gif-25428685

Tenor

flint idol Oct 28, 2024, 9:24 PM

#

in qemu

#

kde froze

#

meme

torpid wigeon Oct 28, 2024, 9:24 PM

#

use xfce like a real linux user

flint idol Oct 28, 2024, 9:24 PM

#

when ur kernel is so bad that when in a vm it causes the host system to die

flint idol Oct 28, 2024, 9:24 PM

#

torpid wigeon use xfce like a real linux user

go away

torpid wigeon Oct 28, 2024, 9:24 PM

#

meme

white mulch Oct 28, 2024, 9:24 PM

#

flint idol kinda sounds like the entirety of uacpi

I mean I didn't copy the interpreter itself KEKW

#

and I didn't ctrl c v any code because it wouldn't work ofc, I rewrote it in c++

flint idol Oct 28, 2024, 9:25 PM

#

🗣️

#

ok uacpi copier

torpid wigeon Oct 28, 2024, 9:26 PM

#

#1230349543623757845 message

flint idol Oct 28, 2024, 9:27 PM

#

shit

#

I forgor to put the license

#

actually

#

if the license is in the file

#

would that count

#

oh infy!

#

it hangs in uacpi_enter_sleep_state(UACPI_SLEEP_STATE_S5);

#

wtf

real pecan Oct 28, 2024, 9:28 PM

#

i mean that doesnt tell me anything

flint idol Oct 28, 2024, 9:28 PM

#

why is it in idle task

#

HOW

real pecan Oct 28, 2024, 9:28 PM

#

lol

flint idol Oct 28, 2024, 9:29 PM

#

IRQS ARE DISABLED

#

AHHHHHHHH

#

HOWWWWWWW

white mulch Oct 28, 2024, 9:40 PM

#

regarding the framebuffer if I understood it right linux has some code that uses vga to set the video mode that it calls on wake https://github.com/torvalds/linux/blob/81983758430957d9a5cb3333fe324fd70cf63e7e/arch/x86/realmode/rm/wakemain.c#L84 (look at the files with this __videocard macro https://github.com/torvalds/linux/blob/81983758430957d9a5cb3333fe324fd70cf63e7e/arch/x86/boot/video.h#L81)

#

idk if this works on all gpus tho because at least newer intel igpus don't advertise themself as vga compatible

flint idol Oct 28, 2024, 9:56 PM

#

white mulch regarding the framebuffer if I understood it right linux has some code that uses...

yeah I'll pass

#

for now at least

real pecan Oct 28, 2024, 9:58 PM

#

who needs a framebuffer

flint idol Oct 28, 2024, 9:58 PM

#

exactly

#

it's not like I'll ever use obos

#

and if I do I can just shutdown

real pecan Oct 28, 2024, 9:59 PM

#

lol

flint idol Oct 28, 2024, 9:59 PM

#

flint idol and if I do I can just shutdown

instead of suspend

flint idol Oct 29, 2024, 12:15 AM

#

I am somehow killing kvm

#

BRO HOW THE FUCK

#

IS THISS FUCKING HAPPENING

flint idol Oct 29, 2024, 12:44 AM

#

well luckily (or perhaps unluckily)

#

it doesn't happen when I don't suspend

#

omg

#

uacpi_kernel_stall

#

yields

#

well

#

something is utterly fucked

flint idol Oct 29, 2024, 12:55 AM

#

flint idol `uacpi_kernel_stall`

oh btw

#

@real pecan uacpi_kernel_stall might not be able to be used in uacpi_enter_sleep_state. if someone ever uses the hpet (or any external timer source for that matter) for stall/sleep, then it probably will break if PTS shutdown the timer source.

