it computes the offset within page
#Astral
analog berry
then adds it to the mapping
so we dont know the initial address after that
or well i guess u could round down to page size
grand shadow
yeah just round down to page size
I can probably make it a todo to add some more proper unaligned support in vmm_* or something
but it'd just do whatever you'll do but in the vmm code
analog berry
yeah that should work i think
grand shadow
wait wdym
grand shadow
oh bruh I thought input variables in inline asm wouldn't get modified?
yeah looking at the disassembly that is true
I guess its ub?
idk
but yeah just save it in a variable or something you can merge it with your pr
or wait
add rdi to clobber ig
doesnt seem to compile if you do so just make it a variable yeah
analog berry
ill fix it
also the vmm is bugged unfortunately
but uacpi makes it pretty far into init
uACPI almost alive on Astral
as u can see it page faults on an address it had mapped earlier and never unmapped
analog berry
I guess its ub?
its just incorrect as per the spec for memset
grand shadow
ik, talking about the asm thing
grand shadow
but uacpi makes it pretty far into init
how'd you map the page
analog berry
void *uacpi_kernel_map(uacpi_phys_addr physical, uacpi_size length) {
uintmax_t pageoffset = (uintptr_t)physical % PAGE_SIZE;
void *virt = vmm_map(NULL, ROUND_UP(length, PAGE_SIZE), VMM_FLAGS_PHYSICAL, ARCH_MMU_FLAGS_READ | ARCH_MMU_FLAGS_NOEXEC, (void*)ROUND_DOWN(physical, PAGE_SIZE));
__assert(virt);
printf("Mapped %lu as %p (%zu bytes)\n", ROUND_DOWN(physical, PAGE_SIZE), virt, ROUND_UP(length, PAGE_SIZE));
return (void *)((uintptr_t)virt + pageoffset);
}
void uacpi_kernel_unmap(void *ptr, uacpi_size length) {
uintptr_t addr = (uintptr_t)ptr;
vmm_unmap((void*)ROUND_DOWN(addr, PAGE_SIZE), ROUND_UP(length, PAGE_SIZE), 0);
printf("Unmapped %p (%zu bytes)\n", (void*)ROUND_DOWN(addr, PAGE_SIZE), ROUND_UP(length, PAGE_SIZE));
}
also it seems to be trying to write to it?
grand shadow
you're missing the ARCH_MMU_FLAGS_WRITE there
analog berry
i have no idea
ill have to see where it dies
during late init
in uacpi
but where idk
analog berry
how would i do that
grand shadow
how I usually do it when gdb doesnt do it for me with bt is just
x/2a ctx->rbp in gdb and follow that
but I'd hope rsp has the return address
so x/1a ctx->rsp might help
analog berry
lets see
0xffff80007fc8dbe8: 0xffffffff8002c874 <pci_read8+20>
oh
pci_init is called like
after everything
interesting placement but ok
grand shadow
you can probably put it before, I just never ran into an issue with it
analog berry
anyways uACPI on Astral is officially alive!
grand shadow
yippe
analog berry
can u believe Astral allocator is good enough that it has survived the uacpi stress test first try
grand shadow
if it survived building gcc, running llvmpipe and all the shit I've thrown at it I would imagine it'd survive uacpi
analog berry
lol fair
grand shadow
the obos and nyaux curse hasn't hit me.....
analog berry
anyways i still have tons to cleanup and reorganize, but for the most part it works great
so thats cool
so things that went wrong while porting:
- broken memset (ill fix it in my PR)
- SPINLOCK_INIT works on a value directly, even though the rest of *_INIT work on pointers
- readonly mapping (my rartedness)
- pci_init happening after like literally everything
also had to add a few utility api but thats mostly misc stuff
prime juniper
Based
prime juniper
And re the wine question: outside of wine specific patches, should workβ’οΈ
prime juniper
took basically about 2:30 hours to get it to work
Damn that was a good base you were working with then yeah
analog berry
yeah astral has pretty much most primitives already available
plus i stole a bit of code from managarm and adapted it 
prime juniper
Astral: powered by mlibc, Limine, uACPI and a tiny bit of Managarm (in random order)
analog berry
lmao
prime juniper
But nice work
analog berry
yeah i mean especially for a solo project astral is in a pretty good place
prime juniper
Absolutely
grand shadow
ty
only took me a shitty real mode bit os, a shitty protected mode os, and to rewrite the old version of astral
obsidian pivot
but uacpi makes it pretty far into init
so obos more stable than astral
obsidian pivot
anyways uACPI on Astral is officially alive!
oh nvm
analog berry
Nah that was my bug
obsidian pivot
the obos and nyaux curse hasn't hit me.....
YET
analog berry
The mapping should've been rw
obsidian pivot
the oberrow curse can hit anyone
if it's gotten to the likes of keyronex a couple weeks ago
analog berry
The rbtree bug
obsidian pivot
if it's gotten to the likes of keyronex a couple weeks ago
it can probably get to anyone
the oberrow curse has been with me since my 2nd kernel
analog berry
The skill issue curse π
grand shadow
The rbtree bug
considering just making mlibc's malloc use astral's slabs cuz of it
jk
that'd be bad
wild marsh
i still don't think the bug is strictly in frg::slab_pool
obsidian pivot
now if you'll excuse me I have some debugging to do
grand shadow
i still don't think the bug is strictly in frg::slab_pool
I have no idea how to even begin debugging that stuff its gonna sit in bug hell for quite a while
wild marsh
i still don't think the bug is strictly in frg::slab_pool
i wouldn't be surprised if the total amount of heap allocations (both in kernel and user space) was in the millions by the time mammogram gets to weston
analog berry
considering just making mlibc's malloc use astral's slabs cuz of it <:meme:57544...
That would be based
grand shadow
it could be gcc corrupting something
or idk
it can reliably be reproduced it'd just be super annoying to debug
analog berry
Just sounds like a 10 hour printf debugging session followed by a one liner fix
I've done those before
Specifically for rb tree too
Id just dump the tree at every state change and then analyze the log
grand shadow
oh god
my setresuid permission checking might be a bit fucked
it was, indeed, borked
the fuck is it even trying to achieve here lol
guess the linux filesystem id stuff? no idea
guess I should make that a no-op then
grand shadow
I think setresgid() is part of the function to open it
and thus it just passes the error down
well awkward!
obsidian pivot
I think setresgid() is part of the function to open it
ah
fleet osprey
?
wtf
you can have sudo insult you? that's hilarious
I just imagine now
$ sudo touch grass
[sudo] password for user:
Sorry, try again.
[sudo] password for user:
Sorry, try again.
