Google hpet spec
#Nyaux
kind root
surreal path
https://www.intel.com/content/dam/www/public/us/en/documents/technical-specifica...
the font is so ugly
thorn bramble
you are interested in the main counter register
surreal path
my eyes
surreal path
you are interested in the main counter register
okay
thats offset addr + 0x0f0
ok
acpi_gas?
thorn bramble
yes, hpet can technically be used over port io
surreal path
what
surreal path
oh
thorn bramble
that tells you how to address it
it contains the address_space_id
which you should probably assert that its memory mapped io
because fuck you if your hpet isnt memory mapped
surreal path
it contains the address_space_id
what?
thorn bramble
GOOD
surreal path
yea i see its a structure but what is address_space_id supposed to represent
thorn bramble
acpi_hpet *hpet = table;
kassert(hpet->address.address_space_id == ACPI_AS_ID_SYS_MEM);
uintptr_t hpet_base_addr = hpet->address.address;```the address space
in which the hpet "operates"
i guess
it can be port io
memory mapped io
pci config space
surreal path
the address space
oh so its like an ID that tells you what type of address space it is
oh get u
thorn bramble
if you read the code you would have realized
lol
there is literally defines right above that tell you all that
thorn bramble
its ok as long as u learn something along the way
thorn bramble
dont forget it should be a volatile uint64_t read
also you should try and print the hpet address, make sure it's present and valid
surreal path
dont forget it should be a volatile uint64_t read
yea i did
surreal path
also you should try and print the hpet address, make sure it's present and valid
i did check if the hpet exists
and checked if it is memory mapped io or not
HPET BASE IN PHYS: FED00000
HPET MAIN COUNTER RN: 0
uint64_t hpet_base = hpet->address.address;
volatile uint64_t poll = *(volatile uint64_t*)(hpet_base + hhdm_request.response->offset + 0x0f0);
serial_print("HPET MAIN COUNTER RN: ");
serial_print(itoa(wow, poll));
serial_print("\n");
asm("cli"); // disable interrupts
for (;;)
{
asm ("hlt");
}
this is what i do
is the main counter register not ticking or somthin?
??
rigid fable
as other people said already, why don't you implement printf?
and then use some trick to redirect either to terminal or "serial" at will
like a debug define ifdef
rigid fable
as other people said already, why don't you implement printf?
and even if you want to have separate functions, just do this instead of having to call itoa every time
and if you pass the base to itoa you can easily print hex, dec and binary or whatever you want
surreal path
its not that big of a deal to me even tho i have nanoprintf and i could just use it, i just prefer using itoa like a cursed manic
kind root
```C
uint64_t hpet_base = hpet->address.address;
volatile uint64_t poll = *(vola...
Have you verified that the counter is 64 bits
Read the spec or osdev
surreal path
spec says the register is 64 bits
kind root
Ye
rigid fable
```C
uint64_t hpet_base = hpet->address.address;
volatile uint64_t poll = *(vola...
cannot he just use cli hlt together instead of having hlt in the loop? Correct me if I am wrong
thorn bramble
omg do not halt in this loop
surreal path
it can be a 32 bit hpet though
ok ill check
rigid fable
eh
thorn bramble
right he just didnt want it to run out of this function
nvm
i thought it was supposed to be the busy loop for calibration
kind root
No
surreal path
oh
kind root
Read the wiki page
surreal path
Read the wiki page
wiki page just shows the structure
doesnt tell me what register bit width is used for
kind root
Scroll ffs lol
surreal path
oh
kind root
There's also a config register
surreal path
yea i see
kind root
That u must check
surreal path
yea 64 bit
and i see that the config register PROBS has the main counter hlt'd rn
i will check
yea hpet counter is not running
its halted
i will enable and see
WOWIE!!!
thats a timer!!!
it goes VERY fast btw
the main counter
lol
kind root
Yes
surreal path
okay so i have the START of the hpet counter
set the lapic timer to divide by 2
set the inital count to 0xffffffff
do i just do a while loop until lapic timer counter is 0?
surreal path
okay how do i do that
thorn bramble
use the hpet to figure out if the time has already passed or not
surreal path
use the hpet to figure out if the time has already passed or not
if the timer has already passed?
