uDRM - a portable Direct Rendering Manager | Operating System Development | Page 2

maiden birch Dec 13, 2024, 10:38 PM

#

good idea

tawdry trench Dec 13, 2024, 10:38 PM

#

i'm lost

robust abyss Dec 13, 2024, 10:38 PM

#

#ifndef __UAPI_PCI_H__
#define __UAPI_PCI_H__

#ifndef UAPI_NEEDS_PCI
#error u dont need me
#endif

#endif

#

even tho even tho

#

I guess agragtor header is a bit nicer

#

since you can in your kernel_api.h do the defines and then include

maiden birch Dec 13, 2024, 10:40 PM

#

#ifndef UAPI_PCI_VERSION
#define UAPI_PCI_VERSION 1
#endif

#if UAPI_PCI_VERSION != 1
#error uAPI headers mismatch, make sure all libraries are using the same version
#endif

@tawdry trench thoughts?

robust abyss Dec 13, 2024, 10:40 PM

#

ok I talked myself into the agragator header design ngl

maiden birch Dec 13, 2024, 10:40 PM

#

robust abyss since you can in your kernel_api.h do the defines and then include

yeah thats the idea

maiden birch Dec 13, 2024, 10:41 PM

#

robust abyss ok I talked myself into the agragator header design ngl

lol

#

Marvin did we lose u or

robust abyss Dec 13, 2024, 10:41 PM

#

~~just ignore everything I said~~

tawdry trench Dec 13, 2024, 10:42 PM

#

uhhhhhhhhhhhhhhhhhhhhhhh

maiden birch Dec 13, 2024, 10:44 PM

#

Here's a summary:

a u project layout

|- include
    |- udrm
        |- internal
            |- {uAPI module}
        |- kernel_api.h

kernel_api.h header aka the aggregator

$ cat include/udrm/kernel_api.h
#define UAPI_WANTS_MEMORY
#include <udrm/internal/uapi/memory.h>

#define UAPI_WANTS_PCI
#include <udrm/internal/uapi/pci.h>

#define UAPI_WANTS_IRQ
#include <udrm/internal/uapi/irq.h>

an api header

#ifndef UAPI_PCI_VERSION
#define UAPI_PCI_VERSION 1
#endif

#if UAPI_PCI_VERSION != 1
#error uAPI pci.h mismatch, make sure all libraries are using the same version
#endif

#ifndef UAPI_PCI_INCLUDED
#define UAPI_PCI_INCLUDED

#ifndef UAPI_WANTS_PCI
#error pci.h header included but not actually needed
#endif

void pci_idk();

#endif

tawdry trench Dec 13, 2024, 10:45 PM

#

what's the UAPI_*_INCLUDED for

maiden birch Dec 13, 2024, 10:45 PM

#

it can be dropped ig

robust abyss Dec 13, 2024, 10:45 PM

#

I think the kernel_api.h I would make would be more like ```c
#define UAPI_WANTS_MEMORY
#define UAPI_WANTS_PCI
#define UAPI_WANTS_IRQ
#include <udrm/internal/uapi/uapi.h>

maiden birch Dec 13, 2024, 10:46 PM

#

so one giant header?

robust abyss Dec 13, 2024, 10:46 PM

#

the included is the include guard no?

maiden birch Dec 13, 2024, 10:46 PM

#

yeah but we dont need it

#

since duplicate declarations are fine

tawdry trench Dec 13, 2024, 10:46 PM

#

just use this as include guard

robust abyss Dec 13, 2024, 10:46 PM

#

maiden birch so one giant header?

which just has ```c
#ifdef UAPI_WANTS_PCI
#include <uapi/pci.h>
#endif

maiden birch Dec 13, 2024, 10:46 PM

#

fair

robust abyss Dec 13, 2024, 10:46 PM

#

so you don't need to include all the headers manually

tawdry trench Dec 13, 2024, 10:47 PM

#

robust abyss which just has ```c #ifdef UAPI_WANTS_PCI #include <uapi/pci.h> #endif ``` and s...

yea this tbh

maiden birch Dec 13, 2024, 10:47 PM

#

do we drop the include guard

robust abyss Dec 13, 2024, 10:47 PM

#

of the WANTS_PCI?

#

inside the pci.h

#

probably

#

or well

maiden birch Dec 13, 2024, 10:47 PM

#

UAPI_PCI_INCLUDED

#

this

robust abyss Dec 13, 2024, 10:47 PM

#

oh that one

#

hmmmm

maiden birch Dec 13, 2024, 10:47 PM

#

it doesnt hurt i guess but its useless

robust abyss Dec 13, 2024, 10:47 PM

#

its a bit cleaner to have it seprate

maiden birch Dec 13, 2024, 10:47 PM

#

what if we have constants or enums or stuff

#

can that also be duplicate

robust abyss Dec 13, 2024, 10:48 PM

#

structs can't

maiden birch Dec 13, 2024, 10:48 PM

#

yeah then the guard must be kept nvm

robust abyss Dec 13, 2024, 10:48 PM

#

yeah you must guard

maiden birch Dec 13, 2024, 10:48 PM

#

anyway let me redo the summary

#

Here's a summary:

a u project layout

|- include
    |- udrm
        |- internal
            |- {uAPI module}
        |- kernel_api.h

kernel_api.h header aka the aggregator

$ cat include/udrm/kernel_api.h
#define UAPI_WANTS_MEMORY
#define UAPI_WANTS_PCI
#define UAPI_WANTS_IRQ
#include <udrm/internal/uapi/uapi.h>

uapi.h header

#ifdef UAPI_WANTS_MEMORY
#include "memory.h"
#endif

#ifdef UAPI_WANTS_PCI
#include "pci.h"
#endif
...

an api header

#ifndef UAPI_PCI_VERSION
#define UAPI_PCI_VERSION 1
#endif

#if UAPI_PCI_VERSION != 1
#error uAPI pci.h mismatch, make sure all libraries are using the same version
#endif

#ifndef UAPI_PCI_INCLUDED
#define UAPI_PCI_INCLUDED

#ifndef UAPI_WANTS_PCI
#error pci.h header included but not actually needed
#endif

void pci_idk();

#endif

#

do we agree as far as this is concerned

robust abyss Dec 13, 2024, 10:51 PM

#

looks good

tawdry trench Dec 13, 2024, 10:51 PM

#

looks good

maiden birch Dec 13, 2024, 10:51 PM

#

nice

maiden birch Dec 13, 2024, 10:52 PM

#

maiden birch Here's a summary: 1. a u project layout ``` |- include |- udrm |- i...

@agile halo take a look once u have time

#

now we have the harder job of working on the apis and carefully splitting them into headers

tawdry trench Dec 13, 2024, 10:54 PM

#

nooo

robust abyss Dec 13, 2024, 10:56 PM

#

#include "uacpi/kernel_api.h"

#

:^)

maiden birch Dec 13, 2024, 10:56 PM

#

so im thinking:

malloc.h -> malloc/calloc/free
map.h -> map/unmap
io.h -> pio r/w, raw memory r/w
work_queue.h -> create_workqueue, submit_work, idk
pci.h -> pci r/w
pci_bar.h -> get_bar, map_bar, idk, bus_master shit
irq.h -> irq alloc, affinity
msi.h -> msi related stuff
log.h -> logging

maiden birch Dec 13, 2024, 10:56 PM

#

robust abyss #include "uacpi/kernel_api.h"

hm?

robust abyss Dec 13, 2024, 10:56 PM

#

whats the need for multiple work queues and not let the kernel manage it?

maiden birch Dec 13, 2024, 10:57 PM

#

the drm driver needs a lot of those

median yoke Dec 13, 2024, 10:57 PM

#

maiden birch so im thinking: 1. malloc.h -> malloc/calloc/free 2. map.h -> map/unmap 3. io.h ...

What if a device does not use MSI, how would a library then register an irw

maiden birch Dec 13, 2024, 10:57 PM

#

and careful low level management for them as well

median yoke Dec 13, 2024, 10:57 PM

#

*irq

robust abyss Dec 13, 2024, 10:57 PM

#

like, what is the difference between using one work queue and two work queues?

median yoke Dec 13, 2024, 10:57 PM

#

One

maiden birch Dec 13, 2024, 10:57 PM

#

wdym?

#

we need a work queue per gpu device for example

robust abyss Dec 13, 2024, 10:58 PM

#

what is physically the difference between creating two different work queues and just submitting into one work queue

maiden birch Dec 13, 2024, 10:58 PM

#

contention

#

imagine having 2 gpus

#

very common

#

and also uacpi work spammed there too for example

robust abyss Dec 13, 2024, 10:58 PM

#

yeah but you will have contention at the scheduler eventually either way no?

median yoke Dec 13, 2024, 10:58 PM

#

maiden birch wdym?

Like, if a device uses legacy irq pins instead of MSI, how would the u whatever library get an irq

maiden birch Dec 13, 2024, 10:58 PM

#

robust abyss yeah but you will have contention at the scheduler eventually either way no?

well one work queue can only be dequeued on one cpu at a time

#

multiple can run in parallel

maiden birch Dec 13, 2024, 10:59 PM

#

median yoke Like, if a device uses legacy irq pins instead of MSI, how would the u whatever ...

idk yet

robust abyss Dec 13, 2024, 10:59 PM

#

so you expect each work queue to essentially be a single thread?

maiden birch Dec 13, 2024, 10:59 PM

#

yeah

#

it could be implemented entirely in udrm as well

#

but it needs a lot of primitives for that

#

like atomics, linked lists etc

#

tbh maybe just make the header thread.h

robust abyss Dec 13, 2024, 11:00 PM

#

it feels like a weird abstraction tbh

maiden birch Dec 13, 2024, 11:00 PM

#

also i forgot about mutex.h and spinlock.h

robust abyss Dec 13, 2024, 11:00 PM

#

like trying to combine threads and dpcs into one

median yoke Dec 13, 2024, 11:00 PM

#

We should just make it based off OBOS' api /j

maiden birch Dec 13, 2024, 11:01 PM

#

well linux bottom halves

#

same thing

robust abyss Dec 13, 2024, 11:01 PM

#

linux is not really the thing you want to look at for inspiration given they have like 3 mechanisms for bottom halfs lol

maiden birch Dec 13, 2024, 11:01 PM

#

true

robust abyss Dec 13, 2024, 11:01 PM

#

threaded irqs, work queues, tasklets

maiden birch Dec 13, 2024, 11:01 PM

#

maybe just a create_thread api would be enough

agile halo Dec 13, 2024, 11:01 PM

#

maiden birch <@379250891838193667> take a look once u have time

looks good

robust abyss Dec 13, 2024, 11:01 PM

#

and there are probably more

agile halo Dec 13, 2024, 11:02 PM

#

maiden birch idk yet

what about a pci irq alloc/dealloc/enable api similar to the one I have in uhda? (and its also similar to linux)

maiden birch Dec 13, 2024, 11:02 PM

#

irq.h

#

then i guess we should do thread.h with low level threading primitives and build whatever we want on top

