OBOS (not vibecoded) | Operating System Development | Page 29

thick jolt Dec 23, 2024, 1:10 AM

#

bru

flint idol Dec 23, 2024, 1:10 AM

#

or should I just save the timers I use

thick jolt Dec 23, 2024, 1:11 AM

#

ignore

real pecan Dec 23, 2024, 1:11 AM

#

thick jolt bru

Nyaux suspend support when

thick jolt Dec 23, 2024, 1:11 AM

#

real pecan Nyaux suspend support when

when pci

#

there will be suspend

flint idol Dec 23, 2024, 1:11 AM

#

if that happens

thick jolt Dec 23, 2024, 1:11 AM

#

after

#

it will

flint idol Dec 23, 2024, 1:11 AM

#

I will be muting that thread

thick jolt Dec 23, 2024, 1:11 AM

#

wait seriously

#

nooo

flint idol Dec 23, 2024, 1:11 AM

#

because of the amount of activity of nyauxmaster complaining

#

about wake not working

thick jolt Dec 23, 2024, 1:11 AM

#

😭 💀

flint idol Dec 23, 2024, 1:11 AM

#

and wake crashing

#

and whatnot

real pecan Dec 23, 2024, 1:12 AM

#

Yeah that one is going to be tough

#

No one can spoonfeed that

thick jolt Dec 23, 2024, 1:12 AM

#

bro

#

i dont get spoonfeed

#

stop saying that

#

its not even funny seriously

real pecan Dec 23, 2024, 1:12 AM

#

Jkjk

thick jolt Dec 23, 2024, 1:12 AM

#

okay

flint idol Dec 23, 2024, 1:12 AM

#

implementing suspend is going into the wild west of acpi

#

the actual power management part

#

it's no longer fun and games when you implement suspend

thick jolt Dec 23, 2024, 1:13 AM

#

it will def be pain but yk ill pull through\

flint idol Dec 23, 2024, 1:13 AM

#

you won't

thick jolt Dec 23, 2024, 1:13 AM

#

i will because ill do something u wont

flint idol Dec 23, 2024, 1:13 AM

#

I already have it..?

thick jolt Dec 23, 2024, 1:13 AM

#

work on other aspects of the kernel while suspend isnt working

#

meme

#

then like keep debugging suspend

#

but still working on other things

flint idol Dec 23, 2024, 1:14 AM

#

flint idol implementing suspend is going into the wild west of acpi

Advanced Configuration and Power Interface

thick jolt Dec 23, 2024, 1:14 AM

#

yes will be very scary

flint idol Dec 23, 2024, 1:14 AM

#

do you know what a power resource is?
that's what I thought, no

thick jolt Dec 23, 2024, 1:14 AM

#

💀

real pecan Dec 23, 2024, 1:15 AM

#

flint idol Advanced Configuration and **Power Interface**

Technically these last two words are on the east side

thick jolt Dec 23, 2024, 1:15 AM

#

ummm actually i know what that is

#

🤓

flint idol Dec 23, 2024, 1:15 AM

#

_ON, _OFF, _PSW, _S3W, _S3D, _PRW, _DSW

#

those are the methods I remember having to evaluate to setup wake gpes

#

also _PSn, _PRn with n being a D state

#

then I needed an entire smp trampoline just to wake up to a bare minimum long mode state

#

along with some 50 lines of x86_64-specific wake code

#

to get stuff running again

#

which will be increasing since I'm adding HPET restore code

#

as well as another ~50 lines of x86_64 specific code to wake the other CPUs again

thick jolt Dec 23, 2024, 1:19 AM

#

for restoring the hpet do u just need to enable the main counter again?

flint idol Dec 23, 2024, 1:19 AM

#

no

thick jolt Dec 23, 2024, 1:19 AM

#

okay then what do u need to do

flint idol Dec 23, 2024, 1:19 AM

#

you need to save everything

#

about it

thick jolt Dec 23, 2024, 1:19 AM

#

ah okay

flint idol Dec 23, 2024, 1:19 AM

#

at least everything you need

#

ig you can skip init of some timers

#

since it has multiple comparators

flint idol Dec 23, 2024, 1:20 AM

#

flint idol as well as another ~50 lines of x86_64 specific code to wake the other CPUs agai...

then PCI stuff needs to be restored

#

which needed a REVAMP of my pci interface

#

I also needed to add suspend/wake callbacks to drivers

#

just for it to still be partially incomplete

#

because I have no framebuffer

#

if that hasn't scared you off, idk what will

#

wait

#

how would I restore const HPET fields

#

thonk

#

maybe I just don't

real pecan Dec 23, 2024, 1:24 AM

#

Yes

flint idol Dec 23, 2024, 1:26 AM

#

so I need to restore:

HPET.generalConfig
HPET.mainCounterValue
HPET.timer0
any other HPET comparators I use

#

I also don't restore IOAPIC state

#

silly me

real pecan Dec 23, 2024, 1:31 AM

#

Who needs that chad

flint idol Dec 23, 2024, 1:31 AM

#

me nooo

real pecan Dec 23, 2024, 1:32 AM

#

Does it work with hpet restoration?

flint idol Dec 23, 2024, 1:33 AM

#

nah

#

I need IOAPIC for GPEs

#

so I will add restore code for that soon™️

#

and you should probably go to sleep

real pecan Dec 23, 2024, 1:34 AM

#

flint idol nah

Wait wdym

#

Does it hang still

flint idol Dec 23, 2024, 1:34 AM

#

it still hangs

real pecan Dec 23, 2024, 1:34 AM

#

Strange

#

Anyways yeah continue tmrw

flint idol Dec 23, 2024, 1:34 AM

#

yup

#

can't believe I'm going to sleep at the same time as infy (he's 8 hours ahead of my timezone)

#

infy challenge: have sleep schedule, challenge level: impossible

real pecan Dec 23, 2024, 1:35 AM

#

Lol

flint idol Dec 23, 2024, 1:35 AM

#

~~but hey who needs sleep when you have osdev~~

real pecan Dec 23, 2024, 1:35 AM

#

Thought u were in Europe

flint idol Dec 23, 2024, 1:35 AM

#

no I'm in canada

real pecan Dec 23, 2024, 1:36 AM

#

Did you move recently?

flint idol Dec 23, 2024, 1:36 AM

#

no I've been in canada for like

#

my entire life

real pecan Dec 23, 2024, 1:36 AM

#

Ah

#

Lol

#

Gn then

flint idol Dec 23, 2024, 1:36 AM

#

gn

#

it's like -10 degrees outside

#

quite warm

#

ultrameme

real pecan Dec 23, 2024, 1:36 AM

#

nooo

real pecan Dec 23, 2024, 1:37 AM

#

flint idol infy challenge: have sleep schedule, challenge level: impossible

Vacation

flint idol Dec 23, 2024, 1:37 AM

#

fair

#

when I was on vacation in uae, I would go to sleep at 3 am

#

in saudi I had a normal-ish sleep schedule

#

mainly in mecca since I think I wanted to be able to get up for the dawn prayer at the sacred mosque

#

but alas, I was only able to make it once

#

another time I woke while the prayer was happening

#

so I used my phone to record the reciter

real pecan Dec 23, 2024, 1:39 AM

#

Oh well

flint idol Dec 23, 2024, 1:39 AM

#

yeah

#

gn now ig

real pecan Dec 23, 2024, 1:39 AM

#

Yup

flint idol Dec 23, 2024, 1:40 AM

#

flint idol so I used my phone to record the reciter

I would send it here, but the file size is too big for discord

thick jolt Dec 23, 2024, 2:08 AM

#

hi omar

#

#1230349543623757845 rn 😡

flint idol Dec 23, 2024, 2:08 AM

#

Hai

flint idol Dec 23, 2024, 12:22 PM

#

flint idol so I will add restore code for that soon™️

soon will be now

#

time to add I/O APIC restoration code

weak kestrel Dec 23, 2024, 12:26 PM

#

flint idol time to add I/O APIC restoration code

suspend/wake-up on osdev wiki when

flint idol Dec 23, 2024, 12:26 PM

#

soon™️

weak kestrel Dec 23, 2024, 12:26 PM

#

how much do you have left?

flint idol Dec 23, 2024, 12:26 PM

#

I need to write I/O APIC restoration code

#

and fix this bug with wake

#

on my laptop

flint idol Dec 23, 2024, 12:27 PM

#

flint idol and fix this bug with wake

which is the deadlock infy and I were debugging yesterday

#

TODO: Rewrite I/O APIC code to not use bitfields

flint idol Dec 23, 2024, 12:29 PM

#

flint idol I need to write I/O APIC restoration code

as well as ACPI NVS saving code

#

nvm, that's for S4 (hibernate) only

real pecan Dec 23, 2024, 12:55 PM

#

flint idol nvm, that's for S4 (hibernate) only

no

flint idol Dec 23, 2024, 12:55 PM

#

ok

real pecan Dec 23, 2024, 12:56 PM

#

https://elixir.bootlin.com/linux/v6.13-rc3/source/drivers/acpi/sleep.c

sleep.c - drivers/acpi/sleep.c - Linux source code v...