thorn bramble
lets say you wait for 10ms
make the hpet wait for 10ms
then you know how many lapic timer ticks are in 10ms
lol
surreal path
how do u make the hpet wait for 10 ms 😭
kind root
how do u make the hpet wait for 10 ms 😭
Bro
Think
You have the counter
You have the amount of time it takes to increase by 1
thorn bramble
the capabilities register tells you how much time passed every time it ticks
surreal path
okay
kind root
Wtf
thorn bramble
bit math
lmao
i dont think that works
even if it did it would be wrong
shift the register value right by 32 and then mask it
surreal path
shift the register value right by 32 and then mask it
thats what im doing
thorn bramble
no LMAO
thorn bramble
no
thorn bramble
yes that is what i said
surreal path
kk
surreal path
TIME PER TICK: 10000000
thorn bramble
10ns
surreal path
yep
thorn bramble
you can use that to measure time now lol
surreal path
now im gonna activate the two brain cells in my brain to figure out how to use this 
thorn bramble
figure out how much you want to wait
surreal path
probs 10ms
thorn bramble
TIME PER TICK: 10000000
convert this to a common unit that's big enough
so you can convert both to nanoseconds
1ms is 10000000ns
10 * that
surreal path
okay
thorn bramble
loop while (current hpet - start hpet) * period_in_ns < 10 * that
lol
congratulations your kernel is now 10ms slower
kind root
Tbh if we keep spoonfeeding him like that its not gonna get him very far
thorn bramble
true
surreal path
Tbh if we keep spoonfeeding him like that its not gonna get him very far
yea dont spoonfeed im trying to understand
let my two brain cells activate
thorn bramble
ok i mute thread i go try and understand vmem.pdf
surreal path
surreal path
ok i mute thread i go try and understand vmem.pdf
kk
kind root
And jm gonna sleep
surreal path
And jm gonna sleep
alr gn
thorn bramble
good night
kind root
Thank you
surreal path
npp
i figured out how to do it without any help
now my kernel can successfully wait 10ms
😎
thorn bramble
you're waiting 100ms
surreal path
you're waiting 100ms
what???
thorn bramble
10*10000000
surreal path
oh
thorn bramble
10(ms)*10000000(10ms in ns)
surreal path
☠️
thorn bramble
one zero too much
thorn bramble
you might want to put a pause in that loop body
surreal path
you might want to put a pause in that loop body
alr
thorn bramble
or actually you might not
might mess with the calibration process on real hw
idk how the lapic timer frequency changes with power state changes
surreal path
or actually you might not
putting a pause causes the lapic to send an interrupt sometimes
which is VERY weird
thorn bramble
what lol
thorn bramble
you probably didn't mask it
thorn bramble
you should use 4 or 8
surreal path
ok sure
thorn bramble
also mask is bit 16
not 15
bit 15 is reserved
bits 18..17 are the timer mode (one shot, periodic, tsc deadline)
surreal path
also mask is bit 16
yea i fixed
how did u know that was the issue
is bro in my pc
thorn bramble
you sent the code earlier
tawdry mirage
you had it set to 1???
why is divisor 1 too fast?
thorn bramble
honestly idk, it just feels like it 
i love spreading misinformation
but clearly it timed out sometimes, so it is too low
thorn bramble
how would any of us know
thorn bramble
try setting up a periodic timer with that interval * 10 or something
see if you get an interrupt every 100ms
surreal path
alr sure
thorn bramble
its pretty easy to eyeball 100ms imho
surreal path
YEA IT FEELS LIKE 100ms!
!!!!
like it feels RIGHT on that dot
wowie!
lets tick every uhh
1 second
surreal path
YEA IT FEELS LIKE 100ms!
no it feels like every second
i actually dont know
i need a stopwatch lmao
yea its doing every second
bruh
oh i calibrated wrong
okay i THINK thats 100ms now
okay it is
cool
ok what do i do now that i have a timer
and a "functioning" apic driver
ig
tawdry mirage
ok what do i do now that i have a timer <:trl:829316641191886888>
time things
surreal path
time things
like what time things
tawdry mirage
i meant time as a verb but that works too
surreal path
i meant time as a verb but that works too
will i hook this up to my scheduler
wouldnt that be a funny!
