which line pitust what do i change to what
#Nyaux
edgy pilot
which line pitust what do i change to what
I don't think that's the issue
haughty kite
idk im not the expert
haughty kite
I don't think that's the issue
it is the issue
edgy pilot
it is the issue
an infinite loop leads to a page fault?
surreal path
oh nvm
kind root
just tail call optimize it
haughty kite
its mutual recursion
edgy pilot
what do those functions do
haughty kite
what do those functions do
some multiplication shit
kind root
wtf
haughty kite
oh wait nvm
its my bug
ok sorry
its still kinda broken
but its not that bad
ooops
surreal path
its my bug
what
surreal path
oh
surreal path
i nuked my cc-runtime
haughty kite
still an LLVM bug tho
surreal path
💀
surreal path
still an LLVM bug tho
fun
edgy pilot
still an LLVM bug tho
time for a bug report then
edgy pilot
dealing with mints shitty makefile
surreal path
huh
haughty kite
yeah
edgy pilot
and what is the bug exactly u keep going back and fourth
check your slab free
maybe you free an incorrect thing, allocate it again and overwrite some data
haughty kite
ok so i dont want to say anythig but im finding it really hard to blame nyaux here
@kind root this doesn't seem like correct decomp for item_array_clear
(this matches disassembly)
at -O0 btw
ah hmm wait
okay nevermind this is probably not a uacpi bug
edgy pilot
<@353593948523986945> this doesn't seem like correct decomp for item_array_clear
it's technically just a hint
haughty kite
"hint" fucking what
edgy pilot
another joke
haughty kite
yeah idk why its no initialized
probably his skill issue on misusing UACPI_SIZED_FREES
can you mangle symbols so that this can't happen?
surreal path
how
haughty kite
can you mangle symbols so that this can't happen?
like use two different symbol names or something
edgy pilot
gcc -> g++ 
haughty kite
should probably
yeah
surreal path
you gotta define UACPI_SIZED_FREES for all translation units
sorry translation units?
edgy pilot
every file
edgy pilot
add -DUACPI_SIZED_FREES to cflags
haughty kite
that fix makes it get to this
surreal path
RAHHH
kind root
nyaux cmon we went over this
surreal path
that fix makes it get to this
476ms its faster now
haughty kite
probably because of some shit i added to make it more likely to catch issues :^)
edgy pilot
that fix makes it get to this
that doesn't look like the latest uacpi commit
elder shoal
edgy pilot
lmao
elder shoal
476 ms faster than astral
surreal path
haughty kite
476 ms faster than astral
faster or slower?
edgy pilot
476 ms faster than astral
slower?
surreal path
kill urself gcc
i cannot figure this out chat
haughty kite
kill urself gcc
you didnt include it
surreal path
uhhh i did
elder shoal
faster or slower?
a secret third thing
edgy pilot
is limine.h included here
surreal path
yes
elder shoal
send all of elf.h
surreal path
#include <limine.h>
#include <stdint.h>
#define EI_NIDENT 16
#define Elf64_Addr uint64_t
#define Elf64_Off uint64_t
#define Elf64_Half uint16_t
#define Elf64_Word uint32_t
#define Elf64_Sword int
#define Elf64_Xword uint64_t
#define Elf64_Sxword signed long int
typedef struct {
unsigned char e_ident[EI_NIDENT];
Elf64_Half e_type;
Elf64_Half e_machine;
Elf64_Word e_version;
Elf64_Addr e_entry;
Elf64_Off e_phoff;
Elf64_Off e_shoff;
Elf64_Word e_flags;
Elf64_Half e_ehsize;
Elf64_Half e_phentsize;
Elf64_Half e_phnum;
Elf64_Half e_shentsize;
Elf64_Half e_shnum;
Elf64_Half e_shstrndx;
} Elf64_Ehdr;
typedef struct {
Elf64_Word sh_name; /* Section name */
Elf64_Word sh_type; /* Section type */
Elf64_Xword sh_flags; /* Section attributes */
Elf64_Addr sh_addr; /* Virtual address in memory */
Elf64_Off sh_offset; /* Offset in file */
Elf64_Xword sh_size; /* Size of section */
Elf64_Word sh_link; /* Link to other section */
Elf64_Word sh_info; /* Miscellaneous information */
