sleep :^)
#uACPI - a portable and easy-to-integrate ACPI implementation
vast kestrel
fiery turtle
well these are from netbsd right
they arent part of the C memory model
like as far as I understand, anything not adhearing to the C model works only because of implementation details
vast kestrel
basically
and I guess technically if you choose std=gnuXX it could decide on a different memory model
but I doubt it is ever formalized lol
flat badge
i mean, adding a r-r and w-w barrier for seqlocks is probably fine
like netbsd does according to the docs
but for the majority of the code i'd stick to the C11 memory model
as it's the de-facto standard now
also, it makes sense, it can be implemented efficiently and it was actually designed around existing implementations instead of just growing over time (as the linux model did)
if Linux was started today, it'd also use mostly the C11 model
fiery turtle
netbsd is having membar_producer/consumer() for these
i might steal producer/consumer tbh as this is exactly what i need i think
thanks for the link
is netbsd available on bootlin?
damn i think not
it only has freebsd
ok lol netbsd just compiles it to a no-op on x86
whats interesting is i dont see a compiler barrier being inserted there
or is function call itself considered a compiler barrier?
ARM does DMBST
but also no compiler barrier
very strange
flat badge
actually on aarch64 there is no difference between r-rw and r-r
afaict
it's both dmb ld isn't it?
@loud ice ^
fiery turtle
is function call considered a compiler barrier?
jaunty fox
is function call considered a compiler barrier?
yes, if it's not inlined
flat badge
or i guess dmb ishld
fiery turtle
yes, if it's not inlined
ah ok then it makes sense
does netbsd not do LTO?
although assembly cant be ltoed probably
flat badge
the easiest compiler barrier is asm volatile ("" : : : "memory");
fiery turtle
the easiest compiler barrier is `asm volatile ("" : : : "memory");`
yeah i know, i was talking about netbsd, it implements barriers in separate assembly files
jaunty fox
lto is not very popular for kernels
fiery turtle
no C code at all
flat badge
but i'd argue that you shouldn't rely on compiler barrier anyway except for very specific stuff (such as per-CPU atomics maybe)
jaunty fox
it's pure poison to observability and debuggability
fiery turtle
but i'd argue that you shouldn't rely on compiler barrier anyway except for very...
well if you define your own barrier you have no way of telling the compiler that it is one unless u use one of these
or a non-inline function call i guess
flat badge
yes that's true
fiery turtle
it's pure poison to observability and debuggability
true
but it gives u free 5-10% perf usually out of nowhere
flat badge
actually on aarch64 there is no difference between r-rw and r-r
so the only mainstream arch where acquire and r-r are different at the hw level is riscv/mips, i guess?
that's probably also why the C11 model has no equivalent of r-r and w-w
fiery turtle
its also very easy to make a mistake with r-r
like u have to actually understand what you're doing insanely well
flat badge
yeah r-r and w-w don't compose very well with stuff around it
since they do not establish transitive orderings with other accesses
fiery turtle
thats why smp_rb and friends suck
flat badge
like, for a acquire-release pair on the same var, you can be sure that everything since it doesn't escape it
but that can't be emulated with r-r and w-w
fiery turtle
like im sure that there are a lot of places in linux drivers using smp_rb expecting r-rw
which it is on a lot of arches as well
torpid root
but it gives u free 5-10% perf usually out of nowhere
that's why I use it for uacpi points optimised builds 
fiery turtle
probably very few people writing simple drivers understand the diff at all
fiery turtle
that's why I use it for uacpi points optimised builds <:meme:575445094589661215>
thats the way
flat badge
the most cursed assumption of the linux model is that plain accesses are atomic if aligned properly
fiery turtle
or i guess `dmb ishld`
loud ice
lto is not very popular for kernels
i think XNU is LTO'd
fiery turtle
i mean for production builds why not LTO
free points and you're not gonna be debugging it
jaunty fox
i think XNU is LTO'd
they won't be dtracing it then
loud ice
because developers are bad at writing code lol
vast kestrel
^^
loud ice
but linux is not, and windows has special handling from msvc because of their UB
fiery turtle
ofc
loud ice
nvm then
fiery turtle
linux has supported lto since forever
vast kestrel
the build system supports it
vast kestrel
I think the perf increase is minimal in real world benchmarks
fiery turtle
nah they managed to make it work
loud ice
okay neat
fiery turtle
but yeah
vast kestrel
at least for most common drivers
jaunty fox
i am influenced by bryan cantrill's philosophy around performance engineering
loud ice
yeah r-r and w-w don't compose very well with stuff around it
when do you even need r-r?
on x86 its a compiler fence
on arm you dont want a fence you want a consume ordering
jaunty fox
he couldn't stand things like -fomit-frame-pointer because it impeded observability
fiery turtle
he couldn't stand things like `-fomit-frame-pointer` because it impeded observab...
which is disabled by default on new ubuntus iirc(?)
