pmOS - microkernel OS for RISC-V, x86, and LoongArch | Operating System Development | Page 3

carmine nacelle Dec 19, 2024, 5:33 PM

#

Is it 3 banks of 32 (64) registers?

#

The driver should program the DebounceTmrOut/DebounceTmrOutUnit/DebounceCntrl bits according to
the data passed by BIOS through ASL code

carmine nacelle Dec 19, 2024, 6:06 PM

#

Is it me or is Linux's driver a bit fucked?

#

#

(AMD spec)

#

Thonk

#

This stuff is reserved...

#

Oh nevermind

#

I can't read

carmine nacelle Dec 19, 2024, 7:14 PM

#

@twilit talon the driver is cooking 🧑‍🍳

#

Btw, linux does this

if (pin <= 255) {
        char ev_name[8];
        sprintf(ev_name, "_%c%02X",
            agpio->triggering == ACPI_EDGE_SENSITIVE ? 'E' : 'L',
            pin);
        if (ACPI_SUCCESS(acpi_get_handle(handle, ev_name, &evt_handle)))
            handler = acpi_gpio_irq_handler;
    }
    if (!handler) {
        if (ACPI_SUCCESS(acpi_get_handle(handle, "_EVT", &evt_handle)))
            handler = acpi_gpio_irq_handler_evt;
    }

#

(the only part that's left is to call uacpi)

#

Although ACPI says

Interrupt based event signaling follows a similar methodology, a generic event device (GED) is declared which in turn
describes all interrupts associated with event generation. The interrupts are listed in a _CRS object. When an interrupt
is asserted the OSPM will execute the event method (_EVT) declared in the GED object specifying the interrupt
identifier as a parameter. In this way the interrupt can be associated with specific platform events

#

A GPIO controller is modeled as a device in the namespace, with _HID or _ADR and _CRS objects, at a minimum. Op-
tionally, the GPIO controller device scope may include GeneralPurposeIO OpRegion declarations ( Section 5.5.2.4.5)
and GPIO interrupt-to-ACPI Event mappings (Section 5.6.5.2). Note that for GPIO-signaled ACPI Events, the corre-
sponding event method (e.g. _Exx, _Lxx, or _EVT) must also appear in the target GPIO controller’s scope. For GPIO
event numbers larger than 255 (0xFF), the _EVT method is used.
Each pin on a GPIO Controller has a configuration (e.g. level-sensitive interrupt, de-bounced input, high-drive output,
etc.), which is described to OSPM in the GPIO Interrupt or GPIO IO Connection resources claimed by peripheral
devices or used in operation region accesses

twilit talon Dec 19, 2024, 7:38 PM

#

carmine nacelle

Damn

twilit talon Dec 19, 2024, 7:39 PM

#

carmine nacelle Btw, linux does this ``` if (pin <= 255) { char ev_name[8]; spri...

Looks correct

twilit talon Dec 19, 2024, 8:02 PM

#

carmine nacelle (the only part that's left is to call uacpi)

Did you figure it out?

twilit talon Dec 19, 2024, 8:07 PM

#

carmine nacelle

pin 0 looks correct according to your aml as well

carmine nacelle Dec 19, 2024, 8:14 PM

#

twilit talon Did you figure it out?

I went to eat

carmine nacelle Dec 19, 2024, 8:15 PM

#

twilit talon pin 0 looks correct according to your aml as well

I think they've connected it to gpio so it can have debounce

#

Because my laptop has its power button on its side for no reason

#

And it's acutally handled by the GPIO controller lol

#

Do I just call uacpi_eval with "_EVT(0)" ?

twilit talon Dec 19, 2024, 8:17 PM

#

yup

#

or u can do it like linux

twilit talon Dec 19, 2024, 8:18 PM

#

carmine nacelle Do I just call uacpi_eval with "_EVT(0)" ?

uacpi_execute

#

which has no ret value

carmine nacelle Dec 19, 2024, 8:19 PM

#

And _EXX and _LXX as well?

twilit talon Dec 19, 2024, 8:19 PM

#

i mean u see what linux does?

#

run _EXX or _LXX if one exists

#

otherwise run _EVT

carmine nacelle Dec 19, 2024, 8:20 PM

#

Yes

#

But do I also call execute with no arguments?

#

Like what's the difference between eval and execute

twilit talon Dec 19, 2024, 8:21 PM

#

just simpler signature

#

uacpi_status uacpi_eval_simple(
    uacpi_namespace_node *parent, const uacpi_char *path, uacpi_object **ret
)
{
    return uacpi_eval(parent, path, UACPI_NULL, ret);
}

uacpi_status uacpi_execute(
    uacpi_namespace_node *parent, const uacpi_char *path,
    const uacpi_object_array *args
)
{
    return uacpi_eval(parent, path, args, UACPI_NULL);
}

uacpi_status uacpi_execute_simple(
    uacpi_namespace_node *parent, const uacpi_char *path
)
{
    return uacpi_eval(parent, path, UACPI_NULL, UACPI_NULL);
}

#

does this help? LULW

carmine nacelle Dec 19, 2024, 8:22 PM

#

Bruh

#

Ok

#

So eval _EXX and if that fails execute _EVT

#

(since it's the same thing)

twilit talon Dec 19, 2024, 8:22 PM

#

is it edge triggered?

carmine nacelle Dec 19, 2024, 8:22 PM

#

Idk

#

It's given by AML

twilit talon Dec 19, 2024, 8:22 PM

#

agpio->triggering

carmine nacelle Dec 19, 2024, 8:22 PM

#

twilit talon Dec 19, 2024, 8:23 PM

#

yeah

#

so get that triggering value

carmine nacelle Dec 19, 2024, 8:23 PM

#

Lol, this means that the power button is literally almost wired to the interrupt line

twilit talon Dec 19, 2024, 8:23 PM

#

yes

carmine nacelle Dec 19, 2024, 8:23 PM

#

Linux does sprintf

twilit talon Dec 19, 2024, 8:23 PM

#

do u understand the logic now?

carmine nacelle Dec 19, 2024, 8:23 PM

#

Yes

twilit talon Dec 19, 2024, 8:23 PM

#

nice

carmine nacelle Dec 19, 2024, 8:23 PM

#

I just didn't know which functions to call

twilit talon Dec 19, 2024, 8:24 PM

#

_EVT is generic, aka a fallback

#

_LXX or _EXX are specialized

#

btw one extra difference

#

_L/EXX doesnt take any args

#

_EVT takes in an integer

#

so u want to pass it uacpi_object_create_integer(pin_idx)

carmine nacelle Dec 19, 2024, 8:26 PM

#

Yeah

#

Ok 10 minuts and this will be done

#

(famous last words)

twilit talon Dec 19, 2024, 8:27 PM

#

lmao

carmine nacelle Dec 19, 2024, 8:31 PM

#

Can I do some sort of get_handle/eval serapately, so I distinguish between non-existing nodes and errors?

twilit talon Dec 19, 2024, 8:31 PM

#

uacpi_namespace_node_find

carmine nacelle Dec 19, 2024, 8:31 PM

#

(though I guess I can do that by the return value)

twilit talon Dec 19, 2024, 8:31 PM

#

or that

carmine nacelle Dec 19, 2024, 8:32 PM

#

UACPI_STATUS_NOT_FOUND ?

#

Or no handler

#

UACPI_STATUS_NO_HANDLER

twilit talon Dec 19, 2024, 8:32 PM

#

why would it be no handler lol

carmine nacelle Dec 19, 2024, 8:32 PM

#

Idk because there's no handler found...................

twilit talon Dec 19, 2024, 8:32 PM

#

which handler

carmine nacelle Dec 19, 2024, 8:34 PM

#

_L00

#

(as an example)

twilit talon Dec 19, 2024, 8:34 PM

#

its edge triggered in your case

carmine nacelle Dec 19, 2024, 8:38 PM

#

Am I supposed to be calling uacpi_object_unref?

twilit talon Dec 19, 2024, 8:39 PM

#

on your created integer? yeah ofc

carmine nacelle Dec 19, 2024, 8:39 PM

#

twilit talon its edge triggered in your case

I'm doing sprintf

twilit talon Dec 19, 2024, 8:39 PM

#

carmine nacelle Am I supposed to be calling uacpi_object_unref?

(after the call right)

carmine nacelle Dec 19, 2024, 8:39 PM

#

(I think so)

#

(I'll push the code once I confirm it works)

#

bruh it's returning 6

#

(I know there is a to text function but I would be debugging things too quickly in that case)

twilit talon Dec 19, 2024, 8:43 PM

#

NOT_FOUND

carmine nacelle Dec 19, 2024, 8:43 PM

#

I think I know why...

#

I think it's working...

twilit talon Dec 19, 2024, 8:48 PM

#

holy shit

#

yes its working

#

Switch (ToInteger (Arg0))
{
    Case (0x00)
    {
        M000 (0x3900)
        M460 ("    Notify (\\_SB.PWRB, 0x80)\n", 0x00, 0x00, 0x00, 0x00, 0x00, 0x00)
        Local0 = \_SB.PCI0.LPC0.EC0.HWAK /* External reference */
        If (((Local0 & 0x04) == 0x04))
        {
            Local0 &= 0xFFFB
            \_SB.PCI0.LPC0.EC0.HWAK = Local0
            Notify (\_SB.LID0, 0x80) // Status Change
            Sleep (0x01F4)
        }
        ElseIf (((Local0 & 0x20) == 0x20))
        {
            Local0 &= 0xFFDF
            \_SB.PCI0.LPC0.EC0.HWAK = Local0
            Notify (\_SB.LID0, 0x80) // Status Change
            Sleep (0x01F4)
        }
        ElseIf (((Local0 & 0x50) == 0x50))
        {
            Local0 &= 0xFFAF
            \_SB.PCI0.LPC0.EC0.HWAK = Local0
            Sleep (0xC8)
            If ((\_SB.PCI0.LPC0.EC0.ADPT == 0x01))
            {
                Notify (\_SB.ADP0, 0x80) // Status Change
                Notify (\_SB.PCI0.LPC0.EC0.BAT0, 0x80) // Status Change
            }
            Else
            {
                Notify (\_SB.ADP0, 0x80) // Status Change
                Notify (\_SB.PCI0.LPC0.EC0.BAT0, 0x80) // Status Change
            }
        }
        ElseIf (((Local0 & 0x80) == 0x80))
        {
            Local0 &= 0xFF7F
            \_SB.PCI0.LPC0.EC0.HWAK = Local0
            Notify (\_SB.PCI0.LPC0.EC0.BAT0, 0x80) // Status Change
        }
        Else
        {
            Local0 &= 0xFFEF
            \_SB.PCI0.LPC0.EC0.HWAK = Local0
            Notify (\_SB.PWRB, 0x80) // Status Change
        }

#

Event this huge if is handled correctly

#

u get into the last branch

carmine nacelle Dec 19, 2024, 8:49 PM

#

Why wouldn't it be

twilit talon Dec 19, 2024, 8:50 PM

#

uacpi bugs potentially

#

this means all init up until this point was correct

#

pretty fucking cool

#

this is like cutting edge hw too

#

    /*
     * Modern hardware uses power button devices instead of the fixed event.
     * Search for them here and hook AML notifications.
     */
    uacpi_find_devices(ACPI_HID_POWER_BUTTON, [](void*, uacpi_namespace_node *node, uint32_t) {
        uacpi_install_notify_handler(node, handlePowerButtonNotify, nullptr);
        return UACPI_ITERATION_DECISION_CONTINUE;
    }, nullptr);

This is what's left for you

carmine nacelle Dec 19, 2024, 8:51 PM

#

I've pushed the code

#

Ok I'll try it

#

If it turn off it's game over

twilit talon Dec 19, 2024, 8:51 PM

#

carmine nacelle

@hexed acorn he actually got it

carmine nacelle Dec 19, 2024, 8:51 PM

#

📎 pmOS.iso

twilit talon Dec 19, 2024, 8:51 PM

#

try on your laptop ^^

twilit talon Dec 19, 2024, 8:52 PM

#

carmine nacelle I've pushed the code

nope

carmine nacelle Dec 19, 2024, 8:53 PM

#

twilit talon Dec 19, 2024, 8:53 PM

#

gitlab L

carmine nacelle Dec 19, 2024, 8:53 PM

#

Does gitlab use caching cdn?