Elf64_Xword sh_addralign; /* Address alignment boundary */
Elf64_Xword sh_entsize; /* Size of entries, if section has table */
} Elf64_Shdr;
typedef struct {
Elf64_Word st_name;
unsigned char st_info;
unsigned char st_other;
Elf64_Half st_shndx;
Elf64_Addr st_value;
Elf64_Xword st_size;
} Elf64_Sym;
typedef struct {
Elf64_Word p_type;
Elf64_Word p_flags;
Elf64_Off p_offset;
Elf64_Addr p_vaddr;
Elf64_Addr p_paddr;
Elf64_Xword p_filesz;
Elf64_Xword p_memsz;
Elf64_Xword p_align;
} Elf64_Phdr;
typedef struct {
Elf64_Sxword d_tag;
union {
Elf64_Xword d_val;
Elf64_Addr d_ptr;
} d_un;
} Elf64_Dyn;
extern volatile struct limine_kernel_file_request kernelfile;
Elf64_Ehdr *get_kernel_elfheader();
uint64_t get_kerneL_address();
a link would be better tbh
surreal path
fair
elder shoal
#if LIMINE_API_REVISION >= 2
struct limine_executable_file_request {
#else
struct limine_kernel_file_request {
#endif
uint64_t id[4];
uint64_t revision;
#if LIMINE_API_REVISION >= 2
LIMINE_PTR(struct limine_executable_file_response *) response;
#else
LIMINE_PTR(struct limine_kernel_file_response *) response;
#endif
u mind pull requesting ur checks btw
elder shoal
what revision did yo set
surreal path
2
elder shoal
stare at this code until you figure it out
haughty kite
you fucking opted into it
haughty kite
its called an mp request now lol
elder shoal
real smp is when all cores boot up at the same time and theres no concept of a bsp 
surreal path
@haughty kite did u pr ur checks yet
runs nicely btw
can u do that pitust
or at least
send the code over
i still am getting 131 ops
and like 12 seconds uacpi init
i need that speed increase lol
pleaseeee
surreal path
132*
molten grotto
you can have arch specific memcpy/memset and then a generic fallback with attribute weak
edgy pilot
haughty kite
i need that speed increase lol
the speed increase is running on a fast laptop i think lol
0 changes and it still runs in 56
oh also, -smp 2
edgy pilot
oh also, -smp 2
what would that do if the kernel is running on a single core
molten grotto
and like 12 seconds uacpi init
how do you manage to get that bad score lol
haughty kite
what would that do if the kernel is running on a single core
make qemu not do stupid shit
surreal path
you can have arch specific memcpy/memset and then a generic fallback with attrib...
im adding that
now js memset
rep stosd or wtv
molten grotto
yes
molten grotto
because not all memcpys have 8 byte aligned sizes? and also rep movsb/stosb are the only optimized ones iirc
molten grotto
I can't understand how can it be 131/s
its like 6k x slower than my kernel for an example
surreal path
I can't understand how can it be 131/s
REALLY Dunno either
molten grotto
unless its your timer somehow but if it also takes a visibly long time then its probably something else
surreal path
its also visibly a long time
so not my timer
oh
i just enabled kfree
@kind root @haughty kite
wtf
1137 ops per second
LETSS FUCKING GOOO
kind root
surreal path
where is nyaux on the letherboards
kind root
do u want to be on the scoreboard
kind root
sure
surreal path
put me onnnn
elder shoal
can someone with an 12th gen i9 run astral so the number can be high 
orchid dawn
I can run it on an i3-12100f, should be close enough
molten grotto
I only have 13 gen i5
cinder plinth
you're not ready I will be on top of the leaderboard soon
molten grotto
I only have 13 gen i5
I don't think there is like a huge difference between the i's after 5 anyway, they are just way more pricey for a very little gain
kind root
the key is to have bad timekeeping
surreal path
these guys are funny
molten grotto
to make it faster
remove all the checks there are
cinder plinth
remove all the checks there are <:meme:575445094589661215>
nah just using object caching
kind root
<@353593948523986945> u taking that disrespect?