#

or whatever udrm wants

robust abyss Dec 13, 2024, 11:03 PM

#

like, it can be fine to create the work queue primitive in uapi, I just feel that its so specific in how it wants to be implemented that it might as well be part of the provided shim

maiden birch Dec 13, 2024, 11:04 PM

#

we can have shared internal code that builds on top of uapi

#

which is nice

robust abyss Dec 13, 2024, 11:04 PM

#

yeah

#

the kernel is the one that needs to compile it tho, since otherwise you would have multiple defs of objects

maiden birch Dec 13, 2024, 11:04 PM

#

yeah ofc

#

it would be a yet another subproject somehow

agile halo Dec 13, 2024, 11:10 PM

#

agile halo what about a pci irq alloc/dealloc/enable api similar to the one I have in uhda?...

or well not exactly the same because the one in uhda is limited to one irq, something like ```c
typedef uapi_handle uapi_pci_device; // idk, it would be nicer to pass a handle instead of an address to these functions (and then you don't necessarily need a way to associate an address with a pci device struct)
typedef uapi_handle uapi_pci_irq;

uapi_error uapi_pci_alloc_irqs(uapi_pci_device dev, uapi_pci_irq_types preferred_types, u32 min, u32 max, u32 count);
void uapi_pci_free_irqs(uapi_pci_device dev);
uapi_error uapi_pci_get_irq(uapi_pci_device dev, u32 num, uapi_pci_irq irq);

void uapi_pci_install_irq_handler(uapi_pci_device dev, uapi_pci_irq irq, uapi_irq_handler irq_handler);
void uapi_pci_enable_irq(uapi_pci_device dev, uapi_pci_irq irq, bool enable);

maiden birch Dec 13, 2024, 11:11 PM

#

i was thinking of a uapi_pci_device_open(address)

#

and then all apis would operate on that handle

median yoke Dec 13, 2024, 11:13 PM

#

maiden birch and then all apis would operate on that handle

I mean can't the address just be the handle

maiden birch Dec 13, 2024, 11:15 PM

#

the kernel will likely have an object representing that device

#

@agile halo what was the reason u went with this design initially?

robust abyss Dec 13, 2024, 11:17 PM

#

yeah handles are better, worst case the kernel will implement it by returning its handle as an address

agile halo Dec 13, 2024, 11:17 PM

#

maiden birch <@379250891838193667> what was the reason u went with this design initially?

the fact that my kernel doesn't really have any way to associate an address to a device (I mean it could be added but I haven't had a reason for it)

maiden birch Dec 13, 2024, 11:18 PM

#

robust abyss yeah handles are better, worst case the kernel will implement it by returning it...

true true

#

yeah there u go

#

u would force the kernel to have an address -> object lookup table

robust abyss Dec 13, 2024, 11:18 PM

#

also I think I prefer ```c
typedef struct uapi_pci_device {
uapi_handle_t handle;
} uapi_pci_device_t;

#

for type safety

maiden birch Dec 13, 2024, 11:19 PM

#

yeah maybe

agile halo Dec 13, 2024, 11:20 PM

#

maiden birch i was thinking of a `uapi_pci_device_open(address)`

tbh this also requires a lookup table, I am not sure if there are any other ways to do it without needing it than to directly provide the handle to the driver init function

maiden birch Dec 13, 2024, 11:21 PM

#

disregard this

#

its the other way around

#

we will give the kernel the list of properties to match a compatible device

robust abyss Dec 13, 2024, 11:21 PM

#

just require uacpi and fully use acpi for all object stuff :^)

maiden birch Dec 13, 2024, 11:21 PM

#

then it will call into e.g. udrm_device_probe

#

and give it the handle

agile halo Dec 13, 2024, 11:21 PM

#

ah yeah address from handle

maiden birch Dec 13, 2024, 11:22 PM

#

robust abyss just require uacpi and fully use acpi for all object stuff :^)

lol

robust abyss Dec 13, 2024, 11:22 PM

#

also rename io.h to generic_address.h and remove all the apis that are included under it :^)

maiden birch Dec 13, 2024, 11:23 PM

#

lmao

#

#1316457641353936956 message

#

this is the convo

#

so yeah i think it makes sense to make the kernel just pass an opaque handle to udrm_device_probe

#

that can then be used to address the device and do r/w with it

robust abyss Dec 13, 2024, 11:23 PM

#

given you don't want to design the kernel but just the driver api yeah

maiden birch Dec 13, 2024, 11:24 PM

#

but we will need raw pci api anyways, because uacpi and aml

#

but thats just read/write

median yoke Dec 13, 2024, 11:25 PM

#

how do we know that someone won't put a backdoor in one of these libs

maiden birch Dec 13, 2024, 11:25 PM

#

read the code

robust abyss Dec 13, 2024, 11:25 PM

#

have it signed by infy

median yoke Dec 13, 2024, 11:25 PM

#

maiden birch read the code

what if it's really well hidden

robust abyss Dec 13, 2024, 11:25 PM

#

smart_FB

maiden birch Dec 13, 2024, 11:25 PM

#

read it really well

median yoke Dec 13, 2024, 11:25 PM

#

actually, how do we know infy didn't put a backdoor in uACPI

robust abyss Dec 13, 2024, 11:25 PM

#

read it after macro expansion

maiden birch Dec 13, 2024, 11:25 PM

#

uacpi has async https apis to send all info

median yoke Dec 13, 2024, 11:25 PM

#

xz utils backdoor repeat

robust abyss Dec 13, 2024, 11:26 PM

#

infy casually checking if uacpi running under obos and finding symbols and shit to have a network backdoor

maiden birch Dec 13, 2024, 11:26 PM

#

median yoke xz utils backdoor repeat

that would mean u have to run some uacpi scripts

#

it doesnt have any

agile halo Dec 13, 2024, 11:27 PM

#

also for the pci irq api I am not sure if passing the device handle to all those functions is necessary, for an example the irq object handle could store the device handle in the kernel impl if its needed for eg. enabling a pci irq (though when the device is passed directly then the handle may be able to be implemented without allocations as eg. a pci irq number for the device)

median yoke Dec 13, 2024, 11:27 PM

#

actually when uAntiCheat

#

and uCheatEngine

maiden birch Dec 13, 2024, 11:27 PM

#

agile halo also for the pci irq api I am not sure if passing the device handle to all those...

yeah idk

#

we'll have to design it

maiden birch Dec 13, 2024, 11:29 PM

#

median yoke and uCheatEngine

feel free to make a repo under uDrivers/

#

meme

tawdry trench Dec 13, 2024, 11:31 PM

#

maiden birch read the code

galaxybrain

vital ibex Dec 14, 2024, 1:35 AM

#

not very many

#

only ones that attempt to reach posix compatibility

#

there are many that don't fall under that category

coral basalt Dec 14, 2024, 6:19 AM

#

yeah, ODR is one definition rule, you can declare anything as many times are you want, as long as the decls are the same. You can even declare the same thing differently in some circumstances, if you crave a little a excitement.

#

you can get some crazy bugs too if you somehow get an implicit declaration of a function (the return type defaults to int).

#

granted, your compiler should warn you about that

coral basalt Dec 14, 2024, 6:26 AM

#

maiden birch i was thinking of a `uapi_pci_device_open(address)`

I would be a big fan of this, it would let me remove a hack I have in place specifically for uacpi's pci accesses.

#

given that memory and io space accesses have map/unmap functions before/after use, it would be nice for pci (and anything else that fits into that category) to follow that style. Imo its a nice hook for the kernel to be able to track resource usage.

tawdry trench Dec 14, 2024, 9:30 AM

#

alright I'll start creating the virtio headers now

#

can't do any real setup yet because of uapi

maiden birch Dec 14, 2024, 10:22 AM

#

https://github.com/uDrivers/uAPI/tree/first-batch-of-headers

GitHub

GitHub - uDrivers/uAPI at first-batch-of-headers

Common set of headers used by all uDrivers. Contribute to uDrivers/uAPI development by creating an account on GitHub.

#

I added all the stuff we talked about into the readme

#

it's currently a PR

#

Take a look before i start adding new headers as well

#

@tawdry trench

#

I guess we also need a platform/ directory in there

#

similar to this idea from uacpi

tawdry trench Dec 14, 2024, 10:30 AM

#

maiden birch it's currently a PR

ah im blind

tawdry trench Dec 14, 2024, 10:31 AM

#

maiden birch Take a look before i start adding new headers as well

ill review in a few mins

maiden birch Dec 14, 2024, 10:31 AM

#

uacpi has platform-specific headers that can be overriden by the host if they want to

#

take a look how it works there and if u agree with this idea as well

tawdry trench Dec 14, 2024, 10:33 AM

#

how does overriding work

#

like defines for renaming?

maiden birch Dec 14, 2024, 10:33 AM

#

#ifdef UACPI_OVERRIDE_TYPES
#include "uacpi_types.h"
#else
// the actual header
#endif

#

so the host would create a uapi_types.h header

#

and define UAPI_OVERRIDE_TYPES

maiden birch Dec 14, 2024, 10:34 AM

#

tawdry trench like defines for renaming?

for libc specifically I allow both

tawdry trench Dec 14, 2024, 10:37 AM

#

ah

#

yea this is nice

tawdry trench Dec 14, 2024, 10:53 AM

#

reviewing the thing now

maiden birch Dec 14, 2024, 10:54 AM

#

nice

tawdry trench Dec 14, 2024, 10:56 AM

#

LGTM

maiden birch Dec 14, 2024, 10:57 AM

#

#error foo.h API version mismatch, make sure all libraries are using the base typo

#

the same

#

and should probably be u-lbiraries

tawdry trench Dec 14, 2024, 10:57 AM

#

or uDrivers

maiden birch Dec 14, 2024, 10:58 AM

#

yeah

#

theres already this

#

so how do we spell it

#

u-drivers or uDrivers or u drivers

tawdry trench Dec 14, 2024, 10:58 AM

#

uDrivers

#

like the repo

maiden birch Dec 14, 2024, 10:59 AM

#

technically not all u drivers have to live in the org

#

a u-driver is anything that uses uapi

tawdry trench Dec 14, 2024, 10:59 AM

#

true

#

just drivers

#

meme

maiden birch Dec 14, 2024, 10:59 AM

#

use uapi in native drivers too 😡

tawdry trench Dec 14, 2024, 11:00 AM

#

clueless

maiden birch Dec 14, 2024, 11:00 AM

#

so uDrivers?

tawdry trench Dec 14, 2024, 11:01 AM

#

ye

maiden birch Dec 14, 2024, 11:01 AM

#

fixed

#

one thing that i dont like

#

e.g. in uacpi not a single internal header is propagated into public headers

#

vice versa, sure

#

maybe uapi should be in udrv/

shut sorrel Dec 14, 2024, 11:03 AM

#

maiden birch use uapi in native drivers too 😡

then it should define the format that'll be used for loading them

maiden birch Dec 14, 2024, 11:03 AM

#

maiden birch maybe uapi should be in udrv/

but then this breaks the assumption about headers in udrv/ intended to be included standalone

