shkwve: an incomplete kernel/maybe bootloader/UEFI thing/maybe something else if i get bored again | Operating System Development | Page 3

untold basin Dec 17, 2024, 1:13 PM

#

#

and it decodes shit

#

even though im a bootloader and so it would be totally fine for me to use the boot protocol KEKW

#

i should use arbitrary self types more

trim seal Dec 17, 2024, 1:15 PM

#

yeah exactly

vapid echo Dec 17, 2024, 1:22 PM

#

untold basin

so usb isnt as hard as people say?

#

well I guess HID is probs the easiest

untold basin Dec 17, 2024, 1:23 PM

#

vapid echo so usb isnt as hard as people say?

define "not as hard as people say"

#

it took me quite a while, and my usb stack is horrible

#

"stack"

#

and i think theres a race in there

vapid echo Dec 17, 2024, 1:23 PM

#

untold basin define "not as hard as people say"

idk I feel like when I hear xhci people make it out to be like impossible

untold basin Dec 17, 2024, 1:23 PM

#

and it also took me a while

untold basin Dec 17, 2024, 1:23 PM

#

vapid echo idk I feel like when I hear xhci people make it out to be like impossible

yeah no its not that bad

#

its not like a gpu driver or anything

vapid echo Dec 17, 2024, 1:24 PM

#

untold basin and it also took me a while

like... 5 days?

#

thats not a lot

trim seal Dec 17, 2024, 1:25 PM

#

vapid echo idk I feel like when I hear xhci people make it out to be like impossible

if u just make a hardcoded stack without layers of abstraction or it being asynchrounous its just structs on top of structs and stuff

#

so not that hard

untold basin Dec 17, 2024, 1:27 PM

#

trim seal if u just make a hardcoded stack without layers of abstraction or it being async...

mine is mostly async

#

theres like one place where im missing async and thats it

trim seal Dec 17, 2024, 1:28 PM

#

nice

untold basin Dec 17, 2024, 1:28 PM

#

vapid echo like... 5 days?

its my slowest driver

#

in terms of time

#

lol

#

i think so at least?

#

and its still not finished, i need to restructure some logic around to support echi/ochi/uhci and shit

#

and i want to support dual-role dwc3

#

also i think im missing an intel quirk that fusl has

trim seal Dec 17, 2024, 1:29 PM

#

yeah thats kinda not that easy because xhci abstracts a lot of stuff away

#

like address allocation

#

etc

untold basin Dec 17, 2024, 1:29 PM

#

trim seal yeah thats kinda not that easy because xhci abstracts a lot of stuff away

nah the big one is that i need to move driver assignment out of xhci

#

address alloc is also a thing

#

but thats gonna be new code and not code i already have written

#

im also missing route string shit for superspeed hubs

#

and hub logic in general

#

and also it wont work on any real hw because i dont allocate scratch buffers

trim seal Dec 17, 2024, 1:32 PM

#

oh yeah hub stuff is like the most complex part

#

and real hw is usually hubs on top of hubs even for native ports

#

especially like PCs

untold basin Dec 17, 2024, 1:32 PM

#

i think my m1 has the two usb3 ports directly attached?

#

because lsusb shows nothing

lofty hearth Dec 17, 2024, 1:33 PM

#

trim seal especially like PCs

or the rpi4 usb2 ports :^)

untold basin Dec 17, 2024, 1:33 PM

#

lol

#

ugh thats gonna suck

trim seal Dec 17, 2024, 1:33 PM

#

I know managarm still doesnt work with my pc because it has stuff behind hubs

untold basin Dec 17, 2024, 1:33 PM

#

lmfao

#

i think i have a usb hub somewhere i could test on

trim seal Dec 17, 2024, 1:34 PM

#

also e.g. my laptops touchscreen is behind a fake hub i think

untold basin Dec 17, 2024, 1:34 PM

#

i see i see

lofty hearth Dec 17, 2024, 1:34 PM

#

maybe i should try debugging hubs again during the christmas break

#

im still kinda thinking about attaching a logic analyzer to the usb hub to see what happens when it responds with an usb transaction error

untold basin Dec 17, 2024, 1:46 PM

#

lofty hearth im still kinda thinking about attaching a logic analyzer to the usb hub to see w...

you have a logic analyzer?

#

damn

lofty hearth Dec 17, 2024, 1:47 PM

#

untold basin you have a logic analyzer?

a shitty 8 channel 24MHz one, which should be borderline barely enough for usb fs

#

i could also use my pi pico as one ig

untold basin Dec 17, 2024, 1:47 PM

#

fair

lofty hearth Dec 17, 2024, 1:51 PM

#

actually, there are projects implementing fake usb hubs with rpi pico pio

#

i could also try that to see what happens there

untold basin Dec 17, 2024, 1:56 PM

#

okay time to design my Seat/IODev abstraction

untold basin Dec 17, 2024, 2:16 PM

#

so i guess there are different kinds of devices

#

theres serial

#

both CDC and hw serial

#

and then theres like GUI serial

#

which would i guess be handled by a GUISerial iodev?

#

or it could be a "proper" gui menu or something?

#

hmm

short kraken Dec 17, 2024, 2:58 PM

#

based

short kraken Dec 17, 2024, 2:59 PM

#

untold basin you have a logic analyzer?

so do I

#

I also have working hub code in FUBS

untold basin Dec 17, 2024, 3:00 PM

#

nice

short kraken Dec 17, 2024, 3:00 PM

#

I will write hub code for FUSL soon too

untold basin Dec 17, 2024, 3:00 PM

#

unfortunately its propriatery and a shitlang

#

so

untold basin Dec 17, 2024, 3:00 PM

#

short kraken so do I

figured

short kraken Dec 17, 2024, 3:00 PM

#

untold basin unfortunately its propriatery and a shitlang

lmao

#

you're only half right, for now

untold basin Dec 17, 2024, 3:00 PM

#

short kraken lmao

which part can you disagree with 😛

short kraken Dec 17, 2024, 3:01 PM

#

untold basin which part can you disagree with 😛

yes

untold basin Dec 17, 2024, 3:01 PM

#

its definitely a shitlang

short kraken Dec 17, 2024, 3:01 PM

#

I mean I wouldn't call it shit

#

but it most certainly isn't open source

untold basin Dec 17, 2024, 3:01 PM

#

i dont think that makes it a shitlang

#

also its closed

#

which is like, a key feature of a shitlang

short kraken Dec 17, 2024, 3:02 PM

#

untold basin also its closed

dw it's gonna have a proper, source first license later

untold basin Dec 17, 2024, 3:02 PM

#

so not even oss :^)

short kraken Dec 17, 2024, 3:02 PM

#

no, the entire idea is to do better than oss

untold basin Dec 17, 2024, 3:02 PM

#

i mean shkwve doesnt have a license at all atm

short kraken Dec 17, 2024, 3:02 PM

#

shitcode

#

proprietary shit

#

:^)

untold basin Dec 17, 2024, 3:02 PM

#

ok time to copy paste dual mit apache2

short kraken Dec 17, 2024, 3:03 PM

#

have fun getting your shit stolen by big corp

untold basin Dec 17, 2024, 3:03 PM

#

short kraken have fun getting your shit stolen by big corp

my code is too bad for that lol

short kraken Dec 17, 2024, 3:03 PM

#

:^)

untold basin Dec 17, 2024, 3:03 PM

#

also id be happy if a big corp stole my code

#

it would mean its actually good!

#

ok i got bored of looking up how to dual apache mit my code

#

ill just do whatever limine does

#

actually i think ill change my mind again

#

its mit now

#

uh

#

what license is FUSL

short kraken Dec 17, 2024, 3:09 PM

#

untold basin what license is FUSL

none yet, we'll see

untold basin Dec 17, 2024, 3:09 PM

#

you should just make it 0bsd

#

or dual mit+apache

#

(the last one gives a patent grant which is important for some big corpos)

short kraken Dec 17, 2024, 3:11 PM

#

untold basin you should just make it 0bsd

this is the main candidate rn

#

but I may do something spicy

vapid echo Dec 17, 2024, 3:11 PM

#

0bsd is public domain but bsd

#

right

untold basin Dec 17, 2024, 3:12 PM

#

its public domain but its actually legally sound

vapid echo Dec 17, 2024, 3:13 PM

#

L public domain

#

L cuck license!!!!

#

gnu will rule the world!!!