[sudo] password for user:
sudo: Wow, you really are terrible at this.
based
grand shadow
only needed me to like, fix some setresuid and setresgid stuff and then implement MSG_WAITALL for recvmsg()
grand shadow
and it works on the rootless xorg
soon you literally wont have to run as root anymore if you set up the image right (based)
fleet osprey
nice!
grand shadow
also I love how quake runs p smooth now that I unshittyfied the ringbuffer and (x86_64) memcpy impls
also gnu hello configure and make -j2 work with the user stuff
it errors make install when not sudoing
but works when sudoing
fleet osprey
also I love how quake runs p smooth now that I unshittyfied the ringbuffer and (...
have you tested how that stacks up to the same machine under Linux with llvmpipe? from the looks of it, you're pretty close
that honestly looks exactly like I how would expect llvmpipe to run if I was running it on my PC under Linux
genuinely great stuff
grand shadow
thats not on llvmpipe, thats on tyr-quakes renderer
but tyr-glquake isnt too far off
I'd need to try it out
fleet osprey
that's sw rendering? ah
grand shadow
that's sw rendering? ah
this is with llvmpipe
idk if the graphical glitches are tyr-quake, llvmpipe or something else
fleet osprey
ah, so a little slower than I would expect from my experience, but then again, your machine isn't as fast as mine, so that kinda makes sense
regardless, it's not even that far off from what I would expect, so my comment still applies
wild marsh
this is with llvmpipe
the graphical glitches might be due to the initial fpu state being wrong?
quake uses long double for some math where it needs more precision
although idk if that's used by glquake or only the software renderer
grand shadow
although idk if that's used by glquake or only the software renderer
It only happens in glquake
robust geyser
usb_xhci_doorbell_write off 0x0000, val 0x00000000
usb_xhci_fetch_trb addr 0x0000000077e70000, CR_ENABLE_SLOT, p 0x0000000000000000, s 0x00000000, c 0x00002401
usb_xhci_slot_enable slotid 1
usb_xhci_queue_event v 0, idx 0, ER_COMMAND_COMPLETE, CC_SUCCESS, p 0x0000000077e70000, s 0x01000000, c 0x01008401
usb_xhci_fetch_trb addr 0x0000000077e70010, CR_ENABLE_SLOT, p 0x0000000000000000, s 0x00000000, c 0x00002401
usb_xhci_slot_enable slotid 2
usb_xhci_queue_event v 0, idx 1, ER_COMMAND_COMPLETE, CC_SUCCESS, p 0x0000000077e70010, s 0x01000000, c 0x02008401
usb_xhci_fetch_trb addr 0x0000000077e70020, TRB_RESERVED, p 0x0000000000000000, s 0x00000000, c 0x00000000```
woo
i can finally go sleep
fervent fjord
goodnight!
analog berry
Work on astral uacpi integration shall continue later today
The init order is kinda weird so I'll have to refactor that a bit
So pci is available earlier
Also ill have to throw out all astral table parsing which was x86 specific anyway
warped plinth
```
usb_xhci_doorbell_write off 0x0000, val 0x00000000
usb_xhci_fetch_trb addr 0...
interested in seeing this upstreamed, i rlly wanna run astral on my laptop, and the ps2 code kept killing it lol
grand shadow
```
usb_xhci_doorbell_write off 0x0000, val 0x00000000
usb_xhci_fetch_trb addr 0...
Holy shit epic
grand shadow
Also ill have to throw out all astral table parsing which was x86 specific anywa...
Yeah you can just nuke that
grand shadow
interested in seeing this upstreamed, i rlly wanna run astral on my laptop, and ...
I need to see what Im doing wrong regarding ps2 init
So that at least the keyboard or smth works
analog berry
Yeah its not like u have a choice on laptops
carmine swallow
ps2 is very annoying with its mixed user input and command responses (and also sometimes returning ack byte and selftest result in the other order or something stupid like that anyway)
grand shadow
The only places I saw it work on real hardware were my laptop (mouse too) and desktop (kb only)
Everywhere else it either locks up fails the selftest or doesnt have any input
flat copper
selftest is broken with most ps2 emulation
grand shadow
selftest is broken with most ps2 emulation
How do you even detect that, is it on that one acpi table
grand shadow
I might just atp take a look at a more stablished kernel's ps2 driver and see what theyre doing
analog berry
I put a lot of effort into stabilizing it for my old kernel
It worked everywhere and pretty sure it didn't have any workarounds for broken self test
Its just that you have to be careful with racing with user data plus have good timeouts everywhere
grand shadow
It worked everywhere and pretty sure it didn't have any workarounds for broken s...
Do you have the link so I can take a look at it
neon crane
@grand shadow I have a test machine that returned a bitflipped config from the i8042 controller, to detect that you'd have to check if bits 3 and 7 are 1
analog berry
Do you have the link so I can take a look at it
UltraLegacy on my GitHub, but be careful cause it's c++
neon crane
PS/2 Controller Configuration Byte
Commands 0x20 and 0x60 let you read and write the PS/2 Controller Configuration Byte. This configuration byte has the following format:Bit Meaning
0 First PS/2 port interrupt (1 = enabled, 0 = disabled)
1 Second PS/2 port interrupt (1 = enabled, 0 = disabled, only if 2 PS/2 ports supported)
2 System Flag (1 = system passed POST, 0 = your OS shouldn't be running)
3 Should be zero
4 First PS/2 port clock (1 = disabled, 0 = enabled)
5 Second PS/2 port clock (1 = disabled, 0 = enabled, only if 2 PS/2 ports supported)
6 First PS/2 port translation (1 = enabled, 0 = disabled)
7 Must be zero
grand shadow
<@215823874615541761> I have a test machine that returned a bitflipped config fr...
Oh wtf
neon crane
this might be the case on your end but I'm not sure
grand shadow
UltraLegacy on my GitHub, but be careful cause it's c++
I can read some c++ ty
analog berry
<https://github.com/iProgramMC/Boron/blob/master/drivers/i8042prt/source/main.c#...
Thats curious
analog berry
I can read some c++ ty
Nice, it worked on at least 5 laptops and a few desktops
analog berry
Nice, it worked on at least 5 laptops and a few desktops
Oh also it supported multiplexing for up to 5 ps2 devices
Like they have on thinkpads with that joystick and a trackpad + keyboard
neon crane
actually I'm slightly curious if it still happens on that machine
so it seems it doesnt happen
but I could have sworn it did
analog berry
Maybe it did due to a skill issue of some sort
neon crane
anyway that hack was carried over twice or thrice "by folklore" from the first ps/2 mouse driver I wrote
grand shadow
Wrt the setuid bit on a chown(), I might just follow what linux does and clear it every time a chown is done
Originally I was going to check if the process is privileged and keep it if it is, but meh
It gets simpler this way
obsidian pivot
out of curiosity, how much memory does astral when it is booted to a bash shell
grand shadow
out of curiosity, how much memory does astral when it is booted to a bash shell
Last time I tried it needed like 50 mb allocated to boot but from counting the pages used since pmm init it seems like it was less, more like in the 30 mbs
Boots to fvwm on 128 mb and its pretty much a usable system for anytging that isnt like quake at 256 (aka I was able to build gnu hello lol)
grand shadow
Last time I tried it needed like 50 mb allocated to boot but from counting the p...