hes right
but uacpi is pretty well tested at this point too
flat nymph
I have Ryzen 9 5900X
unkempt relic
Acpica
Maintainer got convicted of criminal damage
In all seriousness though although I favour using community projects anyway
My choice to use uacpi would have been the same regardless
Acpica is a body barely living outwith the foetid confines of the Linux kernel and maintained by 1 or fewer people
tawdry mirage
speaking of acpica, it's kinda funny how managarm used acpica, lai, and uacpi at various points in time
kind root
speaking of acpica, it's kinda funny how managarm used acpica, lai, and uacpi at...
lol true
surreal path
https://tenor.com/view/pink-panther-code-explode-line-vscode-gif-127165665406425...
real!
surreal path
say hi to nyaux mascot
(real)
this is a joke obviously
but rate my art
0-10
elder shoal
Furry
orchid dawn
but rate my art
eh, I'm probably not much better lol
5
orchid dawn
eh, I'm probably not much better lol
5
in my defense I made this at one point many years in the past (mockup demonstrating how I could replace my broken CPU cooler with a USB fan ductaped to the bottom of the PSU )
worldly condor
steady flume
say hi to nyaux mascot
nyaux fursona reveal
frigid cliff
say hi to nyaux mascot
My friends said “10, excellent”
surreal path
TODO list for today:
abstract smp away to architecture specific code
abstract pagemap and mapping functions to architecture specific code
if we have more time
scheduler
flat nymph
What is there to abstract in SMP?
surreal path
its just the vmm then that needs to be abstracted
yea this is pretty much how i abstract everything
i think its a good way to be honest
so what im gonna do for abstracting away page table things to archiecture yk
im gonna have something like
typedef struct {
uint64_t *root; // on arm this could just be a pointer to the large page table or whatever idfk
} pagemap_t;
void arch_init_pagemap(pagemap_t *take);
void arch_destroy_pagemap(pagemap_t *take);
void arch_map(pagemap_t *take);
void arch_map_hhdm(pagemap_t *take);
void arch_map_framebuffer(pagemap_t *take);
void arch_map_memorymap(pagemap_t *take);
void *arch_map_vmm_region(pagemap_t *take, uint64_t length, uint64_t vmm_region_base); // returns the base of the vmm region
void arch_unmap_vmm_region(pagemap_t *take, uint64_t length, uint64_t vmm_region_base);
having these as seperate functions gives me more control on the archiecture specific code itself
surreal path
im gonna have something like
```C
typedef struct {
uint64_t *root; // on arm ...
@hexed mango what do u think
kind root
no
surreal path
why
kind root
too high level
flat nymph
why arch map_hhdm
flat nymph
just have arch_map which takes a bunch of options (?)
surreal path
some archiectures might not have pages in 4096 bytes
flat nymph
It doesn't matter
kind root
and
surreal path
some archiectures might not support recursion
hexed mango
<@1223455220948930650> what do u think
sorry, a bit too busy to give it a detailed look atm
flat nymph
some archiectures might not support recursion
wdym
surreal path
yea
flat nymph
it's obnoxious anyway
flat nymph
some architectures have inverted page tables
surreal path
yea
exactly
so thats why
im making it this high level
i dont wanna make any assumpations
flat nymph
What I had to implement when porting to riscv was map, unmap and unmap range
(and I've also made my unmap functions take context for tlb shootdowns, but that's an implementation detail i'd suppose)
also some architectured don't even have page tables
as in a page table in form of a radix tree
finite summit
The only APIs I remember using often in the VMA are:
Map (which is more like modify PTE at this point)
Unmap
Query page info
finite summit
The only APIs I remember using often in the VMA are:
Map (which is more like mod...