Elixir Cross Referencer - source code of Linux v6.13-rc3: drivers/acpi/sleep.c

#

/*
 * The ACPI specification wants us to save NVS memory regions during hibernation
 * and to restore them during the subsequent resume.  Windows does that also for
 * suspend to RAM.  However, it is known that this mechanism does not work on
 * all machines, so we allow the user to disable it with the help of the
 * 'acpi_sleep=nonvs' kernel command line option.
 */
static bool nvs_nosave;

void __init acpi_nvs_nosave(void)
{
    nvs_nosave = true;
}

#

/*
 * The ACPI specification wants us to save NVS memory regions during hibernation
 * but says nothing about saving NVS during S3.  Not all versions of Windows
 * save NVS on S3 suspend either, and it is clear that not all systems need
 * NVS to be saved at S3 time.  To improve suspend/resume time, allow the
 * user to disable saving NVS on S3 if their system does not require it, but
 * continue to save/restore NVS for S4 as specified.
 */
static bool nvs_nosave_s3;

#

kinda annoying

#

but thats suspend for u

flint idol Dec 23, 2024, 12:58 PM

#

yeah

#

I'll add a command line option because linux does
no-nvs-save-suspend

#

or something like that

real pecan Dec 23, 2024, 12:59 PM

#

ye

flint idol Dec 23, 2024, 12:59 PM

#

linux calls it nvs-nosave-s3

#

so I'm copying that into obos

#

and I might as well also add a config option for it

#

option (OBOS_NVS_SAVE_S3 "Enable compile-time support for saving NVS before S3" ON)

real pecan Dec 23, 2024, 1:04 PM

#

nice

flint idol Dec 23, 2024, 1:15 PM

#

struct acpi_nvs {
    void* saved_region;
    uintptr_t phys;
    size_t size;
};
struct acpi_nvs* saved_nvs;```

#

@real pecan ULTRA_MEMORY_TYPE_NVS is acpi nvs right? not some weird uefi thing

#

it seems like it, but I'm just making sure

real pecan Dec 23, 2024, 1:22 PM

#

ye

flint idol Dec 23, 2024, 1:22 PM

#

kthx

#

are there any PCs where there are multiple NVS regions?

#

surely there are, right?

real pecan Dec 23, 2024, 1:23 PM

#

probably

thick jolt Dec 23, 2024, 1:23 PM

#

i believe so but i may be wrong

flint idol Dec 23, 2024, 1:27 PM

#

Error 503 Backend fetch failed

#

elixir died

thick jolt Dec 23, 2024, 1:28 PM

#

rip

real pecan Dec 23, 2024, 1:28 PM

#

its been dead for multiple days

#

it takes 10 refreshes to load anything

flint idol Dec 23, 2024, 1:28 PM

#

well I always got github

thick jolt Dec 23, 2024, 1:29 PM

#

real pecan its been dead for multiple days

infy if u were a black hole like ur pfp what would u do

flint idol Dec 23, 2024, 1:29 PM

#

linux treats NVS ram as IO memory

#

calls acpi_os_get_iomem on it

#

to get its address

#

then memcpy's that in the buffer it allocated

#

for each nvs region

#

and restore does the same thing, backwards

#

@real pecan what kernel api does uacpi use to timeout loops, get_nanoseconds_since_boot?

#

like aml loops

real pecan Dec 23, 2024, 1:44 PM

#

i mean what other api could it even use

flint idol Dec 23, 2024, 1:44 PM

#

idk

#

I will return zero from that in case the hpet is still dead somehow

#

and a loop isn't timing out

#

during _WAK

real pecan Dec 23, 2024, 1:44 PM

#

/*
 * Calculate the expiration point for this loop.
 * If a loop is executed past this point, it will get aborted.
 */
block->expiration_point = uacpi_kernel_get_nanoseconds_since_boot();
block->expiration_point +=
    g_uacpi_rt_ctx.loop_timeout_seconds * UACPI_NANOSECONDS_PER_SEC;

flint idol Dec 23, 2024, 1:45 PM

#

or maybe not zero

#

uhh

#

1000000000

#

no

#

UINT64_MAX

real pecan Dec 23, 2024, 1:45 PM

#

it doesnt matter which constant u return

#

since its a constant

flint idol Dec 23, 2024, 1:45 PM

#

no, because UINT64_MAX+any number would overflow

real pecan Dec 23, 2024, 1:45 PM

#

and

flint idol Dec 23, 2024, 1:45 PM

#

but then that number is < UINT64_MAX

#

so the loop times out

#

immediately

real pecan Dec 23, 2024, 1:46 PM

#

why would the loop timeout

#

max + any number is some number

#

if u do this calculation again u get the same number

#

or wait

#

yeah ur right

flint idol Dec 23, 2024, 1:46 PM

#

[uACPI][INFO]: successfully loaded 1 AML blob, 1726 ops in 0ms (avg 1726000000000/s)

#

woo

real pecan Dec 23, 2024, 1:47 PM

#

but this will make it so any while that must run at least twice will not run at all

flint idol Dec 23, 2024, 1:47 PM

#

take that mammogram

real pecan Dec 23, 2024, 1:47 PM

#

or will run once i guess

#

so this might break something

flint idol Dec 23, 2024, 1:47 PM

#

I only set the loop timeout before calling wake_from_sleep

#

or whatever the function was called

#

uacpi_context_set_loop_timeout(3);
uacpi_wake_from_sleep_state(UACPI_SLEEP_STATE_S3);```

#

I assume the default is zero?

#

no it's 30

real pecan Dec 23, 2024, 1:49 PM

#

zero so it insta aborts everything yes LULW

#

do u do full nvs restore now?

flint idol Dec 23, 2024, 1:50 PM

#

yes

#

    if (!OBOS_GetOPTF("nvs-nosave-s3"))
    {
        // Save ACPI NVS regions.
        // Why? Some versions of windows do, so I guess we have to :)

        // Read the memory map.
        uintptr_t index = 0;
        for (obos_pmem_map_entry* entry = MmS_GetFirstPMemMapEntry(&index); entry; )
        {
            if (entry->type == ULTRA_MEMORY_TYPE_NVS)
                saved_nvs_count++;

            entry = MmS_GetNextPMemMapEntry(entry, &index);
        }

        saved_nvs = OBOS_NonPagedPoolAllocator->ZeroAllocate(OBOS_NonPagedPoolAllocator, saved_nvs_count, sizeof(struct acpi_nvs), nullptr);

        index = 0;
        size_t nvs_index = 0;
        for (obos_pmem_map_entry* entry = MmS_GetFirstPMemMapEntry(&index); entry; )
        {
            if (entry->type == ULTRA_MEMORY_TYPE_NVS)
            {
                saved_nvs[nvs_index].region_address = Arch_MapToHHDM(entry->physical_address);
                saved_nvs[nvs_index].size = entry->size;
                saved_nvs[nvs_index].saved_region = OBOS_NonPagedPoolAllocator->Allocate(OBOS_NonPagedPoolAllocator,
                                                                                         saved_nvs[nvs_index].size,
                                                                                         nullptr);
                memcpy(saved_nvs[nvs_index].saved_region, saved_nvs[nvs_index].region_address, saved_nvs[nvs_index].size);
                nvs_index++;
            }

            entry = MmS_GetNextPMemMapEntry(entry, &index);
        }
    }```

#

that's to save

real pecan Dec 23, 2024, 1:50 PM

#

have u tested this on real hw?

flint idol Dec 23, 2024, 1:51 PM

#

Yes, I have no idea if it works

real pecan Dec 23, 2024, 1:51 PM

#

wdym

flint idol Dec 23, 2024, 1:51 PM

#

I tested it

#

but idk if it does its job correctly

#

it doesn't triple fault tho

real pecan Dec 23, 2024, 1:54 PM

#

but i guess it doesnt help with hangs?

flint idol Dec 23, 2024, 1:54 PM

#

nope

real pecan Dec 23, 2024, 1:54 PM

#

its probably nothing complex like that

#

some simple bug

#

maybe obos using ecam to restore ecam type of bug

flint idol Dec 23, 2024, 1:55 PM

#

that was seabios

real pecan Dec 23, 2024, 1:55 PM

#

MTRRs are restored at this point right?

flint idol Dec 23, 2024, 1:55 PM

#

yes

real pecan Dec 23, 2024, 1:55 PM

#

I said type of bug not literal bug lol

flint idol Dec 23, 2024, 1:56 PM

#

what I have restored at that point:

tss/gdt/idt
LAPIC
caching stuff (MTTRs/PAT)
IOAPIC
NVS
HPET
SMP is restored
scheduler is alive
LAPIC timer is alive

#

maybe there is some obscure device I forgot about

#

that I need to restore

real pecan Dec 23, 2024, 1:57 PM

#

trace logs would really help

#

but alas

flint idol Dec 23, 2024, 1:58 PM

#

yeah

#

what I become the AML interpreter

#

I read _WAK until i see something fishy

real pecan Dec 23, 2024, 1:58 PM

#

u can try

#

it does way too much stuff

flint idol Dec 23, 2024, 1:59 PM

#

stores to regions should never do anything but write to memory/io, right?