#

Anyway, it's mirrored on github

#

https://gitlab.com/mishakov/pmos/-/blob/main/devicesd/generic/gpio/gpio.c?ref_type=heads

GitLab

devicesd/generic/gpio/gpio.c · main · Mikhail Kovalev / pmOS · GitL...

GitLab.com

twilit talon Dec 19, 2024, 8:54 PM

#

btw

#

u literally say write-1-to-clear

#

then proceed to &= it

carmine nacelle Dec 19, 2024, 8:54 PM

#

Where

twilit talon Dec 19, 2024, 8:55 PM

#

reg &= ~PIN_INTERRUPT_MASK;

carmine nacelle Dec 19, 2024, 8:55 PM

#

I'm not

twilit talon Dec 19, 2024, 8:56 PM

#

ah nvm

carmine nacelle Dec 19, 2024, 8:56 PM

#

It's only when !is_interrupt_pin(device, offset + j)

#

(influenced by linux)

#

Anyway

#

What did I want to do

carmine nacelle Dec 19, 2024, 8:56 PM

#

twilit talon ```cpp /* * Modern hardware uses power button devices instead of the fi...

Right

carmine nacelle Dec 19, 2024, 8:56 PM

#

twilit talon ```cpp /* * Modern hardware uses power button devices instead of the fi...

Oof C++

#

My ACPI server is written in C

#

(my OS is weird)

#

(because why not have C++ kernel and C userspace)

#

Ok I'm adding it

hexed acorn Dec 19, 2024, 8:57 PM

#

twilit talon try on your laptop ^^

i can try it in a moment

twilit talon Dec 19, 2024, 8:58 PM

#

- uacpi_resources *resources;
- uacpi_status result = uacpi_get_current_resources(node, &resources);
+ result = uacpi_for_each_device_resource(node, "_CRS", find_amd_gpio_resources, device);
if (result != UACPI_STATUS_OK) {
    printf("Failed to get resources for AMD GPIO device\n");
    goto fail;
}

- uacpi_for_each_resource(resources, find_amd_gpio_resources, device);
- uacpi_free_resources(resources);

carmine nacelle Dec 19, 2024, 8:58 PM

#

Ok

twilit talon Dec 19, 2024, 8:59 PM

#

very based work btw, first hobby AMD gpio driver lol

hexed acorn Dec 19, 2024, 8:59 PM

#

now i gotta one-up that and make the first hobby amd i2c driver halfmemeleft

#

which actually is not amd-specific and is just some designware ip iirc

twilit talon Dec 19, 2024, 9:00 PM

#

true

carmine nacelle Dec 19, 2024, 9:10 PM

#

twilit talon Dec 19, 2024, 9:10 PM

#

very very cool

carmine nacelle Dec 19, 2024, 9:10 PM

#

it shut down the laptop

twilit talon Dec 19, 2024, 9:10 PM

#

Can you update the wiki and add the third way lmao

twilit talon Dec 19, 2024, 9:11 PM

#

carmine nacelle it shut down the laptop

congrats, very good work

carmine nacelle Dec 19, 2024, 9:11 PM

#

fun

twilit talon Dec 19, 2024, 9:11 PM

#

wtf

carmine nacelle Dec 19, 2024, 9:11 PM

#

This needs fixing, but otherwise it works

twilit talon Dec 19, 2024, 9:11 PM

#

what is that interrupt?

carmine nacelle Dec 19, 2024, 9:11 PM

#

Gpio

#

Probably

twilit talon Dec 19, 2024, 9:11 PM

#

why is there no handler?

carmine nacelle Dec 19, 2024, 9:11 PM

#

Because of how my interrupts work

#

Because it configures lapic from userspace, and then installs the handle

twilit talon Dec 19, 2024, 9:12 PM

#

hm

carmine nacelle Dec 19, 2024, 9:12 PM

#

And in between that, it's getting interrupts

#

I need to find the interrupt disable bit I'd suppose

twilit talon Dec 19, 2024, 9:12 PM

#

btw while you're here can u test the wip uacpi PR real quick?

carmine nacelle Dec 19, 2024, 9:13 PM

#

Ok

twilit talon Dec 19, 2024, 9:13 PM

#

im about to merge it soon

#

if your test is succesful

#

git fetch && git checkout thread-safety-p3 in submodule

carmine nacelle Dec 19, 2024, 9:14 PM

#

One sec

twilit talon Dec 19, 2024, 9:14 PM

#

thanks

#

especially EC stuff

#

since that uses the infra that was changed a lot

carmine nacelle Dec 19, 2024, 9:16 PM

#

📎 pmOS.iso

#

(I'm restarting)

#

is this a good test?

#

Though all of the events run on CPU 0

twilit talon Dec 19, 2024, 9:21 PM

#

amazing, thanks

twilit talon Dec 19, 2024, 9:21 PM

#

carmine nacelle Though all of the events run on CPU 0

thats fine

#

i mostly wanted to make sure i didnt mess up by introducing deadlocks or anything like that

carmine nacelle Dec 19, 2024, 9:21 PM

#

And my scheduler quantum is 100ms (I think)

twilit talon Dec 19, 2024, 9:21 PM

#

carmine nacelle is this a good test?

first fully thread safe uacpi test coolmeme

carmine nacelle Dec 19, 2024, 9:22 PM

#

But I create a pthread per event

twilit talon Dec 19, 2024, 9:22 PM

#

carmine nacelle And my scheduler quantum is 100ms (I think)

u get quite a delay between plugging the charger and actually receiving a notify

#

i wonder why that is

carmine nacelle Dec 19, 2024, 9:22 PM

#

Yeah idk

twilit talon Dec 19, 2024, 9:22 PM

#

scheduling memes probably

carmine nacelle Dec 19, 2024, 9:23 PM

#

That might be because the kernel is so slow

twilit talon Dec 19, 2024, 9:23 PM

#

or debounce on hw side

carmine nacelle Dec 19, 2024, 9:23 PM

#

My copy to/from user functions need fixing

#

And they're called all the time for IPC

#

Did obos get fixed?

twilit talon Dec 19, 2024, 9:24 PM

#

fixed?

hexed acorn Dec 19, 2024, 9:28 PM

#

oh this reminds me, now that i've replaced the battery in my laptop i can actually test the charger plugged in/out events

carmine nacelle Dec 19, 2024, 10:07 PM

#

twilit talon fixed?

It wasn't booting on my laptop

twilit talon Dec 19, 2024, 10:07 PM

#

ah no i dont think so

carmine nacelle Dec 19, 2024, 10:07 PM

#

Thonk pmOS on hyper?

twilit talon Dec 19, 2024, 10:08 PM

#

hexed acorn oh this reminds me, now that i've replaced the battery in my laptop i can actual...

thats nice because im still working on event stuff (finishing up rn so ill appreciate more testing)

twilit talon Dec 19, 2024, 10:08 PM

#

carmine nacelle <:Thonk:1271231807945441402> pmOS on hyper?

i mean if u want lol

hexed acorn Dec 19, 2024, 10:08 PM

#

iirc they should be ec queries on my laptop

#

but we'll see

twilit talon Dec 19, 2024, 10:08 PM

#

they're ec queries on mishakov's laptop as well

carmine nacelle Dec 19, 2024, 10:09 PM

#

I want to try the old MacBook tomorrow

hexed acorn Dec 19, 2024, 10:12 PM

#

carmine nacelle

this should boot if i write it to an usb stick right?

#

asking since astral's iso didn't last time i tried it

vale hazel Dec 19, 2024, 10:13 PM

#

hexed acorn this should boot if i write it to an usb stick right?

hey u wanna try something

📎 NyauxKC-x86_64.iso

#

trl

#

daily nyaux shilling

twilit talon Dec 19, 2024, 10:14 PM

#

hexed acorn this should boot if i write it to an usb stick right?

need ventoy 💀

carmine nacelle Dec 19, 2024, 10:15 PM

#

hexed acorn this should boot if i write it to an usb stick right?

Yes

#

I'm using ventoy though

#

I think I had some bad luck booting it on either bios or EFI machines directly from USB drive

#

Btw is there managram iso?

#

I've only found a weird format image

hexed acorn Dec 19, 2024, 10:17 PM

#

yeah we have raw hard disk images only

#

it should in theory work if you just write it to a usb stick or a hard disk

#

assuming that xhci/sata/nvme drivers cooperate

carmine nacelle Dec 19, 2024, 10:18 PM

#

$(ISO): ../kernel/build/kernel ../bootstrapd/bootstrapd.elf ../devicesd/devicesd.elf ../logd/logd.elf ../ns16550/ns16550.elf ../piped/piped.elf ../processd/processd.elf ../terminald/terminald.elf ../vfsd/vfsd.elf ../vfatfsd/vfatfsd.elf ../ahcid/ahcid.elf limine.conf limine
    mkdir -p isodir/boot/limine
    mkdir -p isodir/EFI/BOOT

    cp limine.conf isodir/boot/limine/
    cp -v limine/limine-uefi-cd.bin isodir/boot/limine/
    cp -v limine/limine-bios-cd.bin isodir/boot/limine/
    cp -v limine/limine-bios.sys isodir/boot/limine/
    cp -v limine/BOOTRISCV64.EFI isodir/EFI/BOOT/
    cp -v limine/BOOTX64.EFI isodir/EFI/BOOT/

    cp -v ../kernel/build/kernel isodir/kernel

    cp -v ../bootstrapd/bootstrapd.elf isodir/
    cp -v ../devicesd/devicesd.elf isodir/
    cp -v ../logd/logd.elf isodir/
    cp -v ../ns16550/ns16550.elf isodir/
    cp -v ../piped/piped.elf isodir/
    cp -v ../processd/processd.elf isodir/
    cp -v ../terminald/terminald.elf isodir/
    cp -v ../vfsd/vfsd.elf isodir/
    cp -v ../vfatfsd/vfatfsd.elf isodir/
    cp -v ../ahcid/ahcid.elf isodir/
# cp -v ../test/test.elf isodir

    xorriso -as mkisofs -b boot/limine/limine-bios-cd.bin -no-emul-boot -boot-load-size 4 -boot-info-table --efi-boot boot/limine/limine-uefi-cd.bin -efi-boot-part --efi-boot-image --protective-msdos-label isodir -o $(ISO)

#

This is what I do

hexed acorn Dec 19, 2024, 10:19 PM

#

we don't have iso images mainly since we don't have an iso9660 driver so we couldn't mount the image (and we don't have overlayfs or anything so you'd be stuck with a read-only root)

carmine nacelle Dec 19, 2024, 10:19 PM

#

lol (reasonable)

twilit talon Dec 19, 2024, 10:41 PM

#

@carmine nacelle hey uh sorry for bothering u, I realized I had some possible races in the latest version and so I made some fixes, any way u can re-test with the same test?

carmine nacelle Dec 19, 2024, 10:42 PM

#

Ok

twilit talon Dec 19, 2024, 10:42 PM

#

nice

#

make sure u pull in the latest version since it was force-pushed

#

fetch && checkout should work

carmine nacelle Dec 19, 2024, 10:45 PM

#

Is this alr?

twilit talon Dec 19, 2024, 10:45 PM

#

carmine nacelle Is this alr?

yup

carmine nacelle Dec 19, 2024, 10:46 PM

#

📎 pmOS.iso

#

I think the battery is just slow

twilit talon Dec 19, 2024, 10:49 PM

#

nice

#

yeah probably

#

i mean when you're in windows, how long does it take in UI?

hoary moat Dec 19, 2024, 10:49 PM

#

twilit talon yeah probably

I can test it too if you want

twilit talon Dec 19, 2024, 10:49 PM

#

i dont think u use GPEs for anything?

hoary moat Dec 19, 2024, 10:50 PM

#

Wake from suspend

twilit talon Dec 19, 2024, 10:50 PM

#

oh yeah

#

yeah that would be nice

hoary moat Dec 19, 2024, 10:50 PM

#

Sure

carmine nacelle Dec 19, 2024, 10:50 PM

#

My laptop doesn't have s3

#

Could that be crashing obos?

hoary moat Dec 19, 2024, 10:50 PM

#

carmine nacelle Could that be crashing obos?