And unmap is basically deprecated at this point
flat nymph
And unmap is basically deprecated at this point
why
flat nymph
finite summit
Because the functionality of map can do the same:
And in some places related to swap, it is used in that matter
surreal path
anyways im gonna impl my high level abstraction shit
finite summit
Mainly so I could keep the PTE flags, but unmap the page itself
finite summit
anyways im gonna impl my high level abstraction shit
Why?
That's just reusing code in many places
surreal path
look at my responses above
finite summit
For no reason
flat nymph
some archiectures might not have pages in 4096 bytes
again it doesn't matter
finite summit
some archiectures might not have pages in 4096 bytes
Yet you still say "page size in 4kib"
As a parameter
finite summit
It should at least be in bytes
surreal path
fuck i cannot spell
flat nymph
forget about it
flat nymph
arm can have 1kib pages
surreal path
hold on lemme rethink
finite summit
This is a bad way to abstract it
surreal path
okay whats ur better way
finite summit
Abstract it via map and unmap functions that take a base and physical address
Have the arch tell you the page size
And work in that manner
surreal path
Have the arch tell you the page size
huh
wdym by that
what about mapping the framebuffer with wc enabled
finite summit
wdym by that
Make the platform tell you what the page size is
surreal path
what about bigger pages
finite summit
Map will take flags
Telling it how to map the page
E.g., huge page
E.g., map as x caching mode
finite summit
what about mapping the framebuffer with wc enabled
Afaik only x86 does that
But someone correct me if I am wrong
surreal path
what about unmapping
flat nymph
what about mapping the framebuffer with wc enabled
you pass it a flag that says "wc disabled"
finite summit
what about unmapping
Take in a base
flat nymph
Afaik only x86 does that
you have some cache disabled shenanigans on risc-v and arm
finite summit
you pass it a flag that says "wc disabled"
Maybe wc enabled flag?
surreal path
i need its physical address returned
finite summit
i need its physical address returned
Then do so
surreal path
what about mapping in a loop
finite summit
You can also have a query page info function
finite summit
what about mapping in a loop
Call map in a loop?
finite summit
You can also have a query page info function
That tells you the flags it was mapped with, the physical address, etc.
finite summit
Why?
surreal path
this is how i map regiosn currently
finite summit
forget about 4096
^
surreal path
this is how i map regiosn currently
how am i supposed to change this
flat nymph
PAGE_SIZE constant
finite summit
how am i supposed to change this
You rewrite it
surreal path
oh use a loopppp
oh my god
sureee
but how about
the amount of times u loop over
do u loop in bytes?
like
finite summit
the amount of times u loop over
Division
You take in a size
Then you can convert that to a count of pages (you have a page size constant from the arch)
And loop that many times
finite summit
Then you can convert that to a count of pages (you have a page size constant fro...
Don't forget to round up
surreal path
You take in a size
in map? in what function
finite summit
Idk where is ur loop
surreal path
im kvm_region_deallloc and kvm_region_alloc
finite summit
Then that's where you take in a byte size parameter
Or have the size of the region
In the region
surreal path
i already have that
finite summit
Then use that
surreal path
its in bytes
finite summit
Then you can convert that to a count of pages (you have a page size constant fro...
Do this
And then loop that many times to map/unmap
surreal path
Then you can convert that to a count of pages (you have a page size constant fro...
okay so an arch_convert_bytes_to_page
is this correct?
yes
or no
finite summit
okay so an arch_convert_bytes_to_page
You could have a page size constant so you can do that in an arch agnostic way
Like NYAUX_ARCH_PAGE_SIZE
flat nymph
I mean you can probably have an arch-specific map_range function for optimization
flat nymph
okay so an arch_convert_bytes_to_page
No
finite summit
I mean you can probably have an arch-specific map_range function for optimizatio...