loud ice
on arm you dont want a fence you want a consume ordering
aka ```c
uint64_t magic_zero(uint64_t input) {
uint64_t output;
asm("eor %w[out], %w[in], %w[in]"
: [out] "=r"(output)
: [in] "r"(input));
return output;
}
uint64_t dependent_load_after(_Atomic(uint64_t)* ptr, uint64_t previous_load_result) {
return atomic_load_explicit(ptr + magic_zero(previous_load_result), memory_order_relaxed);
}
webkit uses it anyway (through custom magic)
jaunty fox
it's useful for RCU and other techniques
loud ice
yeah
jaunty fox
there you can usually tolerate simply assurance that dependencies are respected, and full acquire is a worse penalty to pay
hollow elm
consume got deprecated for a reason <:trl:829316641191886888>
didnt it get undeprecated?
fiery turtle
well cppref says deprecated in c++26
fiery turtle
webkit uses it anyway (through custom magic)
they have their own compiler?
loud ice
they have their own compiler?
no, they use clang
loud ice
aka ```c
uint64_t magic_zero(uint64_t input) {
uint64_t output;
asm("eor %w[...
this is pretty much their impl lol
fiery turtle
browsers are hard
jaunty fox
@fiery turtle will ultra os use rcu/smr techniques?
fiery turtle
i mean ofc, how can u avoid RCU in a serious kernel
flat badge
on arm you dont want a fence you want a consume ordering
consume is even harder to reason about though
jaunty fox
i mean ofc, how can u avoid RCU in a serious kernel
you can still get quite far without
dragonfly is replicating or partitioning state between cores, to great effect
flat badge
Managarm's kernel doesn't use RCU yet
fiery turtle
what is SMR btw?
flat badge
we will at some point introduce RCU though
left orbit
shingled magnetic recording
loud ice
Managarm's kernel doesn't use RCU yet
managarm is a microkernel so its not as important
flat badge
since the handle -> descriptor maps are quite costly without RCU
loud ice
it's also not a serious kernel
fiery turtle
shingled magnetic recording
real
loud ice
since the handle -> descriptor maps are quite costly without RCU
can you not do fine grained locking on those?
flat badge
just need to find somebody to implement it who doesn't have IP conflicts
jaunty fox
what is SMR btw?
both a general term for the approach of 'safe memory reclamation' (i.e. what you have to do when you want to do lockess things but haven't a tracing GC) as well as a sometimes term for a specific implementation of the approach invented in freebsd and now also used in macos and cpython
fiery turtle
both a general term for the approach of 'safe memory reclamation' (i.e. what you...
ah
slim panther
Managarm's kernel doesn't use RCU yet
what does that tell you about managarm
fiery turtle
yeah first i need to get over the log ring and then implement the rest of the kernel 
flat badge
can you not do fine grained locking on those?
yeah that'd be possible
jaunty fox
~~what does that tell you about managarm~~ <:meme:575445094589661215>
they've at least prepared for it, there's an rcu ready radix tree in libfrigg
but no reclamation mechanism yet
vast kestrel
I have a friend who made a toy rcu implementation on his own, using the membarrier syscall
jaunty fox
~~just need to find somebody to implement it who doesn't have IP conflicts~~
i just followed the original patent and paper, every jot and tittle
vast kestrel
fiery turtle
there's also a syscall which makes it possible to detect when you have been preempted iirc?
jaunty fox
it's expired now
vast kestrel
there's also a syscall which makes it possible to detect when you have been pree...
rseq I think
fiery turtle
yeah that
vast kestrel
on preemption it moves your rip to a pre-configured place or something alike
so you can handle a preemption path
jaunty fox
it worked nicely for me because i am accustomed to working at dispatch IPL which is straightforwardly equivalent to a Classic RCU read section. but if i had to do threaded RCU i would be in a more dificult position since i think Sleeping RCU is still under patent
vast kestrel
my friend used the membarrier to force a fence on the other threads of the process without the reader needing a hw fence
hollow grotto
it worked nicely for me because i am accustomed to working at dispatch IPL which...
NT implements Sleeping RCU
Maybe the patent has already expired
loud ice
or the engineers decided that if they implement it quickly enough the legal team cant stop them 
or they licensed it
jaunty fox
NT implements Sleeping RCU
do they implement the interface alone or is the implementation the same too?
of note IBM granted that liburcu is allowed to use sleeping RCU and be under LGPL
so in principle you could work with that and share your modifications to liburcu alone
and paul mckenney himself is the instigator in chief of the project to put something which exposes an interface very like Sleeping RCU into C++'s standard, so implementations will soon multiply
loud ice
so in principle you could work with that and share your modifications to liburcu...
