embedded hardware
#Nyaux
flat nymph
kind root
Embedded devs love crappy proprietary solutions to already solved problems
broken depot
Embedded devs love crappy proprietary solutions to already solved problems
Don't you love it when microcontroller XYZ implements the RISC-V PLIC spec but not the standard address space layout for it?
tawdry mirage
for controllers that expose a window into the config space:
- configure the pcie controller to show you the config space of the function you want to show to the os
- have mcfg point to the window, and say that only 1 function is available there
- the os will see the device you wanted to show as XXXX:00:00.0
broken depot
Because I've seen that one too.
tawdry mirage
for controllers that expose a window into the config space:
1. configure the pci...
this is what uefi for the rpi5 does btw
kind root
Hmm but how, you would still have to make sure the entire space is mapped at some address
tawdry mirage
for windows support
thorn bramble
apparently not even original solutions to already solved problems, just copy and paste existing solutions and change them ever so slightly to fuck people over :^)
kind root
for controllers that expose a window into the config space:
1. configure the pci...
So u would only show one device at a time?
kind root
Don't you love it when microcontroller XYZ implements the RISC-V PLIC spec but n...
Yeah like why
broken depot
apparently not even original solutions to already solved problems, just copy and...
Microcontrollers are like this all the time and it frustrates me (an embedded dev as a hobby) to no end.
tawdry mirage
So u would only show one device at a time?
one function even, so multifunction stuff like gpus also doesn't work
surreal path
wait i need to impl a way to exit a thread
like
destroy the thread
from ze queue
broken depot
wait i need to impl a way to exit a thread
Well that would be useful for more than just using uACPI in your kernel
surreal path
yea
kind root
That's why I kinda don't wanna invest into the aarch64 version of the kernel if I start making one, it just sucks and u have to have 9999999 drivers per chip
surreal path
wait i need to impl a way to exit a thread
will ask in #schedulers
kind root
That's why I kinda don't wanna invest into the aarch64 version of the kernel if ...
Maybe if I can get my hands on like an aarch64 framework or something
Idk if that has a sane board
flat nymph
Maybe if I can get my hands on like an aarch64 framework or something
framework is riscv
I think
besides x86
thorn bramble
That's why I kinda don't wanna invest into the aarch64 version of the kernel if ...
just support qemu virt board and nothing else 
kind root
framework is riscv
Even better, less memory caching garbage problems to worry about
flat nymph
raspberry pi
thorn bramble
or one single random version of the rpi
flat nymph
and a lot of proprietary garbage
flat nymph
Isnt the latest pi proprietary af
yeah
with risc-v there's no good hardware
(from what I gathered at least)
I also want something physical for my RISC-V port
But I have a bunch of raspberry pi s
So pmOS ARM port???
tawdry mirage
Even their usb
usb on the pi5 is just xhci iirc? but it's on their weird rp1 chip which has everything smushed into one pcie function at different offsets into the pci bar
kind root
I see much more value in supporting my old abandoned laptops than a pi, at least those have a proper GPU and their own screens and stuff
flat nymph
and a lot of proprietary garbage
though it's kinda remedied by the fact that everyone support them
kind root
usb on the pi5 is just xhci iirc? but it's on their weird rp1 chip which has eve...
Pi4 is xhci, pi5 is proprietary stuff iirc
molten grotto
no pi5 also has a xhci
tawdry mirage
Pi4 is xhci, pi5 is proprietary stuff iirc
no it's also xhci
broken depot
That's why I kinda don't wanna invest into the aarch64 version of the kernel if ...
Sweats in already comitted to doing ARM port for BadgerOS
molten grotto
its a dwc3 tho so no public official specs for that
flat nymph
though it's kinda remedied by the fact that everyone support them
like all bsds and a bunch of small operating systems
kind root
Ah ok then
tawdry mirage
its a dwc3 tho so no public official specs for that
yeah but you don't really need any of the device mode stuff for the usb-a ports
so it's fine
kind root
usb on the pi5 is just xhci iirc? but it's on their weird rp1 chip which has eve...