Never
I am adding some new features to my VMM so I could experiment with something like that to optimize
flat nymph
My kernel does map several pages at once
surreal path
#define NYAUX_ARCH_PAGE_SIZE #if __x86_64__ \
4096 \
#endif \
#define NYAUX_ARCH_HUGE_PAGE_SIZE #if __x86_64__ \
/*2mib basically*/\
#define \
Yeah
surreal path
okay
typedef enum {
TINYPAGE = 0,
SMALLERPAGE = 1,
SMALLPAGE = 2,
PAGE = 3,
HUGEPAGE = 4,
humongousPAGE = 5,
GIGTANICPAGE = 6,
cached = 7,
notcached = 8,
} vmmflags;
how about this
finite summit
Terrible
surreal path
whats better
finite summit
Make it a bitmap
surreal path
wdym
surreal path
explain
finite summit
Which I don't plan on explaining
finite summit
no i mean how flags are inserted
As any other bitmap
finite summit
finite summit
Then it's easy
surreal path
what bit represents what in a bitmap
finite summit
You define that
surreal path
how big of a bitmap does it need to be
oh
wait
yea im starting to understand
something like this
surreal path
```C
typedef enum {
TINYPAGE = 0,
SMALLERPAGE = 1,
SMALLPAGE = 2,
PAGE =...
wait what about framebuffer @finite summit
like i dont understand how that would work on other archectiures
like say
i did in a bitmap
big page, cached
but its not a framebuffer
what does that mean
like
hold on
okay
sorry my question is
what if u want different types of caching thats archiecture spefic yk
how does that work
finite summit
what if u want different types of caching thats archiecture spefic yk
You can define an arch specific range of mapping flags
That arch specific functions can use
Rn I'm walking back home from school
edgy pilot
what if u want different types of caching thats archiecture spefic yk
different set of arch agnostic flags
that are translated to arch specific actions
surreal path
different set of arch agnostic flags
iiii
dont understand
im sorry im just really slow
😭
and stupid too
surreal path
finally some code i can read
edgy pilot
map(..., flags, page_size, caching)
map then translates flags, psize and cache to arch thingies and does stuff
surreal path
what if archiectures dont have write_combining
kind root
-EINVAL -ENOTSUP
surreal path
okay
edgy pilot
wdym default enum
surreal path
like normal = write_back
edgy pilot
normal is a name
surreal path
ohhh okayyy
surreal path
enum caching {
uncacheable,
write_through,
write_combining,
write_back,
normal = write_back,
mmio = uncacheable,
framebuffer = write_combining
};
// flags is a unsigned byte
// read | write | exec | user
// uint8_t flags;
Maybe add some namespace to that?
surreal path
how so
edgy pilot
```C
enum caching {
uncacheable,
write_through,
write_combining,
write_b...
finally some code i can
readcopy
finite summit
E.g., write_back becomes caching_write_back
edgy pilot
yeah there's no enum class in c
molten grotto
how so
all that stuff will end up in the global namespace because of c enums being stupid
tawdry mirage
if we're bike shedding this enum then maybe have separate caching modes for posted mmio and non-posted mmio
surreal path
kill meeee
kind root
mmio_strong
surreal path
edgy pilot
if we're bike shedding this enum then maybe have separate caching modes for post...
???
I have never heard of that term
tawdry mirage
non-posted mmio waits for writes to complete, posted mmio doesn't
surreal path
im js make everything high level this is basically too much work and i really wanna js write the scheduler already so
flame me all you want its my kernel
edgy pilot
it's an osdev group project
tawdry mirage
non-posted mmio waits for writes to complete, posted mmio doesn't
x86 doesn't have the distinction but e.g. arm does (device memory with early-write-acknowledge)
and iirc at least on apple silicon for some mmio you have to use a specific type (iirc map pcie mmio as posted?) or you'll get an asynchronous abort when accessing it
edgy pilot
I don't have much done with the aarch64 side of the kernel yet
basically nothing, but I will keep that in mind when I get around to it
kind root
and iirc at least on apple silicon for some mmio you have to use a specific type...
uncached also works right
tawdry mirage
on aarch64 no
kind root
ah ok
tawdry mirage
normal memory uncached permits write coalescing and reordering
surreal path
FINALLY
AFTER LIKE
WHAT
AN HOUR
I GOT IT WORKING CHAT
MY ABSTRACTION WORKS
orchid dawn
🔥
kind root
LFG
kind root
F
surreal path
F
why is it stuck
maybe it failed switching to the page map
why???