"yeah so what we did is we removed all of liburcu and then reimplemented it but its totally the same project, trust"
the lgpl isnt very clear
but the wording may imply that the patent license is granted for everyone
chrome crow
do they implement the interface alone or is the implementation the same too?
how does it work for patents ? if you change some details , can it be considered a different implementation which doesnt violate anything?
jaunty fox
how does it work for patents ? if you change some details , can it be considered...
no one really knows, there's nothing so totally wanton and ghastly as the world of software patents
legally it's just a mess, terrible
the basic rule is how much financial clout you can bring determines how much they mean anything
loud ice
no one really knows, there's nothing so totally wanton and ghastly as the world ...
the GPL has a patent grant clause
kinda vaguely
which makes this even more complicated
jaunty fox
and the only observable effects of that whole rotten system is a nice numerical KPI that makes managers happy and vast amounts of terrible wastes of time in courts litigating by unpleasant firms
loud ice
i suspect IBM wont sue you though
jaunty fox
i suspect IBM wont sue you though
i think the FSF and maybe others would blow their lid if someone were sued for legitimately and peacefully forking an lgpl project then someone comes in and sues them over it
it would be complete subversion of the licence
loud ice
the LGPL has a (kinda bad) patent grant clause
the question is how that works together with ship of theseusing the library
jaunty fox
the question is how that works together with ship of theseusing the library
it would make a very interesting study for a lawyer, but in the end if IBM comes knocking then you give over
loud ice
i mean its ms
so
given that ms used it i suspect ms legal said its fine
which means they think that either ibm wont sue, they have a licensing deal or they do not infringe
lucid sleet
they won't be dtracing it then
pitust is right, release builds are LTOd by default (but it's disabled for debug and kasan builds)
trim frost
is it normal to get 1M more uacpi points withOUT -M q35 on qemu?
i thought that made the os faster
left orbit
see the amount of opcodes it has to process
the average amount of opcodes/s also depends on what kind of opcodes it's running
if you try it on real hw you will notice that it's probably much slower
because it's running ops which are just more expensive than what's in qemu's q35 machine aml
left orbit
trust with what?
fiery ice
what is this function used for?
left orbit
if you want to compare some arbitrary number against other kernels then yes, use q35
but ops/s are in no way a performance measure
left orbit
what is this function used for?
it Returns the PHYSICAL address of the RSDP structure via *out_rsdp_address
what more do you need?
slim panther
shouldn't the kernel already know that though?
slim panther
or is this kernel_api?
left orbit
it's kernel api bro
fiery ice
kernel_api
slim panther
bro
fiery ice
with?
slim panther
the rsdp?
slim panther
what kinda question is that
fiery ice
set the pointer to the rsdp?
slim panther
rtfm
fiery turtle
And I thought I had good comments for the kernel api 💀 
left orbit
it is good tho
slim panther
it is
left orbit
or at least sufficient
fiery ice
i just needed a confirmation
fiery ice
thx
left orbit
of course it's not 10 lines of information describing how every bit of that output variable should be set
fiery turtle
Lmfao
left orbit
but "put RSDP there" should be enough
slim panther
i just needed a confirmation
uacpi will tell you if it worked
fiery turtle
Use the C assignment operator
Or a return expression
To propagate an integer value to uacpi
fiery turtle
uacpi will tell you if it worked
Either it triple faults or not
slim panther
nah i think if you set the rsdp to garbage it just complains that it cant find correct signatures
fiery turtle
True
trim frost
left orbit
this assumes you have proper exception handling set up
which apparently is a lot to ask from many people in this server
fiery turtle
this assumes you have proper exception handling set up
I think he just failed an assertion
So it didn't crash in fact
slim panther
panic!
trim frost
trim frost
assert dominance
trim frost
why does it start with uacpi_status_?
slim panther
bindgen
trim frost
i use bindgen too
slim panther
show your script
trim frost
fiery turtle
x86 is not reduced hw tho
fiery turtle
>prepend_enum_name
Why do u do it btw
slim panther
it's gated
slim panther
Why do u do it btw
?
fiery turtle
Prepend
fiery turtle
Ah
All public facing api is already prepended so
And even internal api that has global visibility
I try my best not to pollute namespaces
trim frost
x86 is not reduced hw tho
i "borrowed" uacpi-rs's script and edited it to my liking
slim panther
All public facing api is already prepended so
i know i just kinda accepted it and thought there was no way of fixing it
trim frost
it has the reduced hardware thing
slim panther
reduced hw is cringe
slim panther
fiery turtle
u also have barebones mode enabled
how do u even get it to load namespace and stuff
fiery turtle
that code doesnt exist in barebones mode
slim panther
conditional compilation
trim frost
show proof or didnt happen
torpid root
so i should trust -M q35 more
if you mean for the leaderboard, that requires -M q35
trim frost
remove-talc?
slim panther
yes
trim frost
why
slim panther
shit
fiery turtle
also im gonna have my 9950x3D soon so ill probably start doing measurements for the leaderboards myself
trim frost
also im gonna have my 9950x3D soon so ill probably start doing measurements for ...