tawdry mirage
well yeah you probably want to use the timer to drive the scheduler
surreal path
okay
ill make a kernel thread so i can continue executing whatever kernel init code i still need to do
or smthin
wait what else do i need again
i have an interrupt controller now
APIC "driver" (i dont even do anything with the ioapic atm lol)
gdt done, idt done, pmm done, vmm done, acpi tables parsed, not starting up other cpus as im not doing smp FOR THE MOMENT, lapic timer setup, i have a scheduler
a vfs now?
like rn???
have i been doing that much already?
i can go onto a vfs rn????????
what do yall think i should do???
cause i dont know where to go next?
thorn bramble
make a printf
surreal path
bruh
surreal path
think thats correct?
wait i have an idea
im gonna make a sleep function using the hpet
also gonna implement some functions for uacpi ig
difference between stall and sleep is just the format ig?
tawdry mirage
stall is supposed to be tighter ig
thorn bramble
sleeping might involve rescheduling? idk
tawdry mirage
as in, won't potentially take longer
void uacpi_kernel_stall(uacpi_u8 usec) {
auto now = systemClockSource()->currentNanos();
auto deadline = now + usec * 1000;
while (systemClockSource()->currentNanos() < deadline);
}
void uacpi_kernel_sleep(uacpi_u64 msec) {
KernelFiber::asyncBlockCurrent(
generalTimerEngine()->sleepFor(msec * 1000 * 1000)
);
}
or more like it lets you reschedule
it might be a long lasting sleep
like 30 seconds
you dont want to be stuck doing nothing when that happens
surreal path
i mean
my lapic timer will be ticking every 10ms
no matter what so
it wont be STUck there
but i mean wasted cpu cycles so early reschedule ig?
thorn bramble
something like that i guess
surreal path
okay made some ksleep function for uacpi and friends
that uses the main counter of the hpet
to like sleep
kind root
difference between stall and sleep is just the format ig?
stall -> busy spin for very few microseconds (up to 255)
sleep -> actually go to sleep, can be up to a second
surreal path
stall -> busy spin for very few microseconds (up to 255)
how do i do that 
spin for very few microseconds
surreal path
Spin for as little as possible
yea but how tf do i do to spin for a few micro seconds, run a bunch of asm pause instructions in a loop for a bit or smthin?
kind root
yea but how <:nooo:750466870993223750> tf do i do to spin for a few micro second...
Uhh wut
U have the hpet counter?
surreal path
U have the hpet counter?
i use the hpet counter as my ksleep
kind root
Yup
surreal path
Yup
which means checking if the thread has reached its amount of time in ms, assuming a quantum of 10ms (which is what i do), every time the scheduler ticks and switches to the next thread. if the next thread has reached its amount of time its unblocked
correct?
kind root
Yes
surreal path
okay simple then, just store a structure about the thread state. if its blocked the ms its blocked for will be in the structure and stuff
surreal path
Yes
something like
struct thread_state_t
{
int state; // 0 - ready, 1 - blocked
int ms_if_blocked; // ms its blocked for if blocked
}
is this correct?
desert haven
you may want to store the time to wake the thread
rather than than how long its going to sleep for, because then you'd also need to know when it went to sleep
surreal path
you may want to store the time to wake the thread
okay so store the time to wake the thread
in what unit
??
desert haven
your kernel your units
milliseconds and nanoseconds are common
or use a type to represent time in some unit-less way
surreal path
your kernel your units
i guess ill store in ns, and use the hpet counter to check if its like.
able to be unblocked or not
correct r4?
desert haven
there's no single correct way to do this, but this is one way that will work 🙂
surreal path
i just wanna make sure im doing it in like a way that isnt stupid 😭
desert haven
haha yeah fair, nah this sounds fine
surreal path
i guess ill store in ns, and use the hpet counter to check if its like.
time in ns rn + ns to sleep for = ns to wake thread
will use hpet counter (again) to get the current time in ns, im just gonna use hpet counter for everything 😎
man i love the hpet
so easy
didnt even have to touch the yucky pic
surreal path
time in ns rn + ns to sleep for = ns to wake thread
again just making sure, would be a correct way to do this?
