it's making up trb address out of thin air
#OBOS (not vibecoded)
flint idol
500k download speed for managarm servers 🥀
im tryna see if managarm's xhci driver will work here
ok the speeds boosted
flint idol
well managarm works
ahhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhhh
i fucking hate reading C++ code
"ah yes let me have code implcitly call other code and make it hard to track what variables are defined where because FUCK YOU"
uncut narwhal
i fucking hate reading C++ code
looking forward to having obos to read when i eventually get to usb specifically so i wont have to read c++ lmao
flint idol
lmao
reading c++, linux, or bsd source code drives me insane
bsd source being the least out of the three
flint idol
reading c++, linux, or bsd source code drives me insane
although reading obos source code under src/oboskrnl/arch/x86_64 makes me believe I was insane when I wrote it
I wrote like all of the x8664 thing in 2024
and had barely any idea how anything worked
so like now nothing works on a lot of real hw
because bah dumb reasons
knotty totem
bsd source being the least out of the three
least what?
flint idol
least unreadable
knotty totem
ah
honest citrus
linux is much more unreadable imo
it's also just as bad to write
no clue how the upstream maintainers do it without wanting to rip their hair out
flint idol
no clue how the upstream maintainers do it without wanting to rip their hair out
clang-format, duh 
flint idol
man I fuckin hate this city's transit
buses are always late
but the ONE day I am late, it is early
and dont forget the next bus is in 20 minutes
@dense vigil can u get the bus to hurry up pls
dense vigil
this should help
flint idol
ok thx
knotty totem
I realize it's not technically nsfw but it's supposed to put the same image in your mind
knotty totem
It's okay ❤️
I had my finger over that section of the gif (I'm on mobile) and legitimately thought it was just... that
dense vigil
That is not what comes to my mind right away but fair
Mainly the source clip context is horse riding the human like a horse and not in that connotation
dense vigil
I had my finger over that section of the gif (I'm on mobile) and legitimately th...
😭
Take this replacement gif
flint idol
time to continue trying to fix xhci because damn me i guess
flint idol
why the fuck is the xhci controller putting the same trb on the event queue twice
ok better question is why is the xhci controller seeing the same trb twice
flint idol
im dumb
real pecan
lol
flint idol
i forgot to update the dequeue pointer when sending the address device for the 2nd time (after sending a command in between)
flint idol
xhci is cursed
does get descriptor request
device, nah take USB Transaction Error instead
is it possible that these usb2 ports are under control of ehci
flint idol
why do i have to get all the bad errors
dense vigil
is it possible that these usb2 ports are under control of ehci
I wonder what Linux reports under this?
Or managarm/BSD
flint idol
managarm works
perfectly fine
with passthrough enabled
tbh
i never got port reset to work on this hardware
when not doing passthrough
maybe that has something to do with something
flint idol
when not doing passthrough
as in, when doing a raw obos boot
flint idol
today I am going to revise the entire xhci controller init sequence
dense vigil
sudo rm -rf ./repo
git pull https://microslopgithub.com/oberrow/OBOS
blazing python
i fucking give up
blazing python
hard xhci reality
stark sigil
cant go outside
-16 celsius and its getting worse every minute 
flint idol
-16 celsius and its getting worse every minute <:nooo:750466870993223750>
same here in southern ontario
apparently in some places in ontario it's gonna reach -50
(not here thank god)
stark sigil
apparently in some places in ontario it's gonna reach -50
W H A T
flint idol
yea
Dangerously cold air swept across parts of Ontario early Saturday morning, with wind chills nearing –40 degrees Celsius in northern areas.
yesterday my toe nearly froze off
i think
it still hurts
stark sigil
i had to shower in hot water after coming home from school
the weather is insane
i think i got a cold as well
stark sigil
its so cold the sun cant even be a deadly lazer anymore
flint idol
real
blazing python
https://github.com/OBOS-dev/obos-pkgs/tree/master/recipes
you have audio supportt ?
flint idol
yes
blazing python
is this hard
blazing python
i want do audio stuff later for my kernel so
blazing python
not if i use a portable audio driver some1 else made
which api you implemented
flint idol
my own™
im planning to eventually add downstream support for my audio server protocol to sdl and stuff
so shit that uses sdl for audio can work
blazing python
im planning to eventually add downstream support for my audio server protocol to...
isnt oss and pulseaudio (i have no idea can this be started with mlibc) better
flint idol
so shit that uses sdl for audio can work
(i should port SDL2, and also enable the windowing stuff in sdl)
xhci isnt real omfg
sdfkjsdfksdo
there are 10 ports on the controller
on the root hub
ZERO RESET SUCCESSFULLY
FSDOKGJSDLKFMSDOKFDMSLFKSDMF
atp it has to be the ehci ctlr trolling me
uncut narwhal
> Dangerously cold air swept across parts of Ontario early Saturday morning, wit...
dont even need to say the unit when its -40 lol
good luck with your usb hell
flint idol
good luck with your usb hell
i think ive been cursed by the lord himself for usb, no one can save me now
This requirement means that Root Hub port may report a device is connected
(CCS and CSC = ‘1’) before the xHC is running (i.e. HCHalted (HCH) = ‘0’), and that
when software enables the xHC and HCHalted (HCH) transitions to ‘0’, PSCEG
shall be asserted for each port with a connected device, generating a respective
Port Status Change Event. In this case
what a fucking lie
rtfdg,l;fedmdksgl;gjmdsf
dense vigil
apparently in some places in ontario it's gonna reach -50
It’s going near -60f for my friend
And he still has to goto school
flint idol
damn that's cold
dense vigil
This is America too iirc
dense vigil
at least your friend isn't working on xhci
They are doing doctor shit or science shit idk
But like really fucking advanced shit I think
flint idol
This is America too iirc
a lot of states are a further north then south eastern ontario
dense vigil
flint idol
where like the entire country lives
i truthfully have no fucking idea what the problem is
i am literally about to give up
flint idol
what do you think ive been doing 😔
dense vigil
You gotta bull the shit even harder
weary hound
i tried disabling the ehci ctlrs before trying to init the xhci ctlrs
did you poke the right bits in the pci config space to toggle the ports
flint idol
did you poke the right bits in the pci config space to toggle the ports
elaborate
flint idol
did you poke the right bits in the pci config space to toggle the ports
do you mean in teh ehci controller thing?