why didnt it triple fault
bro
hold on more printf debugging fun
lets try this again
like
yea its 0x9000
whats the issue
it got as far as Giving address 0x9000
like
thats literarly it
fucking kill me
it just hangs
doesnt even triple fault
just hangs
well it wouldnt triple fault
idfk
it should work
it works on the vm
why not on real hardware
finite summit
why not on real hardware
have you considered zeroing your memory
surreal path
have you considered zeroing your memory
yes
yes i have
wait
lemme not make ker_map fields uninited at the start
solved absolutely nothing
kill me chat
how tf am i supposed to debug this now
i cant gdb real hardware
it just hangs trying to switch to the pagemap
i genuinely dont know what to do
molten grotto
is it in git
surreal path
yes
molten grotto
missing gitmodules as usual
surreal path
molten grotto
oh wait its not missing
surreal path
yea
molten grotto
its just missing the cc runtime
surreal path
i even tried randomizing
the memory in qemu with /dev/random
worked fine
@molten grotto if u need any help compiling or have questions just ping me
molten grotto
well can you commit the updated gitmodules file
molten grotto
wait is kernel/cc-runtime a symlink to something
molten grotto
ah now it works
surreal path
yea
molten grotto
yeah idk, if you haven't figured it out by tomorrow I might take a look again as I am going to go sleep now
surreal path
yeah idk, if you haven't figured it out by tomorrow I might take a look again as...
okay
surreal path
didnt find anything
went over the paging code itself
didnt find anything obvious
looks fine
molten grotto
maybe it could also be some kind of framebuffer issue
surreal path
not really
framebuffer runs fine
under qemu and kvm
and it runs fine up to switching the page tables
korona said theres a possiblity that the page tables could be wrong
not to a point that kvm cares
but to a point real hardware cares
ive been looking up and down the paging code i havent found anything suspicous
and i certinally cannot attach serial to test if its a framebuffer issue
i dont have a serial port on my laptop lmao
molten grotto
make a quick rtl driver + ethernet + ip if it has an rtl card 
surreal path
lmao
molten grotto
anyway Ill take a look now if I can find anything
molten grotto
I found at least two things (of which the first one is likely unrelated), both your memcpy and memset return wrong pointer and in the code where you map the framebuffer you do ```c
uint64_t page_amount =
align_up(entry->length - disalign, 2097152) / 2097152;
what should i return in my memcpy and memset
molten grotto
the original dest pointer, both rep movsb and rep stosb modify the pointer you give to them
surreal path
I found at least two things (of which the first one is likely unrelated), both y...
whats wrong with this as well
molten grotto
why do you subtract the disalign from the length
surreal path
why do you subtract the disalign from the length
oh
where should i subtract it then
molten grotto
you should add it
surreal path
understandable
my bad thanks for catching this
okay im adding them now to the length
flat nymph
man i hate debugging real hardware
just switch to my kernel 
It's so stable it doesn't have framebuffer output (totally not because of bad design)
edgy pilot
I found at least two things (of which the first one is likely unrelated), both y...
@surreal path please use hex
surreal path
molten grotto
also constants instead of hardcoded numbers everywhere lol
surreal path
ik 😭
molten grotto
the code style is also weird
surreal path
anyways im gonna try this now on my hardware?
will it work?
lets see if it works on a vm first
edgy pilot
anyways im gonna try this now on my hardware?
first kernel to fry a cpu
surreal path
runs fine on a vm
okay
lemme just push the changes so u can pull them qwinci and ill try them on my laptop hold on
alr time to flash my usb
edgy pilot
alr time to flash my usb
install limine on your drive
flat nymph
first kernel to fry a cpu
No #1217009725711847465 message
edgy pilot
no more need for isos
molten grotto
or if you have two pcs/laptops then pxe
edgy pilot
No https://discord.com/channels/440442961147199490/1217009725711847465/131866742...
if only mobile discord could properly jump to linked messages
edgy pilot
JavaScript reboot
molten grotto
maybe Ill try it on my laptop to see if it has the same issue
surreal path
alr
edgy pilot
talk is cheap send patches
literally just ctrl+shift+h
surreal path
what does that do
edgy pilot
it blows your house up
surreal path
😱
edgy pilot
press it in vscode
edgy pilot
incredible discovery
surreal path
@edgy pilot done u happy
edgy pilot
no
kind root
no
surreal path
why
edgy pilot
at least do
#define MIB(x) (0x100000 * (x))
kind root
#define KB 1024ull
#define MB (1024ull * KB)
...
edgy pilot
```c
#define KB 1024ull
#define MB (1024ull * KB)
...