?
surreal path
im unsure if thats correct, ALSO i wanna like make a ps2 keyboard driver or smthin. im guessing i need to read the madt for the ISOs. and then configure the ioapic
surreal path
yes
ok, will implment this into my scheduler later when i need this
wanna make a ps2 keyboard driver of some kind
i believe irq 1 is the ps2 keyboard
should be gsi 1
cause i think ioapic has the legacy irqs identity mapped unless otherwise specified in a interrupt source override?
guessing im gonna need to keep track of the irqs?
like a table
or smthin?
how should i do this?
shut laurel
Like you want to
surreal path
Like you want to
i can design that however i want?
surreal path
surreal path
<:trolley:690075871322964017>
we have printf
😱
GUYS THIS IS A BIG ACHIEVEMENT
WOWIE !
PRINTF????
NO MORE CURSED ITOA??
isnt that crazy !
now yall cant mock this os anymore for not having printf cause we finally have a kprintf 😤
tawdry mirage
i assume you're using nanoprintf for it?
surreal path
i assume you're using nanoprintf for it?
yes
tawdry mirage
good, i don't want to think what a hand-written nyaux printf would look like 
surreal path
good, i don't want to think what a hand-written nyaux printf would look like <:h...
what r u implying 
surreal path
uhhh what else
uhhhhhh
o yeah i wanted to start ACTUALLY setting up the ioapic
which is lit not rlly much all i rlly need to do is create a kind of global struct that shows which irq goes to which gsi, initing using the interrupt source overrides, then whenever i want to install a irq handler for that i just lookup this struct table thingy to get the gsi and shit
i think thats how im gonna design it
ionic jetty
You could just directly read the ISO too
I dont think it’s necessary to have a seperate table for it
surreal path
its fineee
surreal path
You could just directly read the ISO too
dont wanna scan the madt for it each time if theres an iso for an irq each time i want to install a handler
surreal path
Just keep a pointer to the iso
hmm okay, so like an array of iso pointers
ionic jetty
Yeah
You can make the table too for sure it just seems unnecessary
Whats the general outline of your todo list atm
What will be the next things you implement
(Btw, I think you have the most comments on any thread 😂)
surreal path
😭
surreal path
What will be the next things you implement
i dont really know.... my kinda first end goal for the os is to eventually have a unix like userland with doom ported. i dont know where to go im just kinda following mints guide in #curated-resources
ionic jetty
Thats a good one yeah
surreal path
they said vfs next
so maybe after the ps2 kbd driver i will learn what the fuck a vfs is and how to implement one
unless yall got better suggestions on what to do next
ionic jetty
You cannot be calling your function poopXY
surreal path
You cannot be calling your function poopXY<:angr:991436925116350494>
im GUESSING that after i init everything, there is no need for a kthread at all or smthin. we just start scheduling and executing processes and then when we syscall thats the only time we go back to the kernel (or if its from an interrupt like a irq handler or smthin)
unkempt relic
im GUESSING that after i init everything, there is no need for a kthread at all ...
this is broadly true
you can see the kernel as being like a library that coexists in every process
and the usual pattern on most kernels is that most threads are in userspace and only make occasional excursions into the kernel
interrupts make a slight mess of this analogy but still
surreal path
and the usual pattern on most kernels is that most threads are in userspace and ...
so usually threads are in userspace
unkempt relic
so usually threads are in userspace
except for kernel worker threads, those appear in most kernels too
but that's the general principle usually
surreal path
except for kernel worker threads, those appear in most kernels too
kernel worker threads? what would those be used for
unkempt relic
kernel worker threads? what would those be used for
page replacement, driver duties, that sort of thing
surreal path
page replacement, driver duties, that sort of thing
ah okay
been fixing some of my scheduler code
now happening via the lapic timer
😎
each thread has a quantum of 10ms cause yea
works rlly well
currently my scheduler is just some simple round robin thing
each thread holds a pointer to the next, if theres no next go back to start of queue
that sort of thing
currently my scheduler treats everything as a separate thread
which was okay but i probs need to change that now for user mode soon
or actually
not
i should treat everything as separate threads
if for some use case i wanted to find what threads belonged to which "processes", i can simply store a process id in each thread
anyways im actually gonna make a proper kpanic
to make debugging easier
print out every register and everything instead of just printing the rip
then ill work on the ps2 driver
surreal path
😎
thats a lot of gsis
is that common?
for an ioapic to handle that many gsis or smth
i think so
surreal path
not really
im guessing this is smthin with qemu?