weary hound
void switchUsbPortsToXhci(PciDevice *dev) {
debugLogger() << " Switching USB ports to XHCI!" << frg::endlog;
auto io = dev->parentBus->io;
auto usb3PortsAvail = io->readConfigWord(dev->parentBus, dev->slot, dev->function, 0xDC);
io->writeConfigWord(dev->parentBus, dev->slot, dev->function, 0xD8, usb3PortsAvail);
auto usb2PortsAvail = io->readConfigWord(dev->parentBus, dev->slot, dev->function, 0xD4);
io->writeConfigWord(dev->parentBus, dev->slot, dev->function, 0xD0, usb2PortsAvail);
}
let me try that
weary hound
on intel chipsets
with ehci
usb ports are muxed between xhci and ehci, and you can toggle it
flint idol
on intel chipsets
if this works i might donate a buck to managarm
fak
it didnt work
😔
uint64_t usb3_ports_available = 0;
uint64_t usb2_ports_available = 0;
DrvS_ReadPCIRegister(dev->location, 0xdc, 4, &usb3_ports_available);
DrvS_WritePCIRegister(dev->location, 0xd8, 4, usb3_ports_available);
DrvS_ReadPCIRegister(dev->location, 0xd4, 4, &usb2_ports_available);
DrvS_WritePCIRegister(dev->location, 0xd0, 4, usb2_ports_available);
oh wait this needs to be done on an xhci controller
grepogijrd
it got further
kinda
some ports now report CCS
uh nvm
i dont have a dollar to my name
dense vigil
some ports now report CCS
What’s a CCS btw lmfao
flint idol
current connection status
dense vigil
That’s good I presume
flint idol
yes it means a shit is connected on the port
flint idol
```c
uint64_t usb3_ports_available = 0;
uint64_t usb2_ports_available = 0;
DrvS_...
i think the reason it got further than port reset on linux with passthrough is because the linux kernel already did this
torpid wigeon
grepogijrd
is this a dutch word
flint idol
perhaps
weary hound
i think the reason it got further than port reset on linux with passthrough is b...
usb device passthrough is very far removed from actual hw
flint idol
no i did xhci controller passthrough
weary hound
ah
flint idol
now time to figure out why the driver hangs after port reset 
flint idol
could be both, even
flint idol
well all i know rn is that im not getting any goddamn irqs
flint idol
so now it decided it doesnt wanna work anymore
dense vigil
What if USB is actually implemented with quantum mechanics and the only way to be able to make a working driver is to be ordained by the gods and or to be in the proper super position at the time
dense vigil
You may need to do some head clips into a block to get your sub pixel correct
flint idol
then i would be able to diff between now and then
dense vigil
Like super Mario bros style
flint idol
so uh for some reason usb hotplugging stopped working everywhere
flint idol
if only i had decided to commit in the past few days
my next commit will be something to the likes of: "xhci: fix bugs on real hardware", +2000 additions, -5000 deletions
flint idol
January 24th, 2026, at 1940: xhci has progressed a bit further, I am still going crazy from this
weary hound
https://github.com/OBOS-dev/obos/blob/xhci-driver/src/drivers/generic/xhci/xhci.c#L463
what if the xhci pushes enough trbs to wrap around?
flint idol
i told myself i would fix it
but im having problems with the first trb
also i have some new code
so let me push that
flint idol
i told myself i would fix it
(meaning, i am aware of the bug, and i will fix it)
weary hound
but im having problems with the first trb
ah
flint idol
i just pushed like 750 changes
weary hound
another thing is that checking USBSTS.EINT for interrupt pending is not correct if you use MSIs, because iirc the interrupter IP is automatically cleared after the write that triggers the MSI
flint idol
oh
weary hound
actually hm, EINT is RW1C so it's only ever set by the controller
still i wouldn't use it and just check the specific interrupter
in managarm we check ERDP.EHB since that has to be cleared by software after it's done processing stuff from the event ring
flint idol
ok ill try that
weary hound
still i wouldn't use it and just check the specific interrupter
technically with MSI you don't need to check anything since you shouldn't share them and spurious ones shouldn't happen either™
flint idol
im gonna share them anyway whats the worst that could happen
weary hound
well the problem is that they're edge triggered
so you can lose them if multiple happen and you have them masked at the time
flint idol
i see
ill fix that in a second, rn the only driver i am running is the xhci driver so other devices shouldn't be a problem™
oh well time to open managarm code
because well ig damn me, i get interrupts with xhci passthrough on this hardware
(and i also get trb completion code 0x4 because usb has damned me)
weary hound
usb transaction error my beloved
managarm also gets it when trying to talk to stuff behind hubs
and i haven't figured out why yet
although i got some new equipment to try and debug it (cheap usb protocol analyzer from china based on an open source design), but i haven't had the time to do it yet
flint idol
gl
i think im gonna do some polling on the event ring to see if it's an interrupter problem or something else
dense vigil
I wonder if there is a good way to test MSI/X seperate from the USB controller
Like a QEMU device to write to an IO port and it gives you an MSIX
flint idol
AHCI
dense vigil
flint idol
anyway this xhci controller uses MSI
and it seems to not be an interrupter problem
ok so no matter what i set the cycle bit to
it doesnt recognize my trb
stark sigil
msi the company
flint idol
omg
i just want an irq
is that so much to ask for, xhci controller
are you really that lazy, xhci controller
sdflkgsndflksdnfs
stark sigil
i have to do all OHCI, EHCI and then XCHI at some point 
dense vigil
i have to do all OHCI, EHCI and then XCHI at some point <:pain:67904709047471309...