#

e.g., \_SB.PCI0.GFX0.CLID = 0x80000000

#

that would only ever translate to uacpi writing to that memory/ioport, right?

real pecan Dec 23, 2024, 1:59 PM

#

possibly locking the global firmware lock

#

yes

flint idol Dec 23, 2024, 2:00 PM

#

maybe it would help to use uacpi before thread safety

#

if this truly is a deadlock

real pecan Dec 23, 2024, 2:00 PM

#

yeah

#

u can try an older uacpi

#

but the glock was supported before thread safety

real pecan Dec 23, 2024, 2:01 PM

#

flint idol e.g., `\_SB.PCI0.GFX0.CLID = 0x80000000`

is this on obos3

#

or i mean

#

dump 3

flint idol Dec 23, 2024, 2:01 PM

#

dump 3

#

laptop dump in _WAK

real pecan Dec 23, 2024, 2:02 PM

#

OperationRegion (IGDM, SystemMemory, ASLB, 0x2000)
Field (IGDM, AnyAcc, NoLock, Preserve)

Its a non-locked SystemMemory opregion

#

so its literally a volatile r/w

flint idol Dec 23, 2024, 2:03 PM

#

anyway, I'm going to send my uacpi version back in time

real pecan Dec 23, 2024, 2:03 PM

#

yeah

flint idol Dec 23, 2024, 2:03 PM

#

to before thread-safety-p2

#

or even before p1

#

if possible

real pecan Dec 23, 2024, 2:04 PM

#

oh btw

flint idol Dec 23, 2024, 2:04 PM

#

ye

real pecan Dec 23, 2024, 2:04 PM

#

it checks ECON

#

maybe disable your ec driver and retry

#

could be hanging in that

flint idol Dec 23, 2024, 2:05 PM

#

k

real pecan Dec 23, 2024, 2:06 PM

#

although

#

thats just a field in NVS lol

#

its not set by anyone

flint idol Dec 23, 2024, 2:06 PM

#

wAIT

#

wtf

#

not wtf

#

that fixed it

#

guess my EC driver is bad

real pecan Dec 23, 2024, 2:07 PM

#

yeah

#

nvm there is an ec opregion

#

and it does ECOK = 1

flint idol Dec 23, 2024, 2:07 PM

#

cok meme

real pecan Dec 23, 2024, 2:07 PM

#

so i geuss that updates NVS via SMM

#

so ur ec driver is just garbo

flint idol Dec 23, 2024, 2:08 PM

#

real pecan Dec 23, 2024, 2:08 PM

#

your ec also has queries

flint idol Dec 23, 2024, 2:08 PM

#

flint idol Dec 23, 2024, 2:08 PM

#

real pecan your ec also has queries

sounds cool

real pecan Dec 23, 2024, 2:08 PM

#

            Method (_Q27, 0, NotSerialized)  // _Qxx: EC Query, xx=0x00-0xFF
            {
                P80H = 0x27
                Notify (ACAD, 0x80) // Status Change
                Notify (BAT1, 0x80) // Status Change
            }

#

u should get events for power plug/unplug

#

u support events right?

flint idol Dec 23, 2024, 2:09 PM

#

never register any

real pecan Dec 23, 2024, 2:09 PM

#

no like

#

u support running EC queries

#

right?

flint idol Dec 23, 2024, 2:09 PM

#

void onECQuery(uacpi_handle hnd)
{
    // Evaluate whatever the EC wants us to evaluate.
    uint8_t idx = (uint8_t)(uintptr_t)hnd;
    char method[] = { '_', 'Q', 0,0,0 };
    snprintf(method+2, 3, "%02x", idx);
    uacpi_eval_simple(ec_node, method, nullptr);
    uacpi_finish_handling_gpe(ec_gpe_node, ec_gpe_idx);
}```

#

yes

real pecan Dec 23, 2024, 2:10 PM

#

do u get any?

#

on that laptop

flint idol Dec 23, 2024, 2:10 PM

#

idk maybe

real pecan Dec 23, 2024, 2:10 PM

#

its 02X not x

flint idol Dec 23, 2024, 2:10 PM

#

thx

real pecan Dec 23, 2024, 2:11 PM

#

flint idol idk maybe

probably not

#

because u would see it in logs

#

so yeah fix your ec

flint idol Dec 23, 2024, 2:11 PM

#

static uint8_t ec_read_reg(struct acpi_gas* reg)
{
    uint64_t val = 0;
    // Wait until OBF goes to one.
    do {
        uacpi_gas_read(&ec_control_register, &val);
    } while (!(val & EC_OBF));
    // Read the register.
    uacpi_gas_read(reg, &val);
    return val;
}

static void ec_write_reg(struct acpi_gas* reg, uint8_t what)
{
    uint64_t val = 0;
    // Wait until IBF goes to zero.
    do {
        uacpi_gas_read(&ec_control_register, &val);
    } while (val & EC_IBF);

    // Write the register.
    uacpi_gas_write(reg, what);
}

static uint8_t ec_read(uint8_t offset)
{
    ec_write_reg(&ec_control_register, BD_EC);
    ec_write_reg(&ec_data_register, offset);
    return ec_read_reg(&ec_data_register);
}

static uacpi_status ec_write(uint8_t offset, uint8_t value)
{
    ec_write_reg(&ec_control_register, BD_EC);
    ec_write_reg(&ec_data_register, offset);
    ec_write_reg(&ec_data_register, value);
    return UACPI_STATUS_OK;
}```

#

this looks sane to me

real pecan Dec 23, 2024, 2:11 PM

#

no idea

flint idol Dec 23, 2024, 2:11 PM

#

this part

#

oops

#

I write burst disable to command

#

instead of ec read/ec write

#

silly me

flint idol Dec 23, 2024, 2:58 PM

#

idk why

#

but enabling burst in the ec driver

#

doesn't work

flint idol Dec 23, 2024, 3:26 PM

#

@real pecan what does UACPI_InternalOpUnresolvedNameString_L do

#

because obos hangs on it during EC init

#

at least taht's the last thing logged before a hang

#

nooo

real pecan Dec 23, 2024, 3:31 PM

#

probably read from a field that starts with an L

flint idol Dec 23, 2024, 3:31 PM

#

hmm

real pecan Dec 23, 2024, 3:31 PM

#

yeah

#

LDS0

#

Method (_REG, 2, NotSerialized)  // _REG: Region Availability
{
    If (((Arg0 == 0x03) && (Arg1 == One)))
    {
        ECOK = One
        If (ECOK)
        {
            If (LDS0)
            {

#

this is where u hang

flint idol Dec 23, 2024, 3:33 PM

#

infy has cameras in my house confirmed

real pecan Dec 23, 2024, 3:33 PM

#

no

#

im in your walls

flint idol Dec 23, 2024, 3:34 PM

#

it hung in INVOKE_HANDLER

flint idol Dec 23, 2024, 3:34 PM

#

real pecan im in your walls

oh hi then

#

can you come help me debug

#

then

real pecan Dec 23, 2024, 3:34 PM

#

invoke handler just calls into your ec_read

real pecan Dec 23, 2024, 3:34 PM

#

flint idol can you come help me debug

im busy

flint idol Dec 23, 2024, 3:34 PM

#

in my walls?

real pecan Dec 23, 2024, 3:34 PM

#

yes

#

i have the entire house to watch

flint idol Dec 23, 2024, 3:35 PM

#

understandable

thick jolt Dec 23, 2024, 3:51 PM

#

flint idol

nyaux code be liek

flint idol Dec 23, 2024, 3:57 PM

#

it hangs waiting for EC_OBF

short mortar Dec 23, 2024, 3:57 PM

#

uint64_t?

flint idol Dec 23, 2024, 3:57 PM

#

yeah what about it

short mortar Dec 23, 2024, 3:58 PM

#

Registers are 8 bits

flint idol Dec 23, 2024, 3:58 PM

#

yeah

#

but gas_read takes a uint64_t*

#

also I just found the bug

#

I was doing reg = uacpi_gas_read(reg, &val)

#

whoops

#

ec driver still hangs on wake from suspend

flint idol Dec 23, 2024, 4:04 PM

#

flint idol doesn't work

yeah the ec never ACKs it

#

so I worked around that

real pecan Dec 23, 2024, 4:49 PM

#

lol maybe fix the driver instead

flint idol Dec 23, 2024, 5:13 PM

#

real pecan lol maybe fix the driver instead

I do everything I need to do for a burst, so idk

#

could be broke firmware

real pecan Dec 23, 2024, 5:20 PM

#

x to doubt

#

but check linux quirk table

flint idol Dec 23, 2024, 5:20 PM

#

sure

real pecan Dec 23, 2024, 5:22 PM

#

do u get events now at least

flint idol Dec 23, 2024, 7:37 PM

#

real pecan do u get events now at least

no

#

unless

#

these look like lid events

#

at one point I get an ec event

#

which fails for reasons

#

fuck

#

why isn't my laptop booting

#

nvm

flint idol Dec 23, 2024, 7:46 PM

#

flint idol which fails for reasons

was spurious

#

according to my logs

#

linux' ec driver makes no sense

white mulch Dec 23, 2024, 7:58 PM

#

at least you are not getting 5 random power button ec querys when you start your kernel like I do on my laptop troll

flint idol Dec 23, 2024, 7:58 PM

#

at least you don't get an ec query every second on your laptop

#

after fixing a deadlock bug with ec events

#

(me right now)

white mulch Dec 23, 2024, 7:59 PM

#

white mulch at least you are not getting 5 random power button ec querys when you start your...

