in what regard? running Linux or as an architecture/device
#Nyaux
rigid fable
rigid fable
Well Linux runs fine on the PS3. As today you need a jailbroken PS3, since on OFW the otherOS feature was removed. Apart from that you just need a bootloader like Petitboot and a PPC build of Linux. There is also a lot of progress in PS3 RSX (gpu) hardware acceleration
look for T2 SDE
thorn bramble
and the ps4 is a somewhat standard amd apu iirc
rigid fable
and as an architecture it is so complex that still today it is THE hardest arch to emulate
thorn bramble
minus the non standard platform
rigid fable
and the ps4 is a somewhat standard amd apu iirc
yes
thorn bramble
i wonder if it's got acpi
thorn bramble
well, its not "just" a bsd
its gotta be heavily modified
to actually fit their needs
and to run on their weird hardware
but they've done it, people got linux to run on it too
its not that bad apparently
kind root
When you port if of course
molten grotto
one thing which I don't know is whether you can get a framebuffer without actually writing/porting an amd gpu driver on it
elder shoal
Just do it ez
kind root
I'm sure its not rocket science to get a fb
flat nymph
hyper for PPC when
kind root
You dont even use it 
For hyper im mostly fixing bugs or adding tiny qol features, as it is mostly feature complete for my needs
But im not against accepting changes that others might want if there's a need
elder shoal
uBIOS for booting hyper oses without a bootloader when
tawdry mirage
and the ps4 is a somewhat standard amd apu iirc
the southbridge has a lot of familiar peripherals but in a kinda brain damaged package
like it responds to any slot number, and they ran out of functions so a whole bunch of stuff is shoved into one function
unkempt relic
amazing everyone went over to powerpcs for games consoles when the ps3/xb360/revolution came out now everyone is on amd64 (i think, i don't know what that one nintendo uses, the wiiu or whatever it was called)
tawdry mirage
the switch is just an nvidia tegra x1 soc
which is aarch64
the wiiu is ppc since it's a direct upgrade to the wii (and that is a direct upgrade to the gamecube)
iirc
unkempt relic
when the ps3 came out its architecture was supposed to be revolutionary (ibm "cell", 1 ppc core, 8 something-or-others) but that fell very flat
tawdry mirage
problem with revolutionary arches for consoles is that it's hard to port games and have them be properly optimized :^)
unless it's an exclusive ig
finite summit
lies
kind root
okok not literally
rigid fable
problem with revolutionary arches for consoles is that it's hard to port games a...
I always considered that a stupid excuse. If you want your game to run on a machine which offers very good hardware for specific tasks and also want it to run well, why do you expect to do it without effort? Obviously there are time costraints in game development, and budget too, but just do not use excuses. ig not the best argument but I do not have time rn
oh and I am quite sure that the only reason this did not affect the xbox 360 is because it had directX and other stuff from Windows which made porting easier. (let's not talk about the memory pool division)
tawdry mirage
I always considered that a stupid excuse. If you want your game to run on a mach...
i mean time and budget constraints are the obvious cause for this
tawdry mirage
i mean time and budget constraints are the obvious cause for this
but also, i'm having a bit of a hard time thinking up what the spus could help you with
considering they require a bunch of setup before you run code on them
and it's compiled separately afaik so that's extra effort as well
rigid fable
yes
rigid fable
but also, i'm having a bit of a hard time thinking up what the spus could help y...
they acted like shader computing units on modern gpus work
edgy pilot
what we talking about
🖥️
rigid fable
you could run physics, sound, graphics (also with the RSX), AI, particles and so on each on one SPU
actually at the beginning Sony wanted to use two separate CELLs for the cpu and gpu, but for cost issues they ditched the idea and quickly tried to partner up with ATI and then NVidia
surreal path
okay so
what does an Elf loader need to do exactly
like im unsure how to implment this i forgot a ton of things about elf lol
edgy pilot
what does an Elf loader need to do exactly
load the elf 
surreal path
load the elf <:halfmemeleft:715831625233203250>
surreal path
what is an auxval
edgy pilot
struct auxval
{
uint64_t at_entry;
uint64_t at_phdr;
uint64_t at_phent;
uint64_t at_phnum;
};
wha
edgy pilot
surreal path
3.4.1 is talking about fpu state
thorn bramble
3.4.1 is talking about fpu state
read further
dont stop 1 paragraph in
"Initial Stack and Register State"
yeah only talks about fpu state
sure.
surreal path
okay i see
edgy pilot
okay i see
ooh is that fedora?
surreal path
ooh is that fedora?
yea i switched to fedora
edgy pilot
any issues?
surreal path
no
edgy pilot
nice
surreal path
i see
surreal path
and auxilary vectors kind of give
arugments to the interpreter itself
ish
am i understanding correctly
thorn bramble
interpreter*
or rather the program that's running
it just so happens that the program is an interpreter
thorn bramble
check mlibc source
that will tell you the bare minimum of what you need to provide
but generally linux and any other kernel that loads elf executables provides most if not all of these
surreal path
where in mlibc source
thorn bramble
on github?