Because lower than 50 mb and limine ooms
Probs cuz of the background bmp or something
I really should implement stuff to make free work at some point in the future
obsidian pivot
ambiguous
obsidian pivot
I was going to say its even more impressive you got this far without `free()`
who needs free?
like let's be honest
my kernel can work perfectly fine without it /j
it only uses 10mib more memory
tawny oak
you should still free memory though
unless you are the bootloader, then just mark it as 'reclaimable' π
analog berry
I still try to free everything I can in my bootloader to keep the memory map cleaner
grand shadow
The only thing that reclaims limine memory is a reboot
analog berry
Even like page caches and stuff are freed on handover
tawny oak
I still try to free everything I can in my bootloader to keep the memory map cle...
respectable, that is the right way to do it.
analog berry
Thanks
craggy tide
Limine also does that for the record
which is why i am not sure it's even worth it to reclaim bootloader memory given it's a few K? something like that tops
craggy tide
Because lower than 50 mb and limine ooms
limine can definitely work with less than 50mb
grand shadow
I dont bother doing it
grand shadow
limine can definitely work with less than 50mb
Its probably cuz of kernel and background
craggy tide
your bmp is quite heavy
grand shadow
Yeah
craggy tide
last i checked
with no wallpaper (and to minimise memory usage further, with graphics: no on BIOS) it boots down to iirc 8mb
nvm
#osdev-beginner-1 message
it's apparently 6mb with framebuffer and 1.5mb without
neon crane
which is why i am not sure it's even worth it to reclaim bootloader memory given...
seems to be a few hundred pages while running boron
this is the memory map dump while running my OS (it has 64 MB of memory available)
entries with 5 are bootloader reclaimable
in total 377 pages, or 1.5 MB are used by the bootloader
craggy tide
1.5mb counts as "a few K" in my book lol
remember it contains the page tables, GDT, and all of the responses and associated data
at the very least
neon crane
if I can simply reclaim bootloader memory then I should be good
like, move everything that limine gave me and that I use into other memory and free the bootloader reclaimable memory
craggy tide
you can, there is no reason why you couldn't
craggy tide
yeah
grand shadow
theres something really satisfying about glxgears
idk why its fun to stare for a minute at the gears
grand shadow
made nohup work, will be useful for launching stuff from /etc/rc eventually
inland cliff
9950X? Threadripper?
grand shadow
only missing the check for signal() now I think
@analog berry @robust geyser idk if yall were using vmm_changemmuflags but if you were be aware I changed the return value from a bool to an errno once you git pull
I will do that change to vmm_map once yall are done with your stuff
craggy tide
9950X? Threadripper?
13900KS
grand shadow
well shit the ssh port bitrotted on the server side
oops!
gonna look into that tomorrow
neon crane
that's what they said about the pentium 4!
fervent fjord
(and they were right)
lean kettle
Intel is not beating the space heater allegations
Actually no desktop chip that is high performance is gonna beat it, TDP >50W was a mistake 
That's probably the hottest take I've ever dropped in this server
fervent fjord
if your CPU isnt pulling 500W are you even using your hardware
saying this from DDR4 at 12V, room smells like ozone
lean kettle
Only 12v? Real pros power DDR4 directly from AC
fervent fjord
the AC is already going thru my left foot into the PSU
dead man switch like a train but for PC
lean kettle
Try using a jumper cable to connect the DDR4 to the rectified DC voltage from the PSU, it should be close to 300V
neon crane
(and they were right)
(indeed)
fervent fjord
Try using a jumper cable to connect the DDR4 to the rectified DC voltage from th...
ill do it when I get more cable, all of it is being used right now in a life sized statue of josh from twenty one pilots playing vignette from their newest album Clacy
lean kettle
Ah ok, yeah I can see how that's more important yea
lean kettle
The only twenty one pilots songs I recently listened to were ride and heavymetalsoul and that was because YouTube mixes decided I should lol
I used to be a fan of them but tbh I've kinda stopped listening to them
fervent fjord
Ive followed them for a while
a new group to my library ive listened to lately is HEALTH
but its quite different in style
its like industrial rock
lean kettle
I don't really follow bands more than I like specific songs from many bands lol
Although I have listened to a lot of alestorm lol
lean kettle
The only thing YouTube mixes are good for is background noise
And occasionally finding new songs
tawny oak
I did not expect you to be an alestorm fan
lean kettle
I did not expect you to be an alestorm fan
inland cliff
Intel is not beating the space heater allegations
Well my Vega 64 draws up to 370 W and it definitely heats up my space
inland cliff
Only 12v? Real pros power DDR4 directly from AC
wait that's going to distort the signaling!
lean kettle
insert any modern high performance CPU/GPU company here is not beating the space heater allegations
inland cliff
The upside of Intel processors is that due to them crashing and eventually breaking, you can claim a lower average power consumption
grand shadow
For today I will be fixing the openssh port, implementing putgrent in mlibc and implementing the signal() check
After that I think all I can do about multiuser will be done?
craggy tide
space heater
nah, it's fine if you run it with actual power limits instead of without any power limits
carmine swallow
also undervolting helps a lot with temps
craggy tide
yes but it also helps with losing stability lol
i just run it with the default intel profile (performance) at PL1=PL2=253W and IccMAX=307A
and i run the DDR5 at JEDEC 4400 as per spec
carmine swallow
yes but it also helps with losing stability lol
well ig it depends on how much you do, I have been fine with my current undervolt at max load and temps stay below 90c
craggy tide
tbh if you're worried about power and temps just use a 65W SKU stock
i've been really impressed with my 12900
runs all core loads (past tau boost expiry) at around 70c
quiet as hell, cool
sure, obviously it's slower than a K part at 241W but you lose some, you gain some
carmine swallow
yeah that sounds nice, though I am pretty happy with my 13600k so I don't think I am going to change it any time soon
craggy tide
you can change the power/current limits to emulate a non K SKU anyways lol
carmine swallow
yeah ig
inland cliff
i just run it with the default intel profile (performance) at PL1=PL2=253W and I...
isn't that the extreme profile?
analog berry
I will do that change to vmm_map once yall are done with your stuff
Thats fine I can rebase on top of master, feel free to break this
craggy tide
isn't that the extreme profile?