Also wtf is that latter part
flat nymph
and a lot of "let's do baremetal dev" kinds off tutorials/things
kind root
*Sweats in already comitted to doing ARM port for BadgerOS*
Nah I know you are passionate when it comes to this stuff 
tawdry mirage
Also wtf is that latter part
so for the rpi5 they moved a bunch of peripherals off of the main soc onto a second chip (a la southbridge) that's connected over pcie, and that chip is a custom design, so ofc it only has one pcie function with a large bar that has the mmio regions of all the peripherals in it
molten grotto
yeah but you don't really need any of the device mode stuff for the usb-a ports
it would be interesting to make a device mode driver tho
tawdry mirage
the way you're supposed to discover the individual pieces is with acpi/dtb
or just hardcode them if you find a pci device with their vendor:device ids
kind root
Least scuffed embedded design
tawdry mirage
look for device nodes with specific _HID and _CID values
flat nymph
like not native
tawdry mirage
well acpi is only available if you use the uefi firmware for it
kind root
look for device nodes with specific _HID and _CID values
And _ADR
tawdry mirage
well acpi is only available if you use the uefi firmware for it
and that's unofficial to begin with
tawdry mirage
so for the rpi5 they moved a bunch of peripherals off of the main soc onto a sec...
at least the rp1 has some docs available (although they're still the preliminary ones with some details missing)
and what's fun is that it has some arm microcontroller cores that you can run code on :^)
kind root
The southbridge?
tawdry mirage
yeah
kind root
Damn
tawdry mirage
and the arm core has access to all the peripherals, including a dma peripheral, so you can program the southbridge to dma to the host ram or whatever
kind root
So dma would be just code in the sb doing a manual memcpy?
tawdry mirage
it has an actual dma controller too
kind root
Ah
tawdry mirage
but yeah the host physical memory window is mapped into the rp1 address space too
i think the dma peripheral is generic and you just point it to the host ram window for dma across pcie
but this is unrelated to things like xhci etc (although ig you could point xhci into the internal rp1 ram instead of host ram?)
tawdry mirage
yes
kind root
Ah
tawdry mirage
tawdry mirage
but yeah the host physical memory window is mapped into the rp1 address space to...
ah but the arm cores don't have access to it
because the 32-bit addr space is too small so it doesn't fit along with all the peripherals
broken depot
They call the entire cpu an application processor
What? In the RP1?
That's just silly
tawdry mirage
or well, not to the full space, afaics there is an ATU window thats 1G (which i assume is programmed to point somewhere into the host address space)
broken depot
or well, not to the full space, afaics there is an ATU window thats 1G (which i ...
There's probably a register to change the window.
kind root
What? In the RP1?
I think the host cpu is called an AP
And in x86 land an AP is something else
Basically just any cores started by bsp
tawdry mirage
There's probably a register to change the window.
this seems about right
you select which 1G chunk you want to look at
and program that into ATU_REGION
since the direct window is 512G in size, the ATU window is 1G in size, and ATU_REGION is 9 bits long
(1<<9 = 512)
broken depot
Actually calling something an application processor is pretty common (in the MCU space at least) for whatever the developer and/or end user is supposed to program.
kind root
Yeah name collisions as per usual
broken depot
Always.
It gets to a point where I sometimes do a double-take going "this thing doesn't contain XYZ, does it?"
tawdry mirage
since the direct window is 512G in size, the ATU window is 1G in size, and ATU_R...
unfortunately i don't have an rpi5 so i can't test
kind root
Ye
broken depot
Like calling the shitty single-level "page table" that ESP32s have for their memory-mapped FLASH an "MMU".
kind root
Lol
broken depot
That made me do a double-take because I was convinced they hadn't had an MMU.