surreal path
so it needs these?
thorn bramble
doesnt matter
surreal path
doesnt care?
surreal path
okay
thorn bramble
like it doesnt check the order
does it
ok sorry my bad
no it doesn't matter what order they are in
as long as they're all there
surreal path
#[repr(C, packed(8))]
struct Elf64EntryHeader {
e_ident: [i8; 16],
e_type: u16,
e_machine: u16,
e_version: u32,
e_entry: u64,
e_phoff: u64,
e_shoff: u64,
e_ehsize: u16,
e_phentsize: u16,
e_phnum: u16,
e_shentsize: u16,
e_shnum: u16,
e_shstrndx: u16
}
#[repr(C, packed(8))]
struct Elf64ProgramHeader {
p_type: u32,
p_flags: u32,
p_offset: u64,
p_vaddr: u64,
p_paddr: u64,
p_filesz: u64,
p_memsz: u64,
p_align: u64
}
#[repr(C, packed(8))]
struct auxvec {
a_type: i32,
a_val: u32,
a_ptr: u64,
a_fnc:
}
okay i have these structs going
what should a_fnc be
thorn bramble
typedef struct
{
int a_type;
union {
long a_val;
void *a_ptr;
void (*a_fnc)();
} a_un;
} auxv_t;```its supposed to be a union
surreal path
how to do a union in rust lmao
thorn bramble
union
surreal path
seriously?
thorn bramble
go try it
thorn bramble
...
surreal path
😭
thorn bramble
thats why you should learn the
language
first.
#[repr(C)]
union AuxvUn {
a_val: u32,
a_ptr: usize,
a_fnc: usize,
}
#[repr(C)]
struct Auxv {
type_: i32,
un: AuxvUn,
}```no anonymous unions?
surreal path
yea i figured it out
edgy pilot
<insert megamind>
thorn bramble
no ub?
oh wait...
he uses raw pointers everywhere!
you should see his code lol
surreal path
😭
edgy pilot
:D
edgy pilot
the elf file
surreal path
the path?
edgy pilot
whatever you prefer
surreal path
what else
edgy pilot
idk
surreal path
💀
surreal path
i see in ur code u take in base
edgy pilot
address the elf is loaded at
surreal path
i see
edgy pilot
yes
surreal path
aslr?
thorn bramble
what did i say
surreal path
lmao
thorn bramble
just load the darn elf
surreal path
okok
thorn bramble
dont owrry about anything else
kind root
inb4 stuck for a month
molten grotto
I wonder what if I wouldn't have helped diagnose all those issues lol
kind root
u know the answer
thorn bramble
*mut vfs
edgy pilot
static mut
thorn bramble
unsafe
kind root
yes after qwinci debugs it
thorn bramble
say?
surreal path
yes after qwinci debugs it
💀
thorn bramble
it works until you compile with --release
edgy pilot
lmao
surreal path
whats gonna happen in --release
thorn bramble
take a guess
surreal path
actually im gonna try that rn
thorn bramble
the answer is nobody knows
because you just dropped
all the ownership
lifetimes
and pointer provenance info
so llvm can do all that it wants
surreal path
let me try
surreal path
oh
thorn bramble
nah its a shit way out
surreal path
it did
edgy pilot
lmao
surreal path
am i gonna have to do a third kernel rewrite
edgy pilot
in lua
surreal path
no
thorn bramble
if you want it to be in rust, yes
lol
like im not even being mean at this point
its dogshit
surreal path
i guess we are doing a third kernel rewrite
thorn bramble
i wish i could put it nicely
but i cant
you just dont write rust like that
and expect it to work
surreal path
but I want you guys to make sure that ANY code i write is the proper "rust way"
edgy pilot
I would suggest to first learn the language you're going to write a kernel in
thorn bramble
but I want you guys to make sure that ANY code i write is the proper "rust way"
its ur kernel bruh
why should i make sure ur writing it correctly lmao
thorn bramble
also everyone already told you to not write code like that
surreal path
double check
thorn bramble
anywhere where you have to use a raw pointer in rust
you should double check
because you should never have to do that
almost never*
and if you do it should be only for the abstractions that provide safety
edgy pilot
you shouldn't use raw pointers
unless you are accessing some system resource or using a library that requires it
thorn bramble
not even that
you should have abstractions over that too
that interact properly with lifetimes and shit
thorn bramble
rust uses all that info for optimizations
edgy pilot
starting now?