on KS SKUs it's the performance profile
grand shadow
oops it really really wants getsockopt to work and I have that stubbed
IP_OPTIONS in specific
analog berry
got rid of all astral manual ACPI table parsing + added a bunch of new things here and there + improved the glue code
shutdown from the power button now works from Astral as well
fleet osprey
π₯
if you want to send some patches (or a compiled build) I can test this out on some of my real hw
analog berry
we'll test it properly once its merged into astral
obsidian pivot
can astral load drivers by iterating the acpi namespace now
obsidian pivot
dam
grand shadow
Also its very x86 specific
I tried my best π
analog berry
arch_apic_* π
fleet osprey
if there's any chacne at seperating it from Limine I might try my hand at a PPC port
but I never even finished procfs
so no guarantees that I'll ever get anywhere w/ it
analog berry
And literally all of init happening in x86-64/main
obsidian pivot
And literally all of init happening in x86-64/main
that's what obos does too
grand shadow
`arch_apic_*` π
thats only called from x86-64 tho
analog berry
I tried my best π
Im sure it will be way better after its ported to a diff arch
obsidian pivot
(oops)
grand shadow
And literally all of init happening in x86-64/main
I mean thats just for simplicity ig
obsidian pivot
when astral m68k port?
grand shadow
when astral m68k port?
it is very 64 bit specific that wont happen
analog berry
Nah dont get me wrong its fine and it seems to work so
robust geyser
not so hot take, 32-bit sucks
obsidian pivot
yeah but the m68k doesn't suck
winged basin
nothing is very "64-bit specific"
obsidian pivot
I have summoned the fadanoid by saying "m68k"
winged basin
i didn't even try to be nonspecific with bits but it just worked
obsidian pivot
same (almost)
winged basin
endianness is also a paper tiger
robust geyser
cant really have a hhdm on 32-bit systems
winged basin
cant really have a hhdm on 32-bit systems
you can
just reject >~1.5g or so
it will cover most actually existing hardware
obsidian pivot
have two page tables
one has one half of the hhdm
the other has the other half
(not actually)
fleet osprey
but what if I wanted to boot on a Pi 0 W
obsidian pivot
(not actually)
((don't do that, that's a bad idea))
grand shadow
oooh damn openssh wants fucking fd passing in unix sockets cuz of privsep
before it didn't need it cuz it was just running as root
but I guess now I need to implement it?
shouldn't be too hardβ’οΈ
grand shadow
nothing is very "64-bit specific"
I think I assume 64 bit in a few places but idk
I'd really just have to go and see
but before even trying for a 32 bit port I'd try a 64 bit one
grand shadow
yeah I was probably just gonna try aarch64 some time in the future
I would have to learn how that works tho I have never even looked into it
inland cliff
Have fun with dtbs
obsidian pivot
cc @analog berry
analog berry
What
obsidian pivot
they're honestly not too bad imo
.
analog berry
Yeah lets not go there
fleet osprey
at least DTS's aren't, I've never actually parsed a compiled DTB, so ig I can't comment on that, maybe once compile they're actually terrible
but the source really isn't that bad, I've spent way too long staring at them for PowerPC
obsidian pivot
coulda sworn you had a rant about DTBs somewher ein this server
craggy tide
not so hot take, 32-bit sucks
it's actually a hot take now apparently
some virus is spreading around our circles where people find a sudden urge to love doing 32-bit stuff
maybe i should just start banning people spreading 32-bit propaganda
robust geyser
you can if you don't need to map all of the physical memory*
and dont mind wasting half of the virtual address space
craggy tide
but before even trying for a 32 bit port I'd try a 64 bit one
please not you too
grand shadow
please not you too
I'm not even planning on porting it to any other architecture anytime soon let alone a 32 bit one
craggy tide
you can if you don't need to map all of the physical memory*
you kind of do, most mmio is around 3.5-4gb on x86 for example, while most usable memory is lower down towards 0, so unless you don't map a lot of the lower half or you don't map a lot of mmio, you can't do anything
robust geyser
true, but if you dont need an hhdm you could just map physical memory anywhere
craggy tide
well yes
robust geyser
im not tryna be defensive or anything btw
i still dont like 32 bit architectures lol
craggy tide
we're talking about hhdm in particular here, especially the notion that it can be "done" on 32-bit architectures
it can't
at least not in any way where it'd be a true hhdm
you have a higher half which isn't even half, it's usually 1gb
good luck mapping anything substantial in there
which is why 32-bit kernels can't (or at least shouldn't) have a hhdm
which is just one of the things that makes 64-bit handier
we literally spent like 5-6 years trying to get people to make 64-bit kernels as their first and accept that 32-bit is obsolete and should die out
and now, summer 2024, for some reason everyone wants 32-bit kernels
makes me wonder what is going on behind the scenes that i am missing
robust bobcat
makes me wonder what is going on behind the scenes that i am missing
it's a conspiracy
craggy tide
it's a conspiracy
it sure does feel like one
obsidian pivot
and now, summer 2024, for some reason everyone wants 32-bit kernels
I think we've all gotten bored of how easy 64-bit archs are
in terms of how much memory you get
craggy tide
i think you're a big part of the problem
craggy tide
what? no
robust bobcat
I think we've all gotten bored of how easy 64-bit archs are
fake news, 64-bit is a bigger number which means it's more difficult
doubly as difficult
craggy tide
sorry if you got "bored" of 64-bit architectures and actually avoiding a big set of issues with OS development
doesn't mean everyone should start to be pushy about doing 32-bit ports
obsidian pivot
the only reason my kernel remotely has 32-bit support is because the m68k
craggy tide
it would be understandable if it was a future feature that needed work to support
but every day 32-bit architectures die more and more
they are pretty much extinct outside of microcontrollers and other niches
obsidian pivot
true
craggy tide
the only reason my kernel remotely has 32-bit support is because the m68k
the m68k is dead and i do not care how good it was
at least for general computing
obsidian pivot
the only reason my kernel remotely has 32-bit support is because the m68k
(and the only reason I did that was to procrastinate the developement of my vfs)
((maybe that was a bad idea))
(((anyway)))
grand shadow
fd passing is almost done, I'll finish it tomorrow
went to a work event today and that sucked all the energy out of me, even taking a fat nap in the evening didn't help too much
fleet osprey
they are pretty much extinct outside of microcontrollers and other niches
Raspberry Pi 0 is neither a microcontroller, nor a "niche" computer
it may not be common sure, but I wouldn't say it's niche either
craggy tide
i'd argue it's very much microcontroller
no one in their right mind uses rpi0 as a general purpose computer
i am just REALLY tired of all the 32-bit apologists in full swing lately
it really annoys me and it seems like i am not the only one that noticed the anomalous wave of this stuff
fleet osprey
no one in their right mind uses rpi0 as a general purpose computer
I run Linux on it and use it for Linux things
I don't run a GUI on it ofc, but I do use as a general purpose Debian box
you have to actually go out of your way to use it as a microcontroller, and not as a Linux machine
so I'd argue that it's very much closer to a general purpose computer than a microcontroller, even if it can technically be used as one
craggy tide
agree to disagree
warped plinth
went to a work event today and that sucked all the energy out of me, even taking...
fat naps r good tho
inland cliff
at least DTS's aren't, I've never actually parsed a compiled DTB, so ig I can't ...