Impossible

#

It crashes before a new gdt is loaded

carmine nacelle Dec 19, 2024, 10:52 PM

#

twilit talon i mean when you're in windows, how long does it take in UI?

It also takes half a second, but it's hard to know for sure with windows 11

twilit talon Dec 19, 2024, 10:52 PM

#

yeah

carmine nacelle Dec 19, 2024, 10:52 PM

#

Maybe it's instant and gui is slow

twilit talon Dec 19, 2024, 10:53 PM

#

we can check with managarm

#

it should be more or less instant there

#

i mean the irq to work

hoary moat Dec 19, 2024, 10:53 PM

#

@twilit talon it worked on qemu

#

wake-from-suspend

twilit talon Dec 19, 2024, 10:53 PM

#

nice nice

hoary moat Dec 19, 2024, 10:53 PM

#

although maybe it's better to test on real hw

twilit talon Dec 19, 2024, 10:53 PM

#

does wake work on real hw?

hoary moat Dec 19, 2024, 10:53 PM

#

probably

twilit talon Dec 19, 2024, 10:54 PM

#

u never tested?

hoary moat Dec 19, 2024, 10:54 PM

#

I did test

#

but idk where it gets to

twilit talon Dec 19, 2024, 10:54 PM

#

ah right because gpu

#

it assumes you're windows 10 and that u can post the gpu back

hoary moat Dec 19, 2024, 10:54 PM

#

I will make it reboot after wake-from-suspend

hexed acorn Dec 19, 2024, 10:54 PM

#

i think the charger notifications also take like 1 or 2 seconds on my laptop as well

carmine nacelle Dec 19, 2024, 10:54 PM

#

with managram?

hexed acorn Dec 19, 2024, 10:55 PM

#

although that's from plugging/unplugging it to seeing the status change in the taskbar in kde

twilit talon Dec 19, 2024, 10:55 PM

#

did u test pmos?

hexed acorn Dec 19, 2024, 10:55 PM

#

i am currently glued to factorio i will test both pmos and managarm in a moment

carmine nacelle Dec 19, 2024, 10:55 PM

#

on my old laptop it would not update untill battery dipped by 1%

#

in linux

twilit talon Dec 19, 2024, 10:55 PM

#

probably because of a timer to re-read _BIF

#

but no actual notifications

carmine nacelle Dec 19, 2024, 10:56 PM

#

I think it was not recieving events

twilit talon Dec 19, 2024, 10:56 PM

#

yeah thats basically what i said

carmine nacelle Dec 19, 2024, 10:56 PM

#

ah

hoary moat Dec 19, 2024, 10:57 PM

#

twilit talon yeah thats basically what i said

It worked

hoary moat Dec 19, 2024, 10:57 PM

#

twilit talon it assumes you're windows 10 and that u can post the gpu back

What windows version makes it not assume that

twilit talon Dec 19, 2024, 10:58 PM

#

who knows

#

probably XP

twilit talon Dec 19, 2024, 10:58 PM

#

hoary moat It worked

what specifically

carmine nacelle Dec 19, 2024, 10:59 PM

#

It'll be known for sure once uDRM becomes usable 🙂

#

Which Windows even added suspend?

twilit talon Dec 19, 2024, 11:00 PM

#

suspend has existed since the inception of acpi

carmine nacelle Dec 19, 2024, 11:00 PM

#

So like 98?

twilit talon Dec 19, 2024, 11:01 PM

#

somewhere in the 90s

carmine nacelle Dec 19, 2024, 11:01 PM

#

ACPI was first published in 96 according to wikipedia

twilit talon Dec 19, 2024, 11:02 PM

#

yeah

#

back then it was APM or whatever

#

shitty bios helpers

hoary moat Dec 19, 2024, 11:02 PM

#

twilit talon what specifically

Wake from suspend

twilit talon Dec 19, 2024, 11:02 PM

#

on real hw?

hoary moat Dec 19, 2024, 11:02 PM

#

Yes

twilit talon Dec 19, 2024, 11:02 PM

#

damn obos cooking

carmine nacelle Dec 19, 2024, 11:03 PM

#

On my old motherboard, Windows could not be able to wake sometimes

#

And it would clear bios

twilit talon Dec 19, 2024, 11:03 PM

#

lmao

carmine nacelle Dec 19, 2024, 11:03 PM

#

b450

hoary moat Dec 19, 2024, 11:03 PM

#

twilit talon damn obos cooking

Yes

carmine nacelle Dec 19, 2024, 11:03 PM

#

So not even that old and exotic

#

Anyway

#

How do I implement working set nooo

hoary moat Dec 19, 2024, 11:04 PM

#

carmine nacelle How do I implement working set <:nooo:750466870993223750>

Find the paper

carmine nacelle Dec 19, 2024, 11:04 PM

#

I've read it

hoary moat Dec 19, 2024, 11:04 PM

#

Then you have list

#

Of pages you need to have swapped in to run a program efficiently

#

Search pager replacement algorithm for info on how to do that

#

*page

carmine nacelle Dec 19, 2024, 11:06 PM

#

I want to use it for disk caching

hoary moat Dec 19, 2024, 11:06 PM

#

Uh ok

carmine nacelle Dec 19, 2024, 11:06 PM

#

But idk if the same principle would apply

#

And just have all disks be virtually mapped into the kernel

hoary moat Dec 19, 2024, 11:07 PM

#

carmine nacelle And just have all disks be virtually mapped into the kernel

Doesn't sound like an amazing idea

carmine nacelle Dec 19, 2024, 11:07 PM

#

And then the algorithm could efficiently page everything in and out

hoary moat Dec 19, 2024, 11:07 PM

#

You're tryna make a page cache it seems, there are probably better eays

#

*ways

carmine nacelle Dec 19, 2024, 11:07 PM

#

yeah

hoary moat Dec 19, 2024, 11:07 PM

#

carmine nacelle And then the algorithm could efficiently page everything in and out

I also don't think you page out page cache

mellow sleet Dec 19, 2024, 11:07 PM

#

carmine nacelle And just have all disks be virtually mapped into the kernel

me omw to get a fuckton of disks and make it run out of virtual memory

carmine nacelle Dec 19, 2024, 11:07 PM

#

But like not mapped mapped

#

Idk

#

Be represented by memory objects

#

For which I use hash tables

#

of struct page

#

So like I could avoid IPC costs

#

By having all the cache be managed by the kernel

hexed acorn Dec 19, 2024, 11:09 PM

#

power button works in pmos on my laptop

carmine nacelle Dec 19, 2024, 11:09 PM

#

Did it turn off?

hexed acorn Dec 19, 2024, 11:10 PM

#

yes

carmine nacelle Dec 19, 2024, 11:10 PM

#

@twilit talon

hexed acorn Dec 19, 2024, 11:10 PM

#

said itll turn off in 3 seconds, and then it did

twilit talon Dec 19, 2024, 11:10 PM

#

bro thats a huge win

#

so many things have to go right for that power press to reach the handler

carmine nacelle Dec 19, 2024, 11:11 PM

#

And we have different platforms I think

twilit talon Dec 19, 2024, 11:11 PM

#

hexed acorn power button works in pmos on my laptop

what about EC events btw

#

for AC and stuff like that

hexed acorn Dec 19, 2024, 11:14 PM

#

i fucking hate the discord mobile app

#

how can you make something that sucks so bad

#

anyway, ec events work as well, tried brightness up/down and ac plugging in and unplugging (and yeah there is like a 1s delay there)

#

brightness up/down only works once though, then no more notifies happen

twilit talon Dec 19, 2024, 11:15 PM

#

very cool

twilit talon Dec 19, 2024, 11:15 PM

#

hexed acorn brightness up/down only works once though, then no more notifies happen

thats normal because it expects u to do stuff in response

#

same for sleep button

#

there's a timeout there after which it should start working again

#

probably

hexed acorn Dec 19, 2024, 11:15 PM

#

also there is an invalid arg error while running some aml method

carmine nacelle Dec 19, 2024, 11:15 PM

#

my laptop spams it

hexed acorn Dec 19, 2024, 11:15 PM

#

#

ig it's trying to read from the ec opregion before it's been set up?

#

or the ec io rather

twilit talon Dec 19, 2024, 11:16 PM

#

@carmine nacelle your ec_read returns UACPI_INVALID_ARGUMENT?

carmine nacelle Dec 19, 2024, 11:16 PM

#

It could be

#

I copied it from managram

#

And it was checking read and write and returning invalid if none match

#

I need to add printfs I'd suppose

hexed acorn Dec 19, 2024, 11:17 PM

#

welp now to write managarm to the usb stick instead

#

also your iso does work if i just write it to the usb stick

twilit talon Dec 19, 2024, 11:18 PM

#

hexed acorn ig it's trying to read from the ec opregion before it's been set up?

nah when _REG is invoked its already set up

carmine nacelle Dec 19, 2024, 11:18 PM

#

https://gitlab.com/mishakov/pmos/-/blob/main/devicesd/generic/ec.c?ref_type=heads#L259

#

Found EC device at and before discovering GPIO is during its initialization

#

I have ```C
if (!device->initialized) {
return UACPI_STATUS_INVALID_ARGUMENT;
}

#

thonk

twilit talon Dec 19, 2024, 11:20 PM

#

lol

#

no that aint gonna work

#

        uacpi_status result = uacpi_install_address_space_handler(
            address_space_node, UACPI_ADDRESS_SPACE_EMBEDDED_CONTROLLER,
            handle_ec_region, device);

#

initialized must be set to true here

#

before this call

#

as u can see qookies laptop accesses it right there during _REG

#

so u must be ready at this point

carmine nacelle Dec 19, 2024, 11:21 PM

#

I think I can just remove that check

twilit talon Dec 19, 2024, 11:21 PM

#

no no

#

u should just move that installation down probably

#

no reason to do it before _GLK and _GPE check

twilit talon Dec 19, 2024, 11:22 PM

#

carmine nacelle I think I can just remove that check

but yeah i dont think its needed

#

@hexed acorn btw your laptop doesnt have an ECDT right?

carmine nacelle Dec 19, 2024, 11:24 PM

#

if someone wants to test it

📎 pmOS.iso

hexed acorn Dec 19, 2024, 11:24 PM

#

qookie@selenium ~ λ ls /sys/firmware/acpi/tables/E*
ls: cannot access '/sys/firmware/acpi/tables/E*': No such file or directory

twilit talon Dec 19, 2024, 11:24 PM

#

yeah

twilit talon Dec 19, 2024, 11:24 PM

#

carmine nacelle if someone wants to test it

i have like 5 laptops i can test a bit later

#

they're all intel tho

carmine nacelle Dec 19, 2024, 11:25 PM

#

Mine are all AMD

twilit talon Dec 19, 2024, 11:25 PM

#

lmao

#

i dont have a single amd cpu in my home

#

besides ps5 and steamdeck

hoary moat Dec 19, 2024, 11:25 PM

#

twilit talon i dont have a single amd cpu in my home

Same

carmine nacelle Dec 19, 2024, 11:25 PM

#

twilit talon besides ps5 and steamdeck

Steam deck is a PC ultrameme

twilit talon Dec 19, 2024, 11:25 PM

#

it is yeah

#

it should have super fast s3 suspend

#

maybe i should test obos on it

hoary moat Dec 19, 2024, 11:26 PM

#

Good idea

carmine nacelle Dec 19, 2024, 11:28 PM

#

Does it suspend to s3?

hoary moat Dec 19, 2024, 11:29 PM

#

Yes

carmine nacelle Dec 19, 2024, 11:30 PM

#

I've commented out pthread stuff here

📎 pmOS.iso

#

Since uACPI is already a good pthread test galaxybrain

#

I think I know why battery events could be so slow

hoary moat Dec 19, 2024, 11:43 PM

#

twilit talon maybe i should test obos on it

📎 obos.iso

#

definitely should

#

(daily obos shilling complete)

carmine nacelle Dec 20, 2024, 9:18 AM

#

thonk

#

Is i2c hard to implement?