Shhhh you don’t need OHCI
Nobody needs usb 1.1 as the main controller
dense vigil
Fuuuuckkkkk
flint idol
Shhhh you don’t need OHCI
wait until you hear about ehci companion controllers
stark sigil
oh no
flint idol
ehci forces you to implement a UHCI or OHCI controller driver for support for older usb devices
XHCI handles USB1, USB2, and USB3 under the same thing
flint idol
but intel wanted to be lazy with EHCI and decided "have the OHCI or UHCI controller do it"
flint idol
USB is insane, I tell you
dense vigil
Who the FUCK invented this
Who the FUCK did this so I can strangle them when they sleep
flint idol
dumbasses at intel and some other dumbasses at other companies
flint idol
flint idol
i dont care if microsoft had anything to do with this or not
flint idol
fuck micro$oft
flint idol
https://en.wikipedia.org/wiki/Ajay_Bhatt
[REDACTED]
he probably didnt invent the EHCI/OHCI/UHCI interfaces
flint idol
IM NEVER GONNA DANCE AGAIN, GUILTY FEET HAV EGOT NO RHYTHM
THOUGH IT'S EASY TO PRETEND
I KNOW YOURE NO T A FOOL
flint idol
https://www.newgrounds.com/audio/listen/1233600
no ty
flint idol
[January 24th, 2026, 20:56: I have started yelling song lyrics in OBOS thread, i have been driven to absolute insanity because of USB]
flint idol
crash out over toilet music is insane
wha
dense vigil
stark sigil
i cant even find the specification for any of them
dense vigil
i cant even find the specification for any of them
@flint idol had to pirate them
Honestly honestly
I should fucking mirror the USB specs on my site
flint idol
no the xhci spec was free
dense vigil
Copyright be damned
flint idol
no the xhci spec was free
and ehci
stark sigil
<@841762704766140486> had to pirate them
oh
flint idol
didnt check uhci or ohci
stark sigil
this is worse than openfirmware
flint idol
idk if the usb specs are pay walled
stark sigil
this is worse than openfirmware
THATS AT LEAST INDEXED BY GOOGLE
flint idol
but I couldnt find an official spec
flint idol
Oh then what did you pirate
oh it turns out it's all free
idk why i couldnt find the usb3 spec anywhere
dense vigil
oh it turns out it's all free
stark sigil
is xhci usb3, ehci usb2, and ohci and uhci usb1?
dense vigil
Can you send them to me so I can mirror them nicely
If you can find all of them
Just so people can find them 🙏
flint idol
is xhci usb3, ehci usb2, and ohci and uhci usb1?
xhci is: USB3, USB2, USB1
ehci is USB2
UHCI+OHCI (mutually exclusive btw) is: USB1
flint idol
Just so people can find them 🙏
stark sigil
xhci is: USB3, USB2, USB1
ehci is USB2
UHCI+OHCI (mutually exclusive btw) is: US...
how the frick do you determine whether its a uhci or an ohci
flint idol
flint idol
how the frick do you determine whether its a uhci or an ohci
pci class codes
stark sigil
pci class codes
flint idol
as you would any other device
stark sigil
ok honestly i think i have to implement device trees even on x86
on powerpc it was practical but x86 is literally pain
flint idol
Just so people can find them 🙏
would be nice if you could also mirror benjamin lunt's usb guide
it's paywalled
i found it on libgen
flint idol
just like dont tell him
flint idol
just like dont tell him
he is the guy who made fysos and its books
dense vigil
DMCAd tho
flint idol
fine ill do it myself
flint idol
real
dense vigil
Honestly would the fellow even care?
flint idol
helping clueless people like me with USB
flint idol
Why don’t you ask
im not on there either
dense vigil
I cannot find UHCI
flint idol
flint idol
I cannot find UHCI
dense vigil
flint idol
lets just uh ignore that
watch us get ToS banned because piracy is against discord rules
dense vigil
watch us get ToS banned because piracy is against discord rules
yeah im not gonna mirror that one
ehci atleast says this
flint idol
there isn't even an OHCI spec 🥀
dense vigil
sooooo
No license, express or implied, by estoppel or otherwise, to any intellectual property rights is granted herein, except that a license is hereby granted to copy and reproduce this specification for internal use only
this is "internal use" right
flint idol
1.1 i think
dense vigil
even if its for public good
weary hound
wait is UHCI or OHCI 1.1 or 1.0
the 1.1 refers to the spec revision not usb version
i mean aside from the ben lunt book this is all publicly available
(so perhaps stop sharing it)
flint idol
ill hide it on my server so no one can find it
hopefully like no one signs with ftp and finds it
because like that would be terrible
you know
dense vigil
this should be correct?
weary hound
for ohci there is even an open source implementation in spinalhdl :^)
if you want to design your own hardware with ohci
flint idol
<:concern:895722900722696252>
maybe i should run dist-upgrade at this point lol
nah ill just use aptitude it's smarter
dense vigil
Let me get my USB
To put them on my file server
Because I forget the login
And I CBA to change it
Because my desktop at home is logged in
flint idol
i hate the cold but i love the vibe of the cold
dense vigil
Gparted how fucking long does it take to format a USB ext4
flint idol
i hate the cold but i love the vibe of the cold
nvm i hate the cold
next time I have to wait for the bus in the cold im gonna listen to snowy from undertale ost
dense vigil
Why is my laptop overheating now
flint idol
Why is my laptop overheating now <:PSP1G_pspTrollar:1434611394782691429>
punishment for considering pirating the book
dense vigil
This is just the normal spec books
Fat32 worked thank you
Unmounting it crashed plasma shell
flint idol
for tomorrow: play minecraft
for monday: study for math exam
for tuesday: do math exam, stress over math exam
for wednesday: play terraria
for thursday: enjoy my break after exam season
for friday: contemplate on my exam grades
for next Saturday: work on obos?