The worse part may be actually using them in your OS, yknow, in a useful way. Parsing data structures is fun and all (or maybe not) but you still need to do something with it
inland cliff
I think we've all gotten bored of how easy 64-bit archs are
Time to work on 128 bit kernel
inland cliff
Raspberry Pi 0 is neither a microcontroller, nor a "niche" computer
it may not b...
3D printing boards are commonly 32 bit ARM
neon crane
they are pretty much extinct outside of microcontrollers and other niches
osdev is a niche by itself
i disagree with beginners making 32-bit operating systems because the 64-bit x86 platform is genuinely easier than the legacy shit
but if you know what you're doing, imo, feel free
neon crane
I think we've all gotten bored of how easy 64-bit archs are
YOU are probably bored
frankly, I'm not
I want to port my OS to i386 class platforms at some point but that's well after I start writing the userspace for it
neon crane
and now, summer 2024, for some reason everyone wants 32-bit kernels
i think the "wave of people who want 32-bit kernels" is a wave of misdirected people from other osdev spaces
such as the osdev.org forums
errant zephyr
some virus is spreading around our circles where people find a sudden urge to lo...
maybe people like not always taking the easiest path?
and also "everyone" is a stretch
neon crane
i mean sure but beginners in particular should not try to run before they can walk
errant zephyr
ofc
neon crane
also mint's general disdain of 32-bit platforms shouldn't degrade into a ban wave
let the people do what they want and brag about it
you can and should intervene if those people shill 32-bit as "easier"
tawny oak
but if you know what you're doing, imo, feel free
I do agree with you here, but I think there's also something to be said for it being discussed frequently which leads to it becoming a normal topic, and so beginners think its a normal thing to do.
obsidian pivot
YOU are probably bored
nah not really
I was going to elaborate on what I meant by that, guess I forgot to
split gazelle
ffs i really want to leave this server because of mint
mint this is not that serious
i just want to osdev in peace
i don't care what bits the os is made for
i don't care what cpu the os is made for
i just want to talk about all kinds of osdev
no matter if its 64, 32, or even 16 and 8 bit
now my trust with you mint has been broken thanks to your bullshit
tawny oak
ok lets not make this personal - if you want to leave no one is stopping you, but I think we should perhaps take this outside of mathewnd's progress thread.
crude agate
32 bit kernels are dumb child play and have been obselete for 27 years
if u look at old 32 bit windows NT you'll see that it's nothing more than a stupid kids' toy...
this is a huge problem in the osdev community, we need to stop the spread of 32 bit and get beginners into 64 bit projects which are 2 times better
errant zephyr
yeah and clearly the best way to do that is to ban all the people that say otherwise
it's well known that a discord mod must ban all opposing opinions asap
analog berry
i think dwl was memeing a bit
split gazelle
this is a huge problem in the osdev community, we need to stop the spread of 32 ...
yeah
errant zephyr
that is obviously how you raise the activity of your server
obsidian pivot
who fights back
split gazelle
people should be forcibly use 64-bit x86 and use only the limine bootloader
obsidian pivot
and BAN EVERYONE who has EVER written one, or has a port to any 32-bit arch in their kernel
split gazelle
if you use the shitty and obselete multiboot, you will be banned from osdeving
if you even dare to port your os to some random niche failed risc architecture from 1991, you will be executed on public and it will be livestreamed to the entirety of the osdev discord server
every architecture and bitness are banned except x86 and 64-bit
every bootloader and boot protocol are banned except limine
analog berry
every architecture and bitness are banned except x86 and 64-bit
nah all architectures supported by limine are allowed
split gazelle
you are forced to use limine and you will have to accept it
this is ingsoc, every os should be on 64-bit x86 and limine
otherwise, you will be publicly executed live on tv
praise our supreme leader mintsuki
crude agate
i think dwl was memeing a bit
no im dead serious
32 bit is killing the next generation
split gazelle
yeah
crude agate
why do u think none of the gen z can find jobs
split gazelle
we're serious
crude agate
they dont know 64 bit
tawny oak
alright, lets take this out of the progress thread
im going to start muting (I understand that conversation happens naturally, but we are so far removed from astral at this point)
crude agate
alright, lets take this out of the progress thread
to where
tawny oak
#lounge-0
analog berry
Currently making some uacpi patches to make astral init code nicer lmao
tawny oak
nice - how many kernels have you ported uacpi to now?
analog berry
well me personally just managarm at this point (if we're not counting my very early ports to test uacpi on real hw)
hoping to eventually see it in reactos (there was some interest iirc)
crude agate
when is the linux port happening
analog berry
when u submit the patches for it ofc
crude agate
lol
flat copper
go for windows or go home
spare sky
when machines had 4 MB of RAM
32-bit address space seemed as vast in the 70s as a 48-bit address space seems today
flat copper
you can support 32bit easily on machines that have very little memory (so you can do a higher half physmap) and low mmio
crude agate
i wonder if we'll ever actually need a full 64 bit address space
spare sky
maybe i should just start banning people spreading 32-bit propaganda
are there people spreading 32-bit propaganda?
I haven't even noticed a huge increase in 32-bit OSes
obsidian pivot
the only kernels I know of with a port to a 32-bit arch on this server are northport, keyronex, and my kernel
crude agate
Big 32 is waging a war on osdev
obsidian pivot
also maestro, but that was done out of not knowing that 64-bit is a better decision
split gazelle
I haven't even noticed a huge increase in 32-bit OSes
^
mint are talking about nonsense
robust bobcat
mean
spare sky
please be civil
grand shadow
Currently making some uacpi patches to make astral init code nicer lmao
What do you plan on doing
analog berry
What do you plan on doing
this stuff basically #1217009725711847465 message
glue init hooks
lean kettle
Also please don't use the astral thread for this (the weird 32 bit flame war I was too asleep to notice), it's clogging up a actually good progress report thread
grand shadow
this stuff basically https://discord.com/channels/440442961147199490/12170097257...
Ohhh yeah makes sense
Tbh at one point I should look into making that code better so I dont end up with an infinite number of *_init functions there
analog berry
@grand shadow whats the proper way to cleanly terminate a kernel thread remotely if i only have its handle
grand shadow
none currently
neon crane
<@215823874615541761> whats the proper way to cleanly terminate a kernel thread ...
why do you need to do that
flat copper
and you'd need to avoid having locks held when the thread is deleted out of existance
grand shadow
^
analog berry
why do you need to do that
i kind of dont, but htere's a deinit hook so if wanted to implement that properly id need to terimate the threads
analog berry
and you'd need to avoid having locks held when the thread is deleted out of exis...
yeah the proper way would probably be to wait for the thread to kill itself and signal that its now done
also i guess astral doesnt have anything like a workqueue so i have to make an adhoc list or something like that
flat copper
what do you need this for
analog berry
scheduling work from an interrupt handler
flat copper
id just make your own list of entries, since you dont need any of the interaction that a real workqueue needs
analog berry
well it kind of lines up with what workqueues do perfectly, does it not?
u submit some work and u wake the thread
flat copper
but you have more than one workqueue threads with dynamic scaling etc
flat copper
if they're all of mostly the same priority its the best way (so all low priority or also all super high priority)
analog berry
yea
neon crane
scheduling work from an interrupt handler
astral has dpcs doesnt it
analog berry
I dont think dcps are supposed to be sleeping or acquiring locks or take a lot of time
neon crane
I dont think dcps are supposed to be sleeping or acquiring locks or take a lot o...
you should be able to at least acquire spinlocks
what do you need to do that "takes a lot of time"?
analog berry
what do you need to do that "takes a lot of time"?