thorn bramble
to write a kernel in rust
kind root
how are u going to rewrite it if u technically dont know rust still
ur going to end up with the same stuff
surreal path
okay then
thorn bramble
basically yeah
thorn bramble
write rust code
edgy pilot
userspace projects
thorn bramble
write rust code
that doesnt look like you just wrote
tawdry mirage
RIIC++ :^)
thorn bramble
based qookie
elder shoal
do a microkernel, write it in asm, and write brainfuck servers :')
thorn bramble
but id rather not see him struggle with that
surreal path
okay so whats the perfect rust project
that will teach me
everything i need to know of rust
thorn bramble
there isnt one
tawdry mirage
a kernel
edgy pilot
okay so whats the perfect rust project
I love tetris please do tetris
thorn bramble
rust is a complicated language, to learn all of what you need to know you need to really spend some time with it
thorn bramble
you have not
surreal path
i have c habits
surreal path
😭 😭
thorn bramble
you just translated C code to rust brother
you didnt even take into consideration what the language offers
surreal path
okay heres what ill do
ill read through the rust book today overnight
and work all night learning rust
ill come back to yall tomorrow
and we can rewrite nyaux
thorn bramble
i dont think thats a good idea
surreal path
why
thorn bramble
to do it
elder shoal
or have a good night of sleep
elder shoal
and slowly learn rust properly
thorn bramble
its gonna take you months
surreal path
but it will delay my work on nyaux
thorn bramble
to learn rust properly
surreal path
nooo
thorn bramble
to be able to write a good kernel
it took me like more than a year or so to actually figure out how rust works
but probably because im slow
thorn bramble
and i was just used to writing code
you need to get used to programming
idk
its like the same thing
surreal path
i think it could take me 1-2 days if i really spend all my weekend on it
thorn bramble
but it feels like its a completely different thing
lol this guy is not being serious
learn rust in 2 days 😭
surreal path
gn iretq
surreal path
wdym
thorn bramble
continue this rust kernel thing
surreal path
and whether you still want to
i still do
thorn bramble
i mean obviously you arent serious if you think you can learn rust well enough in a couple days to be able to write a good kernel
surreal path
i mean obviously you arent serious if you think you can learn rust well enough i...
okay how long then
thorn bramble
its not about how long
its not an amout of time
that makes you good at a language?
surreal path
but what do i do with nyaux while im fucking
learning rust properly
i cant just
stop working on it
like it'd be the third time id do that shit
thorn bramble
idk but you clearly cant keep working on it if you dont know the language
you are just going to end up writing the same shit code with bunch of raw pointers all around the place
because "oh i cant figure out how this shit works but it works with pointers so ill just do that!"
edgy pilot
rewrite it in c and using raw pointers won't matter :D
thorn bramble
like rust is not a good language if you want to work on the kernel
because you are writing rust, not writing a kernel
idk if you catch my drift
surreal path
okay how abt ill come back in a week, after non stop rust development. I know i cant fully learn a language in that length of time but maybe i could be able to start on the rewrite at least
idfk i really dont wanna be waiting fucking months, osdev is my passion
thorn bramble
why do you even want to use rust
thorn bramble
you need to actually understand the language
to be able to make use of its safety features and shit
thorn bramble
i dont know c++
c(++) as in c/c++
surreal path
again just give me a week, i promise that when i come back ill at least be more competent at the language
maybe make a less shitty kernel
still shitty but less
edgy pilot
component?
surreal path
still shitty but less
and if that kernel ends up being shitty
still
ill rewrite it again
and again
and again
tawdry mirage
you can have tagged unions in c(++)
imo rust enum mogs std::variant 
surreal path
until it fucking is good
thorn bramble
imo rust enum mogs std::variant <:halfmemeright:715831702164996096>
oh yeah definitely
surreal path
again just give me a week, i promise that when i come back ill at least be more ...
is that fine
tawdry mirage
thorn bramble
i liked that
surreal path
is that fine
?
thorn bramble
i mean up to you what you do
but as i said maybe if you want to do osdev you should do it in a language you know relatively well
unless you are fine with "waiting"
surreal path
i mean up to you what you do
that is what im gonna do, if the kernel ends up being shit
still
ill rewrite it again
and continue this cycle
until i know proper rust and write a good fucking kernel
thorn bramble
might take a while
surreal path
we'll see
surreal path
even i dont know 100% of C's features
thorn bramble
c has no features bro
its a dead simple language
you write code
it does what you type
literally
surreal path
look ill just see what i can do
im gonna go and fucking learn rust now
see yall later, ill update you guys on how thats going
thorn bramble
well gl to you then 
surreal path
thorn bramble
i am going to sleep
thorn bramble
im gonna write my kernel in c++ tomorrow
surreal path
niceee
surreal path
😭 😭
tawdry mirage
it does what you type
and then some because you triggered some ub unknowingly
thorn bramble
i mean is there really that much ub in C
for the most part it just does what you want it to do
elder shoal
ill update you guys on how thats going
do it
elder shoal
i mean is there really that much ub in C
once I managed to hit some crazy ub when just writing a tcp checksum function which was crazy
thorn bramble
thats why i am going to use c++
tawdry mirage
i mean is there really that much ub in C
well signed integer overflow, strict aliasing
thorn bramble
damn signed integer overflow sounds funky
idk what strict aliasing even means so i dont have to worry about that one right?