flint idol
nice
dense vigil
for monday: study for math exam
for tuesday: do math exam, stress over math exam...
for tomorrow: fix XHCI driver
for monday: fix XHCI driver
for tuesday: fix XHCI driver
for wednesday: fix XHCI driver
for thursday: cry that XHCI driver isnt working and fix XHCI driver
for friday: fix XHCI driver
for next Saturday: fix XHCI driver
flint idol
fucking hell
my pc isnt booting
well its only on the new kernel version
im gonna take a break from xhci
also after xhci i feel like adding evdev support to obos
would be le fun™, probably
flint idol
ok so im tryna get fvwm3 to work
but it's faulting tryna execute 0x11
in lile the actual fvwm3 executable
flint idol
lmao
but yes
I just got blinded by the sun reflecting off all this snow holy
this snow has gotta be like 3 feet deep
vale nymph
Arent u like from the middle east what
flint idol
Arent u like from the middle east what
yes I am from there but I live in canada
vale nymph
Damn
uncut narwhal
yeah the canada explains why the snow is that deep lol
flint idol
but it's faulting tryna execute 0x11
well delightful...
it is stack corruption occurring as soon
as waitpid() in mlibc tries to return
flint idol
i think it's a bug in my signal implementation
yeah that checks out, it tries to return to 0x11, which is 17, which is SIGCHLD
um trying to fix it causes a triple fault lmfao
obos truly has the best code quality
which is probably a security vulnerability
because like anything can triple fault the kernel by passing an invalid ucontext
💀
it doesn't even set the ss/cs
dense vigil
because like anything can triple fault the kernel by passing an invalid ucontext
Well I mean
It’s not really a SECURITY issue per say
Everyone data is safe
Nothing is sacked
Is more of a DoS attack
sterile sparrow
um trying to fix it causes a triple fault lmfao
lol
sterile sparrow
fvwm3 works now
ooo i should try porting it to my thing too
sterile sparrow
ye
dense vigil
it also happens to be astral's WM
Common astral W
flint idol
blazing python
fvwm3 works now
btw at this point you are able to run i3
flint idol
btw at this point you are able to run i3
i3?
flint idol
never tried it
[ LOG ] thread 28: enter write_sync. blkCount=316, pipe->ptr=0, pipe->in_ptr=0, pipe->size=4096, pipe=ffff9000003c5808
[ LOG ] thread 28: ret from write_sync. blkCount=316, pipe->ptr=316, pipe->in_ptr=0, pipe->size=4096, pipe=ffff9000003c5808
[ LOG ] (thread 29, process 10) syscall Sys_ExecVE returned 0xb (OBOS_STATUS_NOT_FOUND)
[ LOG ] (thread 28, process 9) syscall Sys_FdSeek returned 0x2f (OBOS_STATUS_SEEK_UNALLOWED)
[ LOG ] thread 28: enter write_sync. blkCount=316, pipe->ptr=0, pipe->in_ptr=0, pipe->size=4096, pipe=ffff9000003c5808
[ LOG ] thread 28: ret from write_sync. blkCount=316, pipe->ptr=316, pipe->in_ptr=0, pipe->size=4096, pipe=ffff9000003c5808```wtf
there should be some read log
before the pointer goes back to zero
oh
the pipe is freed
fixed it
/bin/sh didn't exist
KVM Internal Error my beloved
ts consistent too
blazing python
KVM Internal Error my beloved
you did some unspeakable things to kvm 🙏
sterile sparrow
acxcidentally
flint idol
acxcidentally
u have no proof of my spelling mistake
sterile sparrow
xaccidentally
flint idol
i accidentally ported x
sterile sparrow
man i hate it when that happens
dense vigil
it starts but doesnt show the time
What calls is it making for the time
Maybe you don’t have thoes implemented?
Could have it return the time from the BIOS RTC
flint idol
Could have it return the time from the BIOS RTC
a) that's not bios
b) i already have sys_clock_get implemented
flint idol
it seems to get the time from clock 0x1 which i defined as the tsc/kernel time source
something does that, at least
flint idol
currently implementing PTYs
flint idol
TIL xterm's build system is insane
and xterm also just so happens to imagine new functions into existence
nvm
like congratz to the guy who wrote this
he managed to make a completely shit build system
what dumbass wrote this build system, seriously
bro what
config.log says it made a confdefs.h but not one is in sight
nvm that
knotty totem
find it then?
vale nymph
TIL xterm's build system is insane
There is a reason why the repo has "abandon all hope, ye who enter here" in the readme
flint idol
There is a reason why the repo has "abandon all hope, ye who enter here" in the ...
how the actual fuck did you get this to build
it's not detecting any of the installed headers correctly
or well it's detecting them but it isn't going in the config.h file thing
stark sigil
There is a reason why the repo has "abandon all hope, ye who enter here" in the ...
gcc should add this too
flint idol
or well it's detecting them but it isn't going in the config.h file thing
so it never actually realizes it has those goddamn headers
atp i might as well write my own makefile for this shit
vale nymph
how the actual fuck did you get this to build
Im just that good
flint idol
i got it to build using duct tape
ffs
it's trying to use bsd PTYs
oh it doesnt build anymore
wtf
flint idol
why does it think im compiling for goddamn BSD
what the fuck is wrong with this
ok i got it to build again
flint idol
whoever invented the ioctls that work on goddamn file descriptors
fuck you
IO control
not FD control
well i got one half of PTYs working
it can definitely output onto the PTY
but xterm isn't recognizing it got output
like, input works
but xterm is just completely heedless to the fact it got output it needs to print
(which is probably because my PTYs are wrong)
knotty totem
how the actual fuck did you get this to build
old xterm works just fine
(from shortly after the xfree86 fork)
knotty totem
why does it think im compiling for goddamn BSD
idk figure it out lmfao
it shouldn't be hard
flint idol
ok i got it to build again
...