#

also shutdown does not work after waking from S5

#

it times out

#

it could be that I am missing something in my wake code

#

which is weird since I

uacpi_prepare_for_wake_from_sleep_state(UACPI_SLEEP_STATE_S3);
UACPI_ARCH_ENABLE_INTERRUPTS();
uacpi_wake_from_sleep_state(UACPI_SLEEP_STATE_S3);```

#

could be my skill issue though

#

since I do not put any devices to sleep

#

nor wake any devices up

#

but like

#

if you wrote the shit needed to the ports needed

#

then it should shutdown

#

and like this is qemu seabios

#

the most basic environment

#

weird

#

on ovmf it works

#

firmware fuckiness

thick jolt Oct 29, 2024, 1:36 AM

#

flint idol I am somehow killing kvm

ive done worse to kvm

#

oh trust me

flint idol Oct 29, 2024, 1:36 AM

#

KVM Internal Error 3

thick jolt Oct 29, 2024, 1:36 AM

#

oh nvm not that worse

#

😭

flint idol Oct 29, 2024, 1:36 AM

#

so I'm playing chess against some bot right

#

wtf did I get myself into 😭

thick jolt Oct 29, 2024, 1:37 AM

#

bro is cooked

#

😭 🙏

#

i got suspeneded in valorant for 7 hours

#

thats insane

flint idol Oct 29, 2024, 1:38 AM

#

bru

thick jolt Oct 29, 2024, 1:38 AM

#

how

#

i didnt even

#

do anything wrong 😭

#

i only yelled at my teammates

#

everyone does that in this game

flint idol Oct 29, 2024, 1:39 AM

#

probs had baby teammates

thick jolt Oct 29, 2024, 1:39 AM

#

yea maybe one of my teammates reported me

flint idol Oct 29, 2024, 1:39 AM

#

FUCK

#

thick jolt Oct 29, 2024, 1:39 AM

#

rip bro

flint idol Oct 29, 2024, 1:40 AM

#

yeah I'm not very good at chess

thick jolt Oct 29, 2024, 1:40 AM

#

same

flint idol Oct 29, 2024, 1:47 AM

#

oop

#

just stalemated

flint idol Oct 29, 2024, 2:05 AM

#

tomorrow I hope on implementing the driver interface part of suspending

#

as now I shall sleep

#

and maybe working around this bug with shutdown

#

but before I sleep

#

I will test obos on one of my test subjects

flint idol Oct 29, 2024, 2:15 AM

#

flint idol I will test obos on one of my test subjects

suspend works

#

wake from suspend, not so much

#

Notify (PWRB, 0x02) // Device Wake

#

is ignored according to uacpi

#

twice, actually

#

Method (SIOH, 0, NotSerialized)
{
    If (((WKRS == 0x02) && (WKRS != 0xFF))) {}
    Else
    {
        If ((^SIO1.PMS1 & 0x08))
        {
            Notify (PS2K, 0x02) // Device Wake
            Notify (PWRB, 0x02) // Device Wake
        }

        If ((^SIO1.PMS1 & 0x10))
        {
            Notify (PS2M, 0x02) // Device Wake
            Notify (PWRB, 0x02) // Device Wake
        }
    }
}```

#

Method (PS2K._PRW, 0, NotSerialized)  // _PRW: Power Resources for Wake
{
    Return (GPRW (0x1B, 0x03))
}```

#

Name (PRWP, Package (0x02)
{
    Zero, 
    Zero
})
Method (GPRW, 2, NotSerialized)
{
    PRWP [Zero] = Arg0
    Local0 = (SS1 << One)
    Local0 |= (SS2 << 0x02)
    Local0 |= (SS3 << 0x03)
    Local0 |= (SS4 << 0x04)
    If (((One << Arg1) & Local0))
    {
        PRWP [One] = Arg1
    }
    Else
    {
        Local0 >>= One
        FindSetLeftBit (Local0, PRWP [One])
    }
    
    Return (PRWP) /* \PRWP */
}```

#

Name (SS1, Zero)
Name (SS2, Zero)
Name (SS3, Zero)
Name (SS4, One)```

#

wait

#

_S3 doesn't look like it exists on this PC

#

If (SS3)
{
    Name (_S3, Package (0x04)  // _S3_: S3 System State
    {
        0x05, 
        Zero, 
        Zero, 
        Zero
    })
}```

#

acpi_namespace_node* s3 = uacpi_namespace_node_find(uacpi_namespace_root(), "_S3_");
if (!s3)
{
    OBOS_Error("Firmware does not have the _S3 (suspend) sleep state\n");
    return OBOS_STATUS_UNIMPLEMENTED; // bios does NOT support suspend.
}```

#

I will do that

#

this is quite weird

real pecan Oct 29, 2024, 3:08 AM

#

flint idol <@353593948523986945> uacpi_kernel_stall might not be able to be used in `uacpi_...