#

(this is sarcasm)

#

(I prefer permissive licenses anyway)

untold basin Dec 17, 2024, 3:13 PM

#

yeah lol

vapid echo Dec 17, 2024, 3:13 PM

#

funnily enough the only people i've seen complain about ""cuck licenses" are not developers

#

like luke smith

untold basin Dec 17, 2024, 3:14 PM

#

lol

short kraken Dec 17, 2024, 8:59 PM

#

also has anyone spotted how the battery aml works on my laptop

#

it has two batteries, I assume there is a way to switch which one it drains

#

currently it's doing the dumb and is draining the internal battery first, and the swappable one last

#

which honesly is like room temperature iq levels

#

I assume there is a way to tell it which one to use from AML or whatever

trim seal Dec 17, 2024, 9:03 PM

#

short kraken it has two batteries, I assume there is a way to switch which one it drains

oh lol it does indeed

#

it's controlled by a field in the embedded controller

short kraken Dec 17, 2024, 9:04 PM

#

trim seal it's controlled by a field in the embedded controller

via AML or something else?

#

I couldn't find anything in the firmware settings

untold basin Dec 17, 2024, 9:04 PM

#

aml + kernel driver

#

linux probably exposes it via power_supply

short kraken Dec 17, 2024, 9:04 PM

#

ooo

trim seal Dec 17, 2024, 9:04 PM

#

i see aml reading it but idk how its changed, probably via the windows WMI driver which talks to AML

short kraken Dec 17, 2024, 9:05 PM

#

#

I assume I use the things starting with ch

#

?

untold basin Dec 17, 2024, 9:06 PM

#

maybe?

#

idk tbh

short kraken Dec 17, 2024, 9:06 PM

#

hmmmm

untold basin Dec 17, 2024, 9:06 PM

#

charge_behavior i guess?

short kraken Dec 17, 2024, 9:07 PM

#

 ⚡ root@thonk  ~  cat /sys/class/power_supply/BAT1/charge_behaviour
[auto] inhibit-charge force-discharge
 ⚡ root@thonk  ~  cat /sys/class/power_supply/BAT0/charge_behaviour
[auto] inhibit-charge force-discharge

#

I probably should switch it to force-discharge on the one I want to use then?

trim seal Dec 17, 2024, 9:07 PM

#

These are your WMI-customizable options, perhaps one of these controls it?

📎 message.txt

short kraken Dec 17, 2024, 9:11 PM

#

I can't find anything in /sys/class/power_supply related to this

untold basin Dec 17, 2024, 9:12 PM

#

yeah i dont know tbh

untold basin Dec 22, 2024, 11:56 AM

#

okay hmm lets think some more about my device abstraction

#

and how to make it less shit

untold basin Dec 24, 2024, 9:07 PM

#

okay i did some more horrible interrupt stuff

#

lets figure out how to make xhci generate interrupts

untold basin Dec 24, 2024, 10:03 PM

#

okay wow that wasnt even that bad

short kraken Dec 24, 2024, 10:13 PM

#

yeah is ez

untold basin Dec 24, 2024, 10:16 PM

#

totally didnt have to add printfs all over qemu to debug it :^)

short kraken Dec 25, 2024, 12:30 AM

#

untold basin totally didnt have to add printfs all over qemu to debug it :^)

yeah no same

diff --git a/hw/usb/hcd-xhci.c b/hw/usb/hcd-xhci.c
index ad40232eb6..65a074b238 100644
--- a/hw/usb/hcd-xhci.c
+++ b/hw/usb/hcd-xhci.c
@@ -31,7 +31,7 @@

 #include "hcd-xhci.h"

-//#define DEBUG_XHCI
+#define DEBUG_XHCI
 //#define DEBUG_DATA

 #ifdef DEBUG_XHCI
@@ -591,16 +591,20 @@ static void xhci_intr_raise(XHCIState *xhci, int v)
     xhci->usbsts |= USBSTS_EINT;

     if (pending) {
+        DPRINTF("xhci_intr_raise: pending\n");
         return;
     }
     if (!(xhci->intr[v].iman & IMAN_IE)) {
+        DPRINTF("xhci_intr_raise: IMAN_IE\n");
         return;
     }

     if (!(xhci->usbcmd & USBCMD_INTE)) {
+        DPRINTF("xhci_intr_raise: USBCMD_INTE\n");
         return;
     }
     if (xhci->intr_raise) {
+        DPRINTF("xhci_intr_raise: raising interrupt!\n");
         if (xhci->intr_raise(xhci, v, true)) {
             xhci->intr[v].iman &= ~IMAN_IP;
         }

#

I would never do that

untold basin Dec 25, 2024, 8:34 AM

#

short kraken yeah no same ```diff diff --git a/hw/usb/hcd-xhci.c b/hw/usb/hcd-xhci.c index ad...

oh i had a lot more printfs

limber abyss Dec 25, 2024, 2:45 PM

#

short kraken yeah no same ```diff diff --git a/hw/usb/hcd-xhci.c b/hw/usb/hcd-xhci.c index ad...

can you not -d trace:*xhci*

#

?

short kraken Dec 25, 2024, 3:57 PM

#

limber abyss can you not `-d trace:*xhci*`

of course you can but there are no printfs for what I wanted to know

untold basin Dec 25, 2024, 6:37 PM

#

i have added extra trace hooks to qemu on occasion even

#

its actually not that bad

untold basin Dec 26, 2024, 8:59 AM

#

okay i kinda glued up support for my new seat subsystem

#

i think i'll rework how my device discovery works to work similarly to Keyronex's DeviceKit

untold basin Dec 26, 2024, 7:08 PM

#

okay, time to do that

#

it's a bit annoying in rust because no interfaces + no inheritance

#

but it sounds like a better architecture

untold basin Dec 26, 2024, 8:11 PM

#

hmm i still dont know how to design my driver abstraction

#

so i like the fact that DeviceKit uses inheritance

#

which is neat

#

but doesnt work in rust sadly

#

which really sucks

#

traits dont really solve everything

vapid echo Dec 26, 2024, 8:14 PM

#

I did a IO/Kit thing too

#

using C++ like apple

untold basin Dec 26, 2024, 8:14 PM

#

i see i see

#

you aren't using rust though

#

i like the inheritance approaches

#

but they dont work in rust

#

so many things dont work in rust, actually

#

its kinda funny

vapid echo Dec 26, 2024, 8:15 PM

#

no not really

#

mine is pretty much a 1:1 apple thing

untold basin Dec 26, 2024, 8:15 PM

#

wait are you using rust?

vapid echo Dec 26, 2024, 8:15 PM

#

no

untold basin Dec 26, 2024, 8:15 PM

#

ah

#

yeah

#

thats easy then :^)

vapid echo Dec 26, 2024, 8:16 PM

#

the nub thing is neat

#

and plists

#

a PCI driver identifies its provider as "PciDevice"

#

and sets a vendor id/device id

#

and now it's automagically discovered on bus iteration

#

and instantiated

#

I guess in rust you'd use dynamic traits

#

or something

untold basin Dec 26, 2024, 8:19 PM

#

in rust youd do idk some crazy shit probably

untold basin Dec 26, 2024, 8:19 PM

#

vapid echo a PCI driver identifies its provider as "PciDevice"

hmm, thats a neat idea

#

might do that

#

right now i just manually probe

vapid echo Dec 26, 2024, 8:19 PM

#

untold basin hmm, thats a neat idea

it's literally how IO/Kit works

untold basin Dec 26, 2024, 8:19 PM

#

like i apply every probe function

vapid echo Dec 26, 2024, 8:19 PM

#

my model is pretty much 1:1 with IO/Kit

#

except it sucks

untold basin Dec 26, 2024, 8:20 PM

#

untold basin like i apply every probe function

to every new device

#

which is ugly and a bit expensive

#

there are more pressing issues with my design though

vapid echo Dec 26, 2024, 8:20 PM

#

what you do is instantiate a PciDevice "nub" for every pci device on the bus

#

then call probe

#

for each device

untold basin Dec 26, 2024, 8:20 PM

#

what's a nub?

vapid echo Dec 26, 2024, 8:20 PM

#

that's what apple does

vapid echo Dec 26, 2024, 8:20 PM

#

untold basin what's a nub?

it's what sits between an actual device and a bus of some kind

#

iirc

untold basin Dec 26, 2024, 8:20 PM

#

but yeah i think caller-controlled probing is a better approach

vapid echo Dec 26, 2024, 8:20 PM

#

it doesnt do anything but probe for drivers

#

https://developer.apple.com/library/archive/documentation/Darwin/Conceptual/KernelProgramming/IOKit/IOKit.html

I/O Kit Overview

Essential information for programming in the OS X kernel. Includes a high-level overview.

#

"Each nub provides a bridge between two drivers (and, by extension, between two families). It is most common that a driver publishes one nub for each individual device or service it controls. "

untold basin Dec 26, 2024, 8:22 PM

#

ah i see i see

#

the nub concept is neat

vapid echo Dec 26, 2024, 8:29 PM

#

yeah if a driver is loaded at runtime you can just go through unmatched nubs

untold basin Dec 26, 2024, 8:31 PM

#

oh yeah thats true

#

thats also cool tbh

vapid echo Dec 26, 2024, 8:32 PM

#

I cant really think of other uses on the top of my head

untold basin Dec 26, 2024, 8:35 PM

#

ngl

#

the lack of inheritance in rust is super limiting

#

i pretty much need some sort of inheritance system

#

hmm

#

hmmmm

vapid echo Dec 26, 2024, 8:36 PM

#

or do you?