nice
slim panther
you can't do real phys allocations with talc
trim frost
what do you use instead? or did you make your own
slim panther
my own
torpid root
why use talc when you can write your own allocator, find out frigg is way faster and switch to it
slim panther
fuck dependencies
fiery turtle
then find out proxima is faster and use that
slim panther
i don't want to make a lib os
trim frost
yeah write your own bootloader
slim panther
then find out proxima is faster and use that <:trl:829316641191886888>
proxima might as well be using a bump allocator with those speeds
trim frost
scratch that make your own cpu and motherboard so everything is ur own
torpid root
then find out proxima is faster and use that <:trl:829316641191886888>
I was just looking at proxima's allocator :^)
slim panther
yeah write your own bootloader
bootloader != os
fiery turtle
proxima might as well be using a bump allocator with those speeds
thats the thing, a bump allocator is slower
trim frost
dont have to depend on anyones microcode
slim panther
thats the thing, a bump allocator is slower
WHAT
fiery turtle
I was just looking at proxima's allocator :^)
look at the pre-magazines version, its even simpler
same speed
trim frost
then how is proxima so fast
torpid root
magic
slim panther
actually, i haven't looked into it, what allocators does proxima use
for page + malloc
trim frost
or just divide the time given to uacpi by 10 and you will have 10x the ops/s
fiery turtle
then how is proxima so fast
monkuous is really good
fiery turtle
actually, i haven't looked into it, what allocators does proxima use
literally 144 lines
torpid root
i strive to be that
powder that makes you say real:
fiery turtle
100-150 ns per allocation
trim frost
i like how frigg uses mdbook
slim panther
good software
torpid root
powder that makes you say real:
the question is how to become that
trim frost
should i make frigg bindings?
torpid root
no
fiery turtle
just make your own allocator ffs
torpid root
imagine making your own allocator
imagine being based like that
fiery turtle
i cant because im gonna be stealing proxima's design
torpid root
imagine being based like that
instead of copying someone else's c++ code, pasting it in rust and just changing syntax
trim frost
sometimes i make bindings
fiery turtle
Ask chatgpt to convert frigg into rust
trim frost
Ask chatgpt to convert frigg into rust
nah im good
torpid root
lmao
torpid root
sometimes i make bindings
when the code you need to copy is too big
trim frost
well you use flanterm and uacpi too, i just make it accessible in rust
mortal yoke
also you can't really make good bindings for frigg anyway because its header only with templates
slim panther
imagine using flanterm
trim frost
imagine using flanterm
what do you use?
slim panther
wdym errors
fiery turtle
Errors
trim frost
also what does your build script do? because i feel like im compiling llvm
trim frost
bootsrap
torpid root
take a wild guess
uses gcc rust
trim frost
no you dont
slim panther
how so
trim frost
i use bindgen without it bruh
slim panther
wdym
slim panther
bindgen requires libclang
trim frost
you dont have to build it yourself
torpid root
I always read binder instead
slim panther
you dont have to build it yourself
i do because i'm doing cross compilation too in that script
and debian builds are retarded
trim frost
i use arch btw
trim frost
yes
fiery turtle
ever heard of a container
What if he also uses arch in the container
trim frost
right?
slim panther
What if he also uses arch in the container
over which he has no control
loud ice
arch best distro
fiery turtle
over which he has no control
Arch do be like that
fiery ice
what changes between sleep and stall?
torpid root
stall is just spin-pause
sleep is thread sleep
frank canopy
stall is meant to be a busyloop yeah
fiery ice
can i pass a spinlock to this?
loud ice
can i pass a spinlock to this?
no
slim panther
mutexes are not spinlocks
fiery ice
oof
frank canopy
i think some people try but yeah no not the same
torpid root
mutex blocks the thread instead of spinning
trim frost
since no_std
loud ice
i use spin::Mutex and spin::SpinMutex
thats not a mutex
trim frost
im aware
loud ice
also i wouldnt use them
slim panther
i use spin::Mutex and spin::SpinMutex
for me it's a noop but i will put a reschedule hint there
i'll make my own locks
fiery ice
but uACPI is not running on a thread for me
trim frost
but uACPI is not running on a thread for me
same, thats why what i do works the way i do it
trim frost
@slim panther
frank canopy
i run uacpi and all that stuff on the main thread
trim frost
when it takes 10 minutes to build a kernel that isnt linux, you know something is wrong
trim frost
still
slim panther
okay fixed it
trim frost
mine took 14 seconds lol
slim panther
well no offense but you probably don't have that many programs then
trim frost
ik
slim panther
try now
left orbit
menix kernel takes like 5 seconds to build
slim panther
he was talking about bootstrap
trim frost
why am i building linux-headers
left orbit
if ur userspace takes 14 seconds to build then idk what kind of userspace it is
because mlibc wants them
if you enable the linux option
slim panther
why am i building linux-headers
i support some linux interfaces
like reboot, mount and so on
fiery ice
what is ctx here?
slim panther
it literally says in the comment
fiery ice
ctx is passed to the provided handler
slim panther
yes
slim panther
hover over it
trim frost
is my cpu cooked chat?
torpid root
what actually is ctx?
whatever uacpi puts in there
fiery ice
it's a pointer
torpid root
you just pass it along to the function
fiery ice
huh?