tawdry mirage
unsigned is defined to wrap around but signed is ub
tawdry mirage
yeah
thorn bramble
interesting
tawdry mirage
it's not technically guaranteed to wrap to the lowest value
so checks like if (a + 1 < a) just get optimized out for example
because since overflow is ub and can't never happen, +1 means it's always larger
thorn bramble
fun
good thing i always do checks like this by checking against the max value
i guess im somewhat aware of things like that, or maybe the code ends up being cleaner or easier to read that way idk
tawdry mirage
so checks like `if (a + 1 < a)` just get optimized out for example
there's a gcc bug report where someone is complaining about that
so much for people knowing c
thorn bramble
so what about this strict aliasing thing?
are you able to tldr that or is it a bit more complicated
is it with multiple pointers pointing to the same value or something
tawdry mirage
dereferencing a pointer that aliases a value of an incompatible type
thorn bramble
ah so like c uint64_t v = 1337; uint8_t x = ((uint8_t*)&v)[5];
tawdry mirage
well through char it's explicitly allowed
thorn bramble
lmao
tawdry mirage
but *(float*)&my_int
edgy pilot
ah so like ```c
uint64_t v = 1337;
uint8_t x = ((uint8_t*)&v)[5];```
is that ub?
molten grotto
no if uint8_t is unsigned char/char
edgy pilot
oh good
tawdry mirage
but `*(float*)&my_int`
the proper way in c to do that would be with an union
thorn bramble
is the famous fast inverse sqrt snippet ub?
tawdry mirage
the proper way in c to do that would be with an union
and in c++ you'd need memcpy or such
molten grotto
the proper way in c to do that would be with an union
or memcpy
tawdry mirage
is the famous fast inverse sqrt snippet ub?
technically yeah
elder shoal
sweet sweet ub
tawdry mirage
you can disable strict aliasing in c with -fno-strict-aliasing
thorn bramble
nice, now i know why its so "fast"
thorn bramble
you can disable strict aliasing in c with `-fno-strict-aliasing`
i assume that gets rid of "some" "optimizations"
probably nothing significant
molten grotto
yeah
tawdry mirage
yeah it means some optimizations aren't possible
thorn bramble
so tldr instead of type punning actually tell the compiler about the types properly
tawdry mirage
yeah it means some optimizations aren't possible
e.g. with strict aliasing this assert might not fire struct { int x; } y = {0}; *(float *)&y = 2; assert(y.x == 0);
tawdry mirage
not even that, the compiler assumes the value of x is unchanged because the write couldn't have aliased the struct
so it yeets the assert out
thorn bramble
yeah basically what i was getting at
oh wait
yeah that makes sense
i thought it would yeet the write and the assert
tawdry mirage
technically it could do that as well
thorn bramble
ah
so if you made the write volatile what would happen?
would it still yeet the assert?
tawdry mirage
it still could, but i have a hunch it might not?
but i'm not sure what compilers will do, but technically volatile doesn't mean "stop caring about strict aliasing"
thorn bramble
i see
yeah i meant in terms of strict aliasing, sorry
it would still be valid to assume that y.x is still 0
because that write could not have possibly affected the struct blah blah
unkempt relic
with a volatile load of y.x it wouldn't, with a volatile write through a float alias, that's another matter
thorn bramble
yeah i was just wondering what swapping out that float write for a volatile one would do
thats interesting though, ill definitely watch out for any aliasing i might or might not do in the future :')
thanks for breaking that down for me qookie
tawdry mirage
some aliasing is fine though
e.g. a pointer to struct T can be cast to a pointer to the type of it's first member
tawdry mirage
well it's useful for approximating inheritance
thorn bramble
yueah i was about to say
it might be nice in case of like a shared header
can you do it the other way?
tawdry mirage
hmm
thorn bramble
int* -> struct { int x; }*
tawdry mirage
i want to say yes
unkempt relic
in practice you always turn off strict aliasing, but it's an interesting thought to consider trying to get away from it altogether
tawdry mirage
i want to say yes
but now i'm not sure :^)
kind root
in practice you always turn off strict aliasing, but it's an interesting thought...
but my microoptimization... 
molten grotto
lol
unkempt relic
but my microoptimization... <:sadkek:893075479958016020>
even despite knowing better now i keep thinking that guy is lebanese
maybe turkish or syrian
thorn bramble
An object shall have its stored value accessed only by an lvalue expression that has one of the
following types:99)
— a type compatible with the effective type of the object,
— a qualified version of a type compatible with the effective type of the object,
— a type that is the signed or unsigned type corresponding to the effective type of the object,
— a type that is the signed or unsigned type corresponding to a qualified version of the effective
type of the object,
— an aggregate or union type that includes one of the aforementioned types among its members
(including, recursively, a member of a subaggregate or contained union), or
— a character type.