Acpica does it and it works fine, so no

real pecan Oct 29, 2024, 7:12 PM

#

flint idol is ignored according to uacpi

this just means u dont hook an event listener using uacpi_install_notify_handler

#

Notify is basically a syscall from aml to kernel

flint idol Oct 31, 2024, 12:17 AM

#

I apparently need to implement D state shit for wake to work properly

#

#1217009725711847465 message

thick jolt Oct 31, 2024, 2:04 AM

#

i might implement this stuff too into my kernel

#

to be honest

#

im unsure about much of what it does but

#

seems interesting

#

when i fix my page tables being fucked up and when i reach userland and have a bunch of unix programs running

#

that might be something id look into

real pecan Oct 31, 2024, 6:55 AM

#

thick jolt im unsure about much of what it does but

ability to put your kernel to sleep and then resume

flint idol Oct 31, 2024, 10:58 AM

#

thick jolt seems interesting

it indeed is

#

but it's hard to implement

short mortar Oct 31, 2024, 1:16 PM

#

May be I'll try to implement it as well (after finishing with exams, and implementing working set...)

flint idol Oct 31, 2024, 7:18 PM

#

Since I'm going to be "done" soon enough

#

You guys will probably able to take my kernel as a reference when you do it

weak kestrel Oct 31, 2024, 7:19 PM

#

what did you have to do?

flint idol Oct 31, 2024, 7:19 PM

#

At the moment, it is the only kernel using uacpi that has some sort of suspend

#

Find sleep capable acpi namespace nodes

#

Put them into the required D state

#

Mark then as a wake GPE

#

*them

#

Then stop all CPUs

#

Then enter S3

#

On wake

flint idol Oct 31, 2024, 7:22 PM

#

flint idol Then stop all CPUs

Also make sure to set the wake vector

flint idol Oct 31, 2024, 7:22 PM

#

flint idol On wake

When you wake up you get put there

#

On the wake vector

#

In real mode

#

On the bsp

weak kestrel Oct 31, 2024, 7:23 PM

#

so you have to do all the initialisations again

flint idol Oct 31, 2024, 7:23 PM

#

weak kestrel so you have to do all the initialisations again

Yes

#

Move any device back to the old D state

#

After entering long mode

#

Also you probably want to have driver specific wake and sleep functions which you call at respective times

#

You will need to restart the LAPIC timer

real pecan Oct 31, 2024, 7:35 PM

#

flint idol At the moment, it is the only kernel using uacpi that has some sort of suspend

..or any hobby kernel with suspend at all

real pecan Oct 31, 2024, 8:50 PM

#

well given oberrow actually gets it to work properly

flint idol Oct 31, 2024, 8:58 PM

#

real pecan well given oberrow actually gets it to work properly

Other than framebuffer I should be able to get it working correctly

real pecan Oct 31, 2024, 8:59 PM

#

if u get it right fb should get restored on real hw as well

#

especially if u disable modern windows support via osi

#

uacpi_bulk_configure_interfaces(UACPI_INTERFACE_ACTION_DISABLE, UACPI_INTERFACE_KIND_VENDOR)