torpid root
what don't you understand?
trim frost
this is my weird rusty way to do it
fiery ice
like, rn my handler is just a pointer to the function
torpid root
no
fiery ice
torpid root
you make your own handler that calls that function
fiery ice
hmm
fiery ice
this is my weird rusty way to do it
why only 1 handler fn?
trim frost
wdym?
slim panther
confusion
fiery ice
so it just installs 1 interrupt handler?
trim frost
yeah
fiery ice
ohhh
slim panther
what else is it supposed to do
fiery ice
that helps a lot, thx
fiery ice
what else is it supposed to do
i tought it could request more than 1
fiery ice
yes
torpid root
technically it should support installing more than one IRQs, but other than SCI, what would uacpi request to install?
slim panther
that's one (1) value
fiery ice
but from what George is doing won't that get overwritten the next call?
torpid root
yes
trim frost
yes
torpid root
but there won't be a next call
slim panther
i think even if so, uacpi wouldn't do it on the same irq
fiery ice
yeah but still, George's code won't work anyway
trim frost
yeah but the handler_fn and ctx will get overwritten
torpid root
yeah but still, George's code won't work anyway
wdym won't work?
trim frost
in my code
fiery ice
wdym won't work?
gets overwritten
slim panther
which is fine
fiery ice
so it can only actually support 1 irq install
trim frost
if it works dont touch it :]
slim panther
i have irq local context pointers
torpid root
so it can only actually support 1 irq install
there's only one irq
slim panther
so i can just store one ctx per irq
torpid root
apparently you can't use a "closure" as a handler in rust
because local variable and stuff
trim frost
also because of "x86-interrupt" abi
torpid root
do it properly without the attribute
fiery ice
schedule_work is to create new threads, right?
slim panther
also because of "x86-interrupt" abi
💀
trim frost
actually it doesnt really need that
slim panther
i don't use x86-interrupt at all
trim frost
yeah because x86_64 crate
slim panther
wut
slim panther
i dont
trim frost
and small instructions
torpid root
use os;
kernel::run();
trim frost
i dont
thats why
but i can just use the pointer to a function that doesnt use that abi
bruh
why is it on by default
fiery ice
XD
torpid root
but i can just use the pointer to a function that doesnt use that abi
all my homies call it an attribute
not abi
torpid root
true dat
loud ice
the x86_64 crate is shit
trim frost
im aware
trim frost
new and improved version
wait
i gotta change static variables
so i wont need them anymore
slim panther
does interrupt stack frame even include the gprs?
fiery ice
new and improved version
why is irq on eoi hardcoded to 9?
torpid root
I doubt it
torpid root
why is irq on eoi hardcoded to 9?
shouldn't be but that's what it'll be anyway
trim frost
why is irq on eoi hardcoded to 9?
it will always be above 8 and for the pic thats all that matters
slim panther
pic
trim frost
yeah
fiery ice
k
fiery ice
XD
trim frost
this is what the interruptstackframe is
fiery turtle
yeah but the handler_fn and ctx will get overwritten
Uacpi only installs one for SCI and if you manually install a new GPE block (don't worry about it)
fiery ice
wait, you can cause a pagefault on command?
slim panther
lol so it doesn't include the gprs
trim frost
i dont need global vars
torpid root
you can't do it properly with x86-interrupt
slim panther
does x86-interrupt do swapgs?
torpid root
no lol
torpid root
yes
slim panther
me when i corrupt my kernel because a user program tripped a breakpoint
trim frost
i dont need global vars
nvm i do
gentle peak
can i pass a spinlock to this?
spinlocks are a possible implementation of mutexes yes
but if you do that you'll get terrible multi thread scalability
fiery ice
only on that core tho
gentle peak
because when a thread needs to wait for the lock it spends its entire time slice busy waiting
fiery ice
- rn it is not on a thread
fiery ice
is this ok for uACPI mappings?
mortal yoke
is this ok for uACPI mappings?
you should add the offset to the address you return (and also there is a bug in the mapping code)
ah wait nvm, you return the original address
slim panther
c++ grindset
torpid root
I get filling with zeros to 2^n digits, but why that many
left orbit
why even type it out in binary
have constants that you | together
so you can actually read the code
fiery ice
yeah, ik
north hull
so you can actually read the code
what type of dark magic are you suggesting, readable code? nahhh
rustic compass
And I thought I had good comments for the kernel api 💀 <:KEKW:91370837232964404...
i havent really looked at it yet
fiery ice
should i make it set and return the cpu's rFlags?
trim frost
should i make it set and return the cpu's rFlags?
I return 0
fiery ice
k
trim frost
And I've seen others do the same
gentle peak
uacpi doesn't care what you return, as far as it's concerned it's an opaque value that gets passed to unlock_spinlock
it's intended to be used to re-enable interrupts if lock disabled them
fiery ice
why is uacpi passing interrupt hanlders as 32 bit pointers?
mortal yoke
why is uacpi passing interrupt hanlders as 32 bit pointers?