idk whats up with that, I didn't find any quirks within linux for that laptop but what I ended up doing anyway was just flushing ec events to the void at the start until it no longer gives anything to get rid of the spurious power button presses

white mulch Dec 23, 2024, 8:00 PM

#

flint idol at least you don't get an ec query every second on your laptop

I mean its possible that you get them in every few seconds but yeah every second sounds a lot

flint idol Dec 23, 2024, 8:00 PM

#

and each time it

#

is spurious

#

i.e., index=0

#

hmm

#

oopsie

#

still not fixed

#

but I did find a bug

#

unless...

real pecan Dec 23, 2024, 8:05 PM

#

white mulch at least you are not getting 5 random power button ec querys when you start your...

does that happen with uacpi as well?

white mulch Dec 23, 2024, 8:06 PM

#

I am not sure, I haven't booted any uacpi kernel on it enough to be able to tell as it only happens sometimes

real pecan Dec 23, 2024, 8:07 PM

#

sad

white mulch Dec 23, 2024, 8:10 PM

#

I can try to boot managarm on it for a few tens of times sometime to see

flint idol Dec 23, 2024, 8:11 PM

#

damn the EC

real pecan Dec 23, 2024, 8:12 PM

#

or pmos

flint idol Dec 23, 2024, 8:15 PM

#

fnuy field name

weary hound Dec 23, 2024, 8:16 PM

#

cunter

#

CNTR and CTR was too hard

flint idol Dec 23, 2024, 8:16 PM

#

maybe it was first named COUNT

#

but preprocessed to CUNT

#

automod flagged a message sent by oberrow

weary hound Dec 23, 2024, 8:17 PM

#

i fail to see how that would happen

#

COUN would be more likely in that case

flint idol Dec 23, 2024, 8:18 PM

#

preprocessor that removes the 2nd letter of a name until it becomes four letters

#

@short mortar did your ec driver not give events?

#

and if so what wsa the problem

short mortar Dec 23, 2024, 8:23 PM

#

Or we can just assume it was intentional

flint idol Dec 23, 2024, 8:23 PM

#

maybe it's the same problem for me

short mortar Dec 23, 2024, 8:24 PM

#

flint idol and if so what wsa the problem

Idk some of the stuff was wrong

flint idol Dec 23, 2024, 8:24 PM

#

the only events I am getting are spurious

short mortar Dec 23, 2024, 8:25 PM

#

short mortar Idk some of the stuff was wrong

But it was hanging so that's why

#

And then it worked fine

flint idol Dec 23, 2024, 8:26 PM

#

nooo

short mortar Dec 23, 2024, 8:26 PM

#

I'll add framebuffer to kernel

thick jolt Dec 23, 2024, 8:33 PM

#

real pecan <:sad:1291432960947650643>

THATS ME THATS ME THATS ME

#

wait this is better

flint idol Dec 23, 2024, 8:33 PM

#

ok

thick jolt Dec 23, 2024, 8:33 PM

#

sorry

flint idol Dec 23, 2024, 8:33 PM

#

np

thick jolt Dec 23, 2024, 8:34 PM

#

can u tell me what i did wrong so i know not to do in future

flint idol Dec 23, 2024, 8:34 PM

#

you did nothing wrong

thick jolt Dec 23, 2024, 8:34 PM

#

oh okay 😭

flint idol Dec 23, 2024, 8:34 PM

#

otherwise I wouldn't be saying "np"

#

:)

thick jolt Dec 23, 2024, 8:34 PM

#

😭 😭

#

fair

flint idol Dec 23, 2024, 8:34 PM

#

(if it were a problem, I'd say so)

#

I HATE ACPI

#

I HATE ACPI

#

I HATE ACPI

#

I HATE ACPI

#

I HATE ACPI

quaint rivetBOT Dec 23, 2024, 8:34 PM

#

Anti-spam Message

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

thick jolt Dec 23, 2024, 8:36 PM

#

@white mulch

#

@real pecan

flint idol Dec 23, 2024, 8:37 PM

#

thanks

white mulch Dec 23, 2024, 8:37 PM

#

np

flint idol Dec 23, 2024, 8:40 PM

#

wth

#

why is the ec driver giving me 0x00 after enabling burst

#

seems pretty fishy

#

OSPM driver writes the Burst Enable Embedded Controller, BE_EC (0x82) command byte and then the Embedded Controller will prepare to enter the Burst mode. This includes processing any routine activities such that it should be able to remain dedicated to OSPM interface for ~ 1 microsecond.

#

    poll_ibf(device);
    write_reg(&device->control, BE_EC);
    poll_obf(device);
    uint8_t status = read_reg(&device->data);
    if (status != BURST_ACK)
        printf("Failed to enable EC burst mode. Out: %hhx\n", status);

pmOS' code

#

    ec_write_reg(&ec_control_register, BE_EC);
    uint8_t response = ec_read_reg(&ec_data_register);
    if (response != BURST_ACK)
    {
        OBOS_Warning("ACPI: Burst not acknoledged by EC, ignoring. Expected: 0x%02x, got 0x%02x\n", BURST_ACK, response);
        return false;
    }
    return true;```

#

obos' code

#

ec_write_reg/ec_read_reg poll ibf/obf

short mortar Dec 23, 2024, 8:56 PM

#

You need to poll obf before reading anything

flint idol Dec 23, 2024, 8:56 PM

#

I do

#

it's implicit

short mortar Dec 23, 2024, 8:56 PM

#

Oh

flint idol Dec 23, 2024, 9:14 PM

#

@real pecan when uacpi reads from a gas, the raw io port read returns 0x801a56a0, among other values, yet the actual gas read returns 0x00, do you have any idea how this could be?

real pecan Dec 23, 2024, 9:14 PM

#

not sure what that means

flint idol Dec 23, 2024, 9:14 PM

#

so basically

#

my ec driver reads from the gas

#

when uacpi actually reads the io port, the value returned is 0x801a56a0

#

and some other values occasionally

#

but the value given to me by uacpi is 0x00

#

the gas format is:
register bit offset: 0
register bit width: 8
access size: 0
address: 0x62

real pecan Dec 23, 2024, 9:15 PM

#

how can this be the value returned if port width is 1 byte

flint idol Dec 23, 2024, 9:15 PM

#

idk

#

wait I'm stupid

#

I was printing a pointer

real pecan Dec 23, 2024, 9:16 PM

#

no like what are u reading even

#

lmao

flint idol Dec 23, 2024, 9:16 PM

#

real pecan no like what are u reading even

the ec status register

real pecan Dec 23, 2024, 9:17 PM

#

i meant the fact that its a 4 byte value

flint idol Dec 23, 2024, 9:17 PM

#

it isn't, I was printing a pointer by accident

real pecan Dec 23, 2024, 9:17 PM

#

💀

flint idol Dec 23, 2024, 9:19 PM

#

uacpi gets 0xb2 from the raw io read

#

but my ec driver gets 0x00

real pecan Dec 23, 2024, 9:20 PM

#

show code

flint idol Dec 23, 2024, 9:21 PM

#

static uint8_t ec_read_reg(struct acpi_gas* reg)
{
    uint64_t val = 0;
    // Wait until OBF goes to one.
    ec_wait_for_bit(&ec_control_register, EC_OBF, EC_OBF);
    // Read the register.
    uacpi_gas_read(reg, &val);
    return val;
}```

#

reg = ec control register

real pecan Dec 23, 2024, 9:21 PM

#

check the return status as well

flint idol Dec 23, 2024, 9:21 PM

#

flint idol the gas format is: register bit offset: 0 register bit width: 8 access size: 0 a...

with this format

flint idol Dec 23, 2024, 9:21 PM

#

real pecan check the return status as well

good idea

real pecan Dec 23, 2024, 9:21 PM

#

it sounds like its not even initialized

flint idol Dec 23, 2024, 9:22 PM

#

what isn't?

real pecan Dec 23, 2024, 9:22 PM

#

val

flint idol Dec 23, 2024, 9:23 PM

#

it fails

real pecan Dec 23, 2024, 9:23 PM

#

lmao

flint idol Dec 23, 2024, 9:24 PM

#

not found

real pecan Dec 23, 2024, 9:24 PM

#

    if (!gas->address)
        return UACPI_STATUS_NOT_FOUND;

flint idol Dec 23, 2024, 9:25 PM

#

weird

#

omg I am stupid

#

I was incrementing a variable called current_index in the resource iterate callback

#

to figure out if this is the control register's resource or data register's

#

resource

#

I incremented it

#

before

#

switch (*current_index)

#

upon fixing this bug

#

suspend breaks

#

it hangs before going into S3

#

and

#

I stop getting all those spurious events

real pecan Dec 23, 2024, 9:36 PM

#

linux disables ec interrupts prior to going into s3

#

i think

#

maybe ur hanging in that

flint idol Dec 23, 2024, 9:41 PM

#

well it hangs polling IBF

short mortar Dec 23, 2024, 9:51 PM

#

Can you try this?