technically it should be valid to alias a struct pointer using a pointer of the type of any of its fields? or im just misunderstanding because im tired
yeah i might be misunderstanding, not sure what "object" refers to in this case
surreal path
progress so far: i learned a lot more about lifetimes and rusts safety system, im currently writing simple userspace programs with extensive use of rusts advanced features to understand the language well
i dont know the eta on when ill be able to start rewriting the kernel but lets just keep going and see how we go
thorn bramble
mate u dont need to rush
take it easy and learn it properly
dont take shortcuts in your little programs like you did with the kernel
surreal path
i wont
thorn bramble
nice, good luck
surreal path
thanks !
surreal path
great idea
ill do that once i finished the book qookie gave yesterday
turns out linked lists are very useful for learning rust as a language lmao
surreal path
update today
so after finishing the linked list booklet and after properly finishing the rust book i believe im ready to try another rewrite, ive already done well without using static mut and instead wrapping a static variable under a mutex making it "safe", im using lifetimes and generics and shit as well. i never cast a pointer from pointer -> address -> pointer anymore. and i am always compiling in release mode
i am experiencing a problem with my own rust flanterm bindings however
= note: rust-lld: error: /home/Rayan/Desktop/NyauxKT/kernel/target/x86_64-unknown-none/debug/deps/libflanterm_bindings-63cfa04b5ca34f3f.rlib(47dd3966b7af1aa8-fb.o):(function flanterm_fb_init: .text.flanterm_fb_init+0x186e): relocation R_X86_64_32 out of range: 18446744071562110592 is not in [0, 4294967295]; references section '.rodata.builtin_font'
>>> referenced by fb.c:1018 (backends/fb.c:1018)
i dont know if this is a problem with my bindings or
if this is a problem with flanterm on their end
surreal path
no >:)
finite summit
tawdry mirage
```rs
= note: rust-lld: error: /home/Rayan/Desktop/NyauxKT/kernel/target/x86_6...
missing -mcmodel=kernel when compiling flanterm?
surreal path
missing `-mcmodel=kernel` when compiling flanterm?
let me check
surreal path
missing `-mcmodel=kernel` when compiling flanterm?
cargo:warning=cc1: error: code model kernel does not support PIC mode
i cant disable pic
it wont let me
finite summit
use C, that's the solution
tawdry mirage
why can't you disable pic?
flat nymph
It's easier to learn Rust if you know C++
finite summit
it's easier to not learn rust
it's like a parasite that once in your brain, is fatal
tawdry mirage
nice
surreal path
i got the gdt and idt working too
in fact let me test that out
with a page fault lol
done
works well
no static mut
barely any unsafe
time for the PMM
surreal path
pmm done
wide nexus
thorn bramble
@surreal path is your code up on github?
surreal path
<@787341689333350450> is your code up on github?
soon
edgy pilot
oh 3rd rewrite already?
good luck
I started mine and basically stopped
In a day
desert haven
howcome you stopped?
edgy pilot
no motivation
desert haven
fair enough
surreal path
after religious testing of the PMM in rust release mode, the PMM is fine and can remain unchanged mostly (other then the removal of static mut for static behind a mutex)
kind root
did u finish the nes emulator already 
surreal path
kinda got it working to a point where it could parse opcodes and execute them
i didnt implement mappers or the apu or anything tho
kind root
why do u just speedrun everything 
writing one thing u can be proud of is so much cooler than many shitty things
surreal path
im not into writing emulators that much to be fair
kind root
i think its kinda cool when u just write a generic thing that executes opcodes and then all of a sudden u can run other peoples stuff on that code
but yeah i get it
surreal path
nyaux might become assert os
like managarm
i just learned i can use asserts OUTSIDE of tests in rust
i am so happy about this information
kind root
remember to use asserts only for state corruption detection, not invalid external data
surreal path
remember to use asserts only for state corruption detection, not invalid externa...
wym
thorn bramble
assert!(pmm_is_ok)
kind root
^
thorn bramble
not assert!(this_user_pointer_isnt_null)
surreal path
oh gotcha
kind root
or not assert(this_hw_register_returned_0)
thorn bramble
extrapolate this to the entire kernel, but yeah
that too
basically use asserts as sanity checks
thorn bramble
assert!(@surreal path pushed his code to github)
surreal path
the PMM has remained unchanged but the VMM is probs prone to a lot of bugs
surreal path
assert!(<@787341689333350450> pushed his code to github)
let me rewrite the VMM
first
lmao
elder shoal
or not `assert(this_hw_register_returned_0)`
nah go full managarm assert everywhere you gotta assert bro just assert please one more assert please just one more assert and we'll have phd code please bro just one more assert
surreal path
nah go full managarm assert everywhere you gotta assert bro just assert please o...