Lo and behold, they didn't, in fact, have an MMU in the sense any programmer would reasonably understand it.
kind root
Its hard to come up with original names for things ig
broken depot
Names are hard indeed
I have come up with a completely original name for a project exactly once to date.
kind root
Badger?
broken depot
Nobody had named a RISC-V CPU "Boa RISC-V" before I did.
kind root
Nobody had named a RISC-V CPU "Boa RISC-V" before I did.
You have your own cpu?
kind root
Did you run it on an FPGA?
elder shoal
that means good risc v in portuguese
broken depot
The CPUs I made before RISC-V were all custom ISAs (and therefor the disadvantage of not having a toolchain).
broken depot
Did you run it on an FPGA?
Yep. Got it going at 25MHz max.
I know more ticks now that could make it faster but I haven't properly started another CPU project since.
It supports hardware atomics too, which I'm very pleased with.
kind root
Very cool
broken depot
I posted about it in #risc-v back when it was actually a recent thing.
So keep your eyes on there occasionally if you want to hear of the next oneβ’οΈ.
kind root
Yee
broken depot
But this is getting off-topic
broken depot
I try to respect channel's topics and this one's topic is your hobby OS
surreal path
currently waiting on monkous to respond about me trying to understand his scheduler desgn
in #schedulers ofc
surreal path
heres what im going to impl
i will add states to threads like running, ready, zombie
u can create a new thread its ref count will be initally 1, because someone in the freelist is referencing it
thread does thing, maybe later down the line thread might have something more then 1 because idk thread communication idfk how that works
if this new thread calls exit_thread()
it will decrement its ref count and put it in the zombie queue
the reaper thread will check if its ref count is 0
if it is free the resource
is this a good way to impl exiting threads
anyone know?
:(
??
silver yarrow
afaict its alright
surreal path
okokok
surreal path
somehow ive corrupted the stack???
i was cleaning up the scheduler code
stack is corrupted??
surreal path
cant even look at ze stack
thanks gdb
ummmm
srry what
π
does this look corrupt
i think it looks corrupt
uhhh
wheres my fucking
thing
bro
wait hold on
what in the fuckj
oh nvm
flat nymph
You've asked to remind you to fix your sleep schedule
flat nymph
Go to sleep now
surreal path
i cant
sorry
i somehow fucking corrupted the stack
it seems
nvm?
i did not?
okay
oaky
okay
i have found something out
after getting preempted while ur in a thread
something happens and the return address is gone
wtf
???
surreal path
something happens and the return address is gone
so it just stops executing thread code in its entirity
surreal path
isn't it 3am for you?
yea
surreal path
i tried already
plush hearth
try again
surreal path
something happens and the return address is gone
is this supposed to happen
like
what
what happened
plush hearth
you aren't pushing/popping the correct number of values?
surreal path
i should be
yea thats the return address
into kentry
not into kentry
rather
into return_from_kernel_in_new_thread
umm what the sigma
yea thats not right
i have no clue
whats causing this stack corruption
wtfff
no idea whats going on lol
i LITERARLY
just
functioned out the queue crap
π
on github ig
it panics with why am i here
for some reason
only one core panics as well
and timer interrupts are still firing
idfk whats going
on
rigid fable
@surreal path hello my sweetie human π
surreal path
hey
im going to sleep i tried debugging this for a good bit but no 1 im way too tired no 2 im already out of ideas π
night
rigid fable
im going to sleep i tried debugging this for a good bit but no 1 im way too tire...
yes go sleep π‘
surreal path
nvm it works now??
tf
weirddd
nvm not anymore
wtfff
now its not anymore??