tawdry mirage
run info ioapic
surreal path
kk
tawdry mirage
er sorry, info pic
tawdry mirage
how are you getting the number of pins
tawdry mirage
what does | 0xff do
surreal path
surreal path
what does `| 0xff` do
OH
☠️
WHOOPS
should've been and
okay cool 😎
one thing im rlly getting annoyed at is how long it takes for the pmm to init
yea i get it, im storing a freelist node into every fkn page
thats gonna take cycles
but comeon
it takes a whole 3 seconds
3-6
for a 3G
idk how to make this faster
not rlly much i can do
cinder plinth
here's what I do
works just fine
I do a freelist on a per-region basis
instead of every page
surreal path
I do a freelist on a per-region basis
yea i guess thats why it'd be faster but at that point its more bitmap then freelist
surreal path
pmm is integrated too deep with the kernel for me to change it, so if i want to do a pmm region freelist kinda thing, i cant rlly cause it'd take too long to refactor everything
but i can see how that would be better overrall
wish i had did that 😔
cinder plinth
tawdry mirage
does add_region merge adjecent free regions or will you eventually end up with only 1 page long regions
cinder plinth
not from a functionality standpoint
actually not even from a performance standpoint
doesnt matter at all
just means the head switches more often
surreal path
okay well it does at least solve the boot time problem, maybe once i finish this ioapic stuff and before i do a vfs ill refactor my pmm to be like this
cause i really hate waiting 3 seconds everytime i boot a nyaux vm under qemu
surreal path
yea
cinder plinth
should be faster on kvm
tawdry mirage
cause i really hate waiting 3 seconds everytime i boot a nyaux vm under qemu
laughs in managarm
tawdry mirage
laughs in managarm <:meme:575445094589661215>
getting to weston takes 17s in qemu on my laptop
surreal path
should be faster on kvm
its rlly fast until the loading pmm part which is the slow as balls part
everything else is speed tho
surreal path
finite summit
I've accidentally wiped my sda drive while trying to dd too many times
surreal path
ok gonna try booting it
brb
Uh oh
It’s hanging at vmm init
this is why I needed to make a kpanic function
I’m gonna try and get nyaux to run real hardware
most likely hanged somewhere here
probs page faulted cause
my display is 1440p
and the background is a 1080p file
yea thats most likely the reason
i need to somehow "resize it"
to fit whatever framebuffer
i have no idea how to do that lmao
kind root
You cant
kind root
Write a graphics driver
tawdry mirage
just don't set a background image 4head
surreal path
but it looks nice !
i need a splash image!!!
!!!!!
what if i just made a really big res image
would there be a way to downsize it
or no
tawdry mirage
also i'm pretty sure flanterm will just pad the image with black on all the sides if it's too small?
surreal path
also i'm pretty sure flanterm will just pad the image with black on all the side...
no
it doesnt
sadly
wish it did
kind root
would there be a way to downsize it
I think you're wasting your time
surreal path
I think you're wasting your time
wdym
kind root
On useless stuff
surreal path
yea fair
lemme at least see if nyaux can run on real hardware
tho
at the very least
ill disable splash images
wicked loom
also i'm pretty sure flanterm will just pad the image with black on all the side...
nah, you have to do it, the canvas is expected to be = width*height
tawdry mirage
ah my bad
kind root
nah, you have to do it, the canvas is expected to be = width*height
Shouldn't be that hard to extend it to support that probably? Sounds like a convenient feature
surreal path
Shouldn't be that hard to extend it to support that probably? Sounds like a conv...
^^
wicked loom
Shouldn't be that hard to extend it to support that probably? Sounds like a conv...
yeah i mean, it's not hard to automatically pad, but for that it'd need to know the actual size of the image, and now it only knows of a raw canvas
the user is kind of expected to do all paddings, stretching, shrinking, whatever, on the canvas before passing it to flanterm
if you want fancy stuff like this you're probably going to have better luck using ilobilo's port
of the Limine logic that does this, to be freestanding
surreal path
i think i just dd'd my drive....
finite summit
I told ya
finite summit
you at least had the code committed and pushed, right?
finite summit
now it's time for you to reinstall gentoo
surreal path
surreal path
no wait i got it working
i didnt dd i just
messed up my boot partition
fixed now
😎
anyways nyaux page faults here on real hardware for some reason?