📎 pmOS.iso

#

It has a framebuffer in kernel

flint idol Dec 23, 2024, 9:51 PM

#

k

#

it logs shit quickly

#

then the screen goes blank

#

with a cursor

#

at the last line

short mortar Dec 23, 2024, 9:52 PM

#

Can you record what it logs?

flint idol Dec 23, 2024, 9:52 PM

#

Scrolling bug

#

https://tenor.com/view/sus-cat-cat-stare-stare-bruh-cat-really-n-gif-25597234

Tenor

short mortar Dec 23, 2024, 9:52 PM

#

If the screen goes blank then it enters userspace

#

My EC driver might also be hanging...

#

I know what I can try

#

📎 pmOS.iso

#

@flint idol

flint idol Dec 23, 2024, 10:01 PM

#

Sure

real pecan Dec 23, 2024, 10:07 PM

#

there goes pmos stability LULW

short mortar Dec 23, 2024, 10:08 PM

#

The kernel did boot it seems

real pecan Dec 23, 2024, 10:11 PM

#

jkjk

short mortar Dec 23, 2024, 10:13 PM

#

oberrow laptop's fault KEKW

real pecan Dec 23, 2024, 10:14 PM

#

linux doesnt seem to have any quirks for ecs that hang on poll or dont respond with acks

#

whats that laptops model

short mortar Dec 23, 2024, 10:16 PM

#

pmos might hang because of spinlocks skill issue

flint idol Dec 23, 2024, 10:27 PM

#

short mortar pmos might hang because of spinlocks skill issue

ur terminal is fucked

short mortar Dec 23, 2024, 10:27 PM

#

wdym

#

It's flanterm

flint idol Dec 23, 2024, 10:28 PM

#

rn_image_picker_lib_temp_539a5c40-9fb2-4262-8e5d-c9b6cec52df4.jpg

#

then the rest is a blank screen until a cursor on the last line

short mortar Dec 23, 2024, 10:31 PM

#

Does it turn off with power button?

flint idol Dec 23, 2024, 10:31 PM

#

no

#

BUT

#

it does seem to "enter userspace/idle"

#

oh shi

#

it does

#

it takes like 5 seconds

short mortar Dec 23, 2024, 10:33 PM

#

Yeah it waits for 3 seconds

#

So EC potentially works? lol

#

And framebuffer is fucked

flint idol Dec 23, 2024, 10:34 PM

#

I need logs from pmOS to help me debug, but alas

short mortar Dec 23, 2024, 10:34 PM

#

It also logs to serial

flint idol Dec 23, 2024, 10:35 PM

#

..which I don't have

short mortar Dec 23, 2024, 10:35 PM

#

yeah

flint idol Dec 23, 2024, 10:35 PM

#

does it log to xhci halfmemeright

short mortar Dec 23, 2024, 10:35 PM

#

Your power button is through EC, right?

flint idol Dec 23, 2024, 10:35 PM

#

think so

short mortar Dec 23, 2024, 10:35 PM

#

flint idol does it log to xhci <:halfmemeright:715831702164996096>

I can implement that meme

flint idol Dec 23, 2024, 10:35 PM

#

flint idol think so

maybe it also has fixed event

flint idol Dec 23, 2024, 10:35 PM

#

short mortar I can implement that <:meme:575445094589661215>

wait seriously

short mortar Dec 23, 2024, 10:36 PM

#

I don't have usb

#

So like in a month

flint idol Dec 23, 2024, 10:36 PM

#

I think that's a bit too far for me

#

so back to printf debugging I go

short mortar Dec 23, 2024, 10:36 PM

#

Idk what's going on with my framebuffer

flint idol Dec 23, 2024, 10:37 PM

#

someone needs to make uUSB

#

why is it hanging nooo

short mortar Dec 23, 2024, 10:44 PM

#

Could this be wrong?

flint idol Dec 23, 2024, 10:44 PM

#

lgtm

short mortar Dec 23, 2024, 10:44 PM

#

If bpp is less than 8

flint idol Dec 23, 2024, 10:44 PM

#

it isn't

short mortar Dec 23, 2024, 10:45 PM

#

wtf why would it not work

flint idol Dec 23, 2024, 10:45 PM

#

microkernel skill issue

short mortar Dec 23, 2024, 10:45 PM

#

No

flint idol Dec 23, 2024, 10:45 PM

#

meme

short mortar Dec 23, 2024, 10:45 PM

#

microkernel booted

#

It's userspace skill issue

#

Oh right

flint idol Dec 23, 2024, 10:48 PM

#

it hangs waiting on OBF

#

nooo

#

in obos

short mortar Dec 23, 2024, 10:48 PM

#

flint idol does it log to xhci <:halfmemeright:715831702164996096>

How do you log onto that?

#

USB serial controllers?

flint idol Dec 23, 2024, 10:49 PM

#

I guess

short mortar Dec 23, 2024, 10:49 PM

#

Like I have a couple of those...

#

Maybe this will work...

📎 pmOS.iso

flint idol Dec 23, 2024, 10:51 PM

#

now the screen goes blank

#

but the power button still works

short mortar Dec 23, 2024, 10:51 PM

#

I've removed kernel framebuffer...

#

Could it be because of how I map framebuffer?

#

No cache

flint idol Dec 23, 2024, 11:04 PM

#

no

short mortar Dec 23, 2024, 11:05 PM

#

I can do other thing

#

Disable userspace framebuffer

#

📎 pmOS.iso

#

@flint idol

flint idol Dec 23, 2024, 11:13 PM

#

short mortar <@841762704766140486>

woo

#

I get logs

#

from kernel

short mortar Dec 23, 2024, 11:14 PM

#

What does it print

flint idol Dec 23, 2024, 11:15 PM

#

cpu x entering userspace/idle

#

three times

#

then a bunch of tasks exit

#

then it stops

short mortar Dec 23, 2024, 11:18 PM

#

Which tasks exit

short mortar Dec 23, 2024, 11:18 PM

#

flint idol then it stops

It's normal

flint idol Dec 24, 2024, 12:30 AM

#

wait why is my mmap 282 loc

#

+65 loc to find an avaliable address

#

I wonder if it could help (for profiling purposes) to store how long locks are held in stuff like spinlocks and mutexes

#

and pushlocks

flint idol Dec 24, 2024, 1:03 AM

#

obos surprisingly doesn't have too much spinlocks

#

29 spinlocks in total

#

18 mutexes

#

or well 16

#

or well

#

maybe a bit less than these numbers because uacpi kernel api

#

2 semaphores

#

2 rw locks

#

and like three event objects

flint idol Dec 24, 2024, 1:15 AM

#

flint idol I wonder if it could help (for profiling purposes) to store how long locks are h...

I added uint64_t lastLockTimeNS; fields to spinlocks/mutexes

#

this field has the time in nanoseconds of how long the lock was last taken

#

as in, the time between acquire and release, in nanoseconds

#

and it's configurable the config option OBOS_ENABLE_LOCK_PROFILING

#

which configures if it is enabled or not, since it has a big perf hit

#

which is visible

flint idol Dec 24, 2024, 12:39 PM

#

Another day, another gruesome obos debug session

#

I have this crippling fear that I'm gonna remove some wait by accident in the EC driver then accidentally corrupt some proprietary bios state causing the system to hang in the firmware

short mortar Dec 24, 2024, 12:49 PM

#

#1217009725711847465 message

real pecan Dec 24, 2024, 12:51 PM

#

yeah its easy to corrupt firmware with a bad ec driver

#

but it fixes itself after power loss

flint idol Dec 24, 2024, 12:51 PM

#

this shit is so stupid

#

why does it just hang there

#

waiting for OBF

real pecan Dec 24, 2024, 12:52 PM

#

skill issue

#

do u have locking around EC?

flint idol Dec 24, 2024, 12:52 PM

#

almost

real pecan Dec 24, 2024, 12:52 PM

#

because u might get an event at the same time as opregion access

#

i had to fix that for managarm

flint idol Dec 24, 2024, 12:57 PM

#

I added some locking

#

and now it hangs somewhere later

real pecan Dec 24, 2024, 12:57 PM

#

lol

flint idol Dec 24, 2024, 1:00 PM

#

so the read completes then

#

it hangs reading from a gas

#

or well

#

hangs waiting for OBF to become one

short mortar Dec 24, 2024, 1:05 PM

#

I have locks in EC

#

I need to fix the framebuffer...

flint idol Dec 24, 2024, 1:06 PM

#

I might know the problem

#

I wait for OBF trying to read the status register

#

in the control register

#

meaning there isn't any data for me to read

#

from any pending commands

short mortar Dec 24, 2024, 1:07 PM

#

I think you read the control whenever you want

flint idol Dec 24, 2024, 1:07 PM

#

yeah

#

is it the same with writes?