Execute aml?
analog berry
you should be able to at least acquire spinlocks
Thats not what I need
grand shadow
aaaAAAAAAA this took longer than I wanted
analog berry
Thats insane
Gj
@grand shadow is there like a max column width u like
Or do u just not break lines ever
neon crane
wdym?
analog berry
E.g. I limit my lines at 80 chars
neon crane
oh source code line limit
analog berry
Yeah
grand shadow
when I remember to break them I do π usually if its tooooo big I add a line break
analog berry
I see
analog berry
yay my astral work queue works
@grand shadow any objections to these?
idk how they were still missing
grand shadow
<@215823874615541761> any objections to these?
go ahead
tbh idk too I was just too lazy lmao
analog berry
cool np
analog berry
Okay uhhh i think its kinda ready: https://github.com/Mathewnd/Astral/pull/8
i'll leave it to you to make a shutdown program, but acpi_poweroff and acpi_reboot helpers are already there
analog berry
lol np
robust geyser
β οΈ
grand shadow
actually crying
robust geyser
that hashmap thing is criminal
it felt terrible copying that from the nvme driver
but i also couldnt think of anything better π
analog berry
lmaooo
grand shadow
I'll open a pr for some mlibc stuff and then I'll take a look at it
robust geyser
power management stack coming soon?
analog berry
probably a lot more to do before we get into that
it needs to at least use the namespace to load drivers based on the devices present
etc etc
proper meteor
aaaAAAAAAA this took longer than I wanted
incredible
grand shadow
implemented putgrent in mlibc so useradd doesn't die when it creates the group
assertos is no more
grand shadow
xhci would be very epic
robust geyser
it would indeed but im very lazy
also i have no idea how to make the usb stack work with ehci/uhci so that will be very fun
neon crane
i really hated that i had to make a hashmap like the nvme driver lmao
i'll sleep better at night thanks to knowing that i did something better than the project i looked at for parts of my nvme driver
except i already did something better than it which is using an MDL (list of physical pages) instead of issuing several consecutive read requests
grand shadow
@analog berry https://github.com/d-tatianin/Astral/blob/8936a774714745686d8c2e2232e07fe4f80f92ba/kernel-src/io/acpi/acpi.c#L57 when returning an errno I don't negate it
if you didnt do that anywhere else I can fix it when I merge
alloc() also returns zero'd data already (uacpi_kernel_calloc) so I will probably also remove the memset there
craggy tide
if you use the shitty and obselete multiboot, you will be banned from osdeving
i think you took my frustration comments more literally than they were intended to be taken
that said, all of what you said afterwards really came off as built up hatred towards me rather than just a reaction to 1 frustrated comment of mine
analog berry
alloc() also returns zero'd data already (uacpi_kernel_calloc) so I will probabl...
Ohh
analog berry
<@353593948523986945> <https://github.com/d-tatianin/Astral/blob/8936a7747147456...
Will fix
craggy tide
especially the part bringing up Limine for some reason (alongside @analog berry ) as if that is the reason i am frustrated at 32-bit stuff
i've never been frustrated at people using Hyper or BOOTBOOT or whatever else, even mb2, making 64-bit kernels
pointing at Limine over my frustration with an unrelated issue is frankly pretty gross
analog berry
I was memeing lol
craggy tide
you may have, but the other guy seemed genuinely mad
analog berry
He's 15 so
craggy tide
for the record i am pretty sure they were banned before as well
for unrelated reasons
craggy tide
i think you took my frustration comments more literally than they were intended ...
but no just to spell it clearly: not 100% of what i say, especially when i am frustrated, is meant literally. did anyone get banned for making 32-bit OSes? no, obviously not
if i ever did that, lukflug is unironically going to show up at my front door and beat me up
craggy tide
you can
no you can't
even with 1 MB of ram
MMIO ever since PCI has been a thing has always been near 3.5/4GB
neon crane
why not
craggy tide
while free memory has always, almost by rule, been lower down, towards 0
that already stretches almost the entire address space
grand shadow
@analog berry also are uacpi events supposed to wake up only one thread?
craggy tide
pee cee eye has not always been a thing
yeah of course, sure on legacy ISA only systems it's doable
analog berry
<@353593948523986945> also are uacpi events supposed to wake up only one thread?
One work type wakes up one thread
spare sky
on 386SX-class machines, everything definitely below 16 MiB
neon crane
exactly
craggy tide
i specified PCI for a reason
grand shadow
One work type wakes up one thread
ah
cuz there is an event thing already
thats why I was asking about the semaphores being used
lean kettle
this is the astral thread
neon crane
cant really have a hhdm on 32-bit systems
-czapek
you can
-fadanoid
you can't
-mint
lean kettle
just fyi
analog berry
cuz there is an event thing already
Ah maybe I missed them
Are they better than my handrolled thing
craggy tide
this is the astral thread
i am replying to things people raised in here
grand shadow
Are they better than my handrolled thing
they wake up all threads and have a built in timeout
tho I think it depends on being able to target a specific cpu
analog berry
they wake up all threads and have a built in timeout
I guess I don't really need that
craggy tide
> cant really have a hhdm on 32-bit systems
-czapek
> you can
-fadanoid
> you ca...
you can't*
*on any 32-bit system that has MMIO significantly high up in memory like literally every PCI capable system
there
grand shadow
then semaphores should work just fine
neon crane
also you can simply segment it into PCI space and normal space, it wouldn't be a direct map but youd map all accessible memory
pci space cant be more than like 512 meg can it
craggy tide
well yes, but you said it yourself that it wouldn't be a direct map
neon crane
ye
craggy tide
but that's as long as you don't have too much memory, as well
analog berry
then semaphores should work just fine
I think the current thing should work yeah, because it needs very simple logic with one producer one consumer
Or rather many producers potentially but one consumer
grand shadow
I think the current thing should work yeah, because it needs very simple logic w...
and also, is semaphore_reset expected to be called with any threads waiting on it?
analog berry
and also, is semaphore_reset expected to be called with any threads waiting on i...
Good question, probably not but at the same time we should probably guard against that?
What would be the response in that case
grand shadow
either assertosβ’οΈ or wake up all waiting threads with a special reason or something
probably assertos
atm
analog berry
Could we just do nothing and just consider this a spurious wake-up if there's a race?