?
fiery ice
mortal yoke
how is that a 32-bit pointer?
fiery ice
unsigned int is a u32 no?
mortal yoke
its the return value of the interrupt handler function
fiery ice
hmm
calm latch
It's a function pointer
fiery ice
so why is it 32 bit?
calm latch
It's a pointer to a function that returns error encoded as u32
frank canopy
the function it points to returns u32
frank canopy
unsigned int (*)(void *) means that something of the form unsigned int foo (void*) would be a valid target
fiery ice
k, thx
calm latch
Though it does use u32 for GSIs... Probably because ACPI says so
frank canopy
yeah GSIs are 32 bit
frank canopy
i trim what uacpi gives me to remove it
frank canopy
i use trim but same idea yeah
fiery ice
i trim what uacpi gives me to remove it
how?
fiery ice
i use my printf
calm latch
Just don't print newline
fiery ice
that auto adds a \n at the end
trim frost
yeah uacpi adds its own \n
frank canopy
yeah
fiery ice
pain
calm latch
i use my printf
Risky, it might break AML
fiery ice
?
calm latch
(I'm doing the same )
frank canopy
Risky, it might break AML
for logging?
calm latch
that auto adds a \n at the end
This is bad
frank canopy
zig's log you cant turn off its trailing newline so i have to remove it
yeah printf shouldnt have a newline but a log function maybe ig
gentle peak
for logging?
Printf is exposed to AML
calm latch
for logging?
No, I give uACPI my own sprintf impl
trim frost
frank canopy
fiery turtle
Or just don't insert your own new line
frank canopy
zig's log adds a newline so i have to trim and its annoying
fiery ice
Or just don't insert your own new line
doing a bool just for that rn
frank canopy
Or just don't insert your own new line
then id have to completely circumvent zig's logging
for regular printf it should probably not add a newline by default, you can make a println or something for that
trim frost
Or just don't insert your own new line
how about you dont insert your own newline in uacpi
calm latch
uACPI 3.0?
fiery turtle
how about you dont insert your own newline in uacpi
Its a standard convention in C
fiery turtle
then id have to completely circumvent zig's logging
Fair enough yeah
frank canopy
well technically i wouldnt need to because calling the log functions there filters through to my log handler which in a kernel im overriding anyway
but the zig convention is no newline in the call to log
calm latch
You could make it a define
frank canopy
yeah
use the weird c thing where you can concat strings by putting them next to each other and conditionally define a thing to either "\n" or nothing
trim frost
what is UACPI_FIXED_EVENT_RTC?
calm latch
RTC fixed event by uACPI
frank canopy
well the uacpi enum value for the acpi RTC fixed event
which i dont know offhand but its in the spec somewhere im sure
found it
i think this is what that is anyway
trim frost
wait so i can run the kernel, set an alarm, then when i rerun at a different time i can execute some code?
fiery ice
i think so
fiery turtle
No, this is for shutting down your PC
fiery ice
ah
fiery turtle
And then it starts at the programmed time
trim frost
oh
fiery ice
doesn't the BIOS and UEFI do that automatically tho?
fiery turtle
Do what
frank canopy
the event thing lets you easily detect when it starts on alarm
fiery turtle
They don't use the rtc to start the PC, no
frank canopy
and i think also if it alarms while already on?
fiery turtle
Why would they need it
fiery turtle
?
fiery turtle
and i think also if it alarms while already on?
Ye
fiery ice
you can set from the uefi panel to start up your pc at a specific time
frank canopy
idk if the fixed event triggers after a configured alarm boot actually
but the fixed event thing for it lets you call a function of yours when the alarm goes off with the kernel running
fiery turtle
you can set from the uefi panel to start up your pc at a specific time
That's possible yes
trim frost
whats UACPI_FIXED_EVENT_MAX?
fiery turtle
idk if the fixed event triggers after a configured alarm boot actually
Probably yes, because u can use it to determine the source which caused the wake
Not that it matters that much
fiery turtle
I've never tried it personally
fiery ice
time to understand what the names are for
fiery ice
i meant what are the tables for here
trim frost
RSDP (Root System Description Pointer)
The entry point into the ACPI system. It points to either the XSDT or RSDT table, which lists all other ACPI tables.
XSDT (Extended System Description Table)
Contains 64-bit pointers to all other ACPI tables. It's the modern version of RSDT (32-bit).
DSDT (Differentiated System Description Table)
Contains the primary AML (ACPI Machine Language) code used by the OS for managing devices and power. This is where much of the ACPI logic resides.
FACP (Fixed ACPI Description Table)
Provides fixed hardware details like power management timers, control registers, and pointers to FACS and DSDT.
APIC (Advanced Programmable Interrupt Controller table)
Describes the system's interrupt controller configuration. Used for configuring local and I/O APICs for multiprocessor support.
HPET (High Precision Event Timer table)
Details about the HPET hardware, which provides precise timers for event scheduling, replacing older timers like PIT and RTC.