#

I think so..

short mortar Dec 24, 2024, 1:07 PM

#

No

flint idol Dec 24, 2024, 1:07 PM

#

ok

short mortar Dec 24, 2024, 1:07 PM

#

As I understood it

#

But it doesn't hurt to wait

flint idol Dec 24, 2024, 1:08 PM

#

it probably does

short mortar Dec 24, 2024, 1:09 PM

#

No but if the control register isn't buffered the ibf won't be set

#

or obf

#

whichever is host to ec

flint idol Dec 24, 2024, 1:10 PM

#

OBF is ec->host

#

IBF is host->ec

short mortar Dec 24, 2024, 1:11 PM

#

But as I understood it, there are 2 registers

#

Host to EC and EC to host

#

When you write to EC, the register is written to and the bit is set

flint idol Dec 24, 2024, 1:11 PM

#

no, the two registers are EC command/status

#

and EC data

short mortar Dec 24, 2024, 1:11 PM

#

untill it is processed

real pecan Dec 24, 2024, 1:11 PM

#

its actually 4 registers

short mortar Dec 24, 2024, 1:11 PM

#

flint idol no, the two registers are EC command/status

Internal registers that is

real pecan Dec 24, 2024, 1:11 PM

#

its 2 but virtually split into 4

short mortar Dec 24, 2024, 1:12 PM

#

Like length 1 buffer

real pecan Dec 24, 2024, 1:12 PM

#

because reading and writing do completely unrelated things

short mortar Dec 24, 2024, 1:12 PM

#

Anyway

flint idol Dec 24, 2024, 1:12 PM

#

fuck

short mortar Dec 24, 2024, 1:12 PM

#

There is buffer for host to device stuff and there is buffer for device to host stuff

flint idol Dec 24, 2024, 1:12 PM

#

why didn't my laptop boot

vale nymph Dec 24, 2024, 1:13 PM

#

Brickes

#

Bricked

flint idol Dec 24, 2024, 1:13 PM

#

all is fine now

short mortar Dec 24, 2024, 1:13 PM

#

💀

flint idol Dec 24, 2024, 1:13 PM

#

yay

#

ec driver is fixed finally I think

vale nymph Dec 24, 2024, 1:13 PM

#

Ill buy your bricked laptop for 1 dollar

flint idol Dec 24, 2024, 1:15 PM

#

when I plug in and plug out my laptop I get EC evetns/!111!11!

short mortar Dec 24, 2024, 1:15 PM

#

what was it?

#

and is power button with events?

flint idol Dec 24, 2024, 1:15 PM

#

I don't think so

#

I think my laptop uses fixed events for htat

real pecan Dec 24, 2024, 1:16 PM

#

what about suspend

flint idol Dec 24, 2024, 1:16 PM

#

it wakes successfully

real pecan Dec 24, 2024, 1:16 PM

#

damn congrats

flint idol Dec 24, 2024, 1:16 PM

#

and no longer hangs in the ec driver

real pecan Dec 24, 2024, 1:16 PM

#

big work

short mortar Dec 24, 2024, 1:16 PM

#

that's cool

flint idol Dec 24, 2024, 1:17 PM

#

indeed

short mortar Dec 24, 2024, 1:17 PM

#

second working uACPI reference EC driver

#

(or third)

flint idol Dec 24, 2024, 1:17 PM

#

does managarm's driver not work

#

yeah

real pecan Dec 24, 2024, 1:18 PM

#

third

flint idol Dec 24, 2024, 1:18 PM

#

good enough for me

real pecan Dec 24, 2024, 1:18 PM

#

even your nvs restore works i guess

flint idol Dec 24, 2024, 1:18 PM

#

I guess

#

no way to work

#

I meant

#

no way to know

real pecan Dec 24, 2024, 1:20 PM

#

are u going to make an article about suspend to s3?

flint idol Dec 24, 2024, 1:20 PM

#

Yes

real pecan Dec 24, 2024, 1:21 PM

#

nice

flint idol Dec 24, 2024, 1:24 PM

#

whoa

#

I get events when I close/open the lid

#

I get occasional Notifys abouts some device with the HID "VPC2004"

#

and even one from an "Ambient Light Sensor"

#

when I remove the battery, I get a Notify

#

@real pecan is there a way to get uacpi to stop spamming Notify warnings without raising the log level > UACPI_LOG_WARN

real pecan Dec 24, 2024, 1:33 PM

#

install a notify handler to root

flint idol Dec 24, 2024, 1:33 PM

#

thx

#

time to update uacpi

#

test

#

then push

#

bruh why is SMP so slow in obos

#

probably because of my TLB shootdowns

#

I need to make cancellable syscalls soon

#

TODO™️

#

theoretically, it is as shrimple as lowering the IRQL to PASSIVE

#

instead of DISPATCH

#

then also adding a way to cancel waits

#

which I have half of

#

the half I have is signalling the waiting threads

#

the half I need is getting it to notice that it needs to abort what it's doing

#

when it's awaken

flint idol Dec 24, 2024, 2:05 PM

#

flint idol theoretically, it is as shrimple as lowering the IRQL to PASSIVE

maybe that's not related

#

yeah why did I raise the irql to dispatch in my syscall handlers

#

surely there was a good reason to that?

#

if it don't crash when I remove it

#

then it's fine

#

https://tenor.com/view/pink-panther-code-explode-line-vscode-gif-12716566540642579632

Tenor

#

@real pecan I was trying to see if would get the framebuffer restored by firmware after wake from suspend if I set back my OSI to windows XP, but instead it deadlocks before suspend

#

in uacpi, ofc

#

at least I think in uacpi

#

    uacpi_bulk_configure_interfaces(UACPI_INTERFACE_ACTION_DISABLE, UACPI_INTERFACE_KIND_VENDOR);
    uacpi_install_interface("Windows 2001", UACPI_INTERFACE_KIND_VENDOR);```

real pecan Dec 24, 2024, 2:21 PM

#

where does it deadlock

flint idol Dec 24, 2024, 2:22 PM

#

idk, all I'm getting are some TIDs and memory addresses

#

of mutexes

real pecan Dec 24, 2024, 2:22 PM

#

well i've looked at those two functions and there's no way they can deadlock

flint idol Dec 24, 2024, 2:22 PM

#

yeah it's not those

#

this is before suspend

real pecan Dec 24, 2024, 2:23 PM

#

flint idol ```c uacpi_bulk_configure_interfaces(UACPI_INTERFACE_ACTION_DISABLE, UACPI_I...

u do this before namespace_initialize right

flint idol Dec 24, 2024, 2:23 PM

#

after having done that

flint idol Dec 24, 2024, 2:23 PM

#

real pecan u do this before namespace_initialize right

before load

real pecan Dec 24, 2024, 2:23 PM

#

ah ok

flint idol Dec 24, 2024, 2:23 PM

#

first it acquires two locks in a row on the same thread
then another thread deadlocks while trying to acquire the first mutex acquired

#

tid 2: waiting for tid 8 to release mutex x
tid 2: waiting for tid 8 to release mutex y
tid 9: waiting for tid 2 to release mutex x
DEADLOCK

#

but the weird thing is, tid 2 is CPU 0's idle thread

real pecan Dec 24, 2024, 2:24 PM

#

flint idol tid 2: waiting for tid 8 to release mutex x tid 2: waiting for tid 8 to release ...

what you've written out here is not a deadlock

flint idol Dec 24, 2024, 2:24 PM

#

whatever it is

#

it causes it to hang

real pecan Dec 24, 2024, 2:25 PM

#

flint idol but the weird thing is, tid 2 is CPU 0's idle thread

your per_cpu->tid bug?

flint idol Dec 24, 2024, 2:26 PM

#

hmm weird

#

a reboot causes it to PF
another reboot causes it to work as normal

#

but unfotunately, no framebuffer

real pecan Dec 24, 2024, 2:32 PM

#

no framebuffer after reboot that pfs?

flint idol Dec 24, 2024, 2:32 PM

#

nope

real pecan Dec 24, 2024, 2:33 PM

#

how do u know it pfs

flint idol Dec 24, 2024, 2:33 PM

#

it PFed before suspend

#

what if I become windows 2000

#

no framebuffer after suspend :(

#

anyway

#

once uDRM comes out, this problem will be solved

#

I guess it's time for me to make interruptible I/O

real pecan Dec 24, 2024, 2:43 PM

#

flint idol it PFed before suspend

where

flint idol Dec 24, 2024, 2:43 PM

#

idk

#

it's gone now tho

real pecan Dec 24, 2024, 2:43 PM

#

lol

flint idol Dec 24, 2024, 2:44 PM

#

flint idol I guess it's time for me to make interruptible I/O

my I/O system is kinda terrible

#

I think NT has something called an IRP which is basically an I/O request packet

#

and I like that idea

#

my current I/O system is:

io requested
forward to driver, this action is synchronous
you wanted async I/O? oh let me just start a worker thread for you and you give me an event object

#

wait you wanted progress on that async I/O? too bad so sad.