😭 😭
managarm os
more like
assert_eq(managarm)
okay that sounded better in my head
💀
edgy pilot
after religious testing of the PMM in rust release mode, the PMM is fine and can...
religious testing
hollow goblet
explain this atheists
surreal path
vmm status: nearly complete
page table manager part of it is done
im working on the region setup function
thorn bramble
put the code up on github !! ! ! !
also the 6502 emu too
i wanna see how you are doing
thorn bramble
cool stuff
orchid dawn
WE GOT THE VMM WORKING
already better than Techflash OS
nice job
surreal path
i am gaming
alright lol
kind root
what was the bug
kind root
did qwinci fix it
surreal path
no
kind root
qookie?
kind root
surreal path
YESSSS
thorn bramble
alright lol
fortnite with the boys
surreal path
fortnite with the boys
less goo
molten grotto
nyaux nt compatibility when 
surreal path
nyaux nt compatibility when <:troll:796858892985434152>
WE GOTTA DO THAT SOON YESSS
finite summit
WE GOTTA DO THAT SOON YESSS
100%
thorn bramble
instead of nyauxrs
surreal path
😭
surreal path
😭 😭
surreal path
so when nyuax working-set vmm
working set?
finite summit
set of pages required to be paged in for an application to run as efficiently as possible
in short
surreal path
😭
molten grotto
just don't support swap yea lol
surreal path
lol
thorn bramble
who swaps anyway
just buy more ram
and let programs oom instead
clearly the better choice
at least you dont risk swapping to death
elder shoal
Im not too far away from supporting swap in astral Id imagine, shouldnt be too different from a file backed page
finite summit
#1061407633745125397 message
finite summit
Im not too far away from supporting swap in astral Id imagine, shouldnt be too d...
wrong
at least
maybe the swap in/out part can be implemented like that
but to know what you'd swap out, you'd probably want a working-set, and a page replacement algorithm (such as aging) to go with it
finite summit
but note that swap in/out doesn't always go to disk
it also uses standby/dirty
*it also can use
lemme find something
#1141057599584878645 message
surreal path
lets go
surreal path
Yippie :3
surreal path
same as map but just clear the entry
i lit do
pub fn unmap(&self, va: usize) {
let him = unsafe { Self::find_pte(self.rootpagetable as usize, va) };
unsafe { him.write(0) };
println!("{:#x}", unsafe { him.read() });
unsafe {
core::arch::asm!("invlpg [{x}]", x = in(reg) va, options(nostack, preserves_flags))
};
}
unsafe fn find_pte(mut pt: usize, va: usize) -> *mut usize {
let mut shift = 48;
for i in 0..4 {
shift -= 9;
let idx = (va >> shift) & 0x1ff;
let ptab: *mut usize = virt(pt);
if i == 3 {
return ptab.add(idx);
}
let entry = ptab.add(idx);
if entry.read() & VMMFlags::KTPRESENT.bits() == 0 {
return entry;
}
pt = entry.read() & 0x000f_ffff_ffff_f000;
}
unreachable!()
}
BUT IT DOESNT WORK
NO PAGE FAULT
edgy pilot
wdym doesn't work
edgy pilot
write to cr3 again
surreal path
oh
surreal path
write to cr3 again
ohh is that it
surreal path
yes
edgy pilot
it shouldn't
surreal path
oh
okay lets try
error: invalid operand for instruction
--> src/mem/vmm/mod.rs:187:31
|
187 | core::arch::asm!("invlpg {x}", x = in(reg) va, options(nostack, preserves_flags))
| ^
|
note: instantiated into assembly here
--> <inline asm>:2:2
|
2 | invlpg r14
| ^
that might have been correct
edgy pilot
does this dereference? ```invlpg [{x}]```
this
surreal path
i do
edgy pilot
and it does unmap
surreal path
it doesnt
edgy pilot
it just doesn't flush the tlb
surreal path
no i need to fix this
edgy pilot
bruh
surreal path
wait i have an idea
edgy pilot
what's your idea
edgy pilot
what was your idea
surreal path
what was your idea
my idea was to run in debug
kind root
nyaux dont get stuck for 10 seconds impossible challenge
surreal path
nyaux dont get stuck for 10 seconds impossible challenge
😭
yea im actually losing it
kind root
how long until u give up
surreal path
NEVER
kind root
edgy pilot
elder shoal
I cant say much because I was stuck for like hours all because I was calling the wrong gcc builtin (__get_cpuid instead of __get_cpuid_count)
thorn bramble
how long until u give up
he was close to giving up a few months ago
but he miraculously came back
kind root
I cant say much because I was stuck for like hours all because I was calling the...