#

all options suck

#

in uacpi platform/ is a part of public api

#

but since this is one repo we cant split it in halves

maiden birch Dec 14, 2024, 11:04 AM

#

shut sorrel then it should define the format that'll be used for loading them

lol

tawdry trench Dec 14, 2024, 11:05 AM

#

hm

#

we're actually approaching UDI territory

#

which is based

maiden birch Dec 14, 2024, 11:06 AM

#

im not faimiliar with it

tawdry trench Dec 14, 2024, 11:06 AM

#

it's basically what we're doing but way more abstracted (in a bad way)

#

to the point where it's almost impossible to implement in a hobby kernel

maiden birch Dec 14, 2024, 11:07 AM

#

lol

#

wdym?

tawdry trench Dec 14, 2024, 11:08 AM

#

sheer amount of complexity

#

the core spec is 250 pages

#

and that's really just the base

maiden birch Dec 14, 2024, 11:10 AM

#

damn

robust abyss Dec 14, 2024, 11:10 AM

#

Better idea

#

Make uapi be UEFI like

#

Sagiriwesmart

maiden birch Dec 14, 2024, 11:11 AM

#

by uefi-like u mean at least as buggy

robust abyss Dec 14, 2024, 11:11 AM

#

7000 page spec with protocols a core services table and so on

tawdry trench Dec 14, 2024, 11:11 AM

#

infinitedownvote

#

convert uacpi to use udi clueless

maiden birch Dec 14, 2024, 11:11 AM

#

please no

#

i will need a corporation to maintain it

shut sorrel Dec 14, 2024, 11:13 AM

#

OSDev Inc.

robust abyss Dec 14, 2024, 11:13 AM

#

UCorp

#

uInfy

maiden birch Dec 14, 2024, 11:13 AM

#

lmao

tawdry trench Dec 14, 2024, 11:14 AM

#

🥭

robust abyss Dec 14, 2024, 11:20 AM

#

How does udi looks like anyways

tawdry trench Dec 14, 2024, 11:20 AM

#

https://udi.certek.com/Release_1.01/PDF/core_spec_vol1.pdf

#

ch4

#

troll2

robust abyss Dec 14, 2024, 11:22 AM

#

You sure this isn't a kernel spec

#

KurisuSip

maiden birch Dec 14, 2024, 11:22 AM

#

wtf

robust abyss Dec 14, 2024, 11:23 AM

#

It's like taking Linux exported functions and saying "hey look at this generic driver interface"

tawdry trench Dec 14, 2024, 11:23 AM

#

exactly

#

that's why it failed

robust abyss Dec 14, 2024, 11:23 AM

#

Might as well do WDK at that point

tawdry trench Dec 14, 2024, 11:24 AM

#

(ignore the fact that the entirety of gnu was against it)

maiden birch Dec 14, 2024, 11:24 AM

#

driver interface aka random bullshit go

robust abyss Dec 14, 2024, 11:24 AM

#

I mean, it's clearly not random

#

But it's too much

maiden birch Dec 14, 2024, 11:24 AM

#

void udi_channel_anchor (
udi_channel_anchor_call_t *callback,
udi_cb_t *gcb,
udi_channel_t channel,
udi_index_t ops_idx,
void *channel_context );
typedef void udi_channel_anchor_call_t (
udi_cb_t *gcb,
udi_channel_t anchored_channel );

#

wtf is it talking about

robust abyss Dec 14, 2024, 11:24 AM

#

It's fine if the kernel shim does more than just call the entry point on the driver lol

tawdry trench Dec 14, 2024, 11:24 AM

#

maiden birch wtf is it talking about

KEKW

tawdry trench Dec 14, 2024, 1:35 PM

#

@maiden birch rest looks good to me

#

wyd now?

maiden birch Dec 14, 2024, 1:38 PM

#

need to add platform headers

#

but idk how to do it cleanly

#

because of the reasons i mentioned

tawdry trench Dec 14, 2024, 1:41 PM

#

just come up with anything and we can iron it out

maiden birch Dec 14, 2024, 1:42 PM

#

lets just put uapi into the root of include then

#

w/e

tawdry trench Dec 14, 2024, 1:42 PM

#

yes

maiden birch Dec 14, 2024, 1:46 PM

#

maybe we should also add

#ifndef UAPI_KERNEL_API
#error this header shouldn't be included directly, use kernel_api.h
#endif

#

because otherwise directly including the header will result in "this header is not needed..." etc

#

which may or may not be false

tawdry trench Dec 14, 2024, 2:42 PM

#

maiden birch maybe we should also add ``` #ifndef UAPI_KERNEL_API #error this header shouldn'...

probably

#

but doesn't that really convolute the headers

maiden birch Dec 14, 2024, 3:05 PM

#

tawdry trench but doesn't that really convolute the headers

Maybe

#

But they're already convoluted

#

So idk

tawdry trench Dec 14, 2024, 3:06 PM

#

yea who cares

#

anyways, i'm porting the spec structs now

#

that's the one thing i can do while i wait

maiden birch Dec 14, 2024, 3:06 PM

#

Ill get basic headers done today a bit later

tawdry trench Dec 14, 2024, 3:06 PM

#

W

maiden birch Dec 14, 2024, 3:06 PM

#

But u can stub stuff out atm

#

Like add fake uapi_uX types etc

#

Or uapi_kernel_alloc etc

#

Then you can replace those once I add the headers

#

If that's blocking you

tawdry trench Dec 14, 2024, 3:07 PM

#

okay sure

#

@maiden birch do we also want to share some helpers?

#

i have this rn

maiden birch Dec 14, 2024, 3:22 PM

#

Yeah we'll make a union header of those

#

I have a lot in uacpi as well

tawdry trench Dec 14, 2024, 3:28 PM

#

@maiden birch so do we want to go through with the unity builds idea

#

i have nothing against it

maiden birch Dec 14, 2024, 3:44 PM

#

As I mentioned I dont control how the project is built and you shouldn't either imo

#

You just provide a list of c files to build

tawdry trench Dec 14, 2024, 3:45 PM

#

well, i kinda want lsp support bc this is a nontrivial amount of code

maiden birch Dec 14, 2024, 3:45 PM

#

For that I have my tests/ directory and a test runner

#

And a dedicated cmake

tawdry trench Dec 14, 2024, 3:45 PM

#

ah

maiden birch Dec 14, 2024, 3:45 PM

#

So my test runner is like a uacpi host

tawdry trench Dec 14, 2024, 3:59 PM

#

idk if i can do the same

#

the test runner is standalone?

maiden birch Dec 14, 2024, 4:15 PM

#

Yup

tawdry trench Dec 14, 2024, 4:15 PM

#

hm

#

what do you test in uacpi

maiden birch Dec 14, 2024, 4:15 PM

#

Aml dumps

#

And my own test cases

tawdry trench Dec 14, 2024, 4:16 PM

#

ig i can do stuff like edid validation

maiden birch Dec 14, 2024, 4:16 PM

#

You can have a fake config space

#

And emulate a device in userspace

#

Verify that stuff that it submits is expected etc

tawdry trench Dec 14, 2024, 4:17 PM

#

ah true

agile halo Dec 14, 2024, 4:17 PM

#

just make sure that the stuff first works on real hw so you don't emulate wrong behaviour meme

maiden birch Dec 14, 2024, 4:17 PM

#

Virtio GPU is nice for that

maiden birch Dec 14, 2024, 4:18 PM

#

tawdry trench ah true

Or your test runner could be a small test kernel

tawdry trench Dec 14, 2024, 4:18 PM

#

"small"

maiden birch Dec 14, 2024, 4:18 PM

#

That you run in qemu and take screenshots to verify the fb state etc

tawdry trench Dec 14, 2024, 4:18 PM

#

only needs a fully fledged feature set

maiden birch Dec 14, 2024, 4:18 PM

#

Can always have no op free etc

#

A lot of edges to cut for testing purposes

livid sentinel Dec 14, 2024, 4:27 PM

#

maiden birch That you run in qemu and take screenshots to verify the fb state etc

how about pushing the fb thru the serial console or hash the fb and compare?

maiden birch Dec 14, 2024, 4:29 PM

#

@tawdry trench I just realized, u can literally test as part of a Linux module, just unload virtio GPU and load your own instead

maiden birch Dec 14, 2024, 4:29 PM

#

livid sentinel how about pushing the fb thru the serial console or hash the fb and compare?

Sure

tawdry trench Dec 14, 2024, 4:31 PM

#

maiden birch <@173337436989030401> I just realized, u can literally test as part of a Linux m...

true lol

#

uapi linux bindings meme

robust abyss Dec 14, 2024, 4:31 PM

#

You can maybe use the uACPI-OS as a base, given it already runs uacpi

#

It is missing threads but has other stuff

maiden birch Dec 14, 2024, 4:33 PM

#

tawdry trench uapi linux bindings <:meme:575445094589661215>

Yeah

maiden birch Dec 14, 2024, 4:33 PM

#

robust abyss You can maybe use the uACPI-OS as a base, given it already runs uacpi

Or that

#

But that doesnt have mesa

tawdry trench Dec 14, 2024, 4:33 PM

#

uKernel troll2

robust abyss Dec 14, 2024, 4:33 PM

#

True

#

Speedrun to make a Linux clone any%?

tawdry trench Dec 14, 2024, 4:34 PM

#

i guess we could try porting to managarm or astral

#

managarm i have to do either way

robust abyss Dec 14, 2024, 4:34 PM

#

Managarm already has mesa so that would be helpful

tawdry trench Dec 14, 2024, 4:34 PM

#

yeop

maiden birch Dec 14, 2024, 4:35 PM

#

True

tawdry trench Dec 14, 2024, 4:35 PM

#

i'll be in vc if anyone wants to watch me seethe with virtiogpu

#

@maiden birch can we have a common HEADER_START or whatever

#

for c++

#

maiden birch Dec 14, 2024, 4:39 PM

#

tawdry trench <@353593948523986945> can we have a common HEADER_START or whatever

Yes

tawdry trench Dec 14, 2024, 4:39 PM

#

cool

maiden birch Dec 14, 2024, 4:39 PM

#

UAPI_BEGIN_DECLS?

tawdry trench Dec 14, 2024, 4:40 PM

#

sure

#

and END_

maiden birch Dec 14, 2024, 4:40 PM

#

Ye

tawdry trench Dec 14, 2024, 4:41 PM

#

which C standard are we using

#

i say > C11

median yoke Dec 14, 2024, 4:42 PM

#

Ansi

tawdry trench Dec 14, 2024, 4:42 PM

#

but _Static_assert

median yoke Dec 14, 2024, 4:42 PM

#

Jk

robust abyss Dec 14, 2024, 4:53 PM

#

I say C11

#

Sweet sweet stdatomic.h

livid sentinel Dec 14, 2024, 4:54 PM

#

C89

robust abyss Dec 14, 2024, 4:54 PM

#

Just use threads.h :^)