#

you wanted to abort an I/O operation, uhh sorry we don't do that here

#

which, imo, is quite shit

#

luckily, I don't have much I/O

#

in the kernel

#

most of it is abstracted behind the VFS

#

#polls message
pls vote

#

in the meantime while I wait for the poll

#

I will be implementing elf loading

#

which I do have

#

but that's for loading drivers

#

#

/// <summary>
/// Allocates space on a thread's stack. Calling this after the thread has started running can be catastrophic.
/// </summary>
/// <param name="ctx">The thread's context.</param>
/// <param name="size">The of the stack region.</param>
/// <param name="status">The status of the function.</param>
/// <returns>A pointer to the region..</returns>
OBOS_WEAK void* CoreS_ThreadAlloca(const thread_ctx* ctx, size_t size, obos_status *status);

#

I have made this amazing api for pushing stuff to the stack

#

there aren't any platforms where the stack grows upward, right?

vale nymph Dec 24, 2024, 3:22 PM

#

flint idol ```c /// <summary> /// Allocates space on a thread's stack. Calling this after t...

Defining a variable:

flint idol Dec 24, 2024, 3:22 PM

#

no

#

this is meant for stuff like process loading where I need to push auxillary values

#

before the process starts

#

void* CoreS_ThreadAlloca(const thread_ctx* ctx, size_t size, obos_status *status)
{
    void* ret = (void*)(ctx->sp -= size);
    return ret;
}```

#

That should be how it's implemented when the stack grows downward

#

void* CoreS_ThreadAlloca(const thread_ctx* ctx, size_t size, obos_status *status)
{
    void* ret = (void*)ctx->sp;
    ctx->sp += size;
    return ret;
}```
and that's for when it grows upward

#

If I ever need that for future reference

#

these functions are just pseudocode, as I need error checking

flint idol Dec 24, 2024, 4:35 PM

#

I theoretically have an elf loader now

#

except it uses an unimplemented function: Mm_MapViewOfUserMemory

#

which I will implement rn

flint idol Dec 24, 2024, 4:54 PM

#

I kinda completely forgot how my VMM works

#

(oops)

thick jolt Dec 24, 2024, 4:57 PM

#

lmao

flint idol Dec 24, 2024, 5:27 PM

#

which will be a problem

#

if I need to implement a new vmm function

#

SO what I'll do is copy and paste code

#

so this is supposed to be mapping a private user memory region ubase->ubase+size from user_ctx into the kernel's context as shared memory

#

at kbase

#

with prot flags protection

#

possibly locking the user pages, depending on lock_pages

#

                // File page.
                page_info info = {};
                MmS_QueryPageInfo(ctx->pt, (uintptr_t)reg->owner->data+currFileOff, &info, nullptr);
                if (flags & VMA_FLAGS_PRIVATE)
                {
                    // Private.
                    // Moooo (CoW)
                    cow = true;
                }
                isPresent = info.prot.present;
                page what = {.phys=info.phys};
                phys = isPresent ? RB_FIND(phys_page_tree, &Mm_PhysicalPages, &what) : nullptr;
                if (phys)
                    MmH_RefPage(phys);
                rng->cow = cow;
                if (cow)
                {
                    rng->un.cow_type = COW_SYMMETRIC;
                    pc_range->cow = true;
                    pc_range->un.cow_type = COW_SYMMETRIC;
                    pc_range->prot.rw = false;
                    info.prot.rw = false;
                }
                MmS_SetPageMapping(ctx->pt, &info, info.phys, false);```

#

this is what I do for mmap'ing files

#

and reading this I discover a possible bug

thick jolt Dec 24, 2024, 5:31 PM

#

lmaooo

flint idol Dec 24, 2024, 5:31 PM

#

but idk if it's ever going to happen given my current page cache setup

thick jolt Dec 24, 2024, 5:32 PM

#

do u want my pizza btw

flint idol Dec 24, 2024, 5:32 PM

#

yes

#

send it over

thick jolt Dec 24, 2024, 5:32 PM

#

my dad bought me pizza even tho i told him im not hungry

flint idol Dec 24, 2024, 5:32 PM

#

my address is 123 seasme street

thick jolt Dec 24, 2024, 5:32 PM

#

yes i will send it by mail

#

it will arrive rotten

#

trl

flint idol Dec 24, 2024, 5:32 PM

#

jk

#

I live at ||2 Rideau Street, Ottawa, Ontario, Canada||

thick jolt Dec 24, 2024, 5:33 PM

#

hi its me ur uber eats driver

#

open the door

flint idol Dec 24, 2024, 5:33 PM

#

search that address

#

maybe I'm not joking

thick jolt Dec 24, 2024, 5:33 PM

#

hold on

#

💀

#

ah yes easiest way to get shot

flint idol Dec 24, 2024, 5:34 PM

#

I live at the senate of canada

thick jolt Dec 24, 2024, 5:34 PM

#

meme

flint idol Dec 24, 2024, 5:34 PM

#

they've a website about my house

#

https://sencanada.ca/en/about/scb/

SenCanada

The Senate of Canada Building

thick jolt Dec 24, 2024, 5:34 PM

#

@astral mason

#

wheres ur managa

#

manga*

#

and ur os

thick jolt Dec 24, 2024, 5:35 PM

#

flint idol https://sencanada.ca/en/about/scb/

lovely

flint idol Dec 24, 2024, 5:35 PM

#

cc @astral mason

thick jolt Dec 24, 2024, 5:35 PM

#

i js hurt my bone

#

ow

#

very painful

flint idol Dec 24, 2024, 5:36 PM

#

is it broken?

thick jolt Dec 24, 2024, 5:36 PM

#

anyways work on vmm function

#

😡

flint idol Dec 24, 2024, 5:36 PM

#

no.

thick jolt Dec 24, 2024, 5:36 PM

#

flint idol is it broken?

no

flint idol Dec 24, 2024, 5:36 PM

#

then ur fine

thick jolt Dec 24, 2024, 5:36 PM

#

i need 3 million dollars

#

then ill be fine

#

coolmeme

flint idol Dec 24, 2024, 7:14 PM

#

right after I complete this function

#

I will be redoing TLB shootdowns

#

obos thread at 40 stars

flint idol Dec 24, 2024, 7:22 PM

#

flint idol and reading this I discover a possible bug

nah nvm

#

I can only work so much on obos in a day before getting bored

#

nooo

real pecan Dec 24, 2024, 7:43 PM

#

obores

thick jolt Dec 24, 2024, 8:48 PM

#

anything i can contribute to obos

#

anything SMALL

#

please

flint idol Dec 24, 2024, 8:48 PM

#

Let me see

thick jolt Dec 24, 2024, 8:48 PM

#

kk

weak kestrel Dec 24, 2024, 8:49 PM

#

thick jolt kk

src/oboskrnl/vfs/fd.c:152

#

double ;;

thick jolt Dec 24, 2024, 8:49 PM

#

weak kestrel src/oboskrnl/vfs/fd.c:152

oki i go fix

#

trl

#

hold on while i clone obos

weak kestrel Dec 24, 2024, 8:50 PM

#

also

thick jolt Dec 24, 2024, 8:50 PM

#

which branch is the right one

flint idol Dec 24, 2024, 8:50 PM

#

Clone userspace-work

thick jolt Dec 24, 2024, 8:50 PM

#

found it

flint idol Dec 24, 2024, 8:50 PM

#

weak kestrel also

Wha

flint idol Dec 24, 2024, 8:50 PM

#

thick jolt found it

I'll find you a TODO

thick jolt Dec 24, 2024, 8:50 PM

#

oki

flint idol Dec 24, 2024, 8:50 PM

#

that you can fulfill

#

oh dear

thick jolt Dec 24, 2024, 8:51 PM

#

just please dont give me stuff like suspend support

#

nooo

flint idol Dec 24, 2024, 8:51 PM

#

I already have that

#

you can do PCIe

#

should be simple enough

thick jolt Dec 24, 2024, 8:51 PM

#

ive never done pci let alone pcie before

flint idol Dec 24, 2024, 8:51 PM

#

I'll find something else then..

thick jolt Dec 24, 2024, 8:51 PM

#

okay

weak kestrel Dec 24, 2024, 8:51 PM

#

flint idol Wha

grep -nR ";;" .

flint idol Dec 24, 2024, 8:51 PM

#

you can implement interruptible waits

flint idol Dec 24, 2024, 8:52 PM

#

weak kestrel grep -nR ";;" .

oh

thick jolt Dec 24, 2024, 8:52 PM

#

flint idol you can implement interruptible waits

interruptible waits?

flint idol Dec 24, 2024, 8:52 PM

#

https://cdn.discordapp.com/emojis/816178972672131083.gif?size=64

flint idol Dec 24, 2024, 8:52 PM

#

thick jolt interruptible waits?

so you know when you block a thread for a specific purpose

#

you're usually blocked

#

until that thing is fulfilled

thick jolt Dec 24, 2024, 8:52 PM

#

yea?

flint idol Dec 24, 2024, 8:52 PM

#

I want you to make that interruptible from external sources

weak kestrel Dec 24, 2024, 8:53 PM

#

obos

flint idol Dec 24, 2024, 8:53 PM

#

the intended behaviour is that CoreS_WaitOnObject reports OBOS_STATUS_ABORTED

weak kestrel Dec 24, 2024, 8:53 PM

#

do you want multiple bugs introduced in your kernel

flint idol Dec 24, 2024, 8:53 PM

#

weak kestrel do you want multiple bugs introduced in your kernel

I am testing his code extensively

#

dw

thick jolt Dec 24, 2024, 8:54 PM

#

flint idol the intended behaviour is that CoreS_WaitOnObject reports `OBOS_STATUS_ABORTED`

can u point me to the file in which i probs need to edit?