#

My laptop has everything connected over it, by the looks of things

twilit talon Dec 20, 2024, 9:34 AM

#

carmine nacelle Is i2c hard to implement?

nah

#

its a very very simple protocol

carmine nacelle Dec 20, 2024, 9:35 AM

#

I know, but how are its controllers?

#

Also, ACPI talks about smbus

twilit talon Dec 20, 2024, 9:36 AM

#

the problem is its not very documented

#

i think

twilit talon Dec 20, 2024, 9:36 AM

#

carmine nacelle Also, ACPI talks about smbus

it has opregions of this type but they arent used very much

#

for very specific hw maybe

carmine nacelle Dec 20, 2024, 9:37 AM

#

So they are separate things

twilit talon Dec 20, 2024, 9:37 AM

#

what is?

#

Well they're discoverable only via ACPI

#

the hid devices

carmine nacelle Dec 20, 2024, 9:37 AM

#

Smbus and i2c

twilit talon Dec 20, 2024, 9:37 AM

#

SMbus is the i2c controller

#

its just taht aml doesnt talk to it directly usually

carmine nacelle Dec 20, 2024, 9:38 AM

#

But AMDI0010 for example

twilit talon Dec 20, 2024, 9:42 AM

#

carmine nacelle But AMDI0010 for example

amd-specific i2c controller probably

twilit talon Dec 20, 2024, 9:43 AM

#

carmine nacelle But AMDI0010 for example

drivers/acpi/acpi_apd.c:        { "AMDI0010", APD_ADDR(wt_i2c_desc) },
drivers/i2c/busses/i2c-designware-platdrv.c:    { "AMDI0010", ACCESS_INTR_MASK },

#

one of these

#

looks like a tiny driver as well

#

And yeah your touchpad is under one of these:

Scope (_SB.I2CB)
{
    Device (TPNL)
    {
        Name (_HID, "GXTP7936")  // _HID: Hardware ID
        Name (_CID, "PNP0C50" /* HID Protocol Device (I2C bus) */)  // _CID: Compatible ID

#

Scope (_SB.I2CA)
{
    Device (TPD0)
    {
        Name (_HID, "XXXX0000")  // _HID: Hardware ID
        Name (_CID, "PNP0C50" /* HID Protocol Device (I2C bus) */)  // _CID: Compatible ID

#

or i guess touch panel and touch pad

#

is it a touch screen laptop?

#

anyway yeah a lot of HID devices via i2c

carmine nacelle Dec 20, 2024, 10:40 AM

#

twilit talon is it a touch screen laptop?

Yes

#

I have another one

#

Which also has an accelerometer and a bunch of stuff

#

And a pen support

#

Anyways, I think I'm motivated enough to work on page cache

#

So I can finally have a disk driver

#

Or maybe I can implement kernel threads Thonk

#

(for the disk stuff)

twilit talon Dec 20, 2024, 11:09 AM

#

why does that need kernel threads?

carmine nacelle Dec 20, 2024, 11:10 AM

#

To flush the cache

#

Linux uses threads

twilit talon Dec 20, 2024, 11:11 AM

#

but you're a microkernel

#

u can have page cache in userspace

carmine nacelle Dec 20, 2024, 11:11 AM

#

I want to put cache in kernel

#

Idk how I would have that in userspace

twilit talon Dec 20, 2024, 11:11 AM

#

page_cache_server

#

all file mmap requests go through it

carmine nacelle Dec 20, 2024, 11:12 AM

#

They would already go through a server

#

But, the kernel has memory objects and is in charge of virtual memory management

#

Partially I guess

twilit talon Dec 20, 2024, 11:12 AM

#

your server can give back the shared mapping to the process that it owns

carmine nacelle Dec 20, 2024, 11:12 AM

#

So it would be quick to just add a few queues and keep track of cache there

twilit talon Dec 20, 2024, 11:12 AM

#

every page fault would be handled by it by paging stuff from disk

carmine nacelle Dec 20, 2024, 11:13 AM

#

I need to think about it

#

I haven't considered it

twilit talon Dec 20, 2024, 11:13 AM

#

i just think its strange to have it in kernel if your disk drivers are in userspace

carmine nacelle Dec 20, 2024, 11:13 AM

#

And I don't see yet how it can be done cleanly

carmine nacelle Dec 20, 2024, 11:13 AM

#

twilit talon i just think its strange to have it in kernel if your disk drivers are in usersp...

Why

twilit talon Dec 20, 2024, 11:13 AM

#

userfaultfd + shared mappings

#

very easy to do cleanly

#

use SCM_RIGHTS to give it the shared mapping of the file, then the faults would be handled by either reusing an existing page from cache, or COWing it, or paging from disk into cache etc

#

u have userfaultfd right?

#

or similar

carmine nacelle Dec 20, 2024, 11:15 AM

#

Kinda

#

Idk how userfaultd works

#

I mean I had it

#

1.5 years ago

twilit talon Dec 20, 2024, 11:15 AM

#

carmine nacelle Idk how userfaultd works

page faulting thread freezes until user decides what to do

carmine nacelle Dec 20, 2024, 11:15 AM

#

Yes

#

I mean kinda

#

Yes but it's broken

twilit talon Dec 20, 2024, 11:16 AM

#

https://man7.org/linux/man-pages/man2/userfaultfd.2.html

#

just read this ig

twilit talon Dec 20, 2024, 11:17 AM

#

carmine nacelle Why

because you're gonna have to be talking to disk drivers in userspace, which kinda breaks the idea of a microkernel

#

its supposed to be abstracted away

carmine nacelle Dec 20, 2024, 11:23 AM

#

Thonk

carmine nacelle Dec 20, 2024, 11:23 AM

#

twilit talon because you're gonna have to be talking to disk drivers in userspace, which kind...

I'm not really talking to drivers

#

It's page in and page out requests

#

But yeah, with write-protect flag I can make userspace track their page stuff

twilit talon Dec 20, 2024, 11:23 AM

#

how do u know who to send them to

carmine nacelle Dec 20, 2024, 11:24 AM

#

My memory objects (memory regions) have an IPC port (handle) link in them

#

Which stop the kernel from zero allocating memory, and pauses the thread and sends a message to the userspace handler untill the page fault is resolved

#

I had that since the beginning

#

That is easy

#

Now, how do I track dirty pages?

#

And like since the kernel does memory management, how does it flush caches when it runs out of memory and stuff

twilit talon Dec 20, 2024, 11:26 AM

#

carmine nacelle Which stop the kernel from zero allocating memory, and pauses the thread and sen...

what does tihs mean

#

does every process handle their own page faults

carmine nacelle Dec 20, 2024, 11:27 AM

#

No

#

Kernel handles them

#

Idk

#

I can link the code

twilit talon Dec 20, 2024, 11:27 AM

#

carmine nacelle Which stop the kernel from zero allocating memory, and pauses the thread and sen...

but u said this

carmine nacelle Dec 20, 2024, 11:27 AM

#

It's optional

twilit talon Dec 20, 2024, 11:27 AM

#

oh

carmine nacelle Dec 20, 2024, 11:27 AM

#

Like you can have kernel give zeroed pages

#

Or you can have it ask userspace for pages

twilit talon Dec 20, 2024, 11:28 AM

#

yeah idk still feels a bit backwards to make the kernel responsible for it

carmine nacelle Dec 20, 2024, 11:28 AM

#

I mean to fill them

#

I kinda thought that what was mach doing

twilit talon Dec 20, 2024, 11:28 AM

#

maybe, not sure

carmine nacelle Dec 20, 2024, 11:28 AM

#

But then you have minix which manages page tables in userspace

twilit talon Dec 20, 2024, 11:28 AM

#

lol

carmine nacelle Dec 20, 2024, 11:28 AM

#

So like kernel has no say at all

#

Idk

#

I need to think about it

#

I can put it in userspace

twilit talon Dec 20, 2024, 11:29 AM

#

yeah

carmine nacelle Dec 20, 2024, 11:31 AM

#

I think I'm basically doing the same as this https://www.gnu.org/software/hurd/microkernel/mach/external_pager_mechanism.html

#

Minus the

capability designating the task

#

#

      Change manager   Pager m_o_return    store_write
\ _________ (B) (A) (C) ________ (D) _______
S | / Default \ / \ / \ / \
W |<=>| Pager |<=>| Mach |==>| server |<=>| storeio |<=>
A | ___/ _/ _/ _/
P |
/

(A) The ?paging policy is implemented by Mach: servers just implement the mechanism.

(B) Once the kernel has selected a page that it would like to evict, it changes the manager from the server to the default pager. This way, if the server does not deallocate the page quickly enough, it cannot cause a denial of service: the kernel will just later double page it to swap (the default pager is part of the tcb).

(C) Mach then invokes memory_object_return method on the control object. The server is expected to save the page free it in a timely fashion. The server is not required to send a response to the kernel.

(D) The manager then transfers the data to the storeio server which eventually sends it to disk. The device driver consumes the memory doing the equivalent of a vm_deallocate.