one thing i learned also is add a check into your panic handler that its not called recursively
that will save you hours
kind root
but he miraculously came back
miraculous save by qwinic
surreal path
miraculous save by qwinic <:meme:575445094589661215>
wdym
kind root
didnt he like solve tons of bugs for u
finite summit
ILL NEVER GIVE UP
elder shoal
qwinci can you fix ps/2 in astral real harwdwaire for me next pls pls plsss 🥺🥺🥺🥺🥺
finite summit
Astral challenge: unfuck ps/2
challenge level, impossible
obos challenge: uncurse
challenge level: locked
surreal path
qwinci pwtty please 🥺 🥺 🥺 fix this dumb ah bug for me plss 🥺
kind root
obos challenge: uncurse
challenge level: locked
obos challenge: run at least for 2ns before double faulting
challenge level: insanity
astral challenge: find one pc (other than mathewnd's hardware) where ps2 works
challenge level: cock torture
nyaux challenge: unmap a page
challenge level: months of debugging by qwinci
elder shoal
find three pcs where ps2 works*
it works on my laptop and desktop
but ONLY ON MY DESKTOP AND LAPTOP
kind root
corrected
elder shoal
soon I'm gonna be running webkitgtk with 128 cores before ps2 works
kind root
lmao
finite summit
obos challenge: run at least for 2ns before double faulting
challenge level: ins...
ok I don't triple fault
I only panic
I usually only double fault
at max
kind root
corrected
thorn bramble
i got windows to double fault only once in my entire life
by accident that is
not very hard to do deliberately
so you should probably work on that
finite summit
I don't double fault often
only a couple times during the fireworks test because of a vmm bug
but other than that, I don't really crash that much
only hangs mainly
median goblet
set of pages required to be paged in for an application to run as efficiently as...
nop
the working set list is the list of virtual address that have valid PTEs
on x86(_64), a valid PTE is any PTE with the present bit (0) set
finite summit
the working set list is the list of virtual address that have valid PTEs
That is what the denning working set paper said iirc
unkempt relic
the working set list is the list of virtual address that have valid PTEs
this is just a concrete operationalisation of what denning proposed
finite summit
That is what the denning working set paper said iirc
As in, what I said was from that paper
finite summit
this is just a concrete operationalisation of what denning proposed
So you're saying that only I'm doing it that way
unkempt relic
So you're saying that only I'm doing it that way
i also implement it that way
finite summit
Ok good to know
median goblet
this is just a concrete operationalisation of what denning proposed
This is the way I understand working sets
flat nymph
and let programs oom instead
Does managram have swap/working set?
haughty kite
probably not
flat nymph
(I'm wanting/trying to implement it in my OS, but I'm stuck with locks)
thorn bramble
how did managarm get involved in this whole thing lol
no managarm doesn't have swap
magic kernel
Does managram have swap/working set?
no
flat nymph
So OBOS and soon NyauxKT >>> Managram?
thorn bramble
managarm works by allocating memory and praying there is enough physical memory to cover the allocations
quite literally
thorn bramble
So OBOS and soon NyauxKT >>> Managram?
in that regard, sure i guess
haughty kite
managarm works by allocating memory and praying there is enough physical memory ...
i wouldnt call it praying
you can call it an assert its okay
thorn bramble
it literally asserts if there is no free physical memory
elder shoal
managarm works by allocating memory and praying there is enough physical memory ...
Skill issue, real posix oses just sigbus
flat nymph
it literally asserts if there is no free physical memory
What happens in release mode?
haughty kite
and let programs oom instead
i hear thats great for reliability :^)
thorn bramble
What happens in release mode?
it.. asserts
probably
idk
compile and test it yourself if you wanna know
flat nymph
I mean asserts are supposed to be removed when not debugging..?
haughty kite
I mean asserts are supposed to be removed when not debugging..?
no its managarm
asserts are forever
median goblet
it.. asserts
assertions are typically removed in non debug builds. Do you not have a "release" mode?
thorn bramble
idk
i'm not a managarm developer lol
fwiw tho, iirc they use meson debugoptimized
so it's debug release :)
elder shoal
Managarm without asserts is like an angel without its wings
thorn bramble
TRUTH.
magic kernel
it literally asserts if there is no free physical memory
I was working on bubbling that up. Soon™️ that shall no longer be the case
It’s even a draft PR (699)
But that just returns a nullptr and that then needs to be handled back up the stack, that part is still lacking. I’m pretty sure I left for summer vacation and I just drafted it in
thorn bramble
that doesn't mean you don't have UB in your code :)
just the parts using references don't have UB
creating a reference still might be UB
surreal path
THEN HOW DO U MAKE SOMETHING COMPLETELY NOT UB 😭
surreal path
fair
hollow goblet
creating a reference still might be UB
creating a reference is ub unless the normal requirements for references also holds up for the pointer
thorn bramble
yep
hollow goblet
and with strict provenance, integer -> pointer casts are UB which are kinda required for like PA->VA conversion
surreal path
and with strict provenance, integer -> pointer casts are UB which are kinda requ...