#

then re-enable only some shit like very old xp

#

aml will assume u dont have any drivers and do most stuff for u

#

or well, try to at least

flint idol Oct 31, 2024, 9:01 PM

#

Uncommon AML W

real pecan Oct 31, 2024, 9:01 PM

#

idk if qemu checks for it, probably not

flint idol Oct 31, 2024, 9:02 PM

#

Would be one of the first times that something in obos worked as intended on real hw but not qemu

white mulch Oct 31, 2024, 9:05 PM

#

also the bochs vbe stuff that I were thinking about didn't work for restoring the fb, idk what exactly qemu wants you to do to it

real pecan Oct 31, 2024, 9:08 PM

#

could just look at qemu source

white mulch Oct 31, 2024, 9:09 PM

#

actually I think I might know at least one possible reason, the pci bars get reset

real pecan Oct 31, 2024, 9:09 PM

#

lool

flint idol Oct 31, 2024, 9:09 PM

#

Oh ye

#

I could make some sort of hack

#

That looks for the fb in PCI bars

#

And just saves all BARS for them

real pecan Oct 31, 2024, 9:10 PM

#

just do info pci after resume

flint idol Oct 31, 2024, 9:10 PM

#

Then replace it later

real pecan Oct 31, 2024, 9:10 PM

#

and see what it says

white mulch Oct 31, 2024, 9:10 PM

#

real pecan Oct 31, 2024, 9:11 PM

#

yup

#

ideally what u should do in your drivers is when initing a pci bus, walk all devices, and look at their bars: if it'a non-zero bar, try to "claim" it aka allocate it out of unused memory. if that fails, allocate a new addres. Record the allocated/claimed address somewhere in a struct.

#

Then on resume just walk again and re-assign

#

thats how linux does it anyway

white mulch Oct 31, 2024, 9:13 PM

#

the irq fields aren't right either, ig you have to restore those too

real pecan Oct 31, 2024, 9:13 PM

#

if u use them, yeah

#

and bus mastering and stuff

#

its basically pci at reset state

flint idol Oct 31, 2024, 9:18 PM

#

Yeah I need to make some more pci helpers

#

For the driver interface

real pecan Oct 31, 2024, 9:19 PM

#

e.g. https://elixir.bootlin.com/linux/v6.11.5/source/drivers/pci/setup-res.c#L136

setup-res.c - drivers/pci/setup-res.c - Linux source...

Elixir Cross Referencer - source file of Linux (version v6.11.5). Browsed file: /drivers/pci/setup-res.c

#

->resource is an array of BARs

#

it tries to use whatever the bios put there

#

if that fails it manually allocates here https://elixir.bootlin.com/linux/v6.11.5/source/drivers/pci/setup-bus.c#L707

setup-bus.c - drivers/pci/setup-bus.c - Linux source...

Elixir Cross Referencer - source file of Linux (version v6.11.5). Browsed file: /drivers/pci/setup-bus.c

real pecan Oct 31, 2024, 9:21 PM

#

flint idol Yeah I need to make some more pci helpers

at the very least u need a struct per pci devices where u keep the initial values of their bars

weary hound Oct 31, 2024, 9:47 PM

#

white mulch the irq fields aren't right either, ig you have to restore those too

the device doesn't care about the irq field afaik? only the pin matters

#

and the pin is hardwired in the device iirc

white mulch Oct 31, 2024, 9:49 PM

#

ah yeah

#

so its basically only the bars + the usual stuff that you do when you setup a device anyway like enabling bus master

flint idol Nov 1, 2024, 1:58 AM

#

yeah rn the only sort of unified API for PCI devices is finding them

#

and interrupts

flint idol Nov 1, 2024, 1:58 AM

#

flint idol yeah rn the only sort of unified API for PCI devices is finding them

also getting bar sizes, but not bars for some reason

flint idol Nov 1, 2024, 11:35 AM

#

before suspend

#

after wake

#

so yeah

#

one of the vga BARs had the framebuffer

#

which is probably why the framebuffer died

#

what I'm thinking to do is why was suggested to me yesterday

#

but own most PCI devices by default

#

then drivers claim them later

#

I wonder what the difference between ovmf and seabios is

#

because with ovmf, qemu shutdowns after suspend

#

but seabios does not

#

with seabios it doesn't even log it with -d trace:qemu_system_shutdown_request

#

before suspend (on ovmf)