#

But mach is a fat microkernel

twilit talon Dec 20, 2024, 11:35 AM

#

im a bit skeptical about these complex designs now because ive seen the proxima os just come in and demolish the uacpi benchnmark without any of these things

carmine nacelle Dec 20, 2024, 11:36 AM

#

It's not complex

#

Mach is complex for no reason, yes

twilit talon Dec 20, 2024, 11:36 AM

#

but its single threaded ofc so perhaps doesnt mean anything

twilit talon Dec 20, 2024, 11:37 AM

#

carmine nacelle Mach is complex for no reason, yes

was it the fastest microkernel?

carmine nacelle Dec 20, 2024, 11:37 AM

#

It was the slowest microkernel KEKW

twilit talon Dec 20, 2024, 11:37 AM

#

lol

carmine nacelle Dec 20, 2024, 11:37 AM

#

L4 was the fastest

#

But they have a very small kernel and just focus on doing IPC as fast as possible I think

twilit talon Dec 20, 2024, 11:38 AM

#

just steal l4 designs then

carmine nacelle Dec 20, 2024, 11:40 AM

#

But it doesn't matter

#

All microkernels are scheduling, IPC and memory management

#

Mach is 100k LoC (I think)

#

L4 is 10k LoC

carmine nacelle Dec 20, 2024, 11:41 AM

#

carmine nacelle Mach is 100k LoC (I think)

It's just overly complicated

twilit talon Dec 20, 2024, 11:41 AM

#

carmine nacelle L4 is 10k LoC

wtf

#

how is it so tiny

#

hyper is bigger than that lmao

carmine nacelle Dec 20, 2024, 11:42 AM

#

My kernel is 20k LoC

#

I think

twilit talon Dec 20, 2024, 11:42 AM

#

uACPI is 26k 💀

carmine nacelle Dec 20, 2024, 11:43 AM

#

misha@Yoga:~/pmos/kernel$ cloc .
     183 text files.
     183 unique files.
       6 files ignored.

github.com/AlDanial/cloc v 1.90  T=0.45 s (393.7 files/s, 79707.7 lines/s)
-------------------------------------------------------------------------------
Language                     files          blank        comment           code
-------------------------------------------------------------------------------
C++                             61           2702           2172          10193
C/C++ Header                    95           2064           4444           6079
C                                1            700           1938           3672
Assembly                        15            341            523           1101
CMake                            4             20              1             55
JSON                             1              0              0             18
make                             1              5              0             10
-------------------------------------------------------------------------------
SUM:                           178           5832           9078          21128
-------------------------------------------------------------------------------

twilit talon Dec 20, 2024, 11:43 AM

#

do --vcs=git

carmine nacelle Dec 20, 2024, 11:43 AM

#

(dlmalloc is 4k LoC)

carmine nacelle Dec 20, 2024, 11:43 AM

#

twilit talon do `--vcs=git`

?

twilit talon Dec 20, 2024, 11:44 AM

#

cloc . --vcs=git

hoary moat Dec 20, 2024, 11:44 AM

#

twilit talon `cloc . --vcs=git`

With one dash only

twilit talon Dec 20, 2024, 11:44 AM

#

works with two here

carmine nacelle Dec 20, 2024, 11:44 AM

#

I've rm'd smoldtb

carmine nacelle Dec 20, 2024, 11:45 AM

#

carmine nacelle My kernel is 20k LoC

And I think my userspace is 30-40k LoC, excluding external libraries

#

So I'm doing ok

#

misha@Yoga:~/pmos$ cloc . --vcs=git
     738 text files.
     732 unique files.
      65 files ignored.

github.com/AlDanial/cloc v 1.90  T=1.17 s (587.6 files/s, 102243.9 lines/s)
-------------------------------------------------------------------------------
Language                     files          blank        comment           code
-------------------------------------------------------------------------------
C                              233           5923           9078          28607
C/C++ Header                   287           5116          12463          17302
diff                             3            103            417          14857
C++                             75           3321           2393          13312
Assembly                        45            703           1396           2206
make                            23            306             91            766
JSON                             3              0              0            241
Markdown                         4             80              0            227
CMake                           10             68             14            150
YAML                             1              1              0             43
Go                               1              2              0             12
-------------------------------------------------------------------------------
SUM:                           685          15623          25852          77723
-------------------------------------------------------------------------------

#

Fun

#

Somehow I have more C than C++ floshed

#

I think it's pulling something in which should be a submodule

#

13k LoC of assembly is very sus

#

Oh and diff

#

Yeah it should be around 50k LoC

carmine nacelle Dec 20, 2024, 11:50 AM

#

carmine nacelle > Change manager Pager m_o_return store_write > \ _______...

Anyway, what this says is that

#

Process wants memory

#

It requests it from server, which tell the kernel "I'm gonna be managing addresses 0xAAAAAAA - 0xBBBBBBB of process X"

#

When the process (thread) page faults, the kernel stops it and asks the server to read the stuff from disk

#

When the kernel decides to evict the memory it gives the page back to disk driver, to write it onto disk...

#

It should be simple

hoary moat Dec 20, 2024, 11:53 AM

#

carmine nacelle It should be simple

Assuming you know when to evict a page

carmine nacelle Dec 20, 2024, 11:53 AM

#

And avoids gazillion task switches every time someone wants to read something from disk

carmine nacelle Dec 20, 2024, 11:54 AM

#

hoary moat Assuming you know when to evict a page

Yes, that's the most difficult part

#

I think I'll have two queues

hoary moat Dec 20, 2024, 11:54 AM

#

This is what the working set is used for

#

Apply that knowledge here

carmine nacelle Dec 20, 2024, 11:54 AM

#

Yes, and that's what I want to implement

#

But for page cache

#

In the end, it's almost the same thing

hoary moat Dec 20, 2024, 11:55 AM

#

carmine nacelle Yes, and that's what I want to implement

https://github.com/OBOS-dev/obos/blob/userspace-work/src%2Foboskrnl%2Fmm%2Faging.c

GitHub

obos/src/oboskrnl/mm/aging.c at userspace-work · OBOS-dev/obos

I was bored... Contribute to OBOS-dev/obos development by creating an account on GitHub.

#

Here code example

carmine nacelle Dec 20, 2024, 11:55 AM

#

But like then I can just say "move anonymous memory to this disk to swap out"

#

And I have swap

#

But your working set size if fixed

hoary moat Dec 20, 2024, 11:57 AM

#

Note that you also have dirty and standby lists for swapping, and if you use the discord search feature you can find hyenasky explaining it

carmine nacelle Dec 20, 2024, 11:57 AM

#

hoary moat Note that you also have dirty and standby lists for swapping, and if you use the...

Yeah I want to implement that

#

But I don't knwo if I understand it correctly

hoary moat Dec 20, 2024, 11:58 AM

#

carmine nacelle Yeah I want to implement that

#1141057599584878645 message

#

Hyenasky explains it decently I think

#

If not you can look into code, or ask people

carmine nacelle Dec 20, 2024, 12:01 PM

#

Yeah I think the explanation is quite clear

#

What I want though, apart from that, is to have the kernel flush pages periodically, even when there's no memory pressure

#

Like what linux does

#

For mmaping files

hoary moat Dec 20, 2024, 12:01 PM

#

I abstract that all in src/oboskrnl/mm/swap.c

hoary moat Dec 20, 2024, 12:02 PM

#

carmine nacelle What I want though, apart from that, is to have the kernel flush pages periodica...

I mean no one is stopping you

carmine nacelle Dec 20, 2024, 12:02 PM

#

carmine nacelle For mmaping files

I case someone yoinks the power cord

hoary moat Dec 20, 2024, 12:03 PM

#

Sounds like an idea you can do

#

You could probably even implement that easily by keeping the modified page writer always running

#

Or occasionally waking it

#

Like have a timer going for x seconds that wakes it

#

And hopefully that won't cause too much of a performance loss

carmine nacelle Dec 20, 2024, 12:05 PM

#

I can run it in background as a kernel thread

#

But what infy was saying

#

Why am I putting it in kernel if I'm a microkernel

hoary moat Dec 20, 2024, 12:05 PM

#

Idk where you'll put it

#

That's up to you

carmine nacelle Dec 20, 2024, 12:06 PM

#

My argument is that it should be kept close to the paging stuff

hexed acorn Dec 20, 2024, 12:53 PM

#

twilit talon how is it so tiny

l4 is a thin layer over kernel mode

carmine nacelle Dec 21, 2024, 10:38 AM

#

lol

twilit talon Dec 21, 2024, 11:10 AM

#

how so slow

carmine nacelle Dec 21, 2024, 11:24 AM

#

I'm changing memory subsystem

#

Also it's using tcg

carmine nacelle Dec 21, 2024, 11:27 AM

#

carmine nacelle I'm changing memory subsystem

(and my anonymous memory mappings are temporarely bad)

twilit talon Dec 21, 2024, 11:30 AM

#

ah

#

dont fuck this up because I use pmos as an example when someone says "uacpi no workie"

#

meme

dense carbon Dec 21, 2024, 11:33 AM

#

carmine nacelle Mach is complex for no reason, yes

Not for no reason

twilit talon Dec 21, 2024, 11:34 AM

#

Why is it so complex yet so slow

dense carbon Dec 21, 2024, 11:34 AM

#

It is neither

#

now i'm on the computer so i can give this a proper treatment

carmine nacelle Dec 21, 2024, 11:35 AM

#

Ok, I don't know the computer history I guess

#

Mach is older than me

dense carbon Dec 21, 2024, 11:36 AM

#

mach consistently recorded favourable results performancewise (by 1986 already it outperformed 4.3BSD and that's why 4.4BSD imported the mach vmm)

carmine nacelle Dec 21, 2024, 11:36 AM

#

carmine nacelle (and my anonymous memory mappings are temporarely bad)

I'm trying to remove the state from page tables and move that to memory objects
And I currently just store pages in a linked list, which is very slow to traverse on page faults (I'll replace it with hash map once everything else works)

dense carbon Dec 21, 2024, 11:36 AM

#

speaking of VMM mach invented the TLB shootdown and the VM object, two concepts which are everywhere now

carmine nacelle Dec 21, 2024, 11:37 AM

#

dense carbon speaking of VMM mach invented the TLB shootdown and the VM object, two concepts ...

I've implemented tlb shootdowns by/after reading the mach paper

dense carbon Dec 21, 2024, 11:38 AM

#

the entirety of the "slow mach" claim is some guy, a german guy i should point out, take of that what you will, trashtalked its IPC for being asynchronous and semantically rich while his own was not, so he compared the time it takes for a roundtrip on mach v.s. his own kernel, l3, and found mach wanting

#

and nowadays that claim is more based on transmission of trivia than anything else

carmine nacelle Dec 21, 2024, 11:39 AM

#

So microkernels not (necessarely) slow?

dense carbon Dec 21, 2024, 11:41 AM

#

carmine nacelle So microkernels not (necessarely) slow?

there's no reason why they should be

#

at least it depends on how "slow" is meant

#

if you compared a perfectly optimal microkernel to a perfectly optimal monolithic kernel then probably you will observe a small performance differential in the monolithic's favour

#

since even in optimal form of each, IPC will never be so fast as simple syscalls

twilit talon Dec 21, 2024, 11:54 AM

#

dense carbon since even in optimal form of each, IPC will never be so fast as simple syscalls

thats not even necessarily why

#

tight coupling in monolithic kernels allows for more reactive stuff, like nvme switching to poll mode dynamically depending on scheduler load etc

#

this is not practical in a microkernel

#

since all components are abstracted away from each other

#

there are a lot of "hacks" you can do in a monolithic kernel that are relatively simple to reduce latency and stuff

#

in a microkernel this would require a lot of work and thought

dense carbon Dec 21, 2024, 11:59 AM

#

twilit talon tight coupling in monolithic kernels allows for more reactive stuff, like nvme s...

hardware drivers in userland are more of an orthogonal axis, traditionally microkernels don't do that, untraditionally monolithic kernels do

#

even so i think that there is a lack of imagination here

#

actually i thought from the start of this conversation about there being certain difficulties (but not insurmountable obstacles) in providing the kind of responsiveness in microkernels that is more obvious in monolithic kernels

#

like your page cache server idea, it drives a hard bargain, how would it participate in the page replacement policy? the easier and more traditional approach is that the microkernel does that policy and delegates to pagers the mechanism of reading or writing back pages on demand

dense carbon Dec 21, 2024, 12:04 PM

#

twilit talon there are a lot of "hacks" you can do in a monolithic kernel that are relatively...

even so i don't think the ability of a microkernel system to have these sorts of things should be written off

twilit talon Dec 21, 2024, 12:04 PM

#

yeah i mean its definitely possible

#

just way way harder to do and get right

dense carbon Dec 21, 2024, 12:05 PM

#

with respect to synchronisation it really is a bit of a challenge i think

#

file truncation on keyronex for example i am thinking about a lot lately, it is a bit of a hassle because i want to replicate the behaviour on a lot of unixes where (shared mapped) pages beyond the truncation point are thrown out of processes that mapped it (so if they try to access them, they get SIGBUS)