@flat nymph
is this UB
i lit barely change the sched
it works
then i change it a little
it doesnt
silver yarrow
flat nymph
It's potentially normal?
flat nymph
its fucking iretqing into send eoi
If you enable interrupts in kernel
kind root
Nyaux whats the current ec status
surreal path
Nyaux whats the current ec status
well
i couldnt tell u
cause the fucking scheduler killed itself
but if i can get the scheduler working
and i can get things like blocking and exiting threads complete
we will have a more stable ec driver
surreal path
It's potentially normal?
kinda was debugging this then took a break
but
i swear to god
read the source urself even
the interrupts ARE dsiabled
π
i just send
on EOI
one*
thats insane
and then somehow the stack gets corrupted
and idfk whats going on
surreal path
<:trl:829316641191886888>
bru
kind root
cause the fucking scheduler killed itself
F
surreal path
F
yes f indeed
i think its a bug when exting out of a thread
exitingg*
as in
2nd preemption
cause ive seen it enter kentry
once
then some shit happened
idfk
ive been banging my head out
trying to debug this
chat this is so fucking weird
Like sometimes
it just decides
im gonna get stuck in send eoi somehow
and get preempted and then send another eoi and then get preempted
and make my iretq return address be send eoi
and sometimes
stack corrupting after 2nd preemption
@flat nymph no way ur awake
i think i will ask for help in #schedulers or something
now I COULD do that
or sleep
ill do it first
elder shoal
Sleep
surreal path
nahh
flat nymph
<@350748255929499649> no way ur awake
Why wouldn't I be
surreal path
Why wouldn't I be
exactlyyy
flat nymph
Maybe it's not switching stacks as it should idk
surreal path
im wtfing as well
surreal path
Maybe it's not switching stacks as it should idk
no it is
i checked with gdb
something fucky is going on
and all i did was cleanup the queueing code a little bit
flat nymph
Make sure you don't use the whatever mechanism there is to switch to a different stack on interrupt
surreal path
i didnt even touch the context switching
surreal path
Make sure you don't use the whatever mechanism there is to switch to a different...
wdym
surreal path
wdym, rsp0?
i still am switching out rsp0
whats wrong with that
even tried adding dummy bogus shit just to make the stack 16 byte aligned
which changed
NOTHING
i didnt even
change the thread struct
i just
changed the queuing
thats it
im so confused
thorny glen
it was probably always broken and u just did something that revealed the bug
surreal path
it was probably always broken and u just did something that revealed the bug
maybe
but the question is
what IS that bug
@flat nymph
i didnt change anything
it should work
flat nymph
<@350748255929499649>
This looks normal idk
flat nymph
i didnt change anything
You do sti though it shouldn't matter
iretq would ideally do that for you
Not ideally
surreal path
yep
flat nymph
It's noop
flat nymph
Yeah because it doesn't do anything
surreal path
yea so im cooked
because
i have no idea whats going on and my eyes look like bags rn
literal bags lmao
hexed mango
Sleep!!!!!
flat nymph
Ask someone smarter
flat nymph
Sleep!!!!!
It's like 11am in his timezone
surreal path
i didnt sleep since
hexed mango
nyaux grind is foreveer
Ya ok grandpa lets get you back to bed
surreal path
indeed
flat nymph
You enable interrupts right before you do send_eoi()
Thet's why it returns to it
I.e. it's working as expected
rigid fable
i just slept
sigma
surreal path
fair
that sworse
im sorry ur experiencing that
okay disabling all other cpus doesnt change the behavior
but itll be easier to debug now
stack looks like this
flat nymph
Which disappears if I add assert
And ubsan doesn't catch it
surreal path
values of stack pointer rn
it seems like it never rets into return_from_kernel_into_new_thread
not only that
but the timer doesnt seem to fire anymore
i think i know why
i am overwriting the new threads stack
with the interrupt context
flat nymph
Which disappears if I add assert
And if I add -fsanitize=undefined
surreal path
i am overwriting the new threads stack
and because i changed my thread run queue into a double linked list
with the thread pointing to itself
this makes sense
@flat nymph how do i fix this, i wanna fix this without doing some hacky idea
flat nymph
Fix what
flat nymph
Where's the code
flat nymph
Ugh how are you overwriting it?
surreal path
look here
on first context switch it thinks we were previously running an old thread
because something is in the runqueue
and the next pointer points to itself
u understand?