that is so weird
im gonna make a stack trace
thing
for debugging
so i can figure this out more
cant gdb step real hardware ☠️
tawdry mirage
if you have special equipment you can debug real hardware :^)
surreal path
now we have a nice stack trace
surreal path
if you have special equipment you can debug real hardware :^)
no i got this 😎
all i need is me and my stack trace
😎 😎
surreal path
now we have a nice stack trace
wait no
this isnt a pro gamer address
wrong image
uhhhh oh right i know why
no still
my brain
💥
is gone
i dont think thats a valid rbp
to be fair
or maybe it is
tawdry mirage
did you pass -fno-omit-frame-pointer to the compiler
surreal path
did you pass -fno-omit-frame-pointer to the compiler
yep
Ok got the stack trace to work
Gonna debug this now
So I can figure out why it’s doing that on real hardware
okay so line 1040 its faulting
at this line
why hmmmm
surreal path
at this line
(when converting a string arg)
surreal path
seems like hardware doesnt play well with uhhh printing to serial cause anytime i do a printf using serial it page faults?
if i printf with flanterm ctx
it works fine
on real hardware
(i think)
ill test that rq
Very interesting
why does my pc not like this thing
it hates it for some reason
it page faults with this macro enabled
if i dont have the macro enabled some of the printfs look weird
this thing off
vs on
lemme see if the real hardware boots with it off at leasttt
doesnt boot even with it off
still faulting
faulting here
when it calls uacpi_error
its doing it whenever copying the string, most likely unknown_path is something funky
whatever uacpi has set absolute path to
its doing smthin funky
i believe this is more an issue with uacpi itself
probs doesnt work well with my hardware (this laptop is pretty difficult to support, even for linux)
surreal path
@kind root might wanna check this out
if someone can try and run nyaux os on their hardware
that would be greatly appreciated
desert haven
im just going to say that its unlikely uacpi or nanoprintf that are the real cause of the bugs here
for npf you at least only have so many inputs to worry about
desert haven
<:thonk:575445129863888896>
more context would be useful here, particular whats the format and what are you passing to this function.
kind root
i believe this is more an issue with uacpi itself
Nah it's not a uACPI issue
Run managarm on that hardware
If it boots its your skill issue
kind root
probs doesnt work well with my hardware (this laptop is pretty difficult to supp...
There's no such thing
surreal path
more context would be useful here, particular whats the format and what are you ...
I’m not passing anything I’m just starting uacpi
surreal path
Run managarm on that hardware
I’ll try
surreal path
more context would be useful here, particular whats the format and what are you ...
I can’t give too much context cause it’s only reproducible on hardware
That’s different to debug
surreal path
im just going to say that its unlikely uacpi or nanoprintf that are the real cau...
Then this don’t make any sense
Tf is going on
kind root
who knows, a lot of things about your kernel could be broken
surreal path
it’s only happening on real hardware and ONLY whenever uacpi tries to print or smth
kind root
and
surreal path
I’ll disable tracing and see if it solves
kind root
this just means the bug is not reproducable on the simple blob that qemu provides
dump the blob with acpidump -b and feed to the ./test_runner <blob>
or just run managarm
surreal path
Alr
surreal path
downloading
this my dsdt.dat (if u can feed it into test runner on ur system that would be quite epic gamer)
im gonna reboot into managarm and test
kind root
this my dsdt.dat (if u can feed it into test runner on ur system that would be q...
works in test runner
surreal path
works in test runner
is there a way to feed a dsdt to qemu
surreal path
ill try feeding it to nyaux
surreal path
`-acpitable file=...`
path to dsdt?
kind root
yeah
surreal path
kk
surreal path
what difference does it make?