Or does that break something
grand shadow
if you wake the threads up theres a very good chance the thread will release the semaphore when it finishes doing whatever its doing I imagine
talking about its use in general
but it doesn't nescessarily check for the semaphore value when it wakes up (since it assumes it can run)
analog berry
Probably
grand shadow
Good question, probably not but at the same time we should probably guard agains...
is there any way this can happen or is it just a maybe
analog berry
is there any way this can happen or is it just a maybe
AML is allowed to do anything lol
Its a question of whether any pc does that
And that I dont know
Maybe error instead of assert is better
Event is generally not used that often in aml
Let alone resetting it
grand shadow
ah fuck it then I'll just assert
analog berry
U think so?
grand shadow
if I ever run into a machine that has the very specific conditions for this to happen I will get a lottery ticket and fix it
analog berry
if I ever run into a machine that has the very specific conditions for this to h...
Lol
Ill add it to my pr then
grand shadow
@analog berry also do you know how linux/other unices handle stuff like telling init that the power button was pressed?
maybe theres a signal or something
analog berry
<@353593948523986945> also do you know how linux/other unices handle stuff like ...
Yeah I do
Its the protocol they use
wild marsh
god bless
analog berry
I think netlink yeah
wild marsh
that'd be the linux-specific way anyway
robust geyser
but i think i remember something something signals
well, shutdown i mean
not exactly power button
analog berry
but i think i remember something something signals
I mean usually the response is to bring up a logout gui
So idk why u would need something as hard as a signal for that
wild marsh
you could just have a device file that you poll/blocking read from and it returns some data when the power button was pressed if you're not aiming for compat with other systems
analog berry
Lol
Astral can just make a file or something
Same way keyboard events work for example
grand shadow
yeah probably
wild marsh
i mean what i described is basically linux's /dev/input :^)
analog berry
i mean what i described is basically linux's /dev/input :^)
Ye
grand shadow
I guess I could have init polling on that for the button press and some named fifo for user commands like poweroff
or have a special acpi process in the background that calls into that fifo or something
analog berry
Ill take a look how Linux does it once I get back home
wild marsh
some cursory poking in sysfs reveals that the acpi power button is just also in /dev/input/?
analog berry
Oh yeah there u go
grand shadow
/dev/acpi it is then
robust geyser
dear lord
actually that makes sense
u can just broadcast all acpi events there
nice
analog berry
There's also lid besides the power button
wild marsh
some cursory poking in sysfs reveals that the acpi power button is just also in ...
qookie@selenium ~ Ξ» cat /sys/devices/platform/PNP0C0C\:00/firmware_node/input/input0/event0/dev
13:64
qookie@selenium ~ Ξ» ls -lh /dev/input/
total 0
...
crw-rw---- 1 root input 13, 64 Sep 1 14:43 event0
...
/dev/keyboardX node only for acpi events
wild marsh
based
grand shadow
I'll take a small break and then build astral with the uacpi pr
analog berry
Ill update whatever we discussed in a bit
obsidian pivot
while free memory has always, almost by rule, been lower down, towards 0
Tbh though, you should be remapping your mmio elsewhere
analog berry
Yes but thats not what hhdm is about
obsidian pivot
I replied to the wrong msg
obsidian pivot
MMIO ever since PCI has been a thing has always been near 3.5/4GB
menat to respond to this
*meant
analog berry
Poor Matthew lmao
grand shadow
gonna try to compile limine
but first I need mtools and I'm too lazy to add the recipes sooo
mtools built it seems
what
craggy tide
seems like it failed to validate your toolchain
grand shadow
does it need x86_64-elf?
craggy tide
no it should build fine with your host compiler
it tries to compile a small object to test if the toolchain compiler actually works
something like ```sh
echo | gcc -x c -c - -o /dev/null
OH
craggy tide
rip
grand shadow
seems like /dev/{null,zero,random,full} do accept seeks so time to change that
well that was disappointing
craggy tide
interesting
wild marsh
god bless it's trying to poke at libc internals
craggy tide
slow but no hacks moment?
wild marsh
no because the existence of __freadptr gets rid of the gnulib impl :^)
grand shadow
first line too like damn didn't even give me the satisfcation of making it go a tiny bit into the build
wild marsh
who even added __freadptr
craggy tide
it did configure :^)
prime juniper
who even added __freadptr
me likely?
craggy tide
git blame time
wild marsh
could just make it return nullptr unconditionally
grand shadow
what does that function even do I wasn't able to find a man page
wild marsh
gives you a pointer into the internal FILE buffer
and the amount of data that's currently sitting there
grand shadow
could just make it return nullptr unconditionally
I'll see if doing that makes it work
wild marsh
that's what the gnulib SLOW_BUT_NO_HACKS impl does
analog berry
@grand shadow aight i think i fixed everything we have talked about
grand shadow
seems like so?
grand shadow
<@215823874615541761> aight i think i fixed everything we have talked about
I'll take a peek in a sec
seems to actually be building
oop
forgot to build mtools with --disable-nls π
wait no thats not nls
iconv_open on mformat
so it is not disable nls stuff lol
tf
robust geyser
lmaooo
grand shadow
I just wont build uefi-cd then
craggy tide
lol
craggy tide
what if the law requires full warranty
honestly not sure
that's just what all licenses say
grand shadow
wonder what more I really need to be able to slap a "self hosting" label on astral
craggy tide
nothing
analog berry
well u cant compile gcc yet right
grand shadow
well u cant compile gcc yet right
yeah true
craggy tide
i don't think you need to be able to build gcc to count as self hosting
analog berry
how do u host yourself if u cant build the compiler
wild marsh
i mean going by that logic you can't really self host ever because you can't run jinx?
robust geyser
whats the problem with building gcc on astral?
craggy tide
well in my personal definition of self hosting you only really need to be able to build the kernel and deploy it such that you can reboot into the new instance
analog berry
i mean going by that logic you can't really self host ever because you can't run...
wdym?
analog berry
whats the problem with building gcc on astral?
mlibc bug
grand shadow
whats the problem with building gcc on astral?
it segfaults deep in mlibc phd code
analog berry
ah
robust geyser
mint did say she might add a containerless mode tho
grand shadow
well in my personal definition of self hosting you only really need to be able t...
well then I better go add fat fs support
unless I try limine's ext2
robust geyser
well but user ns is the main feature
craggy tide
even if you had linux user ns how do you expect to chroot into debian
robust geyser
true
craggy tide
if you don't support linux compat
craggy tide
unless I try limine's ext2
should work [no guarantees made]
wild marsh
just use a linux distro that uses mlibc then switch out libc.so etc for your kernel's version (assuming you use the linux abis) :^)
craggy tide
ah yes
analog berry
just implement every linux syscall
craggy tide
easy
grand shadow
isnt there like some linux emulators or something
wild marsh
port qemu and use qemu-x86_64
grand shadow
true
craggy tide
does that even work like that?