weak kestrel Nov 1, 2024, 11:48 AM

#

are you going to write an osdev.wiki article once you finish it? meme

flint idol Nov 1, 2024, 11:49 AM

#

that might not even be a bad idea

weak kestrel Nov 1, 2024, 11:49 AM

#

I'm mostly interested in what needs to be done for pci devices

#

I gather everything else is the same

flint idol Nov 1, 2024, 11:50 AM

#

you save BARs and other config registers in teh PCi thing

#

probably includes MSI config

#

and other capabilities

#

you throw the device into some D state

#

you go onto the next device

#

on wake you restore everything and move the devices back to D0

#

also

#

there probably should be some driver-specific way to suspend a device

#

and vice versa

real pecan Nov 1, 2024, 11:51 AM

#

flint idol because with ovmf, qemu shutdowns after suspend

wtf

flint idol Nov 1, 2024, 11:51 AM

#

sorry

#

when I make it shutdown

#

after wake from suspend

#

I worded that terribly KEKW

real pecan Nov 1, 2024, 11:52 AM

#

so your resume is broken on seabios?

flint idol Nov 1, 2024, 11:52 AM

#

pretty much

#

qemu doesn't even log the shutdown attempt

real pecan Nov 1, 2024, 11:53 AM

#

lol

flint idol Nov 1, 2024, 11:54 AM

#

flint idol https://discord.com/channels/440442961147199490/1217009725711847465/130128284834...

marker

white mulch Nov 1, 2024, 11:57 AM

#

also about the pci thing I am not sure what else do you need to do because I tried writing the bars back after suspend but it didn't change them at all (and -trace pci*) showed no writes at all

#

tldr: sleep is a huge pain lol

flint idol Nov 1, 2024, 11:57 AM

#

well you do call the _WAK stuff

#

and prepare for wake

#

might be needed

#

at this point I might as well import seabios

flint idol Nov 1, 2024, 12:14 PM

#

https://docs.kernel.org/power/pci.html

#

see 1.3

#

white mulch Nov 1, 2024, 12:16 PM

#

yeah I did that and then back to d0 after suspend for the wake capable devices (idk if you should do it for all devices tho)

#

also apparently my resume is broken on ovmf but not seabios meme

flint idol Nov 1, 2024, 12:17 PM

#

rather the acpi spec is outright wrong here

#

or my computer's firmware made an oopsie

#

because _PRW > 3, so I think that just means 3 in this table

#

_S3D is 2

#

so I should enter D2

#

this device has no D2 state

#

rather that or I misunderstood the spec somewhere else

#

yeah no

#

there's no _PS2 method

white mulch Nov 1, 2024, 12:20 PM

#

_PRW is the deepest system state that it supports waking from

flint idol Nov 1, 2024, 12:20 PM

#

ye

#

that evaluated to

0xD, 0x4```

#

Method (_PRW, 0, NotSerialized)  // _PRW: Power Resources for Wake
{
    Return (GPRW (0x0D, 0x04))
}```

white mulch Nov 1, 2024, 12:21 PM

#

does it have a _S3W

flint idol Nov 1, 2024, 12:21 PM

#

nop

#

Method (GPRW, 2, NotSerialized)
{
    PRWP [Zero] = Arg0
    Local0 = (SS1 << One)
    Local0 |= (SS2 << 0x02)
    Local0 |= (SS3 << 0x03)
    Local0 |= (SS4 << 0x04)
    If (((One << Arg1) & Local0))
    {
        PRWP [One] = Arg1
    }
    Else
    {
        Local0 >>= One
        FindSetLeftBit (Local0, PRWP [One])
    }
    
    Return (PRWP) /* \PRWP */
}```

white mulch Nov 1, 2024, 12:22 PM

#

then you can enter D2 or D3

flint idol Nov 1, 2024, 12:23 PM

#

flint idol rather the acpi spec is outright wrong here

it says that is _S3W is N/A, _S3D is 2, and _PRW is 3

#

then enter D2

#

unless I missed something

#

in the spec

white mulch Nov 1, 2024, 12:24 PM

#

oh yeah nvm

flint idol Nov 1, 2024, 12:25 PM

#

fucked firmware

#

but all other devices are the same

#

that support wake

#

and have _S3D

#

I hate acpi

#

I hate acpi

#

I hate acpi

#

I hate acpi

#

I hate acpi

quaint rivetBOT Nov 1, 2024, 12:35 PM

#

Anti-spam Message

Possible spam detected for user: oberrow. Please contact a moderator to be unmuted.