i do not have strict provenance enabled
and i will keep it disabled
hollow goblet
i believe MIRI runs stuff with strict provenance or something like that? I don't actually know much about provenance lol
even if it's not exactly strict provenance, MIRI will be REALLY MAD at creating pointers from integers
surreal path
who cares abt what miri thinks
hollow goblet
anyhow if you really wanna "prove" your code is UB-free, extract pieces of it to run in miri
surreal path
im just gonna use references everywhere anyway
hollow goblet
you can mess shit up with references too
hollow goblet
it's not easy to check if you're correct without the right tools
it would be really cool to do a unit test + miri setup of sorts
thorn bramble
i know what im doing tho so
surreal path
slab allocator and pmm done in safe rust
should i take a break
for an hour
or vmm now
thorn bramble
slab isnt safe rust*
surreal path
slab isnt safe rust*
wha
thorn bramble
it’s not
thorn bramble
well it's fucking not!
surreal path
okay chill iretq 😭
thorn bramble
you still use raw pointers
surreal path
ONLY A FEW TIMES
thorn bramble
dereferencing them into references doesnt make it safe
lol
u took ur code
and everywhere u made it into references
that doesnt make it safe
its still the same inherently broken algorithm
surreal path
okay so how ELSE am i supposed too write it
surreal path
its still the same inherently broken algorithm
how is the algorithm broken huh
thorn bramble
ur creating pointers from thin air
surreal path
okay so how ELSE am i supposed too write it
explain @thorn bramble TELL ME
thorn bramble
i just did
surreal path
ur creating pointers from thin air
and how else
am i supposed to write it
then taking pointers from "thin air"
surreal path
bro 💀
thorn bramble
without all this shit
surreal path
whatever man as long as the UB is reduced so it doesnt have weird ah bugs its fine
thorn bramble
you didnt reduce anything
you just asked rust to do borrow checking on raw pointers xd
thorn bramble
yeah as i said yesterday
u did rewrite some parts to be more proper or whatever
few less parts to worry about i guess
hollow goblet
then taking pointers from "thin air"
you can only create pointers from pointers
if you really want it to be 100% safe
surreal path
look i give up rust
hollow goblet
but you're doing kernel work, just assume it's gonna be fine™️
hollow goblet
but you're doing kernel work, just assume it's gonna be fine™️
it will be fine dw
hollow goblet
it will be fine dw
it's basically the same thing as being careful with certain stuff and marking stuff as volatile in C
just don't get it optimized out
finite summit
finally, you have restored normal brain function
hollow goblet
and it's fine
hollow goblet
just don't get it optimized out
or reordered
edgy pilot
yep thats what im gonna do
hollow goblet
rust my beloved
finite summit
boo
surreal path
https://tenor.com/view/my-beloved-programming-cpp-cxx-cplusplus-gif-164713923688...
i dont c++
know
c++
edgy pilot
learn
finite summit
rust my beloved
Rust user detected, opinion rejected
surreal path
even if i already know C
hollow goblet
rust my beloved
i enjoy writing a kernel in rust, it's definitely not something for everyone
edgy pilot
its gonna take months
it's going to be easier to learn c++ coming from c than rust
finite summit
just do C
surreal path
it's going to be easier to learn c++ coming from c than rust
u sure?
edgy pilot
yes im sure
finite summit
C is easier anyway, no vtables to worry about
hollow goblet
Rust user detected, opinion rejected
:(
surreal path
yes im sure
okay ill try it out then
hollow goblet
i don't get why so much hate on rust here
finite summit
you'll end up learning C++ then ending up being confused as to why some stuff don't work
or write C-like C++ code and get flamed
hollow goblet
i don't get why so much hate on rust here
i believe as long as something's main selling point is not blazing fast rust 🚀 it's ok
edgy pilot
not all C++ features can be used from kernel mode easily
yeah I kinda forgot about the "in kernel easily" part
I have most of the things already set up and configured so I don't have to worry about most stuff
hollow goblet
i believe as long as something's main selling point is not blazing fast rust 🚀 ...
like why would you even care about the language you're using other than comfort
i am most comfortable with rust
edgy pilot
finite summit
still doesn't give you some runtime support
(a lot of)
also global constructors are wacky
hollow goblet
does C++ have an equivalent of panic=abort
finite summit
and to be avoided
edgy pilot
does C++ have an equivalent of panic=abort
abort()
terminate()
finite summit
in kernel mode
hollow goblet
abort()
terminate()
ok so C++ exceptions are not usable in freestanding whatsoever?
edgy pilot
you can port llvm or gcc libs, it's not hard
I have done it before with both
a few years ago
hollow goblet
anyhow