NOOOOOO I JUST LOST ALL MY DEBUGGING PROGRESS
BECAUSE I FORGOT TO UNSET A BREAKPOINT FOR Arch_GotoUser
knotty totem
ok, unset it now and then continue
flint idol
no because it went to a different user thread
knotty totem
you'll go to the next point it would've breaked at
flint idol
and it's impossible for me to go back to the old one
because it's also a different process
ok i set a breakpoint on the last place i remember being at
note: charproc.c:6894
flint idol
(which is probably because my PTYs are wrong)
it seems like it's a bug with the PTY's IRPs
also, "no job control in background" 🥀
flint idol
*only works without fvwm3 because i think xorg hates me
dense vigil
*only works without fvwm3 because i think xorg hates me
just use xlibre trust trust it will make eveyrthing better trust
flint idol
accidentally ran cat in a no-job-control shell
what do i do
i should debug why it has no job control
hell is should debug how tf it's working
obos has absolutely no "changing controlling TTY" logic
this working should not be possible
except if like it's because on fork() nothing closes the stdio descriptors
vale nymph
*only works without fvwm3 because i think xorg hates me
ask claude to make a better xorg?
if you cant vibecode x12 you are ngmi
flint idol
ok nvm it does work with fvwm3
i need to fix some bug with ps2
after using the mouse/keyboard long enough it kinda just starts having a stroke
to put it shrimply
also unfortunately doom does not work with X11
😔
it loads the WAD and all
but alas, it doesn't actually run
TODO: get SDL's X11 interface thing working
although that sounds like a problem for future me
im gonna stop doing osdev for the day
dense vigil
if you cant vibecode x12 you are ngmi
That’s called xlibre and they break a shit ton of things
knotty totem
That’s called xlibre and they break a shit ton of things
Xlibre is X420 because they must've been high outta their god damn minds when writing it 😂
dense vigil
Xlibre is X420 because they must've been high outta their god damn minds when wr...
How are they even fucking doing
All I know is the creator is a biggot anti vaxor conspiracy person XD
flint idol
ok nvm it does work with fvwm3
in a bit im gonna fix job control in bash on xterm
no clue why it gets EINVAL
flint idol
yay
blazing python
flint idol
void pty_ref(void* tty) { OBOS_SharedPtrRef(&(((struct pty*)(((struct tty*)tty)->interface.userdata))->ptr)); }
void pty_deref(void* tty) { OBOS_SharedPtrUnref(&(((struct pty*)(((struct tty*)tty)->interface.userdata))->ptr)); }
flint idol
hey that's not fair why cant my fvwm3 look like that :(
flint idol
no way xbps-install is almost working
sterile sparrow
faster than movistar internet
flint idol
for funsies im gonna set MSS (max segment size) to 0xffff
(so much for congestion and stuff)
flint idol
im tryna think of a way to increase IO performance
on the network stack
im thinking of one thing in particular and that is to eliminate as many allocations as possible
i was also thinking of delaying ACKs in the TCP implementation
reasons being:
a) sending packets takes time
b) it is better to send one big ACK as opposed to many different small ACKs
one thing i noticed in packet dumps is that obos is sending a bunch of small ACKs really slowly
i just removed a SINGLE assertion
and that boosted I/O perf 10-100 times more
still slow as shit
but
honest citrus
https://download.microsoft.com/download/1/6/1/161ba512-40e2-4cc9-843a-923143f345...
thanks for the free collection of USB specs btw yall 
dense vigil
thanks for the free collection of USB specs btw yall <:troll_shrug:1428667547288...
https://files.evalyngoemer.com/share/manuals/
here is a good set of other specs too
I plan to get nvme on there too soon
And any others that come up that people need
honest citrus
nice
dense vigil
LMK if there is any others you need me to mirror
flint idol
ok so my problem with TCP right now is that it is not seeing a lot of packets that come to it
causing a lot of retransmissions
dense vigil
More driver issues?
flint idol
causing a lot of retransmissions
causing le slowness
flint idol
More driver issues?
no clue what's causing it to drop all these packets
dense vigil
Or it’s getting them too slow and it’s getting retransmitted causing a snake effect
flint idol
#define RX_QUEUE_SIZE (4096)
ill just do this and see what happens
no clue if the device actually supports that
flint idol
```c
#define RX_QUEUE_SIZE (4096)
```
ill just do this and see what happens <:tr...
(RX_QUEUE_SIZE was 32 before)
flint idol
ok i fixed the ACK bugs
and now imn getting reasonable download speeds
in the 100s of KiB/s
but the kernel dies halfway through
the download speeds were so fast for obos that it exposed a race condition
flint idol
the download speeds were so fast for obos that it exposed a race condition
and also it doesn't read the entire file
so i think i might've accidentally broke something
dense vigil
That’s blazingly fast
You should put on your repo that you have a blazingly fast network stack
And put tons of emojis
flint idol
lol
dense vigil
so i think i might've accidentally broke something
Have you tried
Reading the entire file
flint idol
yeah
dense vigil
And I presume DMA with the network card?