#

https://tenor.com/view/vsause-michael-michael-vsause-vsause-michael-gif-25125732

Tenor

untold basin Dec 26, 2024, 8:36 PM

#

vapid echo or do you?

kinda

#

yes

#

yeah no i pretty much do need it

vapid echo Dec 26, 2024, 8:36 PM

#

I think I could think of a way to do it with dynamic traits

#

but I'm not sure it'd be feasible

untold basin Dec 26, 2024, 8:37 PM

#

dont think you can

vapid echo Dec 26, 2024, 8:37 PM

#

ok hear me out

#

what if you just added C++ code

#

meme

untold basin Dec 26, 2024, 8:37 PM

#

i dont want to switch to c++

#

id rather write some cursed macros or a compiler plugin or something

vapid echo Dec 26, 2024, 8:37 PM

#

yeah but like

untold basin Dec 26, 2024, 8:37 PM

#

you can do it with #[repr(C)]

vapid echo Dec 26, 2024, 8:37 PM

#

you can do both

untold basin Dec 26, 2024, 8:38 PM

#

vapid echo you can do both

no thats like the worst of both worlds

untold basin Dec 26, 2024, 10:04 PM

#

actually i dont even need repr(C)

untold basin Dec 26, 2024, 10:22 PM

#

i think i just need some container_of memes

open osprey Dec 28, 2024, 9:20 AM

#

vapid echo https://tenor.com/view/vsause-michael-michael-vsause-vsause-michael-gif-25125732

"Umm actually sweaty inheritance is a BAD thing" - him

untold basin Jan 6, 2025, 9:32 AM

#

okay i'm doing hid part two now

#

for a meme project im adhd'ing myself to but its also an excuse to fix the hid logic

#

and by "fix" i mean "rewrite"

#

okay im so fucking confused rn

#

i'm trying to parse the hid descriptor for my m1's trackpad

#

their adherence to the spec is best described as "not" afaict

#

unless its linux

#

i THINK this is a real hid device? maybe???

#

it has a bus type of BUS_SPI

#

and it has a vendor/product id that is the same as the apple usb thing

#

okay i give up apple sucks

#

lol

untold basin Jan 6, 2025, 7:29 PM

#

okay, new plan

#

im back to redesiging my device shit

#

the NEW system is designed in a way that is legally distinct from Keyronex' DeviceKit

#

and thus is completely original and do not fact check that

untold basin Jan 6, 2025, 7:30 PM

#

untold basin the NEW system is designed in a way that is legally distinct from Keyronex' Devi...

except its more different

#

basically: there's still a tree of devices, except its not really a tree and its more of a digraph

#

a device can also have a special "parent" node and if you take all the parents together you get a tree

#

the NEW problem i'm facing is the exciting problem that because (like in keyronex) my devices don't have names

#

uh

#

my devices dont have names

#

actually maybe keyronex has names

#

but i dont

#

and i need to assign them somehow

#

ugh

rose sonnet Jan 6, 2025, 7:46 PM

#

UUID? trl

untold basin Jan 6, 2025, 7:46 PM

#

i could

#

but thats a bad solution even for me

rose sonnet Jan 6, 2025, 7:47 PM

#

My kernel just gives indices to everything

#

And devices server

#

operating system*

untold basin Jan 6, 2025, 7:47 PM

#

i guess i dont really need names anyway

open osprey Jan 6, 2025, 7:49 PM

#

untold basin actually maybe keyronex has names

it does, implicitly the class name so i don't guarantee distinctive names

untold basin Jan 6, 2025, 7:50 PM

#

open osprey it does, implicitly the class name so i don't guarantee distinctive names

so it doesn't have names

#

i have a device type too

open osprey Jan 6, 2025, 7:50 PM

#

new keyronex branch (not online) has multiple axes and names per axis, and the idea there will be that names will be unique

untold basin Jan 6, 2025, 7:51 PM

#

i see

untold basin Jan 6, 2025, 9:19 PM

#

#[shkwve_derive::device(name = "slice", traits = [Block])]
struct Partition {
    #[edge]
    parent: DeviceRc<dyn Block>,
    #[prop(uint)]
    start: u64,
    #[prop(uint)]
    len: u64,
}
impl Block for Partition {
    fn size(self: &Device<Self>) -> u64 {
        self.len
    }
    fn read(self: &Device<Self>, offs: u64, buf: &mut [u8]) -> Result<()> {
        self.parent.read(offs + self.start, buf)
    }
}
```okay now i have a custom rust macro for making this kind of workflow work

#

i can also guess the #[edge] and #[prop(uint)] annotations automatically now

#

proc macro is so fun

#

also super cursed

#

but fun too

untold basin Jan 6, 2025, 10:06 PM

#

#[shkwve_derive::device(name = "ramdisk", traits = [Block], extra_props = { size: u64 = size })]
struct Ramdisk {
    data: OwnedBuffer,
}
impl Block for Ramdisk {
    fn size(self: &Device<Self>) -> u64 {
        self.data.len() as u64
    }
    fn read(self: &Device<Self>, offs: u64, buf: &mut [u8]) -> Result<()> {
        self.data.get(offs as usize..offs as usize + buf.len(), buf);
        Ok(())
    }
}
``` okay now i got extra_props working

#

and it does stuff to get &Device<Self> to work on the size callback too

#

so my only slightly scuffed proc macro seems to be working now

glad bough Jan 7, 2025, 12:28 AM

#

untold basin ```rs #[shkwve_derive::device(name = "ramdisk", traits = [Block], extra_props = ...

ayo i like this interface

untold basin Jan 7, 2025, 12:29 AM

#

glad bough ayo i like this interface

yeah thats one of us then meme /j

glad bough Jan 7, 2025, 12:29 AM

#

trl

untold basin Jan 7, 2025, 8:56 AM

#

two devices on my tree now

#

perfect

agile nexus Jan 7, 2025, 9:07 AM

#

Any particular goldfish device?

untold basin Jan 7, 2025, 9:09 AM

#

rtc

#

renamed it to goldfish-rtc

untold basin Jan 7, 2025, 9:30 AM

#

okay now i populate the devicetree into my device model meme

#

also i will not say how long it took me to make those unicode box drawing characters to lay out the right way

#

now i also show the compatible here

trim seal Jan 7, 2025, 10:10 AM

#

untold basin now i also show the compatible here

Now also deref phandles and pretty print what they reference

untold basin Jan 7, 2025, 10:39 AM

#

trim seal Now also deref phandles and pretty print what they reference

thats the phandle of the node

#

also it doesnt really make sense with what the tree is intended to do

trim seal Jan 7, 2025, 10:39 AM

#

ykwim

untold basin Jan 7, 2025, 10:39 AM

#

my dt pretty printer in the console does do this already btw

#

the console is commented out tho until i reimplement a bunch of other shit

trim seal Jan 7, 2025, 10:40 AM

#

yeah this tree dumper is even more overengineered than the keyronex one

untold basin Jan 7, 2025, 10:41 AM

#

lmao

#

i like pretty colors though

trim seal Jan 7, 2025, 10:41 AM

#

yeah im a fan myself

untold basin Jan 7, 2025, 10:41 AM

#

also this is meant to be a debugging facility as well kinda

untold basin Jan 7, 2025, 11:06 AM

#

in terms of random shit that is a total waste of time, i wrote a mac multitouch parser

#

so now i can get the raw finger data

untold basin Jan 7, 2025, 11:22 AM

#

okay so anyway lets go back to figuring out how to make the bus meme work

untold basin Jan 7, 2025, 6:14 PM

#

okay now i have this mildly cursed query DSL: ```rs
for f in query!(>(type: "fdt-node" name: "cpus")>(type: "fdt-node" name: "core0")) {
// whatever logic
}

#

which you can use to find nodes

#

and its a whole bunch of proc macro memes to eventually produce an NFA

#

so the codegen ends up being only a little scuffed as fuck

#

actually the codegen seems surprisingly good

vapid echo Jan 7, 2025, 6:57 PM

#

untold basin and its a whole bunch of proc macro memes to eventually produce an NFA

nfa?

untold basin Jan 7, 2025, 7:07 PM

#

nondeterministic finite automaton

#

its math stuff

#

the important point is that it works

#

especially now that i fixed a bunch of small bugs and added even more things (i can now do query!(... as type) and it will only give you answers that are of that node type which is handy sometimes)

#

and it also has theoretical support for really complex queries

#

like regex-type queries

#

now i have a plic node

#

the thing after the ~ is the node's stable name

#

okay now lets figure out why qemu reports a weird uart clock

glad bough Jan 7, 2025, 7:31 PM

#

untold basin okay now i have this mildly cursed query DSL: ```rs for f in query!(>(type: ...