surreal path
solved it
flat nymph
And if I add -fsanitize=undefined
It was stack overflow
surreal path
laura is poop
edgy pilot
laura is poop
cringe opinion tbh
surreal path
@bleak cedar hi
surreal path
It was stack overflow
mishavkov
happy new year
@molten grotto happy new year
everyone else in the thread
happy new year
molten grotto
happy new year for you too! where I live it was already an hour ago
surreal path
happy new year for you too! where I live it was already an hour ago
π
fair
happy new year though
molten grotto
thanks
surreal path
npnp
plush hearth
everyone else in the thread
Happy new year! (It was 9 hours ago for me)
silver yarrow
im still a good 9 hours away
surreal path
@kind root happy new years
also ec works great on my laptop
so theres that
reason its iffy on other laptops is because i stilll need to impl scheduler things like blocking
im working on it
flat nymph
happy new years!
kind root
Happy new years
kind root
also ec works great on my laptop
Great to hear
surreal path
Great to hear
theres no brightness control events so im thinking that has to do with something else, also it seems like the sensors are routed through the ec
there is lid close and lid open
kind root
Which
surreal path
theres no brightness control events so im thinking that has to do with something...
idk i just read the executed aml method
hold on while i dump my dsdt
kind root
Which sensors
kind root
surreal path
<:thinkinplode:428602393723273219>
ummm i dont know how to dump it in windows
i will reboot to linux i suppose
silver yarrow
lol
surreal path
gives me an error
kind root
Which
thorn bramble
acpidump just works
kind root
Yup
thorn bramble
did bro open it in notepad
kind root
did bro open it in notepad
Yeah probably
surreal path
no i didnt
surreal path
ok
surreal path
iyea works now
@kind root maybe u can read it for me, im not the best when it comes to understanding aml
kind root
Im on my phone about to go sleep in 10 minutes
kind root
What is it that youre trying to understand
surreal path
Scopes and such
kind root
find the sensors
What do u mean when u say sensors
surreal path
What do u mean when u say sensors
i kept getting peridoic spam
for sensors
idk
found it
kind root
its in SSDTs
kind root
GHID
kind root
Method _HID calls it
surreal path
that exists
kind root
On your screenshot
kind root
INTC10A0
kind root
Is the PNP ID of that sensor
surreal path
aahhhh okay so why is the ec getting events from it
kind root
Because some new values come in probably
thorn bramble
i was basing it off the fact it gets passed _UID which is SEN2, so it should return INTC10A1
surreal path
Because some new values come in probably
what
thorn bramble
but i dt understand it fully so
kind root
You can grep the Linux source for that PNP ID
surreal path
okok
kind root
To see which device it is and what it does
thorn bramble
inb4 not in linux
surreal path
it isnt in linux actually
kind root
i was basing it off the fact it gets passed _UID which is SEN2, so it should ret...
You are correct for that specific screenshot
thorn bramble
ah okay
i thought i was being retarded
lmao
turns out i can still follow simple control flow
kind root
Intel(R) Innovation Platform Framework Manager
kind root
i thought i was being retarded
Lmfao
surreal path
here
kind root
This is some device specific proprietary controller basically
Yeah as u can see its for thermals
surreal path
fair
kind root
When do you get this notify?
kind root
Temperature changes I guess
surreal path
yea
kind root
So it asks to re read
surreal path
anyways before i do anything im gonna add some cpu id instructions
as well
as fixing up the scheduler even more
kind root
@glossy walrus BTW a non integer _UID example
broken depot
<@182819561203040256> BTW a non integer _UID example <:meme:575445094589661215>
"Strong typing? What's that?"
kind root
Well technically its allowed to be either string or integer
But yeah idk why they allow that
IMO its stupid
Clearly the word sensor there is redundant
broken depot
In the grand scheme of things it's not even close to the worst thing in standardsβ’οΈ for modern computers.
kind root
In the grand scheme of things it's not even close to the worst thing in standard...