flint idol
pretty much
dense vigil
I wonder if you can have the ISR track the packets coming in and then compare that with the actual result
flint idol
so ur saying i should capture the packets for analyzing purposes
(wireshark)
nah the problem is the tcp implementation is overwriting the buffer
dense vigil
In the ISR just track minimal data and compare it with the result to make sure nothing is lost internally
dense vigil
nah the problem is the tcp implementation is overwriting the buffer
Ahhh makes sense
Have you tried
flint idol
it receives data, it gets more, it overwrites it, or so it appears
dense vigil
Not doing that
flint idol
i always forgot how genius you are
flint idol
Not doing that
out of curiosity, is it 3am where u are
dense vigil
out of curiosity, is it 3am where u are
10pm
But I am very lacking in sleep
And I just dealt with a ton of discord drama
flint idol
i can tell go to sleep man
dense vigil
And I need to unwind a little
I can share more in DMs
But holy shit 😭
How can you tell btw
flint idol
a slightly elevated level of sarcasm
flint idol
aight
flint idol
Sorry if I am intruding btw
i do not mind at all
dense vigil
Thank you
flint idol
i need some funny sarcasm in my life
dense vigil
Dosnt everyone
flint idol
yes
flint idol
size_t nPushed = 0;
OBOS_ENSURE(be32_to_host(hdr->seq) >= con->state.rcv.nxt);
Net_TCPPushReceivedData(
con,
segment_data, segment_length,
be32_to_host(hdr->seq) - con->state.rcv.nxt,
&nPushed);
if (con->state.rcv.furthest < con->state.rcv.nxt)
con->state.rcv.furthest = con->state.rcv.nxt;
if ((be32_to_host(hdr->seq) + nPushed) > con->state.rcv.furthest)
con->state.rcv.furthest = (be32_to_host(hdr->seq) + nPushed);
if (be32_to_host(hdr->seq) == con->state.rcv.nxt)
{
con->state.ack_pending = true;
OBOS_ASSERT(con->state.rcv.furthest >= con->state.rcv.nxt);
con->state.rcv.nxt = con->state.rcv.furthest;
}
if (!con->ack_timer.handler && con->state.ack_pending)
{
con->ack_timer.handler = ack_timer_handle;
con->ack_timer.userdata = con;
Core_TimerObjectInitialize(&con->ack_timer, TIMER_MODE_INTERVAL, 5*1000);
}
well this is the code
and apparently it causes the kernel to ACK more than it actually put in the buffer
stark sigil
```c
size_t nPushed = 0;
OBOS_ENSURE(be32_to_host(hdr->seq) >= con->state.rcv.nx...
magnificent
flint idol
magnificent indeed
stark sigil
it was already there 
flint idol
```c
size_t nPushed = 0;
OBOS_ENSURE(be32_to_host(hdr->seq) >= con->state.rcv.nx...
i fixed the bug, Net_TCPPushReceivedData wasn't working as i thought it would
uh how do i verify the file's contents now
sha256
yeah that's not the same file 💀
the checksums are completely different
and ig because the kernel fucking hates me
i consistently get this crash
when running curl on http
flint idol
well it's reading a raw data file so id hope it's not CRLF
dense vigil
Try downloading a binary file that’s all 0xFF or somthing?
And then you can do a check in the hex editor
Unless you already did that
dense vigil
That’s how TCP works innit? You have to re order it back
flint idol
yeah and that's the bug
dense vigil
so receiving all `0xff`s will not reveal anything
dense vigil
Why don’t you put the data in order 
flint idol
really should've thought of that smh
i have a better idea
while [ `sha256sum $out` != `cat sum` ]
do
curl -Lo $out $url
done```
dense vigil
flint idol
keep on redownloading the file until the checksum is right
dense vigil
Bogosort ahh algorithm
flint idol
```sh
while [ `sha256sum $out` != `cat sum` ]
do
curl -Lo $out $url
done```
(actually this is technically how tcp works)
dense vigil
Do packets not have an order ID?
flint idol
if the receiver got a mismatching checksum, the packet is retransmitted
so technically
it is
but it doesn't really work like that at all
i lied
the lion does not concern himself with data accuracy
flint idol
if only i knew how i could make this buffer work
i need the consumer of the buffer to be able to read from it like a stream
but i need the producer to be able to write to it at any arbitrary offset, with the user only being signaled when there is some complete data (i.e., data at offset zero has been written)
what if i used a bitmap to indicate which regions of the buffer have not been written and which haven't been read
so ig the data structure i am looking for is a: string
flint idol
Why not just have a read and write pointer
the writer needs to be able to write at any arbitrary offset into the future
flint idol
How does reading it like a stream work?
a read() reads the currently available bytes
so like if the buffer has bytes 15-20 received, then a read would block until offsets 0-15 are filled
dense vigil
the writer needs to be able to write at any arbitrary offset into the future
But it can’t overwrite data no?
flint idol
no
flint idol
so like if the buffer has bytes 15-20 received, then a read would block until of...
the writer is also able to write bytes 20-30 for example if it gets those
dense vigil
Why not use a freelist of unwritten blocks or some shit
flint idol
seems pretty not right
idk ill figure it out im wording what i want terribly
maybe instead of writing stuff out of order
dense vigil
Do you know how many packets you get ahead of time?
flint idol
what i need to do is reorder segments so that they are in the right order, then process them in order
dense vigil
Why not make a sparse array you allocate lazily
flint idol
Do you know how many packets you get ahead of time?
definitely not
dense vigil
Damn
flint idol
all i know is that i will get at most rcv.wnd bytes ahead of time
but that says absolutely nothing
flint idol
what i need to do is reorder segments so that they are in the right order, then ...
yeah im not a fan of this solution
dense vigil
What if when you get a block you sort them automatically and if it can’t be put in sorted put them in a linked list elsewhere or some other thing
And then as needed go through it and sort it and put onto the main set
flint idol
idk
maybe a bitmap of "received bytes" could the solution
once there are enough contiguous received bytes from the start of the buffer, signal the reader
but of course a bitmap might not be the best
dense vigil
Yeah I don’t think a bitmap is the best
dense vigil
Like an RB tree?
flint idol
no rb-trees are only self-balanced
also wouldn't fit this, i don't need O(log n) search
i need fast iteration, so a list is likely best suited
but at the same time, a list will need allocation, and allocation might not be the best idea in a critical part of the tcp stack (but that is probably a problem for later)
struct received_segment {
intptr_t offset;
size_t len;
// list info
}```when a received_segment is found with offset=0, it signals the reader, and the reader will read from offset to len of that segment
if i knew how much data im getting in advance this could be easier
like if i knew for a fact i would get receive_buffer.length bytes, then i could simply have the reader block until that many bytes have been written
but alas, i dont have that information
actually wait
i do have that information
not exactly
that information
but what ill do is keep my current approach
but only signal the buffer when the kernel sends an ACK to the data
because ik for a fact that the tcp implementation ACKs when it knows it's emplaced all data
and i did that
and the data is still completely wrong
flint idol
because ik for a fact that the tcp implementation ACKs when it knows it's emplac...
i just figured out
that's wrong
yeah it's probably in my best interest to have the packets be sorted in order of SEQuence, then processed in that order
which sounds pretty complicated for my brain rn
so im gonna sleep
TODO for tomorrow: process packets in the right goddamn order
maybe even TODO for tomrrow: implement TCP SACK (selective ACK)
dense vigil
but at the same time, a list will need allocation, and allocation might not be t...