#

Also static assert is nice

tawdry trench Dec 14, 2024, 4:55 PM

#

ye

hexed flare Dec 14, 2024, 5:11 PM

#

Why not C23 ultrameme

tawdry trench Dec 14, 2024, 5:11 PM

#

not everyone uses that iirc

#

(i do)

hexed flare Dec 14, 2024, 5:12 PM

#

Like if you're writing a hobby os in 2024, why are you using an old compiler ultrameme

tawdry trench Dec 14, 2024, 5:12 PM

#

maybe some weird code generator that can't cope with latest standards

maiden birch Dec 14, 2024, 5:18 PM

#

robust abyss Sweet sweet stdatomic.h

U gotta be kidding lol

tawdry trench Dec 14, 2024, 5:19 PM

#

wat

robust abyss Dec 14, 2024, 5:19 PM

#

hmm?

tawdry trench Dec 14, 2024, 5:19 PM

#

did you not know about that

robust abyss Dec 14, 2024, 5:19 PM

#

stdatomic is the best thing since bread

maiden birch Dec 14, 2024, 5:19 PM

#

Implicit seq cst, overrides default behavior of operators, doesnt work with normal types

#

Absolute crap

robust abyss Dec 14, 2024, 5:19 PM

#

bruh no

#

it has explicit memory ordering

#

the defaults are seq cst but who cares no one uses the defaults

maiden birch Dec 14, 2024, 5:20 PM

#

robust abyss it has explicit memory ordering

Implicit with operators

#

Also u have to use _Atomic and u cant use normal types

robust abyss Dec 14, 2024, 5:20 PM

#

meh, I never use the operator versions

#

I mean, that is the same in C++, you have to use std::atomic<type>

#

of course operator overloading so lol

maiden birch Dec 14, 2024, 5:21 PM

#

Yeah

#

In this case its like java with strings and operator ==

#

Secret special case u have to know about

robust abyss Dec 14, 2024, 5:21 PM

#

but anyone who needs to touch stdatomic.h knows what to do

#

and if he doesn't he failed before even trying

maiden birch Dec 14, 2024, 5:22 PM

#

Anyway not a fan but to each their own

robust abyss Dec 14, 2024, 5:22 PM

#

meh, I only take it for the _explicit versions of stuff

#

and I prefer it over using compiler specific variants

tawdry trench Dec 14, 2024, 5:23 PM

#

cry_glasses currently 4 layers deep into a struct declaration

robust abyss Dec 14, 2024, 5:23 PM

#

the people who use the operator stuff are the same ones who use __sync_synchronize to tell themselves everything is now atomic

maiden birch Dec 14, 2024, 5:23 PM

#

Lol

tawdry trench Dec 14, 2024, 5:24 PM

#

who the fuck made the edid struct

#

#

how to most inconveniently organize data

#

kill me

#

@maiden birch EDID is done

maiden birch Dec 14, 2024, 5:44 PM

#

Nice but thats not how I do packed in uACPI

#

For MSVC compatibility

tawdry trench Dec 14, 2024, 5:44 PM

#

aaaaaaaaaa

maiden birch Dec 14, 2024, 5:44 PM

#

Also everything inside a packed struct is packed

tawdry trench Dec 14, 2024, 5:44 PM

#

pls no macro bullshit 😢

tawdry trench Dec 14, 2024, 5:44 PM

#

maiden birch Also everything inside a packed struct is packed

apparently not

maiden birch Dec 14, 2024, 5:45 PM

#

tawdry trench pls no macro bullshit 😢

Same bullshit but slightly different

#

Take a look at acpi.h in uacpi

tawdry trench Dec 14, 2024, 5:45 PM

#

not that bad ig

#

do we also use UAPI_EXPECT_SIZEOF

maiden birch Dec 14, 2024, 5:46 PM

#

There's also EXPECT_SIZEOF which I recommend u use

maiden birch Dec 14, 2024, 5:46 PM

#

tawdry trench do we also use UAPI_EXPECT_SIZEOF

Its nice to verify spec structs

tawdry trench Dec 14, 2024, 5:46 PM

#

meme

maiden birch Dec 14, 2024, 5:46 PM

#

To make sure u didnt mess up

tawdry trench Dec 14, 2024, 5:46 PM

#

i did this

maiden birch Dec 14, 2024, 5:46 PM

#

tawdry trench <:meme:575445094589661215>

Yeah basically

tawdry trench Dec 14, 2024, 5:46 PM

#

do we typedef or just use struct

#

i'm indifferent really

maiden birch Dec 14, 2024, 5:46 PM

#

I typedef public structs

#

Internal ones are not

#

Also all public macros and functions with external linkage are prefixed with uacpi

#

Same goes for structs

#

But u already do this ig

tawdry trench Dec 14, 2024, 5:52 PM

#

@maiden birch

#pragma once

#include "common.h"

UAPI_BEGIN_DECLS

// Extended Display Identification Data
UAPI_PACKED(struct drm_edid {
    uint8_t header[8];      // 00 FF FF FF FF FF FF 00

    // Vendor & Product Identification
    uint16_t mfr_name;
    uint16_t product_id;
    uint32_t serial_id;
    uint8_t model_week;
    uint8_t model_year;

    // EDID Structure Version & Revision
    uint8_t version;
    uint8_t revision;

    // Basic Display Parameters & Features
    uint8_t video_input;
    uint8_t h_screen_cm;
    uint8_t v_screen_cm;
    uint8_t gamma;
    uint8_t features;

    // Color Characteristics
    uint8_t red_green_low;
    uint8_t blue_white_low;
    uint8_t red_x;
    uint8_t red_y;
    uint8_t green_x;
    uint8_t green_y;
    uint8_t blue_x;
    uint8_t blue_y;
    uint8_t white_x;
    uint8_t white_y;

    // Timings
    struct
    {
        uint8_t time_1;
        uint8_t time_2;
        uint8_t time_mfr;
    } established_timings;

    struct
    {
        uint8_t h_resoution;       // (Horizontal addressable pixels ÷ 8) – 31
        uint8_t aspect_and_ref;       // See 3.19
    } standard_timings[8];

    struct
    {
        // Pixel clock frequency divided by 10000.
        // If 0, don't use `data.pixel`.
        uint16_t pix_clock;
        union
        {
            struct
            {
                uint8_t h_video_low;
                uint8_t h_blank_low;
                uint8_t h_video_blank_high;
                uint8_t v_video_low;
                uint8_t v_blank_low;
                uint8_t v_video_blank_high;
                uint8_t h_sync_offset_low;
                uint8_t h_sync_pulse_width_low;
                uint8_t v_sync_offset_pulse_width_low;
                uint8_t bit_definitons;       // High order bits for the previous 4 fields
                uint8_t h_video_mm_low;
                uint8_t v_video_mm_low;
                uint8_t hv_video_mm_high;
                uint8_t h_border;
                uint8_t v_border;
                uint8_t misc;
            } pixel;

            struct
            {
                uint8_t _reserved1;
                uint8_t tag;
                uint8_t _reserved2;
                union
                {
                    char serial_number[13];
                    char alnum_string[13];
                } data;
            } other;
        } data;
    } detailed_timings[4];

    uint8_t extension_block_count;
    uint8_t checksum;
})
UAPI_EXPECT_SIZEOF(struct drm_edid, 0x80);

UAPI_END_DECLS

maiden birch Dec 14, 2024, 5:52 PM

#

Wrong types but yea

tawdry trench Dec 14, 2024, 5:52 PM

#

huh

#

oh i thought we said to use stdint

maiden birch Dec 14, 2024, 5:53 PM

#

Hm?

#

Well I was gonna import a platform/types.h

#

Which creates uapi_uX based on stdinth

tawdry trench Dec 14, 2024, 5:53 PM

#

.

maiden birch Dec 14, 2024, 5:53 PM

#

But it doesnt have to be

tawdry trench Dec 14, 2024, 5:53 PM

#

but sure

#

i don't mind that

maiden birch Dec 14, 2024, 5:53 PM

#

.

tawdry trench Dec 14, 2024, 5:53 PM

#

what about big endian fields

#

can we have _be types

maiden birch Dec 14, 2024, 5:54 PM

#

What would that do

tawdry trench Dec 14, 2024, 5:54 PM

#

the host can inject some checker macro there

#

like linux does

maiden birch Dec 14, 2024, 5:54 PM

#

Sure yeah we can add explicit endianness types to types.h

tawdry trench Dec 14, 2024, 5:54 PM

#

nice

maiden birch Dec 14, 2024, 5:54 PM

#

uapi_le_u8?

tawdry trench Dec 14, 2024, 5:55 PM

#

uapi_u8_le

maiden birch Dec 14, 2024, 5:55 PM

#

Sure

tawdry trench Dec 14, 2024, 5:58 PM

#

maiden birch Dec 14, 2024, 5:58 PM

#

Nice

#

Is this supposed to be a public struct?

tawdry trench Dec 14, 2024, 5:58 PM

#

yes

#

the host might want to evaluate the EDID

maiden birch Dec 14, 2024, 5:58 PM

#

Ah ok

#

Btw

#

Can we do include <udrm/common.h>?

tawdry trench Dec 14, 2024, 5:59 PM

#

yea this is just for testing rn

#

i have some bullshit defines in there

maiden birch Dec 14, 2024, 5:59 PM

#

Ah

#

I see

tawdry trench Dec 14, 2024, 6:00 PM

#

if only limine didn't pass a null pointer for the EDID i could see if it worked

maiden birch Dec 14, 2024, 6:00 PM

#

Wait also

#

Cant we just do uapi_le8?

#

We dont really need signed le

tawdry trench Dec 14, 2024, 6:01 PM

#

uapi_l8

#

and b8

#

meme

maiden birch Dec 14, 2024, 6:01 PM

#

Late

#

Bait

#

Lol

robust abyss Dec 14, 2024, 6:01 PM

#

Well, the le/be versions are only really needed for packed structs where you assume endianness, the rest could be just uxx

maiden birch Dec 14, 2024, 6:02 PM

#

Thats the idea

tawdry trench Dec 14, 2024, 6:02 PM

#

it's more a hint for the user

robust abyss Dec 14, 2024, 6:02 PM

#

Just don't forget to ntos

tawdry trench Dec 14, 2024, 6:02 PM

#

and/or compile time checker

maiden birch Dec 14, 2024, 6:02 PM

#

We wanted to do uapi_le_u8

robust abyss Dec 14, 2024, 6:02 PM

#

Linux has a checker iirc

tawdry trench Dec 14, 2024, 6:02 PM

#

yep

maiden birch Dec 14, 2024, 6:02 PM

#

Im saying we don't need the u part

tawdry trench Dec 14, 2024, 6:02 PM

#

hm

#

what if you want to convey "this is a signed integer which first has to be converted"

maiden birch Dec 14, 2024, 6:03 PM

#

tawdry trench what if you want to convey "this is a signed integer which first has to be conve...

U aren't really supposed to use them directly

#

U should be doing be_to_cpu etc

robust abyss Dec 14, 2024, 6:03 PM

#

Yeah

#

And if you do it means you pass it without converting it first

maiden birch Dec 14, 2024, 6:04 PM

#

As far as conveying it well

#

Idk

tawdry trench Dec 14, 2024, 6:05 PM

#

how should i be adding the uapi submodule

#

directly into include/?

maiden birch Dec 14, 2024, 6:06 PM

#

Yes

tawdry trench Dec 14, 2024, 6:06 PM

#

fatal: 'include' already exists in the index

#

stare

maiden birch Dec 14, 2024, 6:06 PM

#

Wut

tawdry trench Dec 14, 2024, 6:06 PM

#

do i have to do include/uapi????