True
broken depot
How's the scheduler going so far?
surreal path
uhhh so well currently i added a current_thread ptr much like in obos
threads now have states
and well
i have a zombie thread thing
but i didnt impl the reaper thread yet
first
i need to cleanup the queueing evne more
next
i need to impl KCexit_thread()
then
i need to impl the reaper thread
plug that all in into uacpi for schedule work and such
and yea
broken depot
but i didnt impl the reaper thread yet
What I did in BadgerOS is have one thread called "housekeeping" for doing these kinds of simple tasks that have to be deferred to a different thread.
broken depot
I use it for both the reaping of individual threads as well as processes.
silver yarrow
and when the scheduler gets to it, it cleans up the thread
surreal path
so whats the function of the reaper thread
it will check if the current threads in the zombie queues ref count is 0
if it is
silver yarrow
couldnt the scheduler do this
surreal path
clean up resources
silver yarrow
or is it a thread for perf reasons
broken depot
What's the refcount for?
surreal path
perf reasons
broken depot
For the thread handle?
cinder plinth
couldnt the scheduler do this
Yes but this could induce latency
surreal path
What's the refcount for?
for the thread yes
silver yarrow
yeah
broken depot
Couldn't you, in that case, have a custom free function for your refcount ptr and have said function add your thread to a "to-be-reaped" list?
That seems more efficient than checking all threads in the system to me.
silver yarrow
Yes but this could induce latency
ah, ill put moving this into a thread on todo β€οΈ
surreal path
Couldn't you, in that case, have a custom free function for your refcount ptr an...
good point
thanks for that
cinder plinth
ah, ill put moving this into a thread on todo β€οΈ
I'm only speculating but it does make sense
kind root
Yeah freeing large tasks can be expensive
broken depot
BadgerOS' model is slightly different in that threads have a numeric handle, even in the kernel, so I don't have a need for refcount pointers.
And also it has a join/detach model.
Which is what causes thread cleanup in BadgerOS.
cinder plinth
Based object-based model
surreal path
yep
broken depot
BadgerOS uses numeric handles quite a lot, actually.
All of which in sorted arrays so it can be binary-searched.
silver yarrow
BadgerOS uses numeric handles quite a lot, actually.
wdym by this
broken depot
Handles that could be pointers (e.g. a file handle or a thread handle) are numeric instead and the pointers (or structs if they're small enough) are stored in an array, sorted by ID.
desert haven
You use ids rather than pointers, so every attempt to access the object can be checked
desert haven
Handles that could be pointers (e.g. a file handle or a thread handle) are numer...
Oh sorry I didn't see you were tying
broken depot
You basically read my mind :P
desert haven
I'm in your head :p
broken depot
surreal path
You use ids rather than pointers, so every attempt to access the object can be c...
maybe ill do this in nyaux later down the line
broken depot
so how does this replace refcounting
The join/detach model does that.
silver yarrow
mind explaining that one :p
broken depot
In #1153085124959809616, a thread that stops will be put in the "exited" state until a thread_join() is called with its ID. Threads can also be thread_detach()'ed if you're not interested in the event.
The latter case instructs the scheduler to clean up as soon as the thread exits (i.e. immediately add it to the reaping queue).
When a thread enters the reaping queue, it is officially dead and all that remains is the memory allocated to it (stack and thread control block).
Its ID becomes invalid at that time.
desert haven
maybe ill do this in nyaux later down the line
I'd do it sooner rather than later if you intend to, otherwise you'll have a lot of raw pointer accesses to replace
broken depot
Definitely
I used to have raw pointers for things that now use integer handles.
And even with the few occurrences of the thread API actually being used back when the scheduler re-write started, I had to edit dozens of files just to comply with the new API.
surreal path
hi
just realized something
so basically if i am to kill a thread with the reaper
what about the process
will that ALSO be refcounted?