Why not pre allocate some
You could say keep X blocks pre allocated so when new shit comes in they fill up first
flint idol
if i process packets in the right order to begin with this problem doesnt exist
dense vigil
What happens to out of order packets? I presume you keep those non processed elsewhere
flint idol
more amusing than xhci
flint idol
What happens to out of order packets? I presume you keep those non processed els...
throw em in a sorted list
and process them once there are no more holes
dense vigil
more amusing than xhci <:pain:679047090474713098>
dense vigil
and process them once there are no more holes
flint idol
on my phone there is an audio file named weird.wav that was created at 1:30 am...
i do wonder what this could be
dense vigil
You should use miracle sort btw
flint idol
Listen to it
just did
vale nymph
You should use miracle sort btw
I prefer the quantum immortality sort
where you randomize it and if it is not sorted, you destroy the universe
vale nymph
by the theory of quantum immortality, it will be sorted
flint idol
its just some music I wrote during the winter break cuz I was bored
and thus, I will not be sending it
dense vigil
by the theory of quantum immortality, it will be sorted
Do you have such an implementation made?
vale nymph
yeah but I don't want to share it in fear of someone abusing the destruction of the universe for nefarious means
works pretty well, seeing as we are still here
flint idol
time to listen to 647 instupendo with the lights off starting at the ceiling
nvm i think ill get possessed if I do that
dense vigil
I wonder what would happen with an HTTP server
Like your OS hosting it
And seeings how scrambled it is
flint idol
And seeings how scrambled it is
probably not at all
because the problem is not on transmission but instead on reception
flint idol
true
dense vigil
true <:trl:829316641191886888>
It’s better than microslop
Actually though
When you finish this
Host an HTTP(S) server for your kernel
On your kernel
And have it though NAT to catch most attacks on a buggy stock
Stack
And also use it for people to pen test your stack
dense vigil
I could realistically host smth like that on my vps
I’d probaly slap it on a hetzner box inside a VM so you can control it a lot better if it did get PWNed and the person wasn’t so kind
flint idol
I was also thinking it might be possible to host a site on obos as long as I implement some sort of watchdog timer that resets the system if it detects a hang
dense vigil
I was also thinking it might be possible to host a site on obos as long as I imp...
If any attacker got in there you have no idea
With a VM you can heavily restrict the traffic and sniff it to make sure it’s not doing stupid ahh shit to be able to kill it
flint idol
If any attacker got in there you have no idea
how exactly would an attacker get through there if only ssh and http are open
its not like sshd would be any less capable on obos than it is linux
dense vigil
how exactly would an attacker get through there if only ssh and http are open
Just incase there is any bugs in the networking stack with crafted packets and overflows or other shit
I’d just be slightly concerned with a brand new networking stack etc?
Not like a hetzner box is that expensive for a cheaper one?
blazing python
If any attacker got in there you have no idea
who even need to attack obos server
flint idol
Just incase there is any bugs in the networking stack with crafted packets and o...
I paid attention to overflows while writing the network stack
dense vigil
With people fucking hammering every ssh and http server to find vulns and shit
flint idol
I paid attention to overflows while writing the network stack
and crafted packets
blazing python
I paid attention to overflows while writing the network stack
is network stack not hard
flint idol
the kernel drops any packets that lied about their length
and worst case scenario if it overflows and crashes, the watchdog timer takes over and resets the kernel
dense vigil
I’m just saying take some precautions so you don’t have to fucking explain to your VPS provider if it got pwned lmfao
flint idol
yeah im not actually gonna do it anytime soon
dense vigil
dense vigil
time to properly do out-of-order packet processing
What if you did a dynamic array / vector?
And then you get a packet with order 7, and you put it into slot 7 and if you then get order 27 you expand it more etc
And then if it’s lower than the max they can go into the proper spot
flint idol
What if you did a dynamic array / vector?
And then you get a packet with order 7...
cuz that's not really how sequence numbers work
I would need rcv.wnd amount of slots (0xffff)
for that to work
dense vigil
And that’s expensive
flint idol
indeed
flint idol
uh so ive managed to make the receive speeds and accuracy bad
scraped that idea
instead i will go back to what i originally did
devout niche
i feel like an rb tree for the reassembly queue is not ideal
since the reassembly queue should not have many segments on it in most cases
flint idol
true...
im thinking that instead of having a reassembly queue, i will insert packets into the rx buffer as soon as they're available
send an ACK (and signal the rx buffer) everytime rcv.nxt is advanced
and ill have some specific conditions for advancing rcv.nxt
that i will explain once i know what they are
this will also probably aid me when/if i implement the SACK extension
dense vigil
Oberrow quick question
What’s your block layer interface
How does a file system ask for a block
Because I wana make sure I’m not missing anything obvious
flint idol
the ext2 driver (only existing file system driver) either uses the page cache
or submits an IRP
to the block device
dense vigil
So like it would get a pointer to the page cache entry or would it copy to a buffer?
flint idol
it would get a pointer to a page cache entry
and then do whatever it needed with the block
dense vigil
How do you deal with the page cache entry being invalidated while in use? Would the FS free it back after?
flint idol
The kernel is to shit itself if that happens.
but uh as long as it isn't on a standby list, the pagecache entry, it should be good
flint idol
but uh as long as it isn't on a standby list, the pagecache entry, it should be ...
yeah this is actually like really fucked up
dense vigil
Like the pfndb may say BLOCK_CACHE_FREE
Or
BLOCK_CACHE_USED
flint idol
yeah this is actually like really fucked up
some of the pagecache entries are stored for long-term storage (i.e., for the lifetime of an inode)
dense vigil
And it will only reclaim free ones
flint idol
some of the pagecache entries are stored for long-term storage (i.e., for the li...
if the pagecache entry gets reclaimed then the driver can call it quits
dense vigil
I may do a seperate buddy allocator for the cached blocks to allocate from?
flint idol
if the pagecache entry gets reclaimed then the driver can call it quits
uh to fix this i will use the PHYS_PAGE_LOCKED
in struct page
flint idol
just set that bit until the driver is done with the entry, it should be good™
dense vigil
How should one make the allocator linked list skip over blocks like that l?