#

this reeks of gitification

maiden birch Dec 14, 2024, 6:07 PM

#

What

tawdry trench Dec 14, 2024, 6:07 PM

#

let me try something

maiden birch Dec 14, 2024, 6:07 PM

#

Im not sure what youre doing

tawdry trench Dec 14, 2024, 6:07 PM

#

adding a submodule

maiden birch Dec 14, 2024, 6:07 PM

#

We agreed on the project structure like yesterday

#

How did u forget already lol

tawdry trench Dec 14, 2024, 6:07 PM

#

git submodule add https://github.com/uDrivers/uAPI include/

maiden birch Dec 14, 2024, 6:07 PM

#

Thats not how u uh

tawdry trench Dec 14, 2024, 6:08 PM

#

really?

maiden birch Dec 14, 2024, 6:08 PM

#

cd include

#

Got submodule add link uapi

#

The second arg is the name of the new directory

tawdry trench Dec 14, 2024, 6:08 PM

#

oh

#

well that's what i was asking lol

#

i forgotr

maiden birch Dec 14, 2024, 6:09 PM

#

Yeah

#

Lowercase pls

tawdry trench Dec 14, 2024, 6:09 PM

#

yea ofc

#

maiden birch Dec 14, 2024, 6:09 PM

#

Yup

#

Wait your git files and the license are in SRC?

tawdry trench Dec 14, 2024, 6:10 PM

#

no

shut sorrel Dec 14, 2024, 6:10 PM

#

indentation

maiden birch Dec 14, 2024, 6:11 PM

#

tawdry trench no

Huh

tawdry trench Dec 14, 2024, 6:11 PM

#

why should they be in src

maiden birch Dec 14, 2024, 6:11 PM

#

Your vscode sure looks like they are lol

tawdry trench Dec 14, 2024, 6:11 PM

#

maiden birch Dec 14, 2024, 6:12 PM

#

Ah lmao

tawdry trench Dec 14, 2024, 6:12 PM

#

oh i see what's happening

#

i changed my indent width

#

for the subdirs

#

okay so

#

is this right

#

public headers in <udrm/...>

maiden birch Dec 14, 2024, 6:14 PM

#

Yup

tawdry trench Dec 14, 2024, 6:14 PM

#

internal ones in <udrm/internal/....>

maiden birch Dec 14, 2024, 6:15 PM

#

U dont have any public ones right?

tawdry trench Dec 14, 2024, 6:15 PM

#

well, edid.h at least

maiden birch Dec 14, 2024, 6:15 PM

#

Your indent is impossible for me to understand

tawdry trench Dec 14, 2024, 6:15 PM

#

rofl

#

sec

robust abyss Dec 14, 2024, 6:16 PM

#

Wait edid.h isn't under internal lol

maiden birch Dec 14, 2024, 6:16 PM

#

Apparently

tawdry trench Dec 14, 2024, 6:16 PM

#

marvin@pc0 ~/r/u/udrm (master)> tree include
include
├── uapi
│   ├── LICENSE
│   └── README.md
└── udrm
    ├── common.h
    ├── edid.h
    ├── internal
    │   ├── virtio
    │   └── xe
    └── udrm.h

6 directories, 5 files
marvin@pc0 ~/r/u/udrm (master)>

robust abyss Dec 14, 2024, 6:16 PM

#

Oh it indents with nothing where the > would go

#

That's a weird thing to do

maiden birch Dec 14, 2024, 6:16 PM

#

tawdry trench ``` marvin@pc0 ~/r/u/udrm (master)> tree include include ├── uapi │ ├── LICENS...

Yeah looks good for now barring the stray common thing

tawdry trench Dec 14, 2024, 6:17 PM

#

yep

#

that'll be in uapi

maiden birch Dec 14, 2024, 6:17 PM

#

Yeye

robust abyss Dec 14, 2024, 6:17 PM

#

The assumption is that you would -Iudrm/include right? Would would having multiple uapi dirs in the path play out?

maiden birch Dec 14, 2024, 6:17 PM

#

robust abyss The assumption is that you would -Iudrm/include right? Would would having multip...

No

robust abyss Dec 14, 2024, 6:18 PM

#

Oh? So how is the include supposed to work

maiden birch Dec 14, 2024, 6:18 PM

#

Udrm is the include dir

#

Or rather

#

As if it was inside

robust abyss Dec 14, 2024, 6:18 PM

#

So in your project you would do <edid.h>?

maiden birch Dec 14, 2024, 6:18 PM

#

Oh wait

#

Udrm not uapi

tawdry trench Dec 14, 2024, 6:18 PM

#

i think you two are talking around each other

maiden birch Dec 14, 2024, 6:19 PM

#

Yes you're right

robust abyss Dec 14, 2024, 6:19 PM

#

Smuggy

tawdry trench Dec 14, 2024, 6:19 PM

#

Tomat

maiden birch Dec 14, 2024, 6:19 PM

#

robust abyss The assumption is that you would -Iudrm/include right? Would would having multip...

Udrm/uapi

#

Its under right

robust abyss Dec 14, 2024, 6:19 PM

#

Well, right now in his path it's under his direct include dir

maiden birch Dec 14, 2024, 6:21 PM

#

Wait

#

Yeah that's wrong ofc

#

it must be inside udrm

tawdry trench Dec 14, 2024, 6:22 PM

#

death

robust abyss Dec 14, 2024, 6:22 PM

#

Yeah that's what I remembered

maiden birch Dec 14, 2024, 6:22 PM

#

Good catch

#

We moved it from internal, but its still inside udrm not literal include root

tawdry trench Dec 14, 2024, 6:23 PM

#

so move down to include/udrm/uapi?

maiden birch Dec 14, 2024, 6:23 PM

#

Yup

#

Git mv

tawdry trench Dec 14, 2024, 6:23 PM

#

done

#

include
└── udrm
    ├── common.h
    ├── connector.h
    ├── crtc.h
    ├── edid.h
    ├── encoder.h
    ├── internal
    │   ├── virtio
    │   └── xe
    ├── plane.h
    └── uapi
        ├── LICENSE
        └── README.md

6 directories, 8 files

maiden birch Dec 14, 2024, 6:24 PM

#

Correct

maiden birch Dec 14, 2024, 6:26 PM

#

tawdry trench ``` include └── udrm ├── common.h ├── connector.h ├── crtc.h ├──...

Where's udrm.h?

tawdry trench Dec 14, 2024, 6:26 PM

#

i just removed it

#

i'll put it back

maiden birch Dec 14, 2024, 6:26 PM

#

Ah

tawdry trench Dec 14, 2024, 6:26 PM

#

misclick

robust abyss Dec 14, 2024, 6:35 PM

#

Are you going to put all graphic drivers in this repo?

#

At least the drm part

tawdry trench Dec 14, 2024, 6:35 PM

#

yea

robust abyss Dec 14, 2024, 6:35 PM

#

Nice

tawdry trench Dec 14, 2024, 6:36 PM

#

the idea is to have two layers of abstraction

#

one for just the drivers

#

one for the DRM system

#

because i know at least 2 kernels with their own DRM system

median yoke Dec 14, 2024, 6:36 PM

#

isn't DRM like some sorta syscall thing

tawdry trench Dec 14, 2024, 6:36 PM

#

so that can plug right into there

median yoke Dec 14, 2024, 6:36 PM

#

what even is DRM

robust abyss Dec 14, 2024, 6:36 PM

#

An API

tawdry trench Dec 14, 2024, 6:36 PM

#

direct rendering manager

#

scheduler and manager for rendering with a gpu

median yoke Dec 14, 2024, 6:36 PM

#

it's exposed to userspace via ioctl right

tawdry trench Dec 14, 2024, 6:36 PM

#

yes

maiden birch Dec 14, 2024, 6:37 PM

#

median yoke it's exposed to userspace via ioctl right

/Dev/dri

robust abyss Dec 14, 2024, 6:37 PM

#

You are still expected to use a usermode driver like mesa to actually do stuff

#

Given it only manages contexts and memory management of the GPU

median yoke Dec 14, 2024, 6:37 PM

#

ah ok

maiden birch Dec 14, 2024, 6:38 PM

#

Yeah it only exposes modeset, scanout, memory management, and command submission

#

The rest is userspace

#

And command submission is a GPU specific ioctl

robust abyss Dec 14, 2024, 6:39 PM

#

There are a bunch of GPU specific stuff

maiden birch Dec 14, 2024, 6:39 PM

#

Yeah

median yoke Dec 14, 2024, 6:39 PM

#

which mesa deals with, right?

maiden birch Dec 14, 2024, 6:40 PM

#

Yes

#

And xorg or weston

#

They're DRM masters

#

And can submit admin commands

#

Like modeset

livid sentinel Dec 14, 2024, 6:42 PM

#

on which step are the user programm windows managed?

maiden birch Dec 14, 2024, 6:42 PM

#

maiden birch And xorg or weston

These usually have compositors, which draw windows as textures on a canvas

#

That canvas is then give to GPU for scanout

median yoke Dec 14, 2024, 6:43 PM

#

what about stuff like running shaders

maiden birch Dec 14, 2024, 6:43 PM

#

Hm?

median yoke Dec 14, 2024, 6:43 PM

#

after compiling them or whatever

#

how are they rab

#

*ran

#

on the gpu

#

is that DRM's problem

#

or mesa's

maiden birch Dec 14, 2024, 6:43 PM

#

All the work your application does for drawing is done into its own anonymous buffer

#

Which may be shared with the window manager

#

If you want it displayed on the screen

#

Thats the idea of dri

#

You talk to the GPU directly and it renders stuff into your own buffers

#

But then you talk to the window manager and ask it to take the buffer and display it

robust abyss Dec 14, 2024, 6:44 PM

#

Yeah

maiden birch Dec 14, 2024, 6:44 PM

#

You could also save it as PNG for example

#

Which mesa does for testing

robust abyss Dec 14, 2024, 6:45 PM

#

Idk about Linux user space but in windows when in full screen mode the idea is that you control the framebuffer of the actual display and why it can reduce latency

tawdry trench Dec 14, 2024, 6:46 PM

#

that's why tabbing out kills everything trollHD

robust abyss Dec 14, 2024, 6:46 PM

#

Yeah lol

maiden birch Dec 14, 2024, 6:46 PM

#

robust abyss Idk about Linux user space but in windows when in full screen mode the idea is t...