???
and if its 0 free the process
silver yarrow
something along those lines
surreal path
okok thanks
silver yarrow
if the last thread of a process exits, the process is dead
silver yarrow
Its ID becomes invalid at that time.
invalid meaning its removed from the list of threads?
broken depot
Yup
silver yarrow
seems pretty readable from a quick look
one thing I noticed is that u use a lot of atomic fns
surreal path
well
seems like we've encountered a bug
it seems that trying to free the kernel stack
one of the addresses given to invlpg
is not a physical address
so there is a bug
in the vmm
no wait
thats normal right
wait
thorn bramble
invlpg takes a virtual address?
tlb is keyed by the virtual page not the physical page
thorn bramble
you do something wrong != there's ub
thorn bramble
why do u call everything ub lol
thorn bramble
i don't think you know what ub is
surreal path
ub is undefined behavior
thorn bramble
also, you have ubsan meaning most of the dumb ub your code would cause is caught
the fact your ubsan isn't fatal is another thing but oh well
thorn bramble
ubsan isn't perfect
thorn bramble
but you'=d have to royally screw up if it's something ubsan doesnt catch
read: your code poop go figure out whats fucking up ur stack
put a breakpoint at the ret
and check if the return address if correct
surreal path
okok
thorn bramble
maybe ur literally unmapping ur code
and the invlpg runs fine
but then it flushes taht entry out of the tlb
surreal path
well the invlpg does run fine
thorn bramble
and it dies lol? idfk
yeah no shit
but if u flush that page out of the tlb
and its now unmapped
how is the ret gonna run?
what kind of a pf do you get
on that ret
surreal path
non presnet
thorn bramble
and what is that?
surreal path
0x0
surreal path
hold on
surreal path
not a real address
thorn bramble
holding
thorn bramble
good job
i told u
why are u unmapping ur stack lmaop
its in the hhdm
why would u unmap that
plush hearth
i unmapped my stack
How did you manage to unmap your stack?
surreal path
im trying unmap the thread in the reaper queue
thorn bramble
dont touch the hhdm bruh
surreal path
thorn bramble
only unmap memory you mapped first
plush hearth
Yeah, anything in HHDM memory shouldn't get unmapped
surreal path
yes
plush hearth
Do you even allocate the thread a stack?
plush hearth
Or are you using the one limine gives you?
surreal path
Do you even allocate the thread a stack?
yes
i allocate 2 stacks
for reaper thread and the test thread
im trying to exit
thorn bramble
[i can do an assert to check if reaper kernel stack base is the same as the reap...
whjy would a reaper thread reap itself
fix that
surreal path
i will check if it is doing that
plush hearth
i allocate 2 stacks
How do you allocate the stacks?
surreal path
hold on
plush hearth
Why not just with your vma?
thorn bramble
it does use the vmm
plush hearth
It would be faster
surreal path
no @thorn bramble this assertion passes fine
surreal path
thorn bramble
only thing thats unmapped?
surreal path
yes
thorn bramble
0xffff800d4020c000 < 0xffff800d4020ff38 < 0xffff800d40210000
True
surreal path
nvm
thorn bramble
i can see the register dump
thorn bramble
how do u know its a bug
what is it
im assuming u know what it is
if u say its a bug lol
surreal path
i dont know the exact bug but something to do with this if statement i think when creating regions
i could possibly be having overlapping regions
to verify this i will print all regions in the vmm on the pagemap
surreal path
surreal path
align up to page size + page size
whats wrong with it
thorn bramble
idk nothing i guess lol
surreal path
vmm regions here
thorn bramble
just wondering why ur doing that + page size
surreal path
yea do u see anything wrong with this @thorn bramble
thorn bramble
but kvmm_region_alloc looks fine regardless
maybe a little bit suboptimal
but fine regardless
surreal path
okay
thorn bramble
im joking
thorn bramble
anyway the kernel stack being unmapped is the one being used by the reaper thread
like
lol
idk what to tell u
ur looking in the wrong place
figure out why its trying to unmap that stack
oh i see
surreal path
??