sql!

untold basin Jan 7, 2025, 7:32 PM

#

glad bough sql!

lol

glad bough Jan 7, 2025, 7:32 PM

#

i love dsls

untold basin Jan 7, 2025, 7:32 PM

#

what does clock-frequency do on an ns16550

#

what does this value mean

#

i used to think its the clock

#

and then the baud rate is clock / divisor

#

but that does not match qemu logic

untold basin Jan 7, 2025, 7:35 PM

#

glad bough i love dsls

same

#

thats why i made a shit ass one for matching devices

untold basin Jan 7, 2025, 7:57 PM

#

okay now i have an interrupt which is a child of the interrupt controller

#

with an id

#

and a target which is set to the uart

#

this is the pitust™ advanced™™ architecture®®© for interrupts®® (or PAAI™) (patent not pending)

#

not to be confused with AAI which stands for Advanced Autonomous Industries

#

i still need to do interrupt registration

#

and handling

#

okay

#

now i have registration at least

#

time for handling or something i guess

#

okay now im not getting interrupts anymore fuck

#

ah it was just my stupid code

#

now it totally works

vapid echo Jan 7, 2025, 8:25 PM

#

your thing sounds overengineered

untold basin Jan 7, 2025, 8:27 PM

#

it is

#

my software is often the definition of overengineering

untold basin Jan 7, 2025, 8:44 PM

#

huh i think i found a qemu bug

#

maybe

#

or maybe not idk

#

somehow im getting random spurious serial interrupts if i write to uart in my read interrupt handling logic

#

okay anyway

#

my totally normal and certainly not overengineered input queueing system is done

#

this new device system is way better tbh

#

lets do another driver

#

fwcfg time

#

#

okay my new totally not scuffed i swear fwcfg driver is a thing

#

so, the remaining drivers are:

disk stuff
ramfb
~~pci~~
usb hell

#

lets do ramfb

untold basin Jan 7, 2025, 9:38 PM

#

#

so now there is a ramfb device

#

and it like works and shit

#

time for pci i guess

untold basin Jan 7, 2025, 10:29 PM

#

#

pci

untold basin Jan 7, 2025, 11:01 PM

#

#

NVME works!

#

it was just a tiny bit of tweaking to make the old driver work, really

#

the really annoying one is gonna be xhci

#

both because im planning to redesign some things a bit, part N+1

#

but also because it will be when i have to finally implement hotplug

untold basin Jan 7, 2025, 11:28 PM

#

okay now the size of the NVM namespace is no longer a string

untold basin Jan 8, 2025, 8:37 AM

#

and now i have partitions going

untold basin Jan 8, 2025, 9:00 AM

#

okay now i have a syscon driver

#

so far the new device subsystem is miles better than the old one

untold basin Jan 8, 2025, 9:49 AM

#

okay i think it might be time for virtio

untold basin Jan 8, 2025, 10:39 AM

#

okay now i have virtio-blk and virtio-rng

#

want to do virtio-9p

untold basin Jan 8, 2025, 3:21 PM

#

okay cool

untold basin Jan 8, 2025, 7:05 PM

#

okay i got virtio-9p working

#

hopefully

untold basin Jan 8, 2025, 8:32 PM

#

okay i think im gonna have to implement xhci now

#

again

open osprey Jan 8, 2025, 8:51 PM

#

untold basin okay i got virtio-9p working

what do you think of it?

untold basin Jan 8, 2025, 8:52 PM

#

open osprey what do you think of it?

weird

#

i think that's a good way to describe it

open osprey Jan 8, 2025, 8:52 PM

#

untold basin *weird*

what 9p protocol version are you using?

untold basin Jan 8, 2025, 8:52 PM

#

very simplistic

untold basin Jan 8, 2025, 8:52 PM

#

open osprey what 9p protocol version are you using?

9P2000.L

open osprey Jan 8, 2025, 8:52 PM

#

great

#

you picked the right one at least

untold basin Jan 8, 2025, 8:52 PM

#

lol

#

i picked the one that the qemu ~~example~~ test client used

#

lmao

open osprey Jan 8, 2025, 8:53 PM

#

original 9p would have had you bewildered, 9p2000.U an improvement but still overloads everything

untold basin Jan 8, 2025, 8:53 PM

#

heh

open osprey Jan 8, 2025, 8:53 PM

#

9p2000.L finally dissociates the operations into ones that resemble a conventional vfs

untold basin Jan 8, 2025, 8:53 PM

#

it is the opposite of NFS in many ways

#

i have looked at NFS recently for a ctf chall (and i did not enjoy that experience, and will not be doing nfs again most likely)

untold basin Jan 8, 2025, 11:46 PM

#

okay

#

i think im gonna have to do usb hell now

untold basin Jan 10, 2025, 11:25 AM

#

okay back to xhci

untold basin Jan 10, 2025, 12:39 PM

#

okay now i have a new cleaner impl of usb hid

#

im somewhat tempted to rip off linux evdev

untold basin Jan 10, 2025, 1:19 PM

#

okay time for the "fun" that is pci irqs

random elbow Jan 10, 2025, 1:34 PM

#

I have a question do you develop a kernel or a super bootloader with usb support

untold basin Jan 10, 2025, 1:35 PM

#

totally normal bootloader :^)

#

irq routing sucks

#

ugh

glad bough Jan 10, 2025, 1:37 PM

#

are u also going to do x86 expansion rom emulation

#

at some point

#

would be pog

untold basin Jan 10, 2025, 1:38 PM

#

glad bough are u also going to do x86 expansion rom emulation

no

#

that requires qemu doesnt it

#

or basically qemu

#

i.e. no way lmfao

glad bough Jan 10, 2025, 1:39 PM

#

untold basin that requires qemu doesnt it

no

untold basin Jan 10, 2025, 1:39 PM

#

it just requirse a normal cpu emulator then?

#

pci option roms are x86 code

#

and im not on x86 so i guess id need to emulate it

open osprey Jan 10, 2025, 1:39 PM

#

untold basin that requires qemu doesnt it

a few firmwares do it, you need only basic x86 emulation

untold basin Jan 10, 2025, 1:39 PM

#

and im not that crazy

untold basin Jan 10, 2025, 1:39 PM

#

open osprey a few firmwares do it, you need only basic x86 emulation

hmm

#

still probably not

random elbow Jan 10, 2025, 1:40 PM

#

untold basin totally normal bootloader :^)

waiting for the emulator in your bootloader to run real mode applications

untold basin Jan 10, 2025, 1:40 PM

#

how much cpu emulation do you actually need?

#

anyway for now ill focus on pci bullshit

open osprey Jan 10, 2025, 1:42 PM

#

https://github.com/aik/SLOF

GitHub

GitHub - aik/SLOF: Slimline Open Firware - an implementation of IEE...

Slimline Open Firware - an implementation of IEEE1275 Open Firmware for some POWER ISA systems - aik/SLOF

#

this open firmware implementation uses a package called "x86emu" for it

untold basin Jan 10, 2025, 1:43 PM

#

from coreboot: https://github.com/coreboot/coreboot/tree/main/src/device/oprom/x86emu

#

and its a shitload of code

#

okay maybe its not that bad

#

but still no

#

or at least not anytime soon

#

first i need to figure out how pcie interrupt-map is actually encoded

#

because the spec and bindings are incredibly unclear

open osprey Jan 10, 2025, 1:46 PM

#

untold basin because the spec and bindings are incredibly unclear

the definition from the open firmware spec seems reasonable

untold basin Jan 10, 2025, 1:46 PM

#

oh

#

cool!

#

finally someone writes docs

open osprey Jan 10, 2025, 1:46 PM

#

no guarantees linux didn't shit it up

untold basin Jan 10, 2025, 1:48 PM

#

ah nvm

#

i found the right place in the pci bindings

#

which links to https://www.devicetree.org/open-firmware/bindings/pci/pci2_1.pdf

#

ah

open osprey Jan 10, 2025, 1:49 PM

#

untold basin which links to https://www.devicetree.org/open-firmware/bindings/pci/pci2_1.pdf

great, it was written by sun and friends, maybe it will actually make sense

untold basin Jan 10, 2025, 1:49 PM

#

ehh

#

mostly

#

okay its fine

untold basin Jan 10, 2025, 2:14 PM

#

lets figure out how to make pci irqs work

#

okay time to figure out why my interrupts arent interrupting

#

okay good news

#

i now have these pci-irq devices

#

bad news

#

they do not work correctly

#

holy shit

#

interrupt driven xhci actually works???

