i dont see the fb address printed tho?
#Astral
analog berry
analog berry
lol
fleet osprey
bruh
shitty camera, why does it always apply stupid post processing
hang on
there
analog berry
like i dont get it
fleet osprey
want me to transplant that bin into an ISO and try booting it with ventoy and see if it's any different?
craggy tide
yeah
analog berry
ventoy is just the thing that flashes the usb right
craggy tide
no
analog berry
how can it even affect this
craggy tide
i mean sort of
fleet osprey
no, it's a bootloader
analog berry
oh?
craggy tide
it does things that are more pervasive
fleet osprey
it's a heavily modified GRUB that loads ISOs from exfat / FAT32 at runtime
craggy tide
it's not a nice project
gritty shore
ventoy does chainloading I think?
analog berry
thats very weird
craggy tide
i vented my frustration at ventoy many times in the past
craggy tide
for fucks sake
gritty shore
i will see myself out
analog berry
want me to transplant that bin into an ISO and try booting it with ventoy and se...
yes
craggy tide
it will likely be different
i predict the fb will be marked as usable memory
if not, there's still gonna be signficant changes to the memory map
craggy tide
it's ventoy, i honestly do expect them to be capable of that
analog berry
like linux would crash with that as well
fleet osprey
I actually haven't tried anything else on that specific machine, in UEFI mode, from Ventoy
I can verify if Linux and Windows will crash
analog berry
lets see the map first
craggy tide
yeah i was about to say
i may be wrong in my assumption
which is why we need to see the map
fleet osprey
sorry, I have ADHD lol, I have trouble staying focussed on one specific thing when new things come up
one sec
craggy tide
if the fb isn't marked as usable then the only other option is that the Limine allocator allocated over fb memory but like
i see that as highly unlikely
analog berry
yeah probably not
unless the memory map has some edge case
that limine is not handling well enough
craggy tide
we'll see
analog berry
inb4 its random garbage
fleet osprey
swapping BOOTX64.EFI in the limine-c-template and rebuilding should be a good enough test I assume?
fleet osprey
huh, it's not working in QEMU, despite the sha256 of the BOOTX64.EFI in the ISO matching the one that I downloaded from your message
fleet osprey
ah
craggy tide
you can mount that
craggy tide
aight ty
analog berry
i mean in regards to reserved/usable
fleet osprey
as do I, there's an extra reserved and an extra usable at the end
or not
I'm just blind
nvm
:P
analog berry
yeah
@craggy tide limine bug ig?
At least u have the entire map should be easy to repro
grand shadow
so it wasnt oberrow curse
fleet osprey
welp, anyways, now that that's been noted, back to the curse of the spreadsheet
craggy tide
ehh
i need to look into this
@fleet osprey could you try this? you only have to post the last screen really
fleet osprey
alright, one sec
on Ventoy or direct, or both?
I'll just try both
doesn't print anything when booted direct (intentional?)
fleet osprey
yea, just tested it and it doesn't print anything in ventoy either
craggy tide
remember to also update the .bin file
fleet osprey
I did
craggy tide
alright try the second one
fleet osprey
wait was I supposed to let it boot, I thought it was gonna pop up before the menu again
but accidentally let it boot this time and it prints the mem map (half of it scrolls offscreen though) and then hangs
craggy tide
yeah
you're supposed to let it scroll like that and hang
that's intended behaviour
fleet osprey
ah, my bad, I figured it would be the same where it pops up in the EFI console before the limine menu
but anyways, you need the end of the map it printed?
craggy tide
ty
no this is fine
this is weird
especially given the above premise that the memmap is the same with ventoy and otherwise
that has to not be the case else it wouldn't explain why it works fine without ventoy
fleet osprey
maybe I could test directly chainloading with GRUB?
if it breaks there, that would rule out Ventoy garbage, and mean that it just explodes (somehow?) when chainloading
craggy tide
i wouldn't do random tests without having a clear idea of what we're testing
just not to waste time
fleet osprey
fair
craggy tide
@fleet osprey
fleet osprey
one sec
craggy tide
no probs
fleet osprey
new ss? looks the same
craggy tide
yeah
fleet osprey
craggy tide
this one actually has the fb entry
i made a mistake in the previous one that wouldn't show the fb entry because it was skipped over
fleet osprey
ah
craggy tide
whats important is that it seems fine there
fleet osprey
interesting, that's from Ventoy
I wonder why it wasn't page aligned when booting Astral then
craggy tide
try running it again, but edit the config on the fly to pass the kernel as a module or 2
fleet osprey
k
so just to verify your wording: add a line or 2 of module_path: boot():/boot/kernel?