```
I have that but with constexpr functions
molten grotto
yeah it hangs on my laptop too
surreal path
at least its repoducable
wow i cannot spell
reproutdanble/
reproduct?
product?
molten grotto
reproducible
surreal path
@molten grotto did u find anything
thorn bramble
I have that but with constexpr functions
wow not even operators :^)
edgy pilot
wow not even operators :^)
?
tawdry mirage
""_MB 😋
edgy pilot
1_mb?
thorn bramble
size_t operator ""_KiB(size_t x) {
return x * 1024;
}
size_t operator ""_MiB(size_t x) {
return x * 1024_KiB;
}```wow waht di di xo
edgy pilot
wow waht di di xo
is that english? :p
thorn bramble
what no sleep does to a person
thorn bramble
is that english? :p
a bad attempt at it, yeah
surreal path
@molten grotto anything found yet?
thorn bramble
what u lookin for
surreal path
stupid real hardware bug
hangs on switching to page tables on real hardware but kvm and qemu run just fine
read above for more context
surreal path
fair :)
well we
looked over the paging code, other then a mistake with memcpy and memset and something wrong with the disalign which we fixed and it has solved nothing
still nothing
working
at least the bug is reproducible
thanks for qwinci for teaching me how to spell that word
orchid dawn
@surreal path if you don't know if it's actually continuing after switching the pagemap, simply force a hard reset afterwards
if it resets, it works
if it doesn't reset, it doesn't work 
edgy pilot
stupid real hardware bug
those cpu engineers putting bugs in cpus so nyaux doesn't run on their cpus
surreal path
<@787341689333350450> if you don't know if it's actually continuing after switch...
wdym
hold the power button?
surreal path
oh i even put something there
edgy pilot
load an invalid idt and divide by zero
surreal path
it didnt even care
js continues the hang
no triple fault
no thing
js
stuck on switching the page tables
orchid dawn
did you actually load an invalid idt so that it wouldn't hit your double fault handler
surreal path
no idt is valid
orchid dawn
then why would it triple fault
surreal path
did you actually load an invalid idt so that it wouldn't hit your double fault h...
oh u want me not to load an idt
yea lemme try that
if it triple faults
not a framebuffer issue
orchid dawn
no if it triple faults then it is a framebuffer issue
that means that your code is still running
surreal path
oh
orchid dawn
but can't get to the framebuffer
surreal path
if it doesnt?
orchid dawn
then something else has gone horrifically wrong
surreal path
i see
edgy pilot
you didn't map your framebuffer
orchid dawn
no
I'm at school
surreal path
okay fair
edgy pilot
and vmware
edgy pilot
I'll send a ball stealer goblin to your house otherwise
surreal path
edgy pilot
you have 3 minutes
surreal path
runs fine on vbox
edgy pilot
change the chipset
surreal path
okay
surreal path
we got something
edgy pilot
we got something
that took so long I thought you were changing the chipset on your mobo :P
surreal path
??
edgy pilot
debuggers are for the weak
edgy pilot
look at it better
surreal path
bruy
edgy pilot
bnuy
hexed mango
surreal path
the hell is a guru mediation
wait
now its doing it the moment i try to load my kernel?
tf?
wow
now i dont know if its vbox or my kernel
wait
kind root
its your kernel
surreal path
i gave the ram 4mb
hold on
it was just the ram
limine couldnt fit my kernel in the ram at 4mb
for some reason
well it runs now
just fine
surreal path
limine couldnt fit my kernel in the ram at 4mb
maybe limine should throw an error message at these really low ram levels or smthin
either way
idfk what to do
time to test the usb
thorn bramble
it does
thorn bramble
maybe limine should throw an error message at these really low ram levels or smt...