It wouldn’t be much overhead though
flint idol
if (page->flags & PHYS_PAGE_LOCKED) continue;
dense vigil
Because it won’t need to lock it for long
flint idol
?
dense vigil
Yeah
flint idol
anway if u will excuse me i have a reassembly queue to fix
flint idol
well im getting consistent download speeds and accuracy i believe
(that doesn't mean shit)
literally exactly 143 bytes a second
because for some reason the window isn't growing properly
sterile sparrow
hey not bad, you could use a BBS just fine 
flint idol
so to fix the whole "data isn't right" thing
im gonna double buffer this shit
you read that right
ill have the kernel intermediate receive buffer
that will be flushed to a user "ready buffer"
whenever the kernel sends an ACK
flint idol
that will be flushed to a user "ready buffer"
(and of course this buffer will be dynamic)
is it insane? yes
will it work? idk
devout niche
im gonna double buffer this shit
/*
* We manage rcv_wnd like this: it stays constant until we can't
* putnext() any further data and have to queue it; then, we
* reduce rcv_wnd by as much as we receive. this continues until
* we are free to putnext() again, at which point we reopen the
* window to its full extent.
*/
so i push the data up through STREAMS but if STREAMS says it can't take any more, then i queue it within the TCP driver
it seems to be a fine solution
flint idol
hm
devout niche
i found it interesting we both ended up with a 'double buffer' solution there
note the reopening to full extent
streams is full: data is queued up in the tcp driver in the meantime; streams doesn't allow messages to be sent up anymore until the streams queue reaches a low watermark; when that low watermark is reached, potentially a lot of data queued in tcp can go up at once, and then readers naturally coalesce their reads of this data
flint idol
the double-buffering solution gets me the right data
however it seems to fail to return the right amount of data in cases where the amount of data received in total is < rx window size
flint idol
yeah it's failing to push the full extent of data to userspace for some reason
huh it seems like it's not ACKing everything properly
flint idol
flint idol
im having fun running sudo ss -K dst 192.168.3.2 while sshing into the host from obos
stark sigil
sudo
flint idol
after I come back home today im gonna commit my working changes and implement selective ack
dense vigil
Btw how does OBOS handle MMAPing files from block devices?
I heard how managarm does it. Pretty complex but possibly manageable but I wonder how USB person does it
vale nymph
I would assume so seeing as it has a page cache
dense vigil
Page cache is gonna be hard. Wanting to see how others do it
It’s coming up soonerish on the list
flint idol
Btw how does OBOS handle MMAPing files from block devices?
I heard how managarm ...
pagecache
it refs the page in the pagecache
and then maps the physical address
possibly mapping as CoW if it is a private mapping
and possibly creating new entries if prefaulting is requested
if a pagecache entry does not exist and prefaulting is not requested then on fault, it is created and mapped
dense vigil
How do you handle invalidation of MMAPd files or other things so you can update the needed PTEs
flint idol
How do you handle invalidation of MMAPd files or other things so you can update ...
if any page is reclaimed then the kernel will remap the pte with an ID that the MM uses to know that it needs to fetch this file from disk or this anon memory from swap
and this ID is also used to identify the page in the swap
dense vigil
I mean like if say multible processes have this mapped and you need to edit the pages to fault, how do you keep track of that and to clean them all up?
flint idol
I forgor let me check
dense vigil
Thank you
flint idol
uh iirc I always map pages that are swapped out with a swap allocation id so the kernel never has to keep track of what pages are using a virtual page
I also think that if a pagecache entry has mappedPageCount >0 the kernel will not invalidate it
(not same as refcount btw)
dense vigil
uh iirc I always map pages that are swapped out with a swap allocation id so the...
So a swap allocation in of its self keeps track of the PIDs and the mappings they have for swap/MMAP stuff?
flint idol
// TODO(oberrow): Reclamation of file cache pages.```So then the kernel sees oh this is part of this to send it over here to nuke mappings
flint idol
So a swap allocation in of its self keeps track of the PIDs and the mappings the...
the swap allocation simply holds track of the swapped out page
// represents a swap allocation, as well as the physical page that it uses, if it was already read.
typedef struct swap_allocation
{
LIST_NODE(swap_allocation_list, struct swap_allocation) node;
uintptr_t id; // the key
size_t refs;
page* phys; // if !phys, this page must be read from swap.
struct swap_device* provider;
} swap_allocation;
OBOS_NODISCARD swap_allocation* MmH_LookupSwapAllocation(uintptr_t id);
swap_allocation* MmH_AddSwapAllocation(uintptr_t id);
void MmH_RefSwapAllocation(swap_allocation* alloc);
void MmH_DerefSwapAllocation(swap_allocation* alloc);```and then when the kernel swaps smth out, if there is already a swap allocation it will use that as the id otherwise it'll create a new one
dense vigil
So there currently isn’t a way to reclaim a MMAPd file from a process because it isn’t tracked?
Because many virtual address spaces could refer to such a page
flint idol
So there currently isn’t a way to reclaim a MMAPd file from a process because it...
yeah
flint idol
I also dont like the idea of keeping track of virtual pages in a physical mapping