craggy tide
yeah
craggy tide
try this iso as a whole please
craggy tide
the memory map looks fine
fleet osprey
it looks like what Astral was using as the base of that bad entry, is actually shown as the end address of the fb there?
craggy tide
it doesn't look anything like the astral one
do you think there's a chance you could extract the kernel out of this ISO and inject it into an astral image without changing anything else?
replacing the Astral kernel
the kernel is this
fleet osprey
yeah, hang on, I can do that
fleet osprey
yeah it looks almost the same (less reserved space tho?), and Astral still fails
so then I guess it must be a weird bug with Astral's VMM that somehow only has ever cropped up on one machine, in one boot method, and only when using Ventoy to load an ISO.....
wtf????
grand shadow
I dont change anything from the map entries which wtf?
wait its still not aligned to 4096
fleet osprey
oh wait what I didn't notice that
craggy tide
wait its still not aligned to 4096
this entry is wrong
the question is more so why it's only wrong on the Astral ISO
fleet osprey
well it's just Limine 8.0.9, pulled from GH
craggy tide
also the EFI cursor showing up there is weird
or at least what i assume is the EFI cursor
how old is this machine?
grand shadow
anyways
craggy tide
right
grand shadow
bruh
fleet osprey
true
grand shadow
did you run into this when porting qemu
robust geyser
also the EFI cursor showing up there is weird
\n vs \r\n?
craggy tide
it's flanterm
robust geyser
oh ok
craggy tide
it's supposed to own the framebuffer fully
wild marsh
did you run into this when porting qemu
i don't remember any mutex functions failing
hm i don't see how a mutex can be uninitialized
unless it's locked after being destroyed
maybe add prints to the mutex functions that print the function name and mutex pointer?
grand shadow
added the prints now its compiling again
@carmine swallow @robust geyser @analog berry @fleet osprey btw you probably should do a clean build, I changed the deps of pretty much all packages
narrow turtleBOT
wild marsh
.!unmute @grand shadow
narrow turtleBOT
wild marsh
are mods actually affected by the timout
grand shadow
yeah
couldnt even remove it from myself
seems like it never does anything with that mutex before this
time to peek in monitor.c
{
Monitor *mon;
qemu_mutex_lock(&monitor_lock);
mon = g_hash_table_lookup(coroutine_mon, qemu_coroutine_self());
qemu_mutex_unlock(&monitor_lock);
return mon;
}```i guess it got called before monitor_init_globals?
grand shadow
maybe it's coming from `vreport`?
qemu is compiling rn what is vreport
wild marsh
info/warning/error reporting function that takes a va_list (and there's are corresponding report functions that take var args)
i guess you could just replace !monitor_cur() with true there?
but that's just a guess
oh and apparently print_loc also calls monitor_cur
grand shadow
so I assume it hit some error and is just dying super early to the point it cant even call cur_monitor?
I'll do that to see what it says
wild marsh
that'd be my guess yeah
grand shadow
I removed it from error-report.c but it still seems to happen?
grand shadow
I think I will just build qemu with debug symbols, printf(__builtin_return_address(0) and addr2line
wild marsh
monitor_cur is also called in qemu_vprintf
wild marsh
i fail to see where it'd call it though at a glance
grand shadow
astral so good qemu doenst know how to react
wild marsh
trolled
grand shadow
time to see what it wants to say
(probably wont be pretty)
oh bruh it just wanted to print a warning
wild marsh
lmfao
grand shadow
though it still didn't keep going
do I need pthread_kill?
seems like so
well
I will implement sigpending since its easy and I already have everything needed and then tomorrow I implement the thread signaling
grand shadow
oberrow curse after all?
it was not oberrow curse after all :')
wild marsh
pthread_kill should be easy enough no?
robust geyser
once again astral proven to not be shit
wild marsh
assuming your threads behave more like processes that are in one thread group (aka what it's like on lunix )
grand shadow
tgkill would literally just be loop through the thread list of a process and call signal_signalthread on it if it finds it
I can probably do it today
grand shadow
once again astral proven to not be shit
(citation needed)
wild marsh
i wonder if qemu is gonna also be borked on astral or if that's managarm specific and not mlibc specific
qemu seems to struggle with disk io sometimes from what i can tell
a simple test bootsector can't reproduce it ofc but iirc grub fails to load when i tried booting the debian gnu/hurd disk image
grand shadow
Arhffhfdjdjcndks I think Im doing sigsuspend wrong? Apparently it wants the old signal mask to be restored after the signal returns
Jfc I hate signals
Ill deal with this tomorrow
fleet osprey
btw @analog berry got some ACPI errors for ya, since you were curious about any on my Optiplex
here's some from an HP ProBook 4540s
analog berry
btw <@353593948523986945> got some ACPI errors for ya, since you were curious ab...
Could be normal, did you look at dmesg for this laptop on Linux?
fleet osprey
I can, one sec
analog berry
Acpi dump + dmesg output would be very nice to have
fleet osprey
also I need to swap BIOS -> UEFI for that
Limine has bugged video output in UEFI on this laptop, so that boot I sent there is BIOS
but Linux is installed as UEFI
would that be an issue?
if it is I can probably go install GRUB for BIOS on it
analog berry
would that be an issue?
Could potentially taint the results since its the firmware that does aml patching, but we can try
fleet osprey
alright then I'll go install GRUB to the drive as an MBR then
just will take a little bit more time
analog berry
Nice, thanks!
Its very possible that its just a firmware bug because its relatively harmless, but again could be a sign of some other uacpi problem
craggy tide
Limine has bugged video output in UEFI on this laptop, so that boot I sent there...
holy shit how many machines where Limine is randomly bugged on do you have?
@analog berry btw i fixed the issue from yesterday
analog berry
holy shit how many machines where Limine is randomly bugged on do you have?
Ikr, its a gold mine of bugs
analog berry
<@353593948523986945> btw i fixed the issue from yesterday
What was the problem?
fleet osprey
that's the ProBook, it's the one with the weird pitch that you've known about for a bit
craggy tide
What was the problem?
GCC
analog berry
Huh?
craggy tide
basically a symbol (the image size) from the linker script had a bogus size compared to when using clang
i worked around the issue
analog berry
Why did that only manifest itself on that pc?
craggy tide
by not importing linker symbols that had difference operation in them
well for one it needed GCC
and UEFI
so that already narrows it down in terms of how many people tested it
and then it basIcally manifests as Limine trying to allocate itself in the memory map but the size is really large
analog berry
Is the default limine release built with clang or?
Why was astral even using gcc there
craggy tide
so i guess depending on the memory map layout and depending on the location of the loaded image in memory, it may not have any effect
craggy tide
Is the default limine release built with clang or?
the binary release is compiled with LLVM yes
Astral is using GCC because that is what the Limine Jinx recipe in Astral does
it uses the Astral cross compiler
craggy tide
so basically you had to be quite unlucky (or luckly if you ask me lol) to hit the bug in such way that it would cause issues
also in the meantime i noticed and fixed another bug but that wasn't this bug
in any case
yeah
software be like
analog berry
Nice
fleet osprey
Acpi dump + dmesg output would be very nice to have
actually hang on. I have a dmesg with more in it, because i skill issued the grep
analog berry
Thanks ill take a look
ah i see
uACPI tries to _REG it very early, during namespace load, but it references a symbol thats added later on
This _REG is re-executed again after namespace init anyways
ill see why ACPICA doesnt try to do that, but its not really an issue other than the log spam
fleet osprey
👍 just figured I would let you know since you were looking for error logs
analog berry
yeah definitely
its important to investigate any warning/error for uacpi at this point
supple hatch
i lobe uacpi
love
still waiting for qwinci to work on the uacpi-rs bindings for tables so i can stop using this crappy acpi crate 😭
qwinci great work btw
analog berry
still waiting for qwinci to work on the uacpi-rs bindings for tables so i can st...
but he did yesterday lol
analog berry
i lobe uacpi
thanks
supple hatch
but he did yesterday lol
he did?
analog berry
he literally sent it in your channel
supple hatch
he didnt?
robust geyser
u8042
grand shadow
Astral is gonna be running webkitgtk on real hardware before ps2 works
errant zephyr
dw uPS2 is going to come out any minute
uHID
ps2+usb hid+i2c keyboard/touchpad+applespi
errant zephyr
you need to do everything at once :^)
okay tbf you dont need usb hid in one package
but the rest are pretty much just keyboard and mouse
analog berry
like if i just do HID i need the host of the library to implement usb on their own
and for PS/2 i can just implement an i8042 driver myself
so it kinda idk
makes it a bit chaotic as to what needs to be implemented where
carmine swallow
and usb isn't that bad to implement yourself either, at least its not so annoying as ps2
analog berry
although a separate hid parser library like uHID could be done
it is just another bytecode afterall
maybe i2c could go into uPS2 as well somehow idk
supple hatch
release ups2 first
analog berry
ill start with i8042 ofc
it will have optional uacpi integration as well
for finding port numbers and device presence
analog berry
u wish
carmine swallow
that's not the annoying part, the annoying part is the quirks and initialization + mixed scan data and responses
fervent fjord
nah you guys are cooked for this take
fervent fjord
I have no idea about multiplexing because I just dont, but the rest is trivial even
its bad that I do but I seriously doubt the osdev mettle of someone telling me PS2 keyboards are hard
mice have a bit of silly to them with the different modes of scan for 3 and 4 buttons and touchpads
robust geyser
if you want it to work on the quirkiest hardware possible then it is hard iirc
analog berry
well mathewnd has ported tons of software and has a relatively advanced stable kernel
yet his ps2 doesnt work on real hw after 3 rewrites
so no its not as trivial as you think
fervent fjord
I cannot speak as to his project, because I just dont know, there could be complications from a lot of stuff that are project specific, but seriously, I dont get the difficulty
robust geyser
look at the ps2 driver in infy's old OS
analog berry
the difficulty is you have two ports that have overloaded meaning, the sequence to resetting and detecting devices is also racey, has unspecified timeouts, may reply with random garbage, may not reply at all, may need a different way of resetting, may die because of attempting to detect extensions
errant zephyr
basically its just quirky to death
fervent fjord
yeah but thats standard device fuckery, all devices have their issues like that
errant zephyr
and its a lot of legacy
fervent fjord
never attempt a SATA driver
errant zephyr
and its a device that is implemented by different people in different ways for the past 40 years
analog berry
u have to know which commands are allowed to timeout, which need retries, which need garbage byte discarding, which bytes are not actually garbage, which may hang the controller etc
its a giant mess basically
fervent fjord
but nothing of that is hard, its just tedious
analog berry
ahci is really really bad yeah
but for the most parts its setting specific registers and waiting the right amount of time
for ps2 its way more mental gymnastics
grand shadow
yet his ps2 doesnt work on real hw after 3 rewrites
It doesnt work on real hardware anywhere other than the 2 fucking machines I have available to testing which is crazy
I probably couldve gotten it to work if I could at least test it
analog berry
yeah
fervent fjord
I dont know what you have
grand shadow
Its likely some very specific step in initializing ps2 that I missed
Even nvme works on real hardware lol
analog berry
the point im trying to make is its not that trivial even tho "looool its just 2 ports 
"
robust geyser
i kinda notice that the older the hardware usually the harder/trickier/quirkier it is to drive
analog berry
yeah i mean
its because of all the crust and implementations it has accumulated over the years, as well as extensions and not very good initial design
fervent fjord
people develop less PS2 controllers nowadays
especially with those starting to be baked in CPUs and stuff
so less variance
analog berry
sorta, nobody uses real 8042s anymore ofc, but a lot of laptops still ship builtin keyboard/mouse over ps/2, as well as some modern mobos that still have ps/2 ports
fervent fjord
yeah but back in the day you would have so many manufacturers, nowadays all of those usually live in the super io chip in the mobo which is made by like 3 companies, 1 of them, nuovoton, has like 80% market share
analog berry
yeah but back in the day you would have so many manufacturers, nowadays all of t...
probably fair yeah, id like to hope they reuse that code on all new things
analog berry
ok yeah i cross checked with ACPICA, it doesnt run _REG that early unless the namespace has already been loaded (aka we're in a method call or very late host-invoked table load), so ill add that condition as well
thanks for testing
flat copper
ventoy is just the thing that flashes the usb right
it's a very cranky grub + memdisk setup that injects custom proprietary binary blobs into the kernel (not even kernel modules, it patches the kernel) so that you can run a live linux system from the iso file on fat32
and it's chinese and quite shady (developers block all discussion of non-shit solutions for booting without injecting a potential rootkit)
errant zephyr
it's a very cranky grub + memdisk setup that injects custom proprietary binary b...
i think some of them are modules and oss
tbf
flat copper
the repo is full of prebuilt binaries for god knows what
flat copper
yeah but even if they do they're wildly outdated
errant zephyr
iirc the sources ARE there
ah
fucking great
i mean tbf
i have no idea how to do what they do without doing weird shit
hmm
okay now i have one idea
kinda
errant zephyr
i have one good idea and it requires you to have an amd cpu and a bios that isnt patched for the smm memes
and i have another idea of just running a vm
and faking a disk that way
and another idea of stealing an AP to run a virtio
i have no idea if there is actually any non awful way to do it though
you basically need to convince a distro to mount a volume, losetup a file in that volume, and use that as the root
not sure if there's a way to magic that together even with linux
let alone the bsds
ok well with linux just patch the kernel ig :^)
or patch the initial ramdisk to do