#

finally reached driver parity with old shkwve

#

cool :^)

#

maybe i should have a single pci-irq device

#

instead of a couple

#

meh

#

its fine

#

and that would require me to deal with handling multiple interrupt controllers a bit more and thats annoying

#

also i need to do the remaining virtio devices

#

net, serial, scsi, gpu

#

net and serial require async virtio too

#

anyway

#

i could definitely do an arm port right now

#

my asm is almost fully constrained to arch/rv64

#

and the abi for that is a couple functions

#

plus i'd need a gic driver i guess

#

it would be totally doable though

#

but i cant be fucked so

#

no

#

lets instead work on making input work

untold basin Jan 10, 2025, 3:06 PM

#

actually

#

first ill make xhci hotunplug not crash everything

rose sonnet Jan 10, 2025, 3:07 PM

#

I also couldn't get my IRQ routing code from x86 to work on RISC-V, but I haven't bothered figuring out why it didn't work so idk

untold basin Jan 10, 2025, 3:07 PM

#

i'd think x86 irq routing is totally different?

rose sonnet Jan 10, 2025, 3:08 PM

#

oh you fixed it

untold basin Jan 10, 2025, 3:08 PM

#

because acpi and not whatever the fuck FDT is

rose sonnet Jan 10, 2025, 3:08 PM

#

untold basin i'd think x86 irq routing is totally different?

no? you just use lapic instead of plic

untold basin Jan 10, 2025, 3:08 PM

#

rose sonnet oh you fixed it

yeye it was totally working

#

yeah that part true

#

tbh its all the same

#

except more different

#

i think pci irqs are totally different actually

rose sonnet Jan 10, 2025, 3:08 PM

#

you also have acpi with ovmf

#

then it's the same

#

besides a different (slighlty dumb) controller

#

but supposedly it's still a GSI which you route to somewhere

untold basin Jan 10, 2025, 3:09 PM

#

any time i plug in a new keyboard weird things start happening ! shkwve_bin/src/driver/xhci.rs:435 xhci: unexpected transfer event trb with no transfer started, code Success ! shkwve_bin/src/driver/xhci.rs:435 xhci: unexpected transfer event trb with no transfer started, code Success ! shkwve_bin/src/driver/xhci.rs:435 xhci: unexpected transfer event trb with no transfer started, code Success

rose sonnet Jan 10, 2025, 3:10 PM

#

like my ns16550 interrupts worked, and pci didn't

untold basin Jan 10, 2025, 3:10 PM

#

but only sometimes

#

this is super worrying ngl

#

and if that happens then i dont ever get any keyboard input

#

the xhci fsm probably breaks because my code is ultra fragile

rose sonnet Jan 10, 2025, 3:11 PM

#

qemu is also fragile meme

untold basin Jan 10, 2025, 3:12 PM

#

so i get this

untold basin Jan 10, 2025, 3:12 PM

#

rose sonnet qemu is also fragile <:meme:575445094589661215>

nah its definitely me

#

okay lets enable debug logs

rose sonnet Jan 10, 2025, 3:12 PM

#

1st step of debugging: blame the compiler

untold basin Jan 10, 2025, 3:13 PM

#

those might be useful

untold basin Jan 10, 2025, 3:13 PM

#

rose sonnet 1st step of debugging: blame the compiler

that is not a BAD strategy

#

especially since i had compiler-induced failures in the past

#

okay so the original device was assigned slot 0

#

the second device gets assigned slot 1

#

which is why it works i think

#

the third device reuses slot 0

#

fun!

#

well this is worrying

#

okay so i think the issue is memory corruption -> stack exhaustion

#

yeah so it is trying to free shkwve::driver::usb_hid::USBHID

#

okay

#

i was freeing the wrong thing

#

still getting transfer errors though

#

the transfers themselves are normal and expected

#

i think the issue is that the ring contains a bunch of stale transfers

#

yeahh

#

so now that i have this working

#

and NOW i have feature parity

#

time to reenable console

#

also i want to do fbterm

#

actually fbterm might be first

#

because cool

#

i kinda want to switch from the log crate to my own meme

#

its pretty easy to do

#

and i could make it only compile in the log entries that i opt into at compile time or something idk

rose sonnet Jan 10, 2025, 3:59 PM

#

when are you adding userspace? ultrameme

untold basin Jan 10, 2025, 4:00 PM

#

neverrrr

#

(lol)

#

okay, drawing text, attempt 1

#

okay

#

shkwve fbterm is basically better than flanterm now

#

hmm i think something went wrong still

#

but i dont know if its qemu

#

or my code

#

lets build qemu with qtk

#

yeah i think it was the issue

#

with gtk it looks fine

#

lets build qemu AGAIN because i managed to remove it

#

yeah

#

with a patch to force SDL_WINDOW_ALLOW_HIGHDPI it looks fine

#

tame leaf Jan 10, 2025, 4:14 PM

#

that looks like ass tbh

#

bitmaps >>>

tame leaf Jan 10, 2025, 4:16 PM

#

untold basin and NOW i have feature parity

feature parity with what?

untold basin Jan 10, 2025, 4:18 PM

#

tame leaf feature parity with what?

old shkwve

vapid echo Jan 10, 2025, 4:18 PM

#

tame leaf bitmaps >>>

nah it looks cool

untold basin Jan 10, 2025, 4:18 PM

#

tame leaf that looks like ass tbh

it looks ass because dpi scaling on linux is fucked

vapid echo Jan 10, 2025, 4:18 PM

#

however 2 months after, river still looks better

#

🤷‍♂️

untold basin Jan 10, 2025, 4:19 PM

#

untold basin old shkwve

before i threw out the entire device subsystem

#

and did it again

untold basin Jan 10, 2025, 4:19 PM

#

untold basin it looks ass because dpi scaling on linux is fucked

and qemu isnt helping the situation

untold basin Jan 10, 2025, 4:20 PM

#

untold basin it looks ass because dpi scaling on linux is fucked

it should look like a normal font except ??? linux moment

tame leaf Jan 10, 2025, 4:21 PM

#

that's better i guess

#

so like

#

what are you using to render this?

untold basin Jan 10, 2025, 4:22 PM

#

wdym?

#

its a qemu framebuffer

tame leaf Jan 10, 2025, 4:22 PM

#

the font

untold basin Jan 10, 2025, 4:22 PM

#

-display sdl

#

oh

#

i dont remember

#

its probably source code pro

tame leaf Jan 10, 2025, 4:22 PM

#

how are you printing it

untold basin Jan 10, 2025, 4:22 PM

#

rendered by imagemagick

untold basin Jan 10, 2025, 4:22 PM

#

tame leaf how are you printing it

i have a bitmap

#

i know the bitmap looks fine because i checked

#

also the asahi people render it correctly

#

and it does look fine for them

tame leaf Jan 10, 2025, 4:23 PM

#

bruh i thought you were doing ttf rendering or something

untold basin Jan 10, 2025, 4:23 PM

#

lol no

#

im not crazy

#

its just a prerendered bitmap

tame leaf Jan 10, 2025, 4:23 PM

#

i was sweating

untold basin Jan 10, 2025, 4:23 PM

#

untold basin Jan 10, 2025, 4:24 PM

#

tame leaf i was sweating

lmfao

#

i could port rusttype

#

or something

#

idk

#

theres a lot of rust font rasterizers for some reason

#

im sure at least one of them supports nostd

tame leaf Jan 10, 2025, 4:24 PM

#

Clueless

untold basin Jan 10, 2025, 4:25 PM

#

fontdue is nostd

#

okay then meme

#

though it would blow up my executable size probably

tame leaf Jan 10, 2025, 4:25 PM

#

try it

vapid echo Jan 10, 2025, 4:25 PM

#

#

this is how its supposed to look

#

chad

untold basin Jan 10, 2025, 4:25 PM

#

its already 900k

untold basin Jan 10, 2025, 4:25 PM

#

tame leaf try it

eh i dont really want to, because why

vapid echo Jan 10, 2025, 4:25 PM

#

untold basin theres a lot of rust font rasterizers for some reason

and all of them suck

untold basin Jan 10, 2025, 4:26 PM

#

there's really no point

untold basin Jan 10, 2025, 4:26 PM

#

vapid echo and all of them suck

ah i see

tame leaf Jan 10, 2025, 4:26 PM

#

untold basin its already 900k

that's 3 Limines

untold basin Jan 10, 2025, 4:26 PM

#

tame leaf that's 3 Limines

is it

vapid echo Jan 10, 2025, 4:26 PM

#

untold basin ah i see

rust gui in general sucks

tame leaf Jan 10, 2025, 4:26 PM

#

untold basin is it

roughly

untold basin Jan 10, 2025, 4:26 PM

#

you mean compressed limines?