.
surreal path
it does
weird, just to verify lemme try loading astral at these ram level
if it doesnt throw an error message
not my kernel
anyone got the astral iso rq
found it
so it doenst go into the limine bootloader on astral?
no menu
js straight up says out of memory
but when loading my kernel theres a boot menu but THEN it dies?
thorn bramble
astral has abckground
thorn bramble
build limine-c-template
surreal path
okay
surreal path
oh i just figured it out
it dies trying to init flanterm at 4mb
the video memory is 21mb
so
thats understandable
giving it 1mb of video memory
kinda works
but it seems like i die trying to read my own elf i suppose
which is fine again
so nyaux is fine
so far on vbox
js i dont handle oom very well lol
anyways
im gonna try the usb
hold on
edgy pilot
maybe limine should throw an error message at these really low ram levels or smt...
smh you seem to be blaming everything on anything but yourself
that's not very good, especially when programming
surreal path
smh you seem to be blaming everything on anything but yourself
no im serious
it even worked fine downstairs on my sisters pc?
tho uacpi cried cause the aml was weird
@kind root do u want to see the errors
for uacpi
surreal path
smh you seem to be blaming everything on anything but yourself
like what u want nyaux to run on 4mb, i dont think thats a bug tbh
i know what its crashing
and its my code for not handling oom
thats MY fault
im still blaming myself
in a different way tho
surreal path
thats because u dont implement half the stuff it wants
understandable
but again
why isnt nyaux working on my laptop??
but its working on this pc
wtf
tawdry mirage
screenshot of a photo open in an image vieweer
surreal path
lol
surreal path
this is unacceptable
no not until this bug is gone
boot on my lapto
😭
i booted again into my laptop
same issue
but works fine on the desktop
yea im fuckin ending it all
not only does it work on SOME hardware but not all
but its ONLY on my laptop
so its not even properly reproduicable
fun
how tf does that work
SOME devices the page tables load fine
some devices it hangs
HOW
does that even work
x86 cpus are all the same archiecture
like wtf
is this even a page table issue
what is this
i dont know what to do
surreal path
speaking of do u have ubsan enabled
i tried getting ubsan enabled
i couldnt get it working
well ksan
surreal path
which is?
surreal path
ubsan?
how do i get it working
oh yeah
idk how much thread messages that is anymore
😭
30,841
flat nymph
```cpp
size_t operator ""_KiB(size_t x) {
return x * 1024;
}
size_t operator ...
wtf is this real
surreal path
30 thousand
surreal path
wtf is this real
im going to eat you
and ur pmos
no more microkernel
vine boom
monolithic kernel >
anyways how to get ubsan to work chat
is there somethign i need to add
or
flat nymph
wtf is this real
No I don't know if it's a C++ feature
silver yarrow
is there somethign i need to add
Yep
some hooks
Feel free to rip this
though this is meant for gcc ubsan
clang has more hooks iirc
then u just add this flag
flat nymph
fun
surreal path
https://git.thenest.dev/wux/elysium-os/-/blob/5.0/kernel/src/ubsan.c?ref_type=he...
okay adding them now
silver yarrow
its very easy
surreal path
its very easy
okay i finished the functions
ummm
lol???
why is it stuck
???
theres no fault or anything
in the qemu logs
why is it stuck
huh\
surreal path
how are u outputing from the hooks
hold on ill give u the pastebin or do u prefer github?
what do u prefer?
silver yarrow
eitherway
silver yarrow
does ur kprintf require setup
silver yarrow
its possible there is ub before its setup
id recommend hooking it straight to serial
surreal path
kk
surreal path
oh my god
its mostly complaining about flanterm
@silver yarrow how can i make ubsan ignore flanterm
kind root
stop giving flanterm misaligned addresses
surreal path
when did i give it a misaligned address
i gave it the framebuffer addres
that shouldnt be misaligned
thats literarly what i do @kind root
kind root
idk where it gets the addresses from
surreal path
from limine