torpid wigeon Nov 1, 2024, 12:37 PM

#

i hate acpi

#

i hate acpi

flint idol Nov 1, 2024, 12:38 PM

#

maybe if the D state just doesn't exist

#

I ignore it and move on to calling _PSW

#

or DSW

white mulch Nov 1, 2024, 12:39 PM

#

does the device have a _PS0/_PR0 method?

flint idol Nov 1, 2024, 12:39 PM

#

yes

#

it has _PS0

#

and _PS3

#

but not the required _PS2

#

and I just found another unrelated blob doing the same thing

#

actually the other blob isn't even for a computer that has S3

white mulch Nov 1, 2024, 12:50 PM

#

is there a _PR2? because there is the thing where if there is no explicit _PSx method then you are supposed to use the power resources to do the transition

flint idol Nov 1, 2024, 12:50 PM

#

no _PR2

#

or _PS2

flint idol Nov 1, 2024, 12:50 PM

#

white mulch is there a _PR2? because there is the thing where if there is no explicit _PSx m...

guess I gotta do that then

#

and where do I find said power resources

#

_PRW?

white mulch Nov 1, 2024, 12:52 PM

#

_PRW contains the resources needed for the device to be able to wake the system

#

ah yeah ig you can do it using that + _PR0

flint idol Nov 1, 2024, 12:52 PM

#

but isn't PR0 for D state 0

#

and also I have no PR0 in that object

#

*device

#

only for fans

white mulch Nov 1, 2024, 12:54 PM

#

nooo

white mulch Nov 1, 2024, 12:54 PM

#

flint idol but isn't PR0 for D state 0

yeah but I were thinking about the fact that it said that you should turn off the resources that are no longer needed by the device

#

so like if there would be some resources needed in d0 but not in the wake state (or whatever state you explicitly put the device to) you'd turn those off if I understood it correctly

flint idol Nov 1, 2024, 12:57 PM

#

makes sense

white mulch Nov 1, 2024, 12:59 PM

#

but yeah idk what are you supposed to do in the case of your firmware

flint idol Nov 1, 2024, 1:00 PM

#

maybe

#

journalctl will have something in the logs to guide me

white mulch Nov 1, 2024, 1:01 PM

#

or add debug prints to linux code meme though that would be kinda annoying

flint idol Nov 1, 2024, 1:01 PM

#

I don't feel like compiling the linux kernel

#

journalctl has nothin for me

#

thanks infy for making uacpi_object opaque /s

real pecan Nov 1, 2024, 1:12 PM

#

lmao

flint idol Nov 1, 2024, 1:13 PM

#

thank god it doesn't affect my pnp thing

#

bruh

#

I can't even use packages

real pecan Nov 1, 2024, 1:15 PM

#

uacpi_object_get_package

flint idol Nov 1, 2024, 1:16 PM

#

use of undeclared identifier uacpi_package

real pecan Nov 1, 2024, 1:16 PM

#

read

flint idol Nov 1, 2024, 1:16 PM

#

an object array

#

is what it gives me

#

ok

#

bruh

#

uacpi_object_get_integer returns a status

#

and takes a pointer to the integer

real pecan Nov 1, 2024, 1:19 PM

#

yes??