tame leaf Jan 10, 2025, 4:26 PM

#

no

untold basin Jan 10, 2025, 4:26 PM

#

wtf

sacred spire Jan 10, 2025, 4:26 PM

#

untold basin its already 900k

gaddamn

untold basin Jan 10, 2025, 4:27 PM

#

limine is 200k compressed

#

to be fair

#

this is a debug build

tame leaf Jan 10, 2025, 4:27 PM

#

actually it's more than that

untold basin Jan 10, 2025, 4:27 PM

#

and its only light optimizations

tame leaf Jan 10, 2025, 4:27 PM

#

x86-64 Limine is 250k

#

well Limine does -O2

#

it's not like it does -Os

untold basin Jan 10, 2025, 4:28 PM

#

yeah

#

damn

#

though

#

i have drivers

#

which arent small

#

i mean

#

all i have is drivers, lol

#

and 12kloc of code

tame leaf Jan 10, 2025, 4:29 PM

#

the actual ELF of Limine is 1.2M

#

with -g

untold basin Jan 10, 2025, 4:29 PM

#

im not counting debug symbols

#

680k release

#

yeah so im a lot bigger

#

i wonder why

#

other than rust

tame leaf Jan 10, 2025, 4:30 PM

#

it's 250k after strip

untold basin Jan 10, 2025, 4:30 PM

#

tbh its probably rust

tame leaf Jan 10, 2025, 4:30 PM

#

it's probably rust

untold basin Jan 10, 2025, 4:32 PM

#

yeah

#

plus my liberal use of code generation isnt helping

#

you know what is also not helping

#

the fact that i might have accidentally added a bunch of crap to the binary

#

like uh

#

idk actually

#

ah no i didnt

#

i was worried for a sec i added all of keyronex to my size

tame leaf Jan 10, 2025, 4:34 PM

#

wtf

#

how do you accidentally add all of keyronex to your size

untold basin Jan 10, 2025, 4:36 PM

#

tame leaf how do you accidentally add all of keyronex to your size

i have a thing which puts a bunch of files inside of the bootloader

#

and im measuring the size of the final binary

#

its also not rust

#

its the fact that i ship unwind tables

#

its only 290k of code and data

tame leaf Jan 10, 2025, 4:37 PM

#

still fatter than Limine

untold basin Jan 10, 2025, 4:37 PM

#

true

#

that is all on rust

#

like yeah im not gonna beat limine on size ever

#

not the goal either tbh

untold basin Jan 10, 2025, 4:44 PM

#

vapid echo

how did you take that screenshot btw

#

oh wait there is a "screendump" option in qemu im blind

#

here is mine so far

untold basin Jan 10, 2025, 4:57 PM

#

vapid echo

#

i dont have a convinient way of getting reserved regions

vapid echo Jan 10, 2025, 4:57 PM

#

untold basin how did you take that screenshot btw

grim -g (slurp -d) | wl-copy

untold basin Jan 10, 2025, 4:57 PM

#

ah lol

vapid echo Jan 10, 2025, 4:57 PM

#

something like that

untold basin Jan 10, 2025, 4:57 PM

#

i just used qemu monitor screendump

vapid echo Jan 10, 2025, 4:58 PM

#

didnt know that was a thing

untold basin Jan 10, 2025, 4:58 PM

#

because that reliably produces correct pixels

#

anyway

#

works

#

now is a good time to fix my io crap

#

i have fbterm working now

untold basin Jan 10, 2025, 5:37 PM

#

hmm im not sure how i should handle io

#

and logging and shit

tame leaf Jan 10, 2025, 6:37 PM

#

so wait, how are you printing, now?

untold basin Jan 10, 2025, 7:30 PM

#

tame leaf so wait, how are you printing, now?

still sbi console

#

i have a graphical terminal now implemented

#

but i dont use it yet

#

thats blocked on whatever i come up with to manage devices

lunar rock Jan 10, 2025, 8:23 PM

#

untold basin im somewhat tempted to rip off linux evdev

Do it

untold basin Jan 10, 2025, 8:23 PM

#

lunar rock Do it

hmm

#

i probably will

untold basin Jan 15, 2025, 7:57 PM

#

i have a mildly cursed idea

#

what if i made a scuffed powershell clone for my console thing

#

oh yeah also i have the graphical terminal thing doing output now

#

XHCI/USB HID input too

#

no cursor because idk

#

okay now i have a cursor

#

and i have half broken incremental repaint, so i only redraw the cells that changed

#

okay needs more work

#

oh its my cursor movement stuff

#

ugh

#

oh also i think my arrow key handling is fucked

#

oh i think its my shit ass key repeat "logic"

untold basin Jan 20, 2025, 12:41 PM

#

#

wonderful, tree still crashes it

#

oh right

#

its a utf8 moment

#

ughhh

#

literally everything else works in ascii not utf8

untold basin Jan 20, 2025, 1:07 PM

#

#

now the graphical subsystem has the required special characters to draw these as well

#

okay i redid all the commands except lbp support (which i need to fix up for the new device model still)

#

and fs stuff

#

and time

untold basin Jan 20, 2025, 3:08 PM

#

and also now i have an idea for a new device model

#

so i might rewrite it all again now

#

but this time without as much cursed

untold basin Jan 20, 2025, 3:27 PM

#

hmm what if ECS

#

(probably a bad idea but)

#

also im not sure anymore about my idea where nodes dont have a label

#

nah ill do an arm port instead

#

so arm port step 1

#

lets make the riscv port handle relocation

#

i'll need it for any efi ports i may want to do in the future anyway so might as well

untold basin Jan 20, 2025, 7:15 PM

#

okay so, relocations on riscv "progress"

#

also i think theres still some memory corruption

#

yeah

#

works

untold basin Jan 20, 2025, 8:38 PM

#

for some reason, qemu seems to load my kernel in 32 bit mode???

#

wtf

#

ah i was missing -cpu cortex-a72

#

im getting an undefined insn again, pc=0x40081c90

#

its probably FP

untold basin Jan 20, 2025, 9:46 PM

#

ugh

#

now i have to rewrite some of my dt parsing logic to handle gic shit

untold basin Jan 21, 2025, 10:39 AM

#

okay progress

#

i have fixed my dt parsing logic

#

next step, gic

untold basin Jan 21, 2025, 2:52 PM

#

arm64 shkwve!

#

im polling the gic right now so its not really super great

#

and i need to unmask only the required interrupts instead of all of them

#

okay next steps

#

uh

#

implement vbar

#

and wfi

#

cool

#

i now unmask only the important stuff

#

next step vbar

untold basin Jan 21, 2025, 5:27 PM

#

irq delivery works

sacred spire Jan 21, 2025, 5:49 PM

#

big w

untold basin Jan 21, 2025, 5:52 PM

#

this was mostly pretty easy because my shit was mostly portable

untold basin Jan 21, 2025, 7:37 PM

#

okay same thing as before but with interrupts this time

#

okay i think ill do fs now maybe

trim seal Jan 21, 2025, 7:42 PM

#

amd64 port?

#

i want a pretty dumper for the acpi namespace pleading

untold basin Jan 21, 2025, 7:43 PM

#

lol

#

well pl011 is defo first

#

or maybe i should add an earlycon mechanism

trim seal Jan 21, 2025, 7:48 PM

#

sadbapycat

untold basin Jan 21, 2025, 7:53 PM

#

oh exciting

#

programming the pl011 correctly requires me to engage in clk and pll torture

trim seal Jan 21, 2025, 7:57 PM

#

igpu flashbacks

untold basin Jan 21, 2025, 7:59 PM

#

lmao

#

nah its cool

#

this is only one clock

#

and i can hardcode fixed-clock for now

lofty hearth Jan 21, 2025, 8:23 PM

#

i love clock and pll torture

untold basin Jan 21, 2025, 8:26 PM

#

indeed

#

although you have cheated in eir!

#

you hardcode a 24mhz clock

lofty hearth Jan 21, 2025, 8:27 PM

#

for qemu yeah

#

for rpi4 we do a firmware call to set the clock

#

for uart

untold basin Jan 21, 2025, 8:27 PM

#

wait wtf lmao

lofty hearth Jan 21, 2025, 8:27 PM

#

it's called being lazy

untold basin Jan 21, 2025, 8:27 PM

#

theres just a firmware call to set the clock to whatever you want???