#

and that sounds deceptively easy but is more challenging to synchronise

#

first you need some kind of barrier after which you won't get A) any new read/write requests beyond the new limit, and B) any new mmap'ings of the file beyond the new limit, for instance, then you need to deal with all of the existing mappings

#

you need to keep a mind as well to the theoretically possible situation where you're short of memory but the only place you can reclaim it from is the very file you've locked so you can truncate it

#

and i did think about how much harder this stuff would be if i had a microkernel (or maybe it would be easier, maybe i am halfway along a needlessly complicated path and i can't see the straight path)

twilit talon Dec 21, 2024, 12:11 PM

#

yeah sounds like a very fun puzzle to solve

carmine nacelle Dec 21, 2024, 12:12 PM

#

I've also been thinking about that, but in terms of locks ordering

vale hazel Dec 21, 2024, 12:12 PM

#

twilit talon dont fuck this up because I use pmos as an example when someone says "uacpi no w...

lol

dense carbon Dec 21, 2024, 12:12 PM

#

carmine nacelle I've also been thinking about that, but in terms of locks ordering

i find locking one of the hardest things

#

i spend probably 80% of my time thinking about keyronex thinking about that

twilit talon Dec 21, 2024, 12:13 PM

#

dense carbon i find locking one of the hardest things

its also the least rewarding

carmine nacelle Dec 21, 2024, 12:13 PM

#

I slapped locks on memory objects. It's kinda probably slow with a lot of cores, but solves all the problems

twilit talon Dec 21, 2024, 12:13 PM

#

because your reward is: it works (sometimes when it wouldnt?)

vale hazel Dec 21, 2024, 12:14 PM

#

js slap locks everywhere

#

halfmemeright

carmine nacelle Dec 21, 2024, 12:14 PM

#

Now, with memory object inheritance, the child object (chain) might want to request pages from the parent objects, and parent objects want to unmap the pages from child objects, what do I do with locks

dense carbon Dec 21, 2024, 12:14 PM

#

twilit talon because your reward is: it works (sometimes when it wouldnt?)

if you do it really well you get faster/more scalable

#

that's why it bothers me so often

#

the easy thing is you just have a big kernel lock and a few condition variables for the long waits

dense carbon Dec 21, 2024, 12:15 PM

#

carmine nacelle Now, with memory object inheritance, the child object (chain) might want to requ...

do you permit arbitrarily long chains of memory objects?

carmine nacelle Dec 21, 2024, 12:15 PM

#

vale hazel js slap locks everywhere

Slapping locks is easy. When you start locking several of them at the time is when it becomes really fun

carmine nacelle Dec 21, 2024, 12:15 PM

#

dense carbon do you permit arbitrarily long chains of memory objects?

I want to do that

dense carbon Dec 21, 2024, 12:16 PM

#

what for?

carmine nacelle Dec 21, 2024, 12:16 PM

#

So it can work as CoW for fork

vale hazel Dec 21, 2024, 12:16 PM

#

carmine nacelle Slapping locks is easy. When you start locking several of them at the time is wh...

meme

carmine nacelle Dec 21, 2024, 12:16 PM

#

And stuff like that

#

Forking? Just create two new objects that inherit from each other

dense carbon Dec 21, 2024, 12:16 PM

#

that's one thing i dislike about mach, they had these memory object chains, and there is some use to them and they are probably more general

#

but i find it obtuse and problematic to deal with (freebsd needs something like a "garbage collector" to clear out no-longer needed objects within a chain)

carmine nacelle Dec 21, 2024, 12:17 PM

#

carmine nacelle I want to do that

I haven't implemented it yet, but it seems like a simple and effective idea

dense carbon Dec 21, 2024, 12:18 PM

#

carmine nacelle I haven't implemented it yet, but it seems like a simple and effective idea

#

i found the sunos solution of individually refcounted anon structures nicer

carmine nacelle Dec 21, 2024, 12:18 PM

#

dense carbon but i find it obtuse and problematic to deal with (freebsd needs something like ...

I've thought of that

carmine nacelle Dec 21, 2024, 12:18 PM

#

dense carbon i found the sunos solution of individually refcounted `anon` structures nicer

Is there a paper?

dense carbon Dec 21, 2024, 12:18 PM

#

so that's what the new right arrow does

carmine nacelle Dec 21, 2024, 12:19 PM

#

carmine nacelle I've thought of that

I'm also adding "anonymous" flag to objects

#

So that they can only be mapped once

#

And they can easily decay once the parent object only has one anonymous child object

dense carbon Dec 21, 2024, 12:20 PM

#

carmine nacelle Is there a paper?

netbsd's uvm replicated this while still being a machish vmm in other respects, it's discussed in https://www.usenix.org/legacy/publications/library/proceedings/usenix99/full_papers/cranor/cranor.pdf

#

https://keyronex.readthedocs.io/en/latest/vmm.html#anonymous-mappings i also discuss it here

carmine nacelle Dec 21, 2024, 12:21 PM

#

I think that's what I have at the moment but I'm removing it and going to object chains meme

dense carbon Dec 21, 2024, 12:21 PM

#

there might be some merit to object chains but i can't identify any personally

#

fuchsia ended up with them as well, but a big fuchsia guy complained it's problematic for them

#

(another "coincidental" resemblance of fuchsia to mach)

carmine nacelle Dec 21, 2024, 12:22 PM

#

Like what happens if a process forks several times

#

In chain

dense carbon Dec 21, 2024, 12:23 PM

#

carmine nacelle Like what happens if a process forks several times

the refcount of the anons are incremented by 1

carmine nacelle Dec 21, 2024, 12:23 PM

#

Thonk

#

But I can still do that with memory objects

#

Just having them be created horizontally

dense carbon Dec 21, 2024, 12:23 PM

#

but the chain structure doesn't really match

carmine nacelle Dec 21, 2024, 12:24 PM

#

Yeah

dense carbon Dec 21, 2024, 12:24 PM

#

the individually-refcounted page is a more fit match

#

really what i would say is that fork is just a bit of a nightmare in general and while i like unix i don't much like fork

carmine nacelle Dec 21, 2024, 12:25 PM

#

Oh, the problem is storage

#

Although maybe not really

#

It's implementation details (?)

#

At the moment my page struct can only pertain to one memory object, and has an intrusive list pointer, which allows it to be stored in objects

#

But I guess I can replace that with some other data structure, which would not be intrusive

dense carbon Dec 21, 2024, 12:27 PM

#

carmine nacelle At the moment my page struct can only pertain to one memory object, and has an i...

the trick is the singular anon struct is the "object" it'd belong to

carmine nacelle Dec 21, 2024, 12:28 PM

#

Hmm

#

So like pages are still part of that object, but are referenced by many structs?

dense carbon Dec 21, 2024, 12:29 PM

#

carmine nacelle So like pages are still part of that object, but are referenced by many structs?

there's one vm_anon for every virtual page, no matter how many times it's shared

#

and there's a vm_amap which is comprised of a table of pointers to vm_anons (the vm_amap is a member of an anonymous vm object)

carmine nacelle Dec 21, 2024, 12:31 PM

#

But then, what about swap

#

Thonk it would function exactly the same

dense carbon Dec 21, 2024, 12:33 PM

#

carmine nacelle But then, what about swap

the definition of a vm_anon from old keyronex (the one that could actually run xorg etc) might illustrate:

struct vm_anon {
    kmutex_t mutex;
    uintptr_t pfn_or_drumslot: 63, resident: 1;
    uint32_t refcnt;
};

vale hazel Dec 21, 2024, 12:34 PM

#

dense carbon the definition of a vm_anon from old keyronex (the one that could actually run x...

drumsalot???

#

who drums a lot/

#

im so confused

dense carbon Dec 21, 2024, 12:34 PM

#

you must imagine an anon as having some of the responsibilities of a VM object (here viz. storing either a physical page pointer or a swap location)

dense carbon Dec 21, 2024, 12:34 PM

#

vale hazel drumsalot???

drumslot

vale hazel Dec 21, 2024, 12:35 PM

#

dense carbon drumslot

whats that

dense carbon Dec 21, 2024, 12:35 PM

#

vale hazel whats that

the traditional term for a swapfile location

vale hazel Dec 21, 2024, 12:35 PM

#

dense carbon the traditional term for a swapfile location

ohhhh okayyy

#

and whats pfn

dense carbon Dec 21, 2024, 12:36 PM

#

it's because historically drum meory was used to swap to

dense carbon Dec 21, 2024, 12:36 PM

#

vale hazel and whats pfn

(physical) page frame number

vale hazel Dec 21, 2024, 12:36 PM

#

dense carbon (physical) page frame number

oh okay

#

and i see vm anon is reference counted, why so?

#

is there like a particular reason for it or

dense carbon Dec 21, 2024, 12:38 PM

#

vale hazel and i see vm anon is reference counted, why so?

for fork purposes, if it's referenced by > 1 process, then it should be CoW'd

vale hazel Dec 21, 2024, 12:38 PM

#

dense carbon for fork purposes, if it's referenced by > 1 process, then it should be CoW'd

oh okayy

carmine nacelle Dec 21, 2024, 12:38 PM

#

Ok, I don't quite understand all of the aspects of it, but I need to think about it

#

I still want object chains for filesystems though, but I think that can be combined

carmine nacelle Dec 21, 2024, 12:53 PM

#

dense carbon the definition of a vm_anon from old keyronex (the one that could actually run x...

Hmm, I think I can just put that in my struct page

dense carbon Dec 21, 2024, 12:53 PM

#

carmine nacelle Hmm, I think I can just put that in my struct page

no you can't

carmine nacelle Dec 21, 2024, 12:53 PM

#

Why not

#

Because of swap?

dense carbon Dec 21, 2024, 12:54 PM

#

you need somewhere to store the refcnt when it's swapped out

#

in principle maybe you could have your swap allocator store this instead, i think it might work

carmine nacelle Dec 21, 2024, 12:55 PM

#

Yeah, just store page struct like stuff there

dense carbon Dec 21, 2024, 12:55 PM

#

possible lock ordering oddities because vm_anons are their own thing and can be referenced by multiple vm objects

carmine nacelle Dec 21, 2024, 12:55 PM

#

And replace the pointer to page with that

#

Yeah that's gonna make swap very slow

#

Since I guess I will have to be iterating through all anon regions when swapping pages in and out

#

I wanted to say that I don't like extra memory allocations involved with this

#

But I think this is actually fine...

#

My Page is roughly ```C++
struct Page {
Page *next;
u64 offset;
size_t refcount;
PageType type = PageType::Free;
int flags = 0;
};

#

I think I won't even have to inflate it for this stuff

#

Ok I have to study so I'm gonna be quick and try to implement all of this till the end of today ultrameme

vale hazel Dec 21, 2024, 1:05 PM

#

carmine nacelle Ok I have to study so I'm gonna be quick and try to implement all of this till t...

study now

#

pmos later

#

:)

carmine nacelle Dec 21, 2024, 1:05 PM

#

No

#

I started making changes to memory management yesterday after thinking about it for 3 monts

#

I want closure

carmine nacelle Dec 21, 2024, 1:14 PM

#

carmine nacelle I think I won't even have to inflate it for this stuff

No, I do

dense carbon Dec 21, 2024, 1:21 PM

#

carmine nacelle Since I guess I will have to be iterating through all anon regions when swapping...

if you use global rather than local page replacement, you'd keeping some kind of physical-to-virtual mapping for each page

#

else you'd have to do linux style scanning page tables for references to the page you want to outswap

carmine nacelle Dec 21, 2024, 2:34 PM

#

There's something I don't get

#

And that's anonymous mappings

#

That would create a memory object

#

It would have amap attached to it

#

How do I avoid deduplication of zero pages?