MCFG (Memory-Mapped Configuration Space base address table)
Describes how to access PCI Express configuration space through memory-mapped I/O.
WAET (Windows ACPI Emulated Devices Table)
Optional table for Windows guests in virtual machines; provides hints to improve emulated hardware behavior.
BGRT (Boot Graphics Resource Table)
Contains information about the boot logo displayed by firmware, typically used by Windows to blend the OEM logo during boot.
FACS (Firmware ACPI Control Structure)
Used in conjunction with FACP for power management tasks like S3 resume (sleep/standby states).
oh thx
frank canopy
i meant what are the tables for here
its logging what tables it finds
fiery ice
well, at least it booted up first try
fiery ice
oof
(i still have half of the API not filled in) XD
i need to move uACPI to a thread first
trim frost
multithreading: ❌
bsp for everything: 
calm latch
You can run multiple threads on BSP...
fiery ice
hmm
this is not handling the power btn
i am pressing "power down" on qemu
and no message pops up
fiery turtle
Skill issue
fiery ice
...
fiery turtle
Does your kernel get an irq
fiery ice
sec
fiery turtle
And show your install irq implementation
fiery turtle
Did u unmask pic?
fiery ice
oh wait
frank canopy
and remap for that matter
fiery ice
i did not unmask for that one
PIC is ready
but not unmakes for that one interrupt
how can i get back the irq number when i uninstall the handler?
frank canopy
the ctx thing
well no the out_irq_handle
that thing
you can put anything arbitrary there
fiery turtle
^
fiery ice
k
frank canopy
i allocate a struct for it and stuff the irq number and handler function pointer in there personally but you can do whatever
fiery turtle
Did u implement io read and write properly
fiery ice
i did not add + 0x20
fiery ice
Did u implement io read and write properly
these ones?
fiery turtle
It didn't say anything about omitting any api either
fiery ice
well, time to find how to implement those
fiery turtle
Comments
frank canopy
on x86 theyre dead simple too which is nice
fiery ice
you mean these?
fiery ice
yeah
trim frost
and idk what it is for 32
trim frost
yeah use those
fiery ice
this should've printed an error btw
why?
fiery ice
it didn't tho
fiery turtle
recheck your code
fiery ice
hmm, seems like my OS literally stops before finishing to init uACPI
and it seems to stop at the spinlock
when locking it
or well, it is calling it while the lock is locked
hmm, but it's not uACPI's fault
seems like a deadlock somewhere
found the culprit
it was vmm or pmm
one of the 2
@fiery turtle if i get a Failed to initialize: out of memory
which api call should i look at
fiery turtle
anything with create
fiery ice
but in the kernel_api,?
fiery turtle
well obviously
fiery ice
maybe the alloc functions?
fiery turtle
yeah this wont work
fiery ice
what do i do with these?
fiery turtle
implement
fiery ice
f
fiery turtle
for event u can return any non-null handle
if u want it to be a no-op
for mutex u can also return anything non-null, just make sure its not always the same value
but this wont work on real hardware of course
frank canopy
well it might at first as long as youre strict single threading
calm latch
Simple spinlock is like 4 LoC
fiery ice
yeah this wont work
added those methods and now i get these
trim frost
you messed something up in the kernel api
same thing happened to me
idr how i fixed it though
vast kestrel
I mean, it seems like you ran out of memory, so your allocator function returns null for whatever reason
trim frost
how can i get cpu information other than lapic address from MADT?
vast kestrel
Define cpu information
Most of the stuff you will find via cpuid (vendor, family, name, topology, etc)
And not via acpi
trim frost
Define cpu information
i mean things like how many processors there are, their ids, which one is the bsp, etc
vast kestrel
The lapics are the thing that represents the different cpus, and you can read the current lapic id via an msr iirc, which is how you know which one is the bsp
It should have an entry for each lapic in the system and it's id iirc
trim frost
couldnt find where the id is, only the address but ig i can match it to the address i get from msr
vast kestrel
There is an entry per cpu, you might be parsing the table wrong if you don't find it
The address override can be basically completely ignored and iirc you should only use the msr to access the lapic
gentle peak
i mean things like how many processors there are, their ids, which one is the bs...
the madt lists the lapic ids of each processor, and you can read the bsp's lapic id from its lapic registers
fiery ice
I mean, it seems like you ran out of memory, so your allocator function returns ...
well, now i fixed the out of memory
but these still remain
vast kestrel
Well, print the error code that they return, I am pretty sure every uacpi functions returns an error code
vast kestrel
Could be you stub or implement incorrectly some Io or pci read/write functions?
fiery ice
pci function all return OK status
vast kestrel
fiery ice
but why is the power button failing to install then
vast kestrel
Don't return ok for stubs that have side effects
Because it thinks you are writing but it doesn't
fiery ice
pwr button is PCI?