kind root
DSDT is only one and is pointed to by FADT
SSDTs are infinite amount
also it wont necessarily work and might crash qemu because it does arbitrary MMIO
that is not present in qemu chipset
surreal path
oh
kind root
it has a hex header
surreal path
it has a hex header
i renamed it in the hex editor but its still too big
surreal path
how tf am i supposed to actually debug this then
surreal path
how tf am i supposed to actually debug this then <:nooo:750466870993223750>
i cant just keep reflashing nyaux onto the usb and keep rebooting and partitioning it takes too long
surreal path
yea but i have to reflash the usb everytime with dd and reset the partition
and it takes forever to reboot
surreal path
im gonna dd managarm's image.xz and try it
kind root
wtf
surreal path
fkn hell
surreal path
i think it checks for number of make jobs * 2 GB of ram
even trying to get the binary version
gentoo crys
surreal path
yea genbyebye fuck this im switching distros
tawdry mirage
i think thats what it was called
kind root
yea genbyebye fuck this im switching distros
i know a good distro
its called
windows 11
surreal path
i think i will switch tooooooo arch or smthin
surreal path
who knows, a lot of things about your kernel could be broken
why is it always my kernel
wicked loom
just use arch lmao
surreal path
just use arch lmao
yea i switched to arch
fuck gentoo
all my homies hate gentoo
😤
wicked loom
fr ong
surreal path
!!!!
😤 😤
anyways gonna boot into managarm and see how it fairs with my hardware, i just hope it also page faults so i know its an issue with my hardware and NOT with the kernel 😤
Managarm panics on my hardware too
It’s 10000% a hardware issue
To verify if this is correct can someone test nyaux on their hardware
If no blue screen it’s def this laptop being so weird
anyone got the time to do so?
please??
i have no other device on my other then this laptop 😭
thorn bramble
this is in userspace so well after uacpi init
surreal path
this is in userspace so well after uacpi init
yea a ps2 driver
also i dont think managarm has uacpi tracing everything
surreal path
yea and i have mine to tracing everything
thorn bramble
disable trace logs
surreal path
so thats most likely the issue
surreal path
disable trace logs
yea
thorn bramble
it could be your allocator being shit, you running out of stack, you not mapping things correctly
many things to consider
surreal path
it could be your allocator being shit, you running out of stack, you not mapping...
it probs isnt the allocator cause it works on the vm but not on real hardware
- already had my oberrow hell
thorn bramble
the real hardware has a much more advanced aml to parse
than the one in qemu
parse + execute, sorry
surreal path
yea but heres the thing
in this stack trace
one second
im gonna get addr2line
just lemme fix my compiler (cause this is a new system)
thorn bramble
where is the cr2
thorn bramble
lol
surreal path
lol
thorn bramble
lmao
surreal path
just lemme fix my compiler (cause this is a new system)
still im gonna really quick setup my cross compiler
surreal path
im gonna try 2 things
im gonna disable uacpi tracing full
and im gonna
print out the cr2 in the page fault
if that still happens
kind root
It’s 10000% a hardware issue
Nope, its your kernel issue, as you can see managarm boots fine
surreal path
Nope, its your kernel issue, as you can see managarm boots fine
bruh
kind root
This is just some userspace driver bug
thorn bramble
also your dsdt works fine in the userspace runner
kind root
Both userspace and real hw apparently
surreal path
if this is another allocator issue i swear
at least be an issue with printf or smthin simple and not my allocator
kind root
Could be anything
surreal path
okay my page fault screen prints cr2 now
how do i make the stack trace print the line number and c file name of the faulting instruction
i know some stack traces do that
would make it 1 billion times easier to debug real hardware
kind root
Write a dwarf parser
surreal path
dwarf parser?
kind root
Elf debug info format
surreal path
is that difficult to implement, a dwarf parser?
or is it worth using a freestanding library for that
?
kind root
is that difficult to implement, a dwarf parser?
Yes
surreal path
okay this is difficult to get working
this damn libdwarf shi
ugh all i wanna do is just get the functions names from the instruction pointer, like addr2line does
rigid fable
unless that is also part of the dwalf parser, add a sinple disassembler library to have a disassembly on the panic screen
and write out all the general purpose registers
rigid fable
here you probably made two screnshots one afyer eachother, since you can still see the transation effect
surreal path
im doing something very cursed but i have an idea thanks to this perl script i found
in #resources
well not rlly surprised
time to keep working on that linker script !!!! even tho i never write linker scripts and have zero idea what im doing !!!!
hum
now that doesnt look very epic gamer
oooo
looking fresh
LOOKOKKKK
OH MY GOD
THIS IS BIG BRAIN
kind root
are you using libdwarf or?