#

Well, not the zero, but on the first page fault, it would allocate a page in the memory object, and then it would be copied to the amap

#

Now I have zero page inside the memory object, and another page inside the amap

#

Hmm

A special case of mapping is used for MAP_PRIVATE mmap()’s of a vnode

#

So if a mapping is anonymous, pages are moved instead of being copied?

dense carbon Dec 21, 2024, 2:42 PM

#

carmine nacelle Well, not the zero, but on the first page fault, it would allocate a page in the...

the amap is of the VM object

carmine nacelle Dec 21, 2024, 2:44 PM

#

Yes, but if I understand it correctly, you would have pages "stored" in memory objects, and then amap to hold the privately mapped modified pages

dense carbon Dec 21, 2024, 2:45 PM

#

carmine nacelle Yes, but if I understand it correctly, you would have pages "stored" in memory o...

well if you've modified it you've modified it

#

i don't understand your question

#

as best i can understand you're asking what happens if someone creates an anonymous VM object

#

and then maps it MAP_PRIVATE

#

and then they write to a page and it induces the creation first of a pageful of zeroes in the VM object that was mapped, then a second page privately

#

that's a very bizarre thing for someone to do

#

you can optimise that case explicitly if you want

carmine nacelle Dec 21, 2024, 2:56 PM

#

dense carbon and then maps it MAP_PRIVATE

That's how I wanted to implement anonymous mappings

dense carbon Dec 21, 2024, 2:57 PM

#

carmine nacelle That's how I wanted to implement anonymous mappings

you'd be mapping the object shared

#

not private

carmine nacelle Dec 21, 2024, 2:57 PM

#

But then there's fork stuff, which has to clone it somehow

dense carbon Dec 21, 2024, 2:57 PM

#

carmine nacelle But then there's fork stuff, which has to clone it somehow

that's what the anons are for

#

you make a new object

#

and increment the reference of every anon that the parent's object had

#

the object does not matter one bit in this, the anon alone matters

carmine nacelle Dec 21, 2024, 2:59 PM

#

I don't get it then

dense carbon Dec 21, 2024, 2:59 PM

#

it's very simple

carmine nacelle Dec 21, 2024, 2:59 PM

#

Oh

#

No, still not

dense carbon Dec 21, 2024, 3:00 PM

#

you have an anonymous VM object with an amap

#

you create it when you're asked to allocate some anonymous memory in a process

#

you map it shared

carmine nacelle Dec 21, 2024, 3:00 PM

#

So amaps are unconditional?

#

I though that they were only needed for private mappings

dense carbon Dec 21, 2024, 3:01 PM

#

when you fork, you create a new anonymous object for the child, and foreach parent object->amap entry, you ++ amap[i]->refcnt and you store parent amap[i] into child amap[i]

dense carbon Dec 21, 2024, 3:02 PM

#

carmine nacelle I though that they were only needed for private mappings

they are unnecessary for really shared anonymous VM objects, i.e. ones that are not created to back private memory

#

because really shared anonymous VM objects (tmpfs files for instance) never need to be subject to symmetric cow

carmine nacelle Dec 21, 2024, 3:02 PM

#

What I wanted to do is to not clone memory objects at all when forking

dense carbon Dec 21, 2024, 3:02 PM

#

only to asymmetric cow

carmine nacelle Dec 21, 2024, 3:02 PM

#

carmine nacelle What I wanted to do is to not clone memory objects at all when forking

If the mapping is shared, there is no amap

dense carbon Dec 21, 2024, 3:03 PM

#

carmine nacelle What I wanted to do is to not clone memory objects at all when forking

then you can embed amaps in VM map_entry's instead

carmine nacelle Dec 21, 2024, 3:03 PM

#

When the mapping is private, the same object is mapped again, creating new amap

carmine nacelle Dec 21, 2024, 3:03 PM

#

carmine nacelle When the mapping is private, the same object is mapped again, creating new amap

So them the anonymous memory would have to be private object + private mapping

dense carbon Dec 21, 2024, 3:04 PM

#

carmine nacelle When the mapping is private, the same object is mapped again, creating new amap

if you want to have VM objects for everything then you will have to map VM objects for mmap MAP_PRIVATE as, paradoxically you might think, shared mappings

carmine nacelle Dec 21, 2024, 3:04 PM

#

dense carbon then you can embed amaps in VM map_entry's instead

Yes

dense carbon Dec 21, 2024, 3:04 PM

#

but note i say "shared" not "MAP_SHARED"

dense carbon Dec 21, 2024, 3:04 PM

#

carmine nacelle Yes

if you do so then VM objects are entirely unnecessary for private anonymous memory

carmine nacelle Dec 21, 2024, 3:05 PM

#

dense carbon if you do so then VM objects are entirely unnecessary for private anonymous memo...

That's what I had at the moment

#

But I want to unify it into one mechanism

carmine nacelle Dec 21, 2024, 3:05 PM

#

carmine nacelle That's what I had at the moment

Kind off and without CoW

carmine nacelle Dec 21, 2024, 3:06 PM

#

dense carbon if you do so then VM objects are entirely unnecessary for private anonymous memo...

Hmm that's also certain, I can still have amaps, just no objects

#

Then I don't even need anonymous objects (I think)

carmine nacelle Dec 21, 2024, 3:07 PM

#

dense carbon if you want to have VM objects for everything then you will have to map VM objec...

Why would I

dense carbon Dec 21, 2024, 3:08 PM

#

carmine nacelle Then I don't even need anonymous objects (I think)

they might be useful for tmpfs files and also for mmap() MAP_SHARED anonymous mappings

dense carbon Dec 21, 2024, 3:08 PM

#

carmine nacelle Why would I

some people really like them

carmine nacelle Dec 21, 2024, 3:08 PM

#

dense carbon if you want to have VM objects for everything then you will have to map VM objec...

I'm not really getting it

#

On shared mapping, I skip amaps and get pages from the objects directly

dense carbon Dec 21, 2024, 3:09 PM

#

carmine nacelle On shared mapping, I skip amaps and get pages from the objects directly

if you used VM objects for every mapping, you'd have a kind of anonymous VM object that contained an amap

carmine nacelle Dec 21, 2024, 3:09 PM

#

On private mappings of shared objects, I get pages from objects on reads, and copy them to amap on writes

dense carbon Dec 21, 2024, 3:10 PM

#

at first i assumed (wrongly) that you really wanted to have everything be a VM object

carmine nacelle Dec 21, 2024, 3:10 PM

#

carmine nacelle On shared mapping, I skip amaps and get pages from the objects directly

(I've meant this as something to do in futue)

carmine nacelle Dec 21, 2024, 3:10 PM

#

dense carbon at first i assumed (wrongly) that you really wanted to have everything be a VM o...

That is my idea

#

Or was

#

So then I could have object chains for CoW

#

But yeah I think I can try to do this amap thing instead

dense carbon Dec 21, 2024, 3:11 PM

#

you were the world's first man to want to replace clean simple amaps and anons with object chains

carmine nacelle Dec 21, 2024, 3:11 PM

#

The're not clean and simple as I have them implemented at the moment

#

So I guess I should just refactor what I have so that the private copies could be shared

#

Although I can maybe just change my Page struct to count references for private mappings

#

No, there's swap problem again

#

Or maybe there isn't if I walk page tables...

carmine nacelle Dec 21, 2024, 4:25 PM

#

Ok, I think I'll keep the system that I have at the moment

#

Which is similar in spirit to amaps

#

But I'll also keep what I did yesterday, which is having anonymous mappings have memory objects

#

Since that could provide fast physical -> virtual lookups

carmine nacelle Dec 21, 2024, 4:28 PM

#

carmine nacelle Which is similar in spirit to amaps

(I don't have a separate amap table, so the info will be stored in page tables directly)

dense carbon Dec 21, 2024, 5:00 PM

#

carmine nacelle Since that could provide fast physical -> virtual lookups

the traditional way is to do a pv map

#

that's a list head in the page struct which points to a pv_entry struct stating in which vm map + which offset every page is mapped

#

however you can make an economy instead

#

by instead allocating, for every time you allocate a new region of private mmap'd anonymous memory, some struct describing what the offset is and a list of every process that pages belonging to this region could be in

#

and have your page structs point to that thing

carmine nacelle Dec 21, 2024, 5:03 PM

#

dense carbon that's a list head in the page struct which points to a pv_entry struct stating ...

This kinda feels like a big waste of memory

dense carbon Dec 21, 2024, 5:04 PM

#

carmine nacelle This kinda feels like a big waste of memory

it's perhaps another 24 bytes for every mapping of a page

#

however this alternative solution i propose above should work as well

carmine nacelle Dec 21, 2024, 5:05 PM

#

dense carbon by instead allocating, for every time you allocate a new region of private mmap'...

Or have that struct be the memory object

#

With no edge cases

dense carbon Dec 21, 2024, 5:05 PM

#

since while vm_anons are self-contained, they won't be found at arbitrary offsets - they don't move around

carmine nacelle Dec 21, 2024, 5:05 PM

#

mremap?

dense carbon Dec 21, 2024, 5:06 PM

#

carmine nacelle mremap?

i've just found out what that is

#

strange api

carmine nacelle Dec 21, 2024, 5:07 PM

#

Idk if I even want to support it though since that seems to be a Linux thing

carmine nacelle Dec 21, 2024, 6:02 PM

#

I just remember I wanted to test uACPI

#

#

ultrameme

#

Fun, MacBook is refusing to boot limine with ventoy

#

(though I think the framebuffer was also broken without it)

#

Wtf macs don't have delete button??

#

sony's ec works, but i think it got the power button event twice

#

I can't find the charger for the Asus netbook sadmeme

#

And spectre x360 is refusing to charge

carmine nacelle Dec 21, 2024, 6:25 PM

#

carmine nacelle sony's ec works, but i think it got the power button event twice

@twilit talon is that normal?

vale hazel Dec 21, 2024, 6:25 PM

#

carmine nacelle sony's ec works, but i think it got the power button event twice

why does it take so long to print

carmine nacelle Dec 21, 2024, 6:27 PM

#

Because
a. My IPC is slow (because of really bad implementation)
b. I think I'm mapping framebuffer as non-cacheable

vale hazel Dec 21, 2024, 6:27 PM

#

carmine nacelle Because a. My IPC is slow (because of really bad implementation) b. I think I'm ...

so map the framebuffer as WC

#

easy

#

even i do that in nyaux

carmine nacelle Dec 21, 2024, 6:30 PM

#

!!!

vale hazel Dec 21, 2024, 6:30 PM

#

carmine nacelle !!!

thats a face reveal right there

#

nice hair

carmine nacelle Dec 21, 2024, 6:31 PM

#

Macbooks hace ECDT

#

My new b550 motherboard doesn't have EC

#

sadmeme

#

Need to check with windows

#

Also I need to investigate what took it so long

#

But meh

#

I'm really hoping spectre x360 comes back to life

#

Do I need this?

vale hazel Dec 21, 2024, 6:44 PM

#

dma

#

trl

lunar orchid Dec 21, 2024, 6:45 PM

#

carmine nacelle Macbooks hace ECDT

what is ECDT

left dew Dec 21, 2024, 6:53 PM

#

carmine nacelle Wtf macs don't have delete button??

fn+delete

carmine nacelle Dec 21, 2024, 7:00 PM

#

carmine nacelle I'm really hoping spectre x360 comes back to life

Oh no, it's fully dead

#

Even disconnecting the battery doesn't make it turn on

#

Very sad

#

It was a cool laptop

twilit talon Dec 21, 2024, 7:06 PM

#

carmine nacelle !!!

Nice

carmine nacelle Dec 21, 2024, 7:07 PM

#

lunar orchid what is ECDT

Acpi table

lunar orchid Dec 21, 2024, 7:08 PM

#

carmine nacelle Acpi table

and what it contain

twilit talon Dec 21, 2024, 7:08 PM

#

carmine nacelle Do I need this?