vast kestrel
No idea, but could be that it needs pci access for it
For every stub you have don't return OK if it had a side effect like read/write
Return an unimplemented error
And then you will know when uacpi calls it and expects it to work
fiery ice
lemme try to disable the usb hubs i enabled on qemu
nope, still PCI error
time to write a PCI driver then
fiery turtle
u dont need a pci for power button
u need a proper io_read/write
which yours isnt i guess
fiery ice
i have
fiery turtle
show
vast kestrel
Lol
vast kestrel
Nice
fiery ice
thx infy
fiery turtle
np
fiery ice
"works"
there could be 2 things:
- the handler is not registered correctly
- qemu is not using acpi for power down function
fiery ice
yeah well, i suspected
time to see if it receives the isr at least
no irq is being fired
oof
PICs are unmasked
but no interrupt is generated
and it's trying to register the interrupt at irq 41
fiery ice
yeah well it is 9
i am adding the 0x20 to it
but i may have found the problem
no irqs are being fired
not even PIT
normal interrupts work tho
fiery turtle
whats a normal interrupt
fiery ice
like 0 to 31
fiery turtle
thats called an exception
fiery ice
hmm
qemu pauses on this instruction
prob a crash
but no execption is caught by my interrupt handler
when uacpi calls the stall function
this is my stall function
Cr2 is null
calm latch
qemu pauses on this instruction
This makes no sense
calm latch
did you turn kvm off?
calm latch
then it won't log exceptions
fiery ice
wdym won't log exceptions?
fiery ice
5+ minutes
mortal yoke
what is it doing lol
calm latch
wsl is a vm
fiery ice
and then resume
trim frost
(i am running qemu on wsl)
why?
fiery ice
idk
trim frost
doesnt qemu run natively on windows?
fiery ice
yes
fiery ice
mine starts instantly too, but qemu freezes for a bit and idk why
calm latch
(most of which is printing to screen being slow(
trim frost
mine starts instantly too, but qemu freezes for a bit and idk why
the boot logo thing?
trim frost
weird
mortal yoke
Even emulating RISC-V, my OS takes like a few seconds to fully initialize
tbh if you do stupid stuff like mapping the entire hhdm using 4kb pages + a naive bitmap allocator with no optimizations you can get pretty large slowdown at boot
fiery ice
i map only the entire memmap
but like, the OS does execute i think, bc after it unfreezes a bunch of log instantly appear
mortal yoke
it's still a good idea to use 2mb pages
calm latch
tbh if you do stupid stuff like mapping the entire hhdm using 4kb pages + a naiv...
I don't use HHDM
slim panther
I don't use HHDM <:trl:829316641191886888>
how
calm latch
Temporary mappings
slim panther
no limine?
calm latch
With limine
slim panther
but the only way to access phys mem is with hhdm 😳
calm latch
I mean my kernel also boots with Hyper
calm latch
but the only way to access phys mem is with hhdm 😳
I just don't map it in my page tables
mortal yoke
but that's arguably still using hhdm <:meme:575445094589661215>
well if you allocate the regions you temporarily map the pages to I wouldn't call it a hhdm
mortal yoke
well if you allocate the regions you temporarily map the pages to I wouldn't cal...
or well ig it could still be but like with the definition of hhdm where its all the memory mapped at one static offset not just small chunks mapped at different allocated positions
fiery ice
idk why the qemu app just freezes for no reason, while the OS keeps going in the background
i cannot press any button in the app
any idea why it was crashing there tho?
mortal yoke
run it with -d int -no-reboot -no-shutdown without kvm
fiery ice
and smm off?
mortal yoke
if there is no exception then another thing to check is that interrupts are enabled at the point where you pause nvm I though of hlt lol
fiery ice
i do cli before that
bc it fails here
oof, running with -d int is pain rn
i cannot send the shutdown signal
i'll add it via code
mortal yoke
i cannot send the shutdown signal
why?
calm latch
but that's arguably still using hhdm <:meme:575445094589661215>
Only before I load my page tables
fiery ice
but the OS still runs in the background
calm latch
or well ig it could still be but like with the definition of hhdm where its all ...
I just have a region of 16 pages where I map different addresses
fiery ice
done
only look at the last one
the one before it is the PIT timer
maybe
ok nope
last interrupt is the framebuffer
before it RIP points at my terminal
fiery turtle
Implement pci
fiery ice
oh, that's PCI?
fiery turtle
bc it fails here
If it failed to prepare there's no reason to call enter, it won't work
fiery ice
fiery turtle
Idk, it says unimplemmented, you return it somewhere
fiery turtle
This error is a result of your unimplemented api
fiery ice
no, i was referring to incompatible AML type
left orbit
people refusing to implement the kernel apis is crazy
fiery turtle
no, i was referring to incompatible AML type
Its a bug in your api
left orbit
what's the point of a kernel api anyway if people just don't implement it
i think you should remove it
fiery ice
i still did not create a pci driver
fiery turtle
True
fiery ice
that's why
fiery turtle
Its just annoying when people ask for help but they don't even bother implementing the api and stub most things out
Like yes that won't work
It exists for a reason
fiery ice
If it failed to prepare there's no reason to call enter, it won't work
it did not fail to prepare btw