#

okay now i need to grab a dtb dump

lofty hearth Jan 21, 2025, 8:28 PM

#

https://github.com/raspberrypi/firmware/wiki/Mailbox-property-interface#clocks
https://github.com/raspberrypi/firmware/wiki/Mailbox-property-interface#set-clock-rate

untold basin Jan 21, 2025, 8:29 PM

#

ah the clk and pll torture is powered by a brcm,bcm2711-cprman

#

with a clock cell

lofty hearth Jan 21, 2025, 8:29 PM

#

i haven't looked at the proper™️ way to do it, but the fw interface works well enough to just get uart up and running

untold basin Jan 21, 2025, 8:29 PM

#

and the cprman points to firmware = &/soc/raspberrypi,bcm2835-firmware

lofty hearth Jan 21, 2025, 8:29 PM

#

i also don't know how i want to deal with clocks and resets in managarm

untold basin Jan 21, 2025, 8:30 PM

#

but yeah ill need to do clk and pll torture at some point

untold basin Jan 21, 2025, 8:32 PM

#

lofty hearth i haven't looked at the proper™️ way to do it, but the fw interface works well e...

the fw interface is the proper way to do it

lofty hearth Jan 21, 2025, 8:32 PM

#

the linux driver doesn't talk to the firmware?

untold basin Jan 21, 2025, 8:33 PM

#

you sure?

lofty hearth Jan 21, 2025, 8:33 PM

#

https://elixir.bootlin.com/linux/v6.12.6/source/drivers/clk/bcm/clk-bcm2835.c

clk-bcm2835.c - drivers/clk/bcm/clk-bcm2835.c - Linu...

Elixir Cross Referencer - source code of Linux v6.12.6: drivers/clk/bcm/clk-bcm2835.c

#

C-f firmware only yields some comments

untold basin Jan 21, 2025, 8:33 PM

#

aah

lofty hearth Jan 21, 2025, 8:33 PM

#

only 4 in fact

untold basin Jan 21, 2025, 8:33 PM

#

oh yeah it has a reg

#

so yeah

lofty hearth Jan 21, 2025, 8:33 PM

#

also no hits for mbox/mailbox

untold basin Jan 21, 2025, 8:33 PM

#

you COULD engage in clk and pll torture

#

tho

#

why

#

just do the firmware call lul

lofty hearth Jan 21, 2025, 8:34 PM

#

well it'd be needed for other platforms

untold basin Jan 21, 2025, 8:34 PM

#

i can 100% assure you that the shkwve clk and pll torture subsystem will use a firmware call

untold basin Jan 21, 2025, 8:34 PM

#

lofty hearth well it'd be needed for other platforms

other platforms with a broadcom soc?

lofty hearth Jan 21, 2025, 8:34 PM

#

i'm not too worried about clocks on the pi4, but e.g. i do have a wip port to exynos7870 somewhere

untold basin Jan 21, 2025, 8:35 PM

#

damn

#

i should try targetting one of my phones

#

or the m1

#

probably the m1 tbh

#

since you have a nice uart over a special dongle

lofty hearth Jan 21, 2025, 8:36 PM

#

on the galaxy j7 that i was testing on you can get an uart over the usb port too

untold basin Jan 21, 2025, 8:36 PM

#

yes but you need a different special dongle from the one i have

lofty hearth Jan 21, 2025, 8:36 PM

#

by using an otg dongle that has a special pulldown resistor with a specific resistance

#

it turns D+/D- into RX/TX

untold basin Jan 21, 2025, 8:36 PM

#

same as on my pixel 3a and 5

#

the pixel 5 even has a bug that lets you get code exec in uefi so...

lofty hearth Jan 21, 2025, 8:37 PM

#

but so far i've done without that and instead just reenabled enough of the display controller to get a working framebuffer after the bootloader jumps to the kernel meme

untold basin Jan 21, 2025, 8:37 PM

#

(i think)

#

lmfao

#

just abuse ramoops :^)

#

or set a uefi variable

#

thats a trustzone call i believe

#

so firmware does it for you

lofty hearth Jan 21, 2025, 8:38 PM

#

no uefi here

untold basin Jan 21, 2025, 8:38 PM

#

or at least it is on my phone

#

ah unfun

#

anyway

#

back to getting inspired by the managarm pl011 driver

lofty hearth Jan 21, 2025, 8:38 PM

#

samsung has a proprietary bootloader that's loaded by the bootrom and that just jumps to linux

#

(and handles download mode etc)

untold basin Jan 21, 2025, 8:38 PM

#

ah right samsung moment

lofty hearth Jan 21, 2025, 8:39 PM

#

afaicu it's a single flat binary on one of the partitions

untold basin Jan 21, 2025, 8:39 PM

#

meanwhile, pixels have like 3

#

4?

lofty hearth Jan 21, 2025, 8:39 PM

#

(+ another partition for crypto code and another for config stuff and a tar archive with all the images like the samsung logo etc meme )

untold basin Jan 21, 2025, 8:40 PM

#

abl, abl_payload, hyp, imagefv, tz, uefisecapp, xbl, xbl_config, cdt

lofty hearth Jan 21, 2025, 8:40 PM

#

on samsung phones with exynos socs (at least the older ones) it's sboot, cm, and param

untold basin Jan 21, 2025, 8:42 PM

#

abl has an elf and a uefi firmware volume

#

abl_payload is JUST an efi firmware volume

#

hyp is an elf

#

imagefv is an elf with a uefi firmware volume

#

tz is an elf

#

uefisecapp is an elf

#

xbl is an elf and an uefi firmware volume

#

xbl_config is a flat binary

#

cdt is some magic header + dtb

#

(no it does not contain any devices, cdt is just some provisioning data. obviously.)

#

pl011 driver works too

#

interrupt-driven btw

#

its only a 2 kloc diff

#

now the real question is

#

does it run on riscv still

#

the answer is no

#

okay i fixed THOSE bugs now it works

untold basin Jan 22, 2025, 10:10 AM

#

okay

#

so do i implement more boot protocols

#

or do i implement x86

#

or do i implement more boot protocols for me to be loaded by

#

later today ill try running on m1 (this is not gonna be fun)

#

oh also or do i fix SMP

#

or do i clean up the riscv arch code

#

i think ill fix my fs layer

#

hmm or maybe ill unfuckup my device model

#

by adding an explicit path system back

#

and a multi-axis system

untold basin Jan 22, 2025, 10:28 AM

#

untold basin hmm or maybe ill unfuckup my device model

(again)

untold basin Jan 22, 2025, 6:54 PM

#

okay new plan

#

ill get my central scrutinizer and framework

#

and shkwve on m1

untold basin Jan 22, 2025, 9:45 PM

#

okay

#

time to do the fun thing

#

apple m1!

untold basin Jan 22, 2025, 11:07 PM

#

okay the primary uart is the s5l at 0x235200000

#

#define UTXH     0x020
#define UTRSTAT  0x010
#define UTRSTAT_TXBE    2
    while (!(read32(uart_base + UTRSTAT) & UTRSTAT_TXBE))
        ;

    write32(uart_base + UTXH, c);

#

cool

#

okay i only did env setup

lunar rock Jan 22, 2025, 11:51 PM

#

I'm getting tempted to buy an m1 Mac Mini

untold basin Jan 23, 2025, 8:31 AM

#

lunar rock I'm getting tempted to buy an m1 Mac Mini

lmao

#

i have a 3 day weekend coming up because therres a storm or some shit

untold basin Jan 23, 2025, 9:07 AM

#

for initial bringup i will probably use m1n1 hv

#

so i just need to write(0x235200020, output) to get logs

#

i really need to implement the even newer device subsystem though

untold basin Jan 23, 2025, 9:11 AM

#

untold basin for initial bringup i will probably use m1n1 hv

then when that works (and i can take the mandatory pictures for bragging rights) i'll try getting it working on pure real hw

#

i still defo need to mess around with EL2 stuff

#

since right now i just assume el1

#

also im missing arm paging

#

and i need any boot protocol

#

so i can claim the true bragging rights

untold basin Jan 23, 2025, 10:11 AM

#

ugh i think i have to rework my device model but i dont know how to make the new one

#

hmm actually it might be fine

#

or maybe not idk

#

so the problem is that my device model doesnt really have a great way of naming different nodes

#

and it sprays assumptions about DTs everywhere

#

with no easy way to avoid it

#

hmm maybe i could actually steal a page from managarm's book and have a flat list not a tree

#

pros:

i do not need a tree
easier probably

#

cons:

not a tree
i cant have as many colors
lots of things really are trees, like the acpi/aml tree

untold basin Jan 23, 2025, 10:37 AM

#

ok ill just add names to my devices and call it a day

#

ah wait thta doesnt map to my interface

#

fuck

#

ugh

zenith lion Jan 23, 2025, 11:25 AM

#

untold basin programming the pl011 correctly requires me to engage in clk and pll torture

just program it incorrectly lol

zenith lion Jan 23, 2025, 11:27 AM

#

untold basin tbh its probably rust

you're doing -Zbuild-std or whatever right?