What device is this?

carmine nacelle Dec 21, 2024, 7:08 PM

#

twilit talon What device is this?

My desktop

#

B550

twilit talon Dec 21, 2024, 7:10 PM

#

carmine nacelle My desktop

I mean pnp0200

carmine nacelle Dec 21, 2024, 7:10 PM

#

"DMA controller" according to Windows

twilit talon Dec 21, 2024, 7:10 PM

#

I think its the legacy isa one

carmine nacelle Dec 21, 2024, 7:11 PM

#

Could be

twilit talon Dec 21, 2024, 7:14 PM

#

carmine nacelle Could be

which resources does it use

carmine nacelle Dec 21, 2024, 7:19 PM

#

some hp meme

carmine nacelle Dec 21, 2024, 7:20 PM

#

twilit talon which resources does it use

A bunch of IO ranges

twilit talon Dec 21, 2024, 7:20 PM

#

carmine nacelle some hp meme

nice, your os works on quite a lot of hw

#

and ec seems to work too

carmine nacelle Dec 21, 2024, 7:21 PM

#

It works on everything I've tried so far

twilit talon Dec 21, 2024, 7:21 PM

#

carmine nacelle A bunch of IO ranges

https://wiki.osdev.org/ISA_DMA yeah its this, scroll down to registers

ISA DMA

carmine nacelle Dec 21, 2024, 7:22 PM

#

carmine nacelle It works on everything I've tried so far

No obos curse

mellow sleet Dec 21, 2024, 7:23 PM

#

do you have a ps2 driver

twilit talon Dec 21, 2024, 7:23 PM

#

indeed

carmine nacelle Dec 21, 2024, 7:23 PM

#

mellow sleet do you have a ps2 driver

I don't

#

I did a year ago

twilit talon Dec 21, 2024, 7:23 PM

#

thats where your stability will end LULW

carmine nacelle Dec 21, 2024, 7:24 PM

#

Why

twilit talon Dec 21, 2024, 7:24 PM

#

astral curse

carmine nacelle Dec 21, 2024, 7:24 PM

#

My PS/2 controller was working

twilit talon Dec 21, 2024, 7:24 PM

#

astral curse is ps2 driver breaks on every existing hw but your own

carmine nacelle Dec 21, 2024, 7:24 PM

#

I had been polling it so it would work even with some lost interrupts KEKW

#

I have no idea why they were lost though

carmine nacelle Dec 21, 2024, 7:33 PM

#

carmine nacelle <@353593948523986945> is that normal?

Still, is it normal that it got 2 power button events?

twilit talon Dec 21, 2024, 7:35 PM

#

could be physical damage or bugged firmware

#

usually firmware masks this event so u cant retrigger it

#

its a fixed event in this case not an EC one right?

carmine nacelle Dec 21, 2024, 7:36 PM

#

I think it's ec

twilit talon Dec 21, 2024, 7:37 PM

#

then we can take a look at aml

carmine nacelle Dec 21, 2024, 7:37 PM

#

Oh

#

It's fixed + EC

#

From logs

twilit talon Dec 21, 2024, 7:37 PM

#

ah ok

#

never seen one that supports both

carmine nacelle Dec 21, 2024, 7:38 PM

#

So it've gotten the event, then the fixed one

twilit talon Dec 21, 2024, 7:39 PM

#

makes sense

carmine nacelle Dec 21, 2024, 7:40 PM

#

Why

twilit talon Dec 21, 2024, 7:40 PM

#

u enable both, u receive both

carmine nacelle Dec 21, 2024, 7:40 PM

#

At the same time?

twilit talon Dec 21, 2024, 7:40 PM

#

why wouldnt it be at the same time

carmine nacelle Dec 21, 2024, 7:41 PM

#

Idk firmware should do something

twilit talon Dec 21, 2024, 7:41 PM

#

u should disable the fixed one if ec is supported probably

#

or idk, see how linux decides

#

does your FADT even report fixed pwrb event support?

carmine nacelle Dec 21, 2024, 7:41 PM

#

HP laptop has EC, but uses fixed interrupt

carmine nacelle Dec 21, 2024, 7:41 PM

#

twilit talon does your FADT even report fixed pwrb event support?

Idk

twilit talon Dec 21, 2024, 7:42 PM

#

there's a fadt bit for that

#

i think u can decide based on that

#

Control Method Power Button (V1) : 1

#

this one

#

either way it shouldnt matter because u should have protection from that

#

i can also just spam this button as a user

carmine nacelle Dec 21, 2024, 8:03 PM

#

Btw have you been able to try it on your PCs? meme

twilit talon Dec 21, 2024, 8:04 PM

#

no but i should try

#

can u post the latest iso

carmine nacelle Dec 21, 2024, 8:10 PM

#

#1236769772805554206 message

twilit talon Dec 21, 2024, 8:13 PM

#

thanks

twilit talon Dec 22, 2024, 12:57 PM

#

@carmine nacelle any way u can check kapi-improvements for regressions on some of your hw? I have removed all uacpi_kernel_raw_*

carmine nacelle Dec 22, 2024, 12:57 PM

#

I can check that later

twilit talon Dec 22, 2024, 12:57 PM

#

nice

carmine nacelle Dec 22, 2024, 12:57 PM

#

today

#

I'm a bit busy atm

twilit talon Dec 22, 2024, 12:57 PM

#

no rush

hoary moat Dec 22, 2024, 12:58 PM

#

I can test it if you want

twilit talon Dec 22, 2024, 12:58 PM

#

sure, u can remove the deleted api while at it as well

#

thx

hoary moat Dec 22, 2024, 12:58 PM

#

np

twilit talon Dec 22, 2024, 1:01 PM

#

I really need to find all the chargers from my laptops so i can test pmos on them (and obos as well ig)

hoary moat Dec 22, 2024, 1:01 PM

#

boots on qemu

#

will test my laptop

twilit talon Dec 22, 2024, 1:02 PM

#

shutdown and reboot is what mightve regressed so make sure that still works

hoary moat Dec 22, 2024, 1:03 PM

#

k

#

inited on my laptop

twilit talon Dec 22, 2024, 1:04 PM

#

LETSFUCKINGGOOOOOO

carmine nacelle Dec 22, 2024, 1:05 PM

#

I can explain how to compile pmOS meme

hoary moat Dec 22, 2024, 1:05 PM

#

twilit talon <:LETSFUCKINGGOOOOOO:1003833035688513606>

we have a slight problem

#

it doesn't shutdown

twilit talon Dec 22, 2024, 1:06 PM

#

lmao

#

u sure its not an obos problem?

hoary moat Dec 22, 2024, 1:07 PM

#

checking that rn

twilit talon Dec 22, 2024, 1:07 PM

#

carmine nacelle I can explain how to compile pmOS <:meme:575445094589661215>

i have your iso

carmine nacelle Dec 22, 2024, 1:07 PM

#

But like to test uACPI

twilit talon Dec 22, 2024, 1:07 PM

#

ah

#

yeah

#

i dont have amd and stuff tho but yeah

hoary moat Dec 22, 2024, 1:07 PM

#

luckily for you infy, it hangs in obos' code

twilit talon Dec 22, 2024, 1:07 PM

#

lol

hoary moat Dec 22, 2024, 1:08 PM

#

now it's my problem to find out why

twilit talon Dec 22, 2024, 1:08 PM

#

does it hang in qemu as well?

hoary moat Dec 22, 2024, 1:08 PM

#

nope

#

nooo

twilit talon Dec 22, 2024, 1:08 PM

#

F

hoary moat Dec 22, 2024, 1:08 PM

#

it hangs halting the other CPUs

#

except it broadcasts an NMI

#

so I suspect it's a bug in the handler

carmine nacelle Dec 22, 2024, 1:08 PM

#

How did I come to have the (seemingly) most stable OS here?

hoary moat Dec 22, 2024, 1:10 PM

#

-# I should probably find all times I said "Microkernel L" to you and replace them with "microkernel W"

twilit talon Dec 22, 2024, 1:10 PM

#

skill unissue

hoary moat Dec 22, 2024, 1:11 PM

#

I got it to reproduce with qemu

#

then gdb crashed

twilit talon Dec 22, 2024, 1:12 PM

#

when your os is so stable it crashes gdb

hoary moat Dec 22, 2024, 1:14 PM

#

meme

twilit talon Dec 22, 2024, 1:14 PM

#

raw read stuff is basically this now:

uacpi_status uacpi_system_io_read(
    uacpi_io_addr address, uacpi_u8 width, uacpi_u64 *out
)
{
    uacpi_status ret;
    uacpi_handle handle;

    ret = uacpi_kernel_io_map(address, width, &handle);
    if (uacpi_unlikely_error(ret))
        return ret;

    ret = uacpi_kernel_io_read(handle, 0, width, out);
    uacpi_kernel_io_unmap(handle);

    return ret;
}

#

similar for memory

carmine nacelle Dec 22, 2024, 1:16 PM

#

I just use iopl so my io_map does nothing

#

So I don't think there would be a difference...

twilit talon Dec 22, 2024, 1:16 PM

#

but u still have to ask the kernel to modify iopl no?

carmine nacelle Dec 22, 2024, 1:16 PM

#

Yes, but it's a flag in rflags

twilit talon Dec 22, 2024, 1:16 PM

#

ah lol

hoary moat Dec 22, 2024, 1:16 PM

#

maybe the bug was a deadlock since I would halt other CPUs with a lock held

carmine nacelle Dec 22, 2024, 1:16 PM

#

Which I just set unconditionally

twilit talon Dec 22, 2024, 1:17 PM

#

anyway the idea is there shouldn't be any difference besides less API to implement

carmine nacelle Dec 22, 2024, 1:17 PM

#

There's other mechanism which does allow you to allow individual ports

#

Which I might implement in the future

#

But what I have atm works so it's not a priority

twilit talon Dec 22, 2024, 1:18 PM

#

tbh yeah

#

u should do that

#

https://man7.org/linux/man-pages/man2/ioperm.2.html

hoary moat Dec 22, 2024, 1:18 PM

#

@twilit talon it now hangs in uacpi

twilit talon Dec 22, 2024, 1:18 PM

#

just implement this

hoary moat Dec 22, 2024, 1:18 PM

#

rather that or it hangs in my kernel api

twilit talon Dec 22, 2024, 1:19 PM

#

i can tell u which one it is meme

hoary moat Dec 22, 2024, 1:19 PM

#

carmine nacelle There's other mechanism which does allow you to allow individual ports

can you test shutdown on pmos

#

to make sure I didn't mess up

twilit talon Dec 22, 2024, 1:19 PM

#

does it only hang on real hw or qemu as well

hoary moat Dec 22, 2024, 1:19 PM

#

real hw

#

qemu worked

twilit talon Dec 22, 2024, 1:20 PM

#

technically stuff changed in the PR should work the same way on QEMU and real hw

#

its a 1-byte write to one io port

#

what if u revert uacpi tho

hoary moat Dec 22, 2024, 1:20 PM

#

lemme see

#

infy lucks out again

twilit talon Dec 22, 2024, 1:25 PM

#

obos daily regression debugging

carmine nacelle Dec 22, 2024, 1:33 PM

#

New branch

📎 pmOS.iso

hoary moat Dec 22, 2024, 1:34 PM

#

hoary moat real hw

frick

#

it hangs in prepare for sleep state

carmine nacelle Dec 22, 2024, 1:34 PM

#

Haven't tested it myself

hoary moat Dec 23, 2024, 4:07 PM

#

@carmine nacelle your stuff hangs on my laptop

#pmOS - microkernel OS for RISC-V, x86, and LoongArch