thorn bramble
u set kernel_stack_base to stack + stack size
so when u go to free the thread's stack
u actually free the region after
maybe
not sure
but like
thats waht you do with reaper thread stack at least
u set both base and ptr to the end of the stack
kernel_stack_base should be the base of the allocated region
kernel_stack_ptr should be base + size i guess
surreal path
i know how to fix this
thorn bramble
did it work after u fixed that?
surreal path
hold on
thorn bramble
well dont thank me yet then lol
thorn bramble
i wanna go to sleep and dont really feel like cloning and building nyaux
surreal path
oh no
thorn bramble
did it free a kernel stack?
surreal path
i believe so yes
thorn bramble
well i hope it worked lol
ur vague description of the current state of doings makes it hard to tell
but yw i guess
thorn bramble
lfg
surreal path
4 for u
thorn bramble
sleep schedule not found
surreal path
yea U need to sleep
thorn bramble
ahahahhassdhuawshsdjaudqahjwesiikaeqfgjae
surreal path
HAHAHAHAA
thorn bramble
FUNNY
surreal path
HAHAHA
thorn bramble
KILL ME
surreal path
FUNNY
thorn bramble
godo night
surreal path
goodnight
finite summit
good job
surreal path
yesss!
thorn bramble
fixing nyaux one bug at a time
surreal path
flat nymph
theres no brightness control events so im thinking that has to do with something...
It might be through the keyboard?
surreal path
maybe
flat nymph
(maybe I should fix the pmOS keyboard driver)
surreal path
lol u should
now i will abstract the scheduler a bit
clean up thread creating code
like seriously gotta clean that up
also fix the queueing code i forgot
but BEFORE that
ima add cpu id instruction and print out cpu manufactor
flat nymph
I just have a destroy state
My kernel does the same
flat nymph
Couldn't you, in that case, have a custom free function for your refcount ptr an...
Refcounts are obnoxious
Why have them in threads, when they're either alive or not
flat nymph
BadgerOS uses numeric handles quite a lot, actually.
I also have that
flat nymph
Yeah, anything in HHDM memory shouldn't get unmapped
You can easily avoid HHDM mapping issues by not using HHDM in your kernel
finite summit
Refcounts are obnoxious
is this sarcastic?
flat nymph
You can easily avoid HHDM mapping issues by not using HHDM in your kernel <:trl:...
This is half-sarcastic, but I have to tell everyone that I'm not using HHDM
surreal path
cringe
flat nymph
Refcounts are obnoxious
With threads you have a lot of questions, like what does hold a refcount and what doesn't
Like you have a queue of ready threads, do you hold refcount for that, or maybe you use something similar to std::shared_ptr and use that for linked lists, but then it becomes obnoxious, and what if a thread dies and someone holds a reference to it, etc.
silver yarrow
You can easily avoid HHDM mapping issues by not using HHDM in your kernel <:trl:...
I dont use it anywhere but the VMM
so I can swap out the impl relatively easily
surreal path
guess whos
fucking lpaotp
laptop* broke?
mine!!!
okay so i was trying nobara some linux distro
wow i just crashed
okay
so bascially i ran it right, then it froze and then the screen got all glitched
and then the display was off
and it wouldnt come back
idk what linux did maybe it messed with the nvidia GSP firmware in a way that bricked it
to be fair this nobara distro claimed to have patches around nvidia
so i suspect they might have something to do with this
that or my laptop croaked out from too much osdev
anyways im using the family computer rn. my dad said when he wakes up he will take a look at it
or well
take a look at the fact that i even opened the laptop and disconnected its battery and cmos to which it did NOT fix it
am i even gonna get a new laptop
it cost fucking 2000 euro
i dont think he will buy a new one
he could try to get it repaired
but it would take weeks
and i already used the warnenity
im fucked
im royally fucked rn chat
surreal path
idk what linux did maybe it messed with the nvidia GSP firmware in a way that br...
how do u fucking
mess up a driver this bad
to the point of bricking the GPU
π
haughty kite