kind root
wtf
surreal path
kind root
this won't tell u the line number
surreal path
i no need
kind root
only the function and byte offset
surreal path
im too dumb to get libdwarf working anyway
surreal path
why tf did u make a special phdr for symbols
idk
thorn bramble
use the rodata phdr
surreal path
i removed it anyway after figuring out that was dumb
surreal path
use the rodata phdr
yea im using rodata now
thorn bramble
good
kind root
bro never heard of binary search
surreal path
i havent
the problem is that the rip is pointing the the line number not the beginning of the function
finite summit
it's because you spam it
finite summit
pretty sure there's 1000 messages per one thing you implement
surreal path
yea hey at least i implement things !
at least im understanding what im doing, instead of not understanding and just copying code or smthin like i used to do back in 2020 when i was stupid
surreal path
at least im understanding what im doing, instead of not understanding and just c...
i cant get the debugging symbols to work
im not doing this, this is a waste of time
for the time i've been spending trying to get debugging symbols
i probs would have just fixed the issue happening
💀
and what i was doing was rlly cursed and dumb
boohoo i have to restart
so what
okay first part of stack trace is this line
second is this?
uh huh
here
next part
very weird
surreal path
its always faulting at this instruction
surreal path
when trying to put the absolute path thing into the string
cr2 is 0 when this happens
okay so absolute path is never init'd to something so its not pointing to anywhere
when uacpi_error tries to print out absolute_path
it faults because absolute_path doesnt point to anywhere at all
and this is VERY weird cause clearly <unknown> is there
surreal path
okay so absolute path is never init'd to something so its not pointing to anywhe...
@kind root what do u think is going on here, why is absolute path pointing to nothing. i know nothing about aml interrupters so probs best to ask u if you can know whats happening
kind root
this is unrelated to aml or uacpi
its a bug in your code
corrupting something into oblivion
surreal path
with the memory is that what ur saying
kind root
yeah, even the fact that this gets called
surreal path
okay
kind root
could be your kernel_api impl fucking something up
surreal path
could be your kernel_api impl fucking something up
im reading it
kind root
<@353593948523986945> what do u think is going on here, why is absolute path poi...
like seriously, you clearly see it gets initialized in every branch
how would this be affected by anything
surreal path
how would this be affected by anything
i dont know, its difficult for me to think rn im tired. im not sure how i can debug this since this is an issue only happening with large aml parsing and since the qemu blob is tiny i cant rlly do much
is there any emulators that have large blobs or anything for uacpi to parse or is their absolutely NOTHING i can do
?
surreal path
i actually dont know what to fucking do anymore, i stepped through vbox trying to figure it out but i just couldnt
why is my allocator always the problem, why is my code always shitty why cant i just write good fucking code for once in my life
wicked loom
It’s 10000% a hardware issue
try vinix
finite summit
why is my allocator always the problem, why is my code always shitty why cant i ...
Just become a contributor to OBOS
You can write the code that's simple
while I do all the work
surreal path
It’s my fucking allocator as always
my allocator isn’t even that complex yet it ALWAYS has bugs
I just
wicked loom
it's because you spam it
says YOU, of all people
kind root
^
surreal path
replace it with a bump
I don’t know what a bump is but I guess I’ll look into it
thorn bramble
alloc(size):
result = base
base += size
return result
wicked loom
yeah it's stupid simple
thorn bramble
align size to 8 beforehand tho or smth
so you dont hand out unaligned pointers
free doesnt exist
or well, if ptr + size == base then base = ptr, else you cant free
but there is no point in implementing free for bump
finite summit
says YOU, of all people
His thread was posted 16 days ago, and has nearly 7000 messages
Mine was posted around august of 2023 I believe
and has 6300 messages
thorn bramble
hes just really invested
finite summit
I've also put progress in that thread for my kernel over all rewrites but the first
surreal path
hes just really invested <:halfmemeright:715831702164996096>
I spend my entire day everyday on my OS, every hour. When I come back from school I instantly open vscode and get working. I really enjoy osdev despite how frustrating it can be sometimes
It makes me really happy seeing me do things like creating a kernel and seeing it actually work
brings me a lot of joy
I don’t rlly have a life outside of school lmao
I hope people can look over this thread one day and perhaps learn from my mistakes and from all the great advice everyone here has given to make nyaux actually possible
for their os
wicked loom
the grinder
surreal path
Yep! 😎💪🔥
worldly condor