vale nymph Dec 24, 2024, 8:54 PM

#

You dont even test yours extensively meme

flint idol Dec 24, 2024, 8:54 PM

#

vale nymph You dont even test yours extensively <:meme:575445094589661215>

exactly my point

flint idol Dec 24, 2024, 8:54 PM

#

thick jolt can u point me to the file in which i probs need to edit?

locks/wait.h

#

under src/oboskrnl

thick jolt Dec 24, 2024, 8:54 PM

#

okay

flint idol Dec 24, 2024, 8:54 PM

#

you might also need to make changes to mutexes, semaphores, and events

weak kestrel Dec 24, 2024, 8:54 PM

#

contribute to my kernel

flint idol Dec 24, 2024, 8:54 PM

#

which are also under locks/*

thick jolt Dec 24, 2024, 8:54 PM

#

okay

flint idol Dec 24, 2024, 8:55 PM

#

come back if you have a question on wtf something does

#

or when you're done

#

:)

thick jolt Dec 24, 2024, 8:56 PM

#

yea im so confused

weak kestrel Dec 24, 2024, 8:56 PM

#

weak kestrel contribute to my kernel

https://github.com/ilobilo/ilobilix/tree/rewrite

GitHub

GitHub - ilobilo/ilobilix at rewrite

Second attempt at making an os. Contribute to ilobilo/ilobilix development by creating an account on GitHub.

#

bigplead

flint idol Dec 24, 2024, 8:56 PM

#

thick jolt yea im so confused

my code is too powerful for you

#

meme

thick jolt Dec 24, 2024, 8:57 PM

#

so

#

when u want to wait on an object

#

or thread*

#

u add it to some kind of threadlist

#

am i understanding this

flint idol Dec 24, 2024, 8:57 PM

#

yup

thick jolt Dec 24, 2024, 8:57 PM

#

if i wanted to make this interruptable

#

hmmm

flint idol Dec 24, 2024, 8:57 PM

#

hint: CoreH_SignalWaitingThreads

weak kestrel Dec 24, 2024, 8:58 PM

#

thick jolt so

obosman I suggest you put some empty lines in your code

#

between different parts

flint idol Dec 24, 2024, 8:58 PM

#

weak kestrel obosman I suggest you put some empty lines in your code

I'm trying to in new code I write

#

and I do occasionally go around the codebase adding newlines

vale nymph Dec 24, 2024, 8:59 PM

#

thick jolt so

Are you sleeping with dispagch irql

flint idol Dec 24, 2024, 8:59 PM

#

flint idol hint: `CoreH_SignalWaitingThreads`

most of the infrastructure is already there, it's just that if you interrupt most waits, it falsely reports SUCCESS

vale nymph Dec 24, 2024, 8:59 PM

#

Oh dear

#

Also in one of the if cases you return with dispatch irql

flint idol Dec 24, 2024, 8:59 PM

#

where?

thick jolt Dec 24, 2024, 8:59 PM

#

flint idol most of the infrastructure is already there, it's just that if you interrupt mos...

so CoreH_SignalWaitingThreads is happening in an interrupt?

vale nymph Dec 24, 2024, 8:59 PM

#

The one right before the thread block

thick jolt Dec 24, 2024, 8:59 PM

#

is this correct

flint idol Dec 24, 2024, 9:00 PM

#

vale nymph The one right before the thread block

oh shit you're right

#

thanks mathew

#

nd

flint idol Dec 24, 2024, 9:00 PM

#

thick jolt so CoreH_SignalWaitingThreads is happening in an interrupt?

it happens on release of a lock, or to signal event objects

#

and iirc, it has to be called at <= IRQL_DISPATCH

#

or yells at your face

thick jolt Dec 24, 2024, 9:01 PM

#

i dont know what IRQL_DISPATCH is supposed to mean

#

sorry im asking a ton of stupid questions

flint idol Dec 24, 2024, 9:01 PM

#

nah it's fine

#

this is stranger code to you

#

IRQL_DISPATCH is where all scheduler IRQs are masked out

#

as well as timer IRQs

#

an IRQL is basically a level where certain IRQs are not executed

#

highest is 15 on x86-64, which means masked

thick jolt Dec 24, 2024, 9:02 PM

#

i see

flint idol Dec 24, 2024, 9:02 PM

#

lowest is PASSIVE (0) which means unmasked

#

enum
{
    IRQL_PASSIVE,
    IRQL_DISPATCH = 2,
    IRQL_TIMER = 2,
#if OBOS_ARCHITECTURE_HAS_ACPI
    IRQL_GPE = 8,
#endif
    IRQL_MASKED = 0xf,
}

#

this is the enum simplified

thick jolt Dec 24, 2024, 9:03 PM

#

i get you

vale nymph Dec 24, 2024, 9:03 PM

#

Its meant to be like the interrupt priority some cpus have

#

So its a per cpu thing

flint idol Dec 24, 2024, 9:03 PM

#

and at a higher IRQL, lower IRQL interrupts are also masked

thick jolt Dec 24, 2024, 9:03 PM

#

flint idol most of the infrastructure is already there, it's just that if you interrupt mos...

why is it falsely reporting success

flint idol Dec 24, 2024, 9:03 PM

#

because it doesn't expect to be interrupted except when it succeeds in getting what it wants

#

because signalling an object really just means waking an arbitrary thread in it

thick jolt Dec 24, 2024, 9:04 PM

#

how does obos know if its wait is getting interrupted exactly

flint idol Dec 24, 2024, 9:04 PM

#

flint idol because it doesn't expect to be interrupted except when it succeeds in getting w...

*signalled

#

how it knows is because it's in a list of threads in the object
and signalling an object just chooses an arbitrary thread to wake

#

so it's like:

waitforobject:
    append(obj->waiting_threads, curr)
    block(curr)
    return success
signalobject:
    thread towake = obj->waiting_threads.head
    remove(obj->waiting_threads, towake)
    ready(towake)
    return```

thick jolt Dec 24, 2024, 9:07 PM

#

mm okay i see

flint idol Dec 24, 2024, 9:07 PM

#

now signal object in the actual kernel can also signal all objects at once

#

as you'll able to see from the signature

thick jolt Dec 24, 2024, 9:07 PM

#

yea i can see that in the while loop

#

with the all bool

weak kestrel Dec 24, 2024, 9:08 PM

#

flint idol so it's like: ``` waitforobject: append(obj->waiting_threads, curr) bloc...

I think you're missing * after thread

#

meme

flint idol Dec 24, 2024, 9:08 PM

#

it's pseudocode

weak kestrel Dec 24, 2024, 9:08 PM

#

ik

thick jolt Dec 24, 2024, 9:09 PM

#

so what we need to do exactly is in waitforobject somehow check if we got signalled, then stop blocking the current thread and return an obos failure

#

am i understanding the problem correct

flint idol Dec 24, 2024, 9:09 PM

#

yup

thick jolt Dec 24, 2024, 9:09 PM

#

or does this sound like complete gibbberish

#

okay

#

is there like a function you already have that can check if the current thread is getting signalled?

#

like some callback or smthin

weak kestrel Dec 24, 2024, 9:09 PM

#

thick jolt so what we need to do exactly is in waitforobject somehow check if we got signal...

set a variable if cancelled and check for it

flint idol Dec 24, 2024, 9:09 PM

#

you specifically need to check if you got signalled and the object is still not ready

thick jolt Dec 24, 2024, 9:09 PM

#

flint idol you specifically need to check if you got signalled *and the object is still not...

yea

flint idol Dec 24, 2024, 9:10 PM

#

because you know you've been signalled as soon as you leave the CoreH_ThreadBlock

#

you need to make it figure out why, instead of it assuming blindly it was because it was signalled successfully

#

and then return the "why" as the status

#

so it's rather it returns OBOS_STATUS_SUCCESS because our wait was successful
or it returns OBOS_STATUS_ABORTED because our wait was aborted

thick jolt Dec 24, 2024, 9:12 PM

#

im probs gonna have to mess with this snippit of code here

#

the actual waiting part

flint idol Dec 24, 2024, 9:13 PM

#

that less so

#

you might need to add a new function

#

void abort_wait(struct waitable_header* obj);```

#

something like that

thick jolt Dec 24, 2024, 9:13 PM

#

why would i need the actualy object itself

#

if im checking IF I GOT

#

interrupted

flint idol Dec 24, 2024, 9:14 PM

#

well to add interruptible waits to the kernel, you first need a way to interrupt a wait

#

brb

thick jolt Dec 24, 2024, 9:15 PM

#

okay

flint idol Dec 24, 2024, 9:15 PM

#

the behaviour to interrupt a wait would probably be slightly different than that of the behaviour to signal a wait

#

but it's up to you to figure it out

thick jolt Dec 24, 2024, 9:16 PM

#

well

#

if i take the object

#

signal it but disable use signaled

#

maybe that could work?

#

no thats stupid

#OBOS (not vibecoded)