#

wait this is freestanding, you're likely already doing it

untold basin Jan 23, 2025, 11:28 AM

#

yes ofc i am

#

for a target named checks notes l0eklpdnp4av9vig5bupeab1fidtpnj3d1hoh91p04rfi

#

so i think its definitely -Zbuild-std

zenith lion Jan 23, 2025, 11:29 AM

#

i think you can get away with not doing it for some freestanding targets because they're tier 1 or whatever

untold basin Jan 23, 2025, 11:29 AM

#

i think l0eklp... is definitely not tier 1

zenith lion Jan 23, 2025, 11:29 AM

#

untold basin so i think its definitely -Zbuild-std

ok that's good at least it's not the entirety of std

untold basin Jan 23, 2025, 11:29 AM

#

yeah ofc

#

the unwind tables can probably be optimized further too tbh

#

they arent particularly optimal at the moment

zenith lion Jan 23, 2025, 11:30 AM

#

untold basin i think l0eklp... is definitely not tier 1

last time i checked the target list x86_64-unknown-none was tier 1 so there is prebuilt std for them? so you could probably not do build-std and that would likely cause the binary size to be huge

untold basin Jan 23, 2025, 11:31 AM

#

zenith lion last time i checked the target list x86_64-unknown-none was tier 1 so there is p...

yes but i dont want to use it

#

also there is no difference between build-std and them building the std for me

zenith lion Jan 23, 2025, 11:31 AM

#

I use custom json target specifications for all my targets which i strongly recommend if already not doing that

untold basin Jan 23, 2025, 11:31 AM

#

yes i am

#

lol

zenith lion Jan 23, 2025, 11:31 AM

#

cool

untold basin Jan 23, 2025, 11:32 AM

#

{
    "abi": "softfloat",
    "arch": "aarch64",
    "code-model": "small",
    "cpu": "cortex-a57",
    "crt-objects-fallback": "false",
    "data-layout": "e-m:e-p270:32:32-p271:32:32-p272:64:64-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128-Fn32",
    "eh-frame-header": true,
    "emit-debug-gdb-scripts": true,
    "features": "+v8a,+strict-align,-neon,-fp-armv8",
    "linker": "rust-lld",
    "linker-flavor": "gnu-lld",
    "llvm-abiname": "softfloat",
    "llvm-target": "aarch64-elf",
    "max-atomic-width": 64,
    "panic-strategy": "abort",
    "supported-sanitizers": [],
    "target-pointer-width": "64",
    "position-independent-executables": true,
    "dynamic-linking": true
}```

#

something like this

#

(this is the arm one the riscv one is similar enough though)

zenith lion Jan 23, 2025, 11:32 AM

#

it's kinda required for non-x86 because there is no default target for most of those freestanding ones iirc?

#

but some people in the rust #os-dev channel don't use a json target specification for x86

untold basin Jan 23, 2025, 11:33 AM

#

cringe

zenith lion Jan 23, 2025, 11:33 AM

#

yep

untold basin Jan 23, 2025, 11:33 AM

#

ill definitely be using one

zenith lion Jan 23, 2025, 11:33 AM

#

what is shkwve supposed to be

untold basin Jan 23, 2025, 11:33 AM

#

a bootloader i guess?

zenith lion Jan 23, 2025, 11:33 AM

#

and am i supposed to pronounce it shockwave

untold basin Jan 23, 2025, 11:33 AM

#

yes

#

lol

untold basin Jan 23, 2025, 11:34 AM

#

untold basin ```json { "abi": "softfloat", "arch": "aarch64", "code-model": "smal...

they are derived from some properties

#

target:
  arch: aarch64
  cpu: cortex-a57
  code-model: small
  data-layout: e-m:e-p270:32:32-p271:32:32-p272:64:64-i8:8:32-i16:16:32-i64:64-i128:128-n32:64-S128-Fn32
  abi: softfloat
  features: +v8a,+strict-align,-neon,-fp-armv8
``` so its actually only this

zenith lion Jan 23, 2025, 11:34 AM

#

did you know you can pass linker args and linker scripts there

untold basin Jan 23, 2025, 11:35 AM

#

in the target json?

zenith lion Jan 23, 2025, 11:35 AM

#

yep

untold basin Jan 23, 2025, 11:35 AM

#

no i did not

#

but why

#

i just generate a cargo config matching the current selected configuration

#

which contains a -Clink-args i need

#

and also a bunch of other flags like -Zinstrument-mcount if required

#

and -Crelocation-model=pie

#

and -Zdirect-access-external-data=yes

zenith lion Jan 23, 2025, 11:36 AM

#

i mostly used it for linker scripts I didn't even touch that many linker args

#

what I did before using limine for risc-v was create a linker script and target for each board (e.g. i had riscv64-qemu.json and stuff)

#

and I wasn't using SBI or anything it was booting from M-mode so it depended on the reset address of the SoC

untold basin Jan 23, 2025, 11:39 AM

#

zenith lion what I did before using limine for risc-v was create a linker script and target ...

i have something like that except 10x more cursed

#

and with more yaml

zenith lion Jan 23, 2025, 11:39 AM

#

omg show me

untold basin Jan 23, 2025, 11:39 AM

#

configs are here https://git.sr.ht/~pitust/shkwve/tree/trunk/item/config

#

the schema is here https://git.sr.ht/~pitust/shkwve/tree/trunk/item/build-src/config.go

#

the parser is here https://git.sr.ht/~pitust/shkwve/tree/trunk/item/build-src/data-ast.go

#

most if not all of the processing is here https://git.sr.ht/~pitust/shkwve/tree/trunk/item/build-src/main.go

zenith lion Jan 23, 2025, 11:40 AM

#

ok yes that's cursed

untold basin Jan 23, 2025, 11:40 AM

#

zenith lion ok yes that's cursed

yes but cool

zenith lion Jan 23, 2025, 11:40 AM

#

yep cool that you got it working the way you wanted

untold basin Jan 23, 2025, 11:40 AM

#

and its layered

zenith lion Jan 23, 2025, 11:41 AM

#

tbh hardcoding where the memory starts and ends in a linker script is more cursed

#

this is an actually good solution

untold basin Jan 23, 2025, 11:41 AM

#

zenith lion tbh hardcoding where the memory starts and ends in a linker script is more curse...

linker scripts?

#

i generate those

#

at compile time

zenith lion Jan 23, 2025, 11:41 AM

#

I wrote them for every board manually :3

untold basin Jan 23, 2025, 11:41 AM

#

lol

untold basin Jan 23, 2025, 11:42 AM

#

untold basin at compile time

i used to hardcode a base in these configs before i switched to self-relocating executables

#

which dont need that

zenith lion Jan 23, 2025, 11:42 AM

#

ooo cool

#

i never really played around with self-relocating

untold basin Jan 23, 2025, 11:42 AM

#

its like 10 lines of asm

#

and a custom relocation table format

#

which is shit and horrible

zenith lion Jan 23, 2025, 11:43 AM

#

so shkwve will be like uboot i imagine

#

what boot protocols does shkwve support?

untold basin Jan 23, 2025, 11:43 AM

#

zenith lion what boot protocols does shkwve support?

none

#

it used to do limine

#

and i want to fix limine

zenith lion Jan 23, 2025, 11:43 AM

#

then a rewrite happened?

untold basin Jan 23, 2025, 11:43 AM

#

and probably add linux boot

untold basin Jan 23, 2025, 11:43 AM

#

zenith lion then a rewrite happened?

yeah i changed how all device stuff works

untold basin Jan 23, 2025, 11:44 AM

#

zenith lion i never really played around with self-relocating

        # BSP code
        adr x10, __blob_base
        adr x11, __rel_start
        adr x12, __rel_size
        ldr x11, [x11]
        ldr x12, [x12]

        add x12, x12, x11
_loop_begin:
        cmp x11, x12
        beq _loop_end
        ldr w13, [x11, x10]
        ldr x14, [x13, x10]
        add x14, x14, x10
        str x14, [x13, x10]
        add x11, x11, 4
        b _loop_begin
_loop_end:

its pretty easy

#

__rel_start and __rel_size are patched by the build tooling

#

to contain a list of u32s that need relocating

zenith lion Jan 23, 2025, 11:44 AM

#

shkwve is like what I would want from my bootloader if I were to roll one myself

untold basin Jan 23, 2025, 11:45 AM

#

lol

zenith lion Jan 23, 2025, 11:46 AM

#

what is that shell thingy?

#

and how powerful is it?

untold basin Jan 23, 2025, 11:46 AM

#

zenith lion what is that shell thingy?

which shell thingy?

#

you mean the console.rs stuff?

zenith lion Jan 23, 2025, 11:47 AM

#

untold basin pl011 driver works too

untold basin Jan 23, 2025, 11:47 AM

#

ah yeah that

#

heres the help command

#

and it works both over serial and usb+framebuffer

zenith lion Jan 23, 2025, 11:47 AM

#

very cool

untold basin Jan 23, 2025, 11:48 AM

#

hmm the serial stuff is a bit fucked

#

actually i think its pretty broken

#

ugh

zenith lion Jan 23, 2025, 11:49 AM

#

oof

untold basin Jan 23, 2025, 11:50 AM

#

only with env tho wtf

#

actually no

#shkwve: an incomplete kernel/maybe bootloader/UEFI thing/maybe something else if i get bored again