flint idol Nov 1, 2024, 1:20 PM

#

yeah

#

just kinda annoying

real pecan Nov 1, 2024, 1:20 PM

#

safety ™️

flint idol Nov 1, 2024, 1:20 PM

#

but who am I to complain

#

it's a free acpi library

real pecan Nov 1, 2024, 1:21 PM

#

what is it supposed to do if u give it a different type or a null pointer

flint idol Nov 1, 2024, 1:21 PM

#

https://cdn.discordapp.com/attachments/1062027635611422860/1297300887508877342/assert.gif

real pecan Nov 1, 2024, 1:22 PM

#

uacpi is built onto the idea that all errors are propagated to the client and it on its own doesnt have the right to decide what to do with them

flint idol Nov 1, 2024, 1:25 PM

#

just made it use the new object api

real pecan Nov 1, 2024, 1:25 PM

#

so it doesnt have bullshit like uacpi_kernel_abort etc

flint idol Nov 1, 2024, 1:25 PM

#

good thing I noticed that you merged the PR before I wrote the D state code

#

otherwise I would've went to play minecraft for 43 days

#

and come back

real pecan Nov 1, 2024, 1:26 PM

#

lol

flint idol Nov 1, 2024, 1:52 PM

#

the 2nd value in _PRW is irrelevant for finding out the d state to put the device in when entering a sleep state

#

only snd/snw matter

#

with n being the sleep state

#

that you wanna enter

#

which basically makes the tables in the acpi spec for each

#

_SnD object

#

the same

#

since some parts of the table say I can enter one of two sleep states

#

I will by default choose the deepest

#

if that does not exist

#

I go a bit shallower

#

etc.

#

typedef enum d_state {
    DSTATE_INVALID = -1,
    DSTATE_0,
    DSTATE_1,
    DSTATE_2,
    DSTATE_3HOT,
    DSTATE_3COLD,
    DSTATE_MAX = DSTATE_3COLD,
} d_state;

// if dry_run is true, then the function does not actually put the device into the state,
// but only checks if it would be able to and returns an apprioriate status.

obos_status OBOS_DeviceSleep(uacpi_namespace_node* dev, d_state new_state, bool dry_run);
obos_status OBOS_DeviceWake(uacpi_namespace_node* dev, d_state new_state, bool dry_run);
d_state OBOS_DeviceGetDStateForWake(uacpi_namespace_node* dev, uacpi_sleep_state state, obos_status* status);```

#

if (snw >= snd)
{
    if (status)
        *status = OBOS_STATUS_MISMATCH;
    return DSTATE_INVALID;
}```

#

I wonder if it makes sense to have a check like this

#

because surely the deepest D state that the device can wake itself in

#

in sleep state n

#

is always deeper or equal than the shallowest state that the device can be in in sleep state n

#


if (snw <= DSTATE_2 && snd == UINT64_MAX)
{
    nStates = 3;
    avaliableStates[0] = DSTATE_2;
    avaliableStates[1] = DSTATE_1;
    avaliableStates[2] = DSTATE_0;
}

if (snd <= DSTATE_2 && snw == UINT64_MAX)
{
    nStates = 1;
    avaliableStates[0] = DSTATE_2;
}

if (snd == DSTATE_2 && (snw >= DSTATE_3HOT && snw != UINT64_MAX))
{
    nStates = 3;
    avaliableStates[0] = DSTATE_3COLD;
    avaliableStates[1] = DSTATE_3HOT;
    avaliableStates[2] = DSTATE_2;
}```

#

idk if I like this logic

#

I mean it probably works

#

but I wanna make it ~~unreadable~~ use less redundant checks

#

typedef enum d_state {
    DSTATE_INVALID = -1,
    DSTATE_0,
    DSTATE_1,
    DSTATE_2,
    DSTATE_3HOT,
    DSTATE_3COLD,
    DSTATE_MAX = DSTATE_3COLD,
} d_state;

// if dry_run is true, then the function does not actually put the device into the state,
// but only checks if it would be able to and returns an apprioriate status.

OBOS_EXPORT obos_status OBOS_DeviceSetDState(uacpi_namespace_node* dev, d_state new_state, bool dry_run);
OBOS_EXPORT obos_status OBOS_DeviceHasDState(uacpi_namespace_node* dev, d_state state);
// Returns DSTATE_INVALID on error, or if the device does not need to be moved in another
// D state to wake us.
// Always check status to make sure.
OBOS_EXPORT d_state     OBOS_DeviceGetDStateForWake(uacpi_namespace_node* dev, uacpi_sleep_state state, obos_status* status);```

#OBOS (not vibecoded)