This is new tech right

#

Iirc before it would copy shit

robust abyss Dec 14, 2024, 6:46 PM

#

I think that's how it always worked in windows

#

When using exclusive full screen mode

maiden birch Dec 14, 2024, 6:46 PM

#

Oh yeah the new thing is shared surfaces

#

Before it would memcpy into compositor

robust abyss Dec 14, 2024, 6:46 PM

#

Yeah

maiden birch Dec 14, 2024, 6:48 PM

#

median yoke or mesa's

Well the drawing is done via opengl and vulkan right

#

Mesa provides implementations of those apis

robust abyss Dec 14, 2024, 6:48 PM

#

Mesa has the shader compiler

maiden birch Dec 14, 2024, 6:48 PM

#

robust abyss Mesa has the shader compiler

U mean like 3 layers

#

Glsl to nim to gpu

robust abyss Dec 14, 2024, 6:49 PM

#

Iirc they even have their entirely own IR and optimizations since they felt llvm wasn't really the best for it

#

Yeah

#

There is some cool blog where they explain why they have their own IR

#

uVulkan when :^)

agile halo Dec 14, 2024, 6:50 PM

#

its also interesting how you get access to the ir inside gl programs KEKW

maiden birch Dec 14, 2024, 6:51 PM

#

robust abyss uVulkan when :^)

uMesa

agile halo Dec 14, 2024, 6:52 PM

#

agile halo its also interesting how you get access to the ir inside gl programs <:KEKW:9137...

or well not officially but you can use glGetProgramBinary to get the ir, modify the ir, recalculate a crc in the header and then create a new program + use glProgramBinary to load the modified ir to there

robust abyss Dec 14, 2024, 6:53 PM

#

Sadge it's not giving you the direct shader binary meant for the exact gpu :^)

maiden birch Dec 14, 2024, 6:53 PM

#

agile halo or well not officially but you can use glGetProgramBinary to get the ir, modify ...

Why does gl even have that

robust abyss Dec 14, 2024, 6:53 PM

#

For faster loading

maiden birch Dec 14, 2024, 6:53 PM

#

Ah

agile halo Dec 14, 2024, 6:53 PM

#

for caching shaders I'd think yeah

maiden birch Dec 14, 2024, 6:54 PM

#

Doesn't even save that much

static prairie Dec 14, 2024, 6:54 PM

#

mesa already does shader caching anyway

agile halo Dec 14, 2024, 6:54 PM

#

I mean it does save the glsl compilation + the nir generation + optimizations that it does in that process

robust abyss Dec 14, 2024, 6:54 PM

#

^

#

And nothing stops you from technically saving the raw binary

#

Like api wise

maiden birch Dec 14, 2024, 6:55 PM

#

robust abyss And nothing stops you from technically saving the raw binary

Hm?

static prairie Dec 14, 2024, 6:55 PM

#

mesa's shader compilation + caching is actually really fast, like when CS2 first came out people were complaining about shader compilation stutters but on mesa it was not perceptible

robust abyss Dec 14, 2024, 6:55 PM

#

Like saving the raw machine code for the GPU

static prairie Dec 14, 2024, 6:55 PM

#

even when skipping steam's shader precompilation

robust abyss Dec 14, 2024, 6:55 PM

#

So many shader cache levels

#

Just use spirv smh

tawdry trench Dec 14, 2024, 6:57 PM

#

spir v supremacy

#

actually, lets use DXBC

#

clueless

maiden birch Dec 14, 2024, 6:57 PM

#

robust abyss Like saving the raw machine code for the GPU

Yeah but no gl api for that

robust abyss Dec 14, 2024, 6:57 PM

#

The GetProgramBinary could give you that no? If the implementor decided to

agile halo Dec 14, 2024, 6:57 PM

#

yes

maiden birch Dec 14, 2024, 6:58 PM

#

Well idk how the spec defines it

median yoke Dec 14, 2024, 6:59 PM

#

maiden birch Well idk how the spec defines it

The format of the program binary written into binary is returned in the variable whose address is given by binaryFormat, and may be implementation dependent. The binary produced by the GL may subsequently be returned to the GL by calling glProgramBinary, with binaryFormat and length set to the values returned by glGetProgramBinary, and passing the returned binary data in the binary parameter.

#

https://registry.khronos.org/OpenGL-Refpages/gl4/html/glGetProgramBinary.xhtml

#

from there

static prairie Dec 14, 2024, 6:59 PM

#

maiden birch Yeah but no gl api for that

glShaderBinary & glSpecializeShader

#

https://www.khronos.org/opengl/wiki/SPIR-V

SPIR-V

robust abyss Dec 14, 2024, 7:00 PM

#

Damn

static prairie Dec 14, 2024, 7:00 PM

#

If you link multiple shader objects in the same program, then either all of them must be SPIR-V shaders or none of them may be SPIR-V shaders. You cannot link SPIR-V to non-SPIR-V shaders in a program.
interesting

robust abyss Dec 14, 2024, 7:00 PM

#

Lmao

maiden birch Dec 14, 2024, 7:01 PM

#

static prairie > If you link multiple shader objects in the same program, then either all of th...

The way mesa does linking is just copy paste I think

#

So makes sense

maiden birch Dec 14, 2024, 7:02 PM

#

static prairie glShaderBinary & glSpecializeShader

interesting

tawdry trench Dec 14, 2024, 7:55 PM

#

@maiden birch https://github.com/uDrivers/uAPI/pull/1/files#diff-b335630551682c19a781afebcf4d07bf978fb1f8ac04c6bf87428ed5106870f5R15-R22

GitHub

Docs & a few base headers by d-tatianin · Pull Request #1 · uDriver...

#

this says i should have uapi in include/udrm/internal/uapi

#

not include/udrm/uapi

maiden birch Dec 14, 2024, 7:56 PM

#

Yeah ill change that

tawdry trench Dec 14, 2024, 7:56 PM

#

oh ok

maiden birch Dec 14, 2024, 7:56 PM

#

Ill get to this pr in about a few hours

tawdry trench Dec 14, 2024, 7:56 PM

#

dope

#

lmk when

maiden birch Dec 14, 2024, 7:56 PM

#

This version is before we agreed on moving out of internal

tawdry trench Dec 14, 2024, 7:57 PM

#

yea i forgot most things yesterday because i was on like 2h of sleep

maiden birch Dec 14, 2024, 7:58 PM

#

This was today lol

tawdry trench Dec 14, 2024, 7:58 PM

#

fml

maiden birch Dec 14, 2024, 7:58 PM

#

Lmao

maiden birch Dec 14, 2024, 10:50 PM

#

alright time to add the goddamn headers

tawdry trench Dec 14, 2024, 10:50 PM

#

lets go

maiden birch Dec 14, 2024, 10:53 PM

#

updated the readme with all of the things we have talked about today

tawdry trench Dec 14, 2024, 10:53 PM

#

cool

maiden birch Dec 14, 2024, 10:53 PM

#

that was a call to re-review

tawdry trench Dec 14, 2024, 10:53 PM

#

on it

maiden birch Dec 14, 2024, 10:53 PM

#

lol

tawdry trench Dec 14, 2024, 10:56 PM

#

https://github.com/uDrivers/uAPI/pull/1/commits/ff65fa1e0834df24f75fdbc084303ab250909f63#diff-b335630551682c19a781afebcf4d07bf978fb1f8ac04c6bf87428ed5106870f5R17-R27

GitHub

Docs & a few base headers by d-tatianin · Pull Request #1 · uDriver...

#

is this correct?

#

is the first macro not a header guard?

maiden birch Dec 14, 2024, 10:57 PM

#

maiden birch maybe we should also add ``` #ifndef UAPI_KERNEL_API #error this header shouldn'...

.

#

actually i cant even make it a header guard

#

because there will be multiple kernel_api.h

#

and this #define UAPI_WANTS_MEMORY should also be under ifndef

#

do we agree?

agile halo Dec 14, 2024, 11:00 PM

#

yeah sounds fine to me

maiden birch Dec 14, 2024, 11:00 PM

#

for cases like

glue.c:

#include <udrm/kernel_api.h>
#include <uacpi/kernel_api.h>

tawdry trench Dec 14, 2024, 11:30 PM

#

sounds good

maiden birch Dec 14, 2024, 11:38 PM

#

@tawdry trench pushed a bunch of base platform headers

tawdry trench Dec 14, 2024, 11:38 PM

#

👍

robust abyss Dec 14, 2024, 11:41 PM

#

Petition to drop msvc support so you can use statement expressions in relevant macros 😔

maiden birch Dec 14, 2024, 11:42 PM

#

i know its tempting but

#

no

#

it would simplify align.h with typeof() as well

robust abyss Dec 14, 2024, 11:42 PM

#

Like at this point even Microsoft are ditching msvc in favor of clang Smuggy

maiden birch Dec 14, 2024, 11:43 PM

#

still kinda the defacto default windows compiler

robust abyss Dec 14, 2024, 11:43 PM

#

Smh

#

I know I am just memeing

maiden birch Dec 14, 2024, 11:43 PM

#

but yeah im sad about the crappy minmax and align

robust abyss Dec 14, 2024, 11:44 PM

#

Btw visual studio actually has built in support for clang these days, full drop in replacement (especially with the clang-cl frontend)

#

As simple as a drop-down option in the build options

#

It's pretty cool

maiden birch Dec 14, 2024, 11:45 PM

#

nice

tawdry trench Dec 14, 2024, 11:45 PM

#

based

#

yet another clang W

robust abyss Dec 14, 2024, 11:45 PM

#

Maybe games will finally stop shipping with code that looks like a GCC 4 debug output

maiden birch Dec 14, 2024, 11:45 PM

#

lmao

robust abyss Dec 14, 2024, 11:46 PM

#

Like man, msvc got an ssa only in the last couple years

#

And before that it didn't even have a proper ast

#

Optimizations that anyone reading the dragon book could do were missing

#

Anyways im waiting for more uapi headers to look at

maiden birch Dec 14, 2024, 11:49 PM

#

ill add a few basic ones rn

#

then we can add the more advanced ones tomorrow

robust abyss Dec 14, 2024, 11:51 PM

#

Are you going to move uacpi to use uapi as well?

maiden birch Dec 14, 2024, 11:51 PM

#

eventually yes

robust abyss Dec 14, 2024, 11:51 PM

#

Nice

maiden birch Dec 14, 2024, 11:51 PM

#

since i have more uprojects planned once i finish it

#

ill have to do that

#

pushed 3 uapi headers

robust abyss Dec 15, 2024, 12:00 AM

#

Won't the sized frees cause weird collisions if some drivers use one and some the other?

maiden birch Dec 15, 2024, 12:00 AM

#

its not up to the drivers to decide

#

its a project-wide client configuration

robust abyss Dec 15, 2024, 12:01 AM

#

Hmmmm I see

#

So it's expected so drivers support both

maiden birch Dec 15, 2024, 12:01 AM

#

yeah

robust abyss Dec 15, 2024, 12:01 AM

#

Nice

maiden birch Dec 15, 2024, 12:01 AM

#

maybe tahts too much we could talk about it

robust abyss Dec 15, 2024, 12:01 AM

#

I assume the same for formatted logs

maiden birch Dec 15, 2024, 12:02 AM

#

yup

#


    UAPI_CACHING_WB,
    UAPI_CACHING_WT,
    UAPI_CACHING_WC,
    UAPI_CACHING_UC,

#

how do u cover NgNrE crap with this

agile halo Dec 15, 2024, 12:08 AM

#

maiden birch maybe tahts too much we could talk about it

tbh for most cases it wouldn't be that hard to support both frees (as the size is known either statically or otherwise) so you can just define eg a macro ```c
#ifdef UAPI_SIZED_FREES
#define uhda_free(ptr, size) uapi_kernel_free(ptr, size)
#else
#define uhda_free(ptr, size) uapi_kernel_free(ptr)
#endif

maiden birch Dec 15, 2024, 12:08 AM

#

literally what uacpi does

#

#ifdef UACPI_SIZED_FREES
#define uacpi_free(mem, size) uacpi_kernel_free(mem, size)
#else
#define uacpi_free(mem, _) uacpi_kernel_free(mem)
#endif

#

just internally

tawdry trench Dec 15, 2024, 12:08 AM

#

rofl

agile halo Dec 15, 2024, 12:10 AM

#

maiden birch how do u cover NgNrE crap with this

is that really necessary tho? like sure it might not be the most optimal to use nGnRnE but eh, I kinda doubt it matters

maiden birch Dec 15, 2024, 12:10 AM

#

ok so heres what linux does

#

ill just give u a TLDR

#

The default ioremap does nGnRE which will work with most devices

#

but it also has a strong _np version

#

which does nGnRnE

#

 * ioremap_np needs an explicit architecture implementation, as it
 * requests stronger semantics than regular ioremap()

#

ill just add a UAPI_CACHING_MMIO and UAPI_CACHING_MMIO_STRONG

#

typedef enum uapi_caching {
    /*
     * Automatically deduce caching type by inspecting the system memory map.
     * (e.g. if the region happens to reside inside normal RAM, map as WB)
     */
    UAPI_CACHING_AUTO = 0,

    UAPI_CACHING_WB,
    UAPI_CACHING_WT,
    UAPI_CACHING_WC,
    UAPI_CACHING_UC,

    UAPI_CACHING_MMIO,

    /*
     * A stronger version of CACHING_MMIO, with an arch-specific implementation.
     * E.g. for the ARM architecture you may map MMIO as follows:
     * CACHING_MMIO -> nGnRE
     * CACHING_MMIO_STRONG -> nGnRnE
     */
    UAPI_CACHING_MMIO_STRONG,
} uapi_caching;

#

thoughts?

#

anyway pushed to pr

tawdry trench Dec 15, 2024, 12:18 AM

#

i am indifferent

maiden birch Dec 15, 2024, 12:18 AM

#

@tawdry trench any other headers u desperately need now that are blocking you?

tawdry trench Dec 15, 2024, 12:19 AM

#

well only IRQ and PCI are really missing

maiden birch Dec 15, 2024, 12:19 AM

#

yeah these ones are non-trivial

agile halo Dec 15, 2024, 12:19 AM

#

phys mem alloc

maiden birch Dec 15, 2024, 12:19 AM

#

we'll have to design how we want to approach that

tawdry trench Dec 15, 2024, 12:19 AM

#

agile halo + phys mem alloc

this

maiden birch Dec 15, 2024, 12:19 AM

#

phys mem i can add right now

#

/*
 * Allocate 2^'order' physical pages with an alignment of 'align'.
 * 'max_phys_addr' specifies the maximum acceptible address of the
 * end of the last allocated page.
 */
uapi_phys_addr uapi_allocate_pages(
    uapi_size order, uapi_size align, uapi_phys_addr max_phys_addr
);
#ifndef UAPI_SIZED_FREES
void uapi_free_pages(uapi_phys_addr);
#else
void uapi_free_pages(uapi_phys_addr, uapi_size order_hint);
#endif

Do we agree with this api

#

maybe order -> plain count

tawdry trench Dec 15, 2024, 12:21 AM

#

what about hinting for huge pages

maiden birch Dec 15, 2024, 12:21 AM

#

align?

tawdry trench Dec 15, 2024, 12:22 AM

#

true

agile halo Dec 15, 2024, 12:23 AM

#

maiden birch maybe order -> plain count

yeah maybe, it could still have the power of 2 guarantee (though ofc its not really needed but most if not all allocated things will be that)

maiden birch Dec 15, 2024, 12:23 AM

#

what if i need like 3 pages tho

#

why guarantee a power of 2?

agile halo Dec 15, 2024, 12:24 AM

#

well yeah ig it can be removed and be up to the kernel

static prairie Dec 15, 2024, 12:24 AM

#

maiden birch why guarantee a power of 2?

anyone with a buddy allocator won't be able to hand you non-power-of-2 amounts of pages

#

contiguous that is

maiden birch Dec 15, 2024, 12:24 AM

#

round up internally?

#

or do we want to make the library responsible for rounding up

agile halo Dec 15, 2024, 12:26 AM

#

idk if that's really necessary, imo the kernel can do that if it only supports that kind of sizes

maiden birch Dec 15, 2024, 12:26 AM

#

like your average osdever uses a bitmap

#

so they will be happy with an exact amount

static prairie Dec 15, 2024, 12:27 AM

#

btw will uapi be the new kernel api for uacpi too?

maiden birch Dec 15, 2024, 12:27 AM

#

ill add it as an optional mode at some point

#

if uapi gains traction

static prairie Dec 15, 2024, 12:27 AM

#

ah

maiden birch Dec 15, 2024, 12:27 AM

#

or maybe sooner since i have plans for more uprojects

#

and integrating two into one kernel will be annoying

#

btw we're gonna have an acpi.h header as well probably for evaluation of acpi methods

#

but optionally i guess

#

i know igpus use acpi extensively for various gpu/screen related events

agile halo Dec 15, 2024, 12:30 AM

#

would it be specific to uacpi or a generic interface?

maiden birch Dec 15, 2024, 12:30 AM

#

nah generic ofc

#

my ego is not that big lol

agile halo Dec 15, 2024, 12:32 AM

#

that could also be nice for abstracting other kernel acpi usage though it kinda depends on how much it would support in terms of eg. node discovery or whatever

maiden birch Dec 15, 2024, 12:33 AM

#

yeah it would be an interesting excercise

#

alright i pushed page allocation as well

#

pci.h and irq.h we'll solve tomorrow since that's very non-trivial

#

@tawdry trench feel free to add extra helpers or w/e while u work on your stuff, and drop as a pr

tawdry trench Dec 15, 2024, 12:36 AM

#

ok

maiden birch Dec 15, 2024, 12:37 AM

#

Are u still reviewing my pr?

agile halo Dec 15, 2024, 12:37 AM

#

theres also the other misc api like spinlocks, mutexes, events, sleeps/stalls that you had in the draft

maiden birch Dec 15, 2024, 12:37 AM

#

yeah ill add those in later, those need a uapi_status which i still havent added

tawdry trench Dec 15, 2024, 12:37 AM

#

maiden birch Are u still reviewing my pr?

im following the discussion rn

maiden birch Dec 15, 2024, 12:37 AM

#

ah

tawdry trench Dec 15, 2024, 12:38 AM

#

i cant comment on the caching stuff but rest lgtm

maiden birch Dec 15, 2024, 12:38 AM

#

since our map callback is ioremap + normal map we kinda have to do this i think

maiden birch Dec 15, 2024, 12:39 AM

#

tawdry trench i cant comment on the caching stuff but rest lgtm

alright thanks, merging then

#

I made a ruleset

#

that requires linear history and prevents force pushes

tawdry trench Dec 15, 2024, 12:41 AM

#

lmao

maiden birch Dec 15, 2024, 12:41 AM

#

aight merged

#

udrm is officially the first project to use uapi

tawdry trench Dec 15, 2024, 12:42 AM

#

https://tenor.com/view/lebron-sunshine-lebron-james-sunshine-sunshine-lebron-you-are-my-sunshine-you-are-my-sunshine-gif-509896211431970290

Tenor

maiden birch Dec 15, 2024, 12:42 AM

#

lmao

#

switching uacpi to use this will be fun

tawdry trench Dec 15, 2024, 12:50 AM

#

(removing the c everywhere)

maiden birch Dec 15, 2024, 12:50 AM

#

Well I want to support both versions

#

So I dont break people's osses

#

Mfw last repo update 3 days ago

#

nooo

agile halo Dec 15, 2024, 12:52 AM

#

meanwhile last uhda update: 1 month ago

maiden birch Dec 15, 2024, 12:52 AM

#

Btw you old URL correctly redirects to the org now

#

Which is nice

maiden birch Dec 15, 2024, 12:53 AM

#

agile halo meanwhile last uhda update: 1 month ago

Any plans to continue it?

tawdry trench Dec 15, 2024, 12:53 AM

#

move uhda to udrivers meme

maiden birch Dec 15, 2024, 12:53 AM

#

That too

agile halo Dec 15, 2024, 12:53 AM

#

maiden birch Any plans to continue it?

yes I just have had other things

tawdry trench Dec 15, 2024, 12:53 AM

#

i can add you as an admin

agile halo Dec 15, 2024, 12:54 AM

#

tawdry trench move uhda to udrivers <:meme:575445094589661215>

sure

tawdry trench Dec 15, 2024, 12:54 AM

#

pog

main crow Dec 15, 2024, 12:54 AM

#

maiden birch So I dont break people's osses

just make a uapi->uacpi compat layer

tawdry trench Dec 15, 2024, 12:55 AM

#

💀

#

just call it abi break

main crow Dec 15, 2024, 12:55 AM

#

that would be the boring thing

#

also does uapi support iommus?

agile halo Dec 15, 2024, 12:56 AM

#

that's up to the kernel?

#

or well idk what do you even mean by that lol

main crow Dec 15, 2024, 12:56 AM

#

kinda isnt no

#

i mean you need to deal with the fact that the device phys addr != host phys addr

#

etc

maiden birch Dec 15, 2024, 12:57 AM

#

We haven't designed that yet

#

But it should

maiden birch Dec 15, 2024, 12:58 AM

#

main crow just make a uapi->uacpi compat layer

Most things can just be a macro rename

#

So yeah

main crow Dec 15, 2024, 12:58 AM

#

maiden birch Most things can just be a macro rename

exactly

maiden birch Dec 15, 2024, 12:58 AM

#

Some callbacks are more generic now

main crow Dec 15, 2024, 12:59 AM

#

also your kernel api is kinda shit for binding authors

maiden birch Dec 15, 2024, 12:59 AM

#

Or hear me out

#

Attribute alias

maiden birch Dec 15, 2024, 1:00 AM

#

main crow also your kernel api is kinda shit for binding authors

U mean that it’s not a vtable?

main crow Dec 15, 2024, 1:00 AM

#

maiden birch U mean that it’s not a vtable?

and you don't pass around context pointers either

maiden birch Dec 15, 2024, 1:00 AM

#

Which context

tawdry trench Dec 15, 2024, 1:00 AM

#

what context

agile halo Dec 15, 2024, 1:00 AM

#

are you talking about uacpi?

main crow Dec 15, 2024, 1:00 AM

#

maiden birch Which context

like a user data pointer that is global to all state

maiden birch Dec 15, 2024, 1:01 AM

#

Huh why would anyone need that

main crow Dec 15, 2024, 1:01 AM

#

its super common in almost all libraries to pass around data

main crow Dec 15, 2024, 1:01 AM

#

maiden birch Huh why would anyone need that

to avoid global state?

maiden birch Dec 15, 2024, 1:01 AM

#

Which global state

tawdry trench Dec 15, 2024, 1:01 AM

#

yea

main crow Dec 15, 2024, 1:01 AM

#

any of it

maiden birch Dec 15, 2024, 1:01 AM

#

Callbacks pass around a user pointer

main crow Dec 15, 2024, 1:01 AM

#

uapi_free_pages does not

#

and neither does uapi_allocate_pages

#

none of the log things do

maiden birch Dec 15, 2024, 1:02 AM

#

Why in the world would that need context

main crow Dec 15, 2024, 1:02 AM

#

maiden birch Why in the world would that need context

so that you know what the real callback is

#

like, think about how youd write an e.g. rust binding for that

maiden birch Dec 15, 2024, 1:03 AM

#

So u sacrifice performance and add useless parameters to everything to satisfy rust binding authors?

main crow Dec 15, 2024, 1:03 AM

#

maiden birch So u sacrifice performance and add useless parameters to everything to satisfy r...

you lose like no performance

#

and anyway you can make that an ifdef

agile halo Dec 15, 2024, 1:03 AM

#

also I don't see an issue with the current api for rust bindings?

maiden birch Dec 15, 2024, 1:03 AM

#

Yeah idk man

agile halo Dec 15, 2024, 1:04 AM

#

like you are going to be having a global allocator/log anyway

main crow Dec 15, 2024, 1:04 AM

#

its ugly

#

also its another indirection

maiden birch Dec 15, 2024, 1:04 AM

#

Why cant the rust implementation call rust_alloc_pages directly

main crow Dec 15, 2024, 1:05 AM

#

maiden birch Why cant the rust implementation call rust_alloc_pages directly

because the bindings crate cant call the "os" crate directly

#

it has to do it through a function pointer

#

because you cant have cycles like that

maiden birch Dec 15, 2024, 1:05 AM

#

Well what current uacpi bindings do is wrap it in a class

#

And that class obviously has function pointers

main crow Dec 15, 2024, 1:06 AM

#

so... you add another indirection?

maiden birch Dec 15, 2024, 1:06 AM

#

Just like u do with context lol

main crow Dec 15, 2024, 1:06 AM

#

true

#

but at least its not global state with context

#

just pass around context pointers its not like that expensive, and the drivers arent that hot anyway

maiden birch Dec 15, 2024, 1:07 AM

#

Like I get the idea, but damn making the whole API ugly af and feeling very un-c-like just to make it better for bindings

#

Idk man

#

Like its a c library first and foremost

tawdry trench Dec 15, 2024, 1:08 AM

#

extern "C" {
    fn uapi_malloc();
}

#

am i missing something

maiden birch Dec 15, 2024, 1:08 AM

#

What is this?

tawdry trench Dec 15, 2024, 1:08 AM

#

is this not all you have to do for rust bindings

maiden birch Dec 15, 2024, 1:08 AM

#

Oh rust

#

Apparently not?

tawdry trench Dec 15, 2024, 1:08 AM

#

i don't get it

main crow Dec 15, 2024, 1:09 AM

#

tawdry trench is this not all you have to do for rust bindings

yes but you cant call to the caller crate

#

okay so like

agile halo Dec 15, 2024, 1:09 AM

#

the talk was about a generic abstraction crate and not a raw bindings crate

main crow Dec 15, 2024, 1:10 AM

#

crate uacpi_bindings {
  fn uacpi_malloc() { /* you cant put anything here to call real_kernel (except a global function pointer which is ugly) */ }
}
crate real_kernel {
  fn stuff() { uacpi_bindings::do_anything() }
}

maiden birch Dec 15, 2024, 1:10 AM

#

Btw would like uacpi_set_context/get_context work? @main crow

main crow Dec 15, 2024, 1:10 AM

#

maiden birch Btw would like uacpi_set_context/get_context work? <@696410219663851552>

that kinds doesnt solve the problem tho

maiden birch Dec 15, 2024, 1:10 AM

#

Hm?

main crow Dec 15, 2024, 1:11 AM

#

the problem isnt that there isn't a way to have global state, because there is

#

the point is that global state bad

#

and you should avoid it if it's trivially possible

maiden birch Dec 15, 2024, 1:11 AM

#

Its abstracted away from you

#

From your safe rust world

#

Without global state

main crow Dec 15, 2024, 1:11 AM

#

still has global state inside

#

and therefore ugly

maiden birch Dec 15, 2024, 1:11 AM

#

The context pointer is the exact same thing lol

#

Just passed through a function

#

How would u set it exactly

#

During uacpi_init?

#

That would save it globally

agile halo Dec 15, 2024, 1:12 AM

#

well uacpi also has other global state

maiden birch Dec 15, 2024, 1:13 AM

#

Its fundamentally a global state library

main crow Dec 15, 2024, 1:13 AM

#

agile halo well uacpi also has other global state

that's cringe too tbh

agile halo Dec 15, 2024, 1:13 AM

#

for drivers ig it would be saved inside the device

main crow Dec 15, 2024, 1:13 AM

#

okay tbf for uacpi it makes sense

#

because you only have one

maiden birch Dec 15, 2024, 1:14 AM

#

Yeah u cant have multiple acpi instances

main crow Dec 15, 2024, 1:14 AM

#

well

#

yeah in normal cases

maiden birch Dec 15, 2024, 1:14 AM

#

Even in non normal cases?

main crow Dec 15, 2024, 1:14 AM

#

maiden birch Even in non normal cases?

what about the non normal case of running tests in parallel 😛

maiden birch Dec 15, 2024, 1:14 AM

#

Well my test cases are run by python

main crow Dec 15, 2024, 1:14 AM

#

anyway imo you should not have global state if the user asks you for

maiden birch Dec 15, 2024, 1:15 AM

#

It can spawn as many runners as it wants

maiden birch Dec 15, 2024, 1:15 AM

#

main crow anyway imo you should not have global state if the user asks you for

Its easy to move global state out into a struct

#

Its just pointless work and redundant parameter to every uacpi function that will always have the same value

main crow Dec 15, 2024, 1:16 AM

#

maiden birch Its just pointless work and redundant parameter to every uacpi function that wil...

you can ifdef around it trivially

maiden birch Dec 15, 2024, 1:16 AM

#

maiden birch Its easy to move global state out into a struct

Most of it already lives in one struct

maiden birch Dec 15, 2024, 1:16 AM

#

main crow you can ifdef around it trivially

U cant

#

Its literally every existing uacpi public api

#

Which there's hundreds of functions

agile halo Dec 15, 2024, 1:17 AM

#

#ifndef UAPI_PASS_CONTEXT
#define UAPI_CONTEXT
#else
#define UAPI_CONTEXT ,void *ctx
#endif

void *uapi_malloc(uapi_size size UAPI_CONTEXT);

main crow Dec 15, 2024, 1:17 AM

#

yeah that

maiden birch Dec 15, 2024, 1:17 AM

#

Won't compile

main crow Dec 15, 2024, 1:17 AM

#

you can put a leading comma on the UAPI_CONTEXT define

agile halo Dec 15, 2024, 1:17 AM

#

oh yeah the comma needs to be inside there

main crow Dec 15, 2024, 1:17 AM

#

and that does compile

maiden birch Dec 15, 2024, 1:18 AM

#

That does

#

Anyway per driver state is fine, but kernel api is global and as such not stateful

#

So I wouldnt go there

#

But we could talk about it later if someone disagrees

hexed flare Dec 15, 2024, 8:32 AM

#

static prairie anyone with a buddy allocator won't be able to hand you non-power-of-2 amounts o...

Why not? You allocate 4 pages and free 1 at the end

hexed flare Dec 15, 2024, 8:36 AM

#

maiden birch so they will be happy with an exact amount

(my PMM for example can allocate any size)

robust abyss Dec 15, 2024, 9:14 AM

#

maiden birch Anyway per driver state is fine, but kernel api is global and as such not statef...

I agree with that, passing useless context parameters everywhere is just not useful, and if it's going to hold it as a global context anyways internally might as well let the kernel do it if it needs to

#

The only case I can think is maybe if you want a different allocator per driver for whatever reason

#

But any place where the context would be saved globally I think you might as well not have it at all

tawdry trench Dec 15, 2024, 2:02 PM

#

@maiden birch how should init happen? do we have a central "initialize everything" function, or should there be specific functions to initialize each driver

maiden birch Dec 15, 2024, 2:03 PM

#

no central init

#

its very driver specific

tawdry trench Dec 15, 2024, 2:03 PM

#

what about drm_core?

#

that also has its own init fn?

maiden birch Dec 15, 2024, 2:03 PM

#

yes

tawdry trench Dec 15, 2024, 2:03 PM

#

ok

maiden birch Dec 15, 2024, 2:04 PM

#

i'd have a one per driver and one per each device

#

so udrm_init and the udrm_device_probe

#

similar to module_init/module_fini and then per device stuff

tawdry trench Dec 15, 2024, 2:04 PM

#

are the probe funcs like the linux pci probe callbacks?

maiden birch Dec 15, 2024, 2:04 PM

#

yes

tawdry trench Dec 15, 2024, 2:04 PM

#

great

maiden birch Dec 15, 2024, 2:05 PM

#

its basically: i found a device that matches your list, try to see if you can initialize it and bind to it

tawdry trench Dec 15, 2024, 2:05 PM

#

yea that's what i thought

#

i'm just going to make up the IRQ/PCI bindings for now until they are finalized

maiden birch Dec 15, 2024, 2:05 PM

#

yeah

tawdry trench Dec 15, 2024, 2:05 PM

#

i'll comment them out

maiden birch Dec 15, 2024, 2:06 PM

#

ill see what edge cases need to be handled in the irq bindings

#

like linux has request_irq with just one number but i know thats not all there is

#

and im not sure how that handles msi

livid sentinel Dec 15, 2024, 2:48 PM

#

uAPI/vm.h:
void *uapi_kernel_map(
uapi_phys_addr addr, uapi_size len,
uapi_caching, uapi_access_type
);
might not work under certain oses if they deny user setable executable memory

maiden birch Dec 15, 2024, 2:49 PM

#

sure, return a null

#

none of the current libraries need executable memory, this is more for the future

#

also like "user" setable? this is a kernel level driver

livid sentinel Dec 15, 2024, 2:50 PM

#

i plan to put it into userspace

#

everything that doesnt have a strong reason for kernel level permissions goes into userspace

maiden birch Dec 15, 2024, 2:51 PM

#

userspace level drivers should still be considered more priveleged than a normal application

livid sentinel Dec 15, 2024, 2:52 PM

#

yea they get access to physical hardware but not access to rwx memory the only thing they can do is ask to get a library loaded

maiden birch Dec 15, 2024, 2:53 PM

#

anyways this is just for the api completeness, no current use cases exist

livid sentinel Dec 15, 2024, 2:54 PM

#

also if i understand correctly this function requests specific physical memory?

#uDRM - a portable Direct Rendering Manager