carmine swallow
port qemu and use qemu-x86_64
doesn't that require a host linux tho?
craggy tide
^
wild marsh
i don't actually know if it just translates syscalls or emulates thm
carmine swallow
because as far as I understand it just translates them
craggy tide
i think it translates them
carmine swallow
but I haven't looked at the source so I can't say for sure
grand shadow
the biggest thing about the "swapping out the kernel" part is that while my root filesystem is on disk the kernel is loaded from the limine cd
craggy tide
why?
grand shadow
it was just easier to make a disk like that
robust geyser
you can write a limine config on astral
regal stone
holy shit
robust geyser
;)
regal stone
limine inside of limine
grand shadow
you can write a limine config on astral
true
craggy tide
limine inside of limine
inside of astral*
regal stone
inside of astral*
ik
robust geyser
port fdisk/parted so you can install from a live cd
regal stone
but technically
craggy tide
then technically i develop limine inside of limine since i boot linux with limine
regal stone
im curious
grand shadow
port fdisk/parted so you can install from a live cd
I wanted to port fdisk but thats util linux
so it probably needs some weird linux api
regal stone
how long has this project been in development
robust geyser
so it probably needs some weird linux api
not really
grand shadow
this rewrite like a year and a few months
robust geyser
youd be surprised
grand shadow
oh huh
carmine swallow
also I think I am going to make the mlibc glibc binary compat thing I made a little less hacky so then you can try it on astral if you want though idk what you could run with it, for factorio you need X11, Xext, GL, Xinerama, Xrandr, Xcursor, alsa (audio devices don't need to be there, just the lib), pulseaudio (copying libs from host might work as a hacky workaround), xcb, Xrender, Xfixes, dbus, Xau, Xdmcp, libsndfile, libasyncns, libflac, libvorbis, libopus, libogg, libmpg123, libmp3lame
grand shadow
download sysroot, format disk, unpack it, install limine
honestly that could work
maybe
I just need the disk formatting tool
I'll try it some point in the future
right now I'm gonna get uacpi merged after I shower
grand shadow
also I think I am going to make the mlibc glibc binary compat thing I made a lit...
dear lord I only have like half of that
prime juniper
well in my personal definition of self hosting you only really need to be able t...
can you develop on your own OS? tools can be shipped from the outside but you need to be able to code, compile and run (either run the entire thing in qemu or reboot into new kernel). If yes, that's self-hosting
grand shadow
@analog berry seems to be working here, was able to build and tell it to power off
is there anything is specific you need me to test? otherwise lgtm
all I have is some style stuff that I'll change once its merged but nice work ty
analog berry
is there anything is specific you need me to test? otherwise lgtm
well not really unless u wanna give real hw a test
obsidian pivot
are you going to start enumerating the acpi namespace for devices now that you have uACPI
it's really cool
analog berry
all I have is some style stuff that I'll change once its merged but nice work ty
thanks! i can change stuf myself if u want
grand shadow
are you going to start enumerating the acpi namespace for devices now that you h...
soon I'll have to write an actual decent driver interface as to not fill the main function with 100 different init calls
once I do that I will certainly use it
but for now I might play around and add a /dev/acpi device
obsidian pivot
I'll just make oDriverInterface 
grand shadow
oOsdev
analog berry
but for now I might play around and add a /dev/acpi device
oh yeah that would be cool
if shutdown worked via gui
obsidian pivot
but for now I might play around and add a /dev/acpi device
what would it do though
robust geyser
you can poll on it
and figure out
if someone pressed the power button
or closed the lid
or whatever
i imagine for now it would only be the power button
grand shadow
and probably I'd have you be able to write some character into it or something to tell the computer to shutdown
analog berry
and probably I'd have you be able to write some character into it or something t...
u dont want like a shutdown(2)?
obsidian pivot
I'll just make oDriverInterface <:ultrameme:729026898666979428>
if I do ever do that, it would support dynamically loaded elf files
robust geyser
that would be a better idea
grand shadow
u dont want like a shutdown(2)?
nothing stops me from making sys_reboot in mlibc just open /dev/acpi on x86 or something
analog berry
ig
grand shadow
reboot(2) seems very linuxy and I want to be able to just
echo "1" > /dev/acpi
or something
robust geyser
i think /dev/acpi is fine
analog berry
keep in mind
u will need to expose e.g. battery devices
fan devices
etc
tons of stuff
to userspace
obsidian pivot
a directory would be better since acpi is a namespace
robust geyser
/sys
grand shadow
do you have any docs for uacpi
obsidian pivot
he has docs in header files
analog berry
if u port a gui for showing battery indicators that would be cool
obsidian pivot
in forms of comments
analog berry
do you have any docs for uacpi
headers are all documented + osdev wiki has tons of stuff
it has like driver interace examples and stuff
obsidian pivot
headers are all documented + osdev wiki has tons of stuff
btw is uacpi_namespace_for_each_node_depth_first public api
now
analog berry
of course
obsidian pivot
because months ago you said my thing used private api
like before you had public api
analog berry
it used to be private probably
grand shadow
hm I think for now I will just have a very simple /dev/acpi and once I start enumerating devices (both pci and acpi) I will redo it
since it'd be super simple and work just for stuff like the power button
and other simple things
analog berry
For modern real hw u will also want an EC driver
since thats how power button is routed there
obsidian pivot
or you can be lazy like me and say TODO: Implement EC driver
(7 months todo by now)
grand shadow
sync; sync; echo 5>/dev/acpi
urinonically what I will do in init lmao
sync();
sync();
write(acpifd, WHATEVER_I_CHOOSE, 1);
analog berry
why are two sync's needed?
analog berry
lol
robust geyser
sync is not allowed to complete while any other sync is in progress
so thats why you usually double sync
analog berry
its async?
grand shadow
on astral sync() is blocking
robust geyser
nice
grand shadow
posix just leaves it at a The writing, although scheduled, is not necessarily complete upon return from sync().
and I decided that its just safer to wait until its done
grand shadow
tho I also need to implement kill(-1, signal) for it to make sense
robust geyser
VERSIONS
According to the standard specification (e.g., POSIX.1-2001), sync() schedules the writes, but may return before the actual writing is done. However
Linux waits for I/O completions, and thus sync() or syncfs() provide the same guarantees as fsync() called on every file in the system or filesystem reβ
spectively.
hmm
i always thought sync on linux was async
analog berry
i guess its just posix stupidity
robust geyser
i vividly remember that sync would just return immediately
while the 2nd one would block
grand shadow
probably shitBSD 0.01 didn't block and they decided to leave it at that
obsidian pivot
only one way to find out
analog berry
probably shitBSD 0.01 didn't block and they decided to leave it at that
u mean ShitOS3
obsidian pivot
isn't that dwl's os
analog berry
its up for grabs
obsidian pivot
@stark walrus do you want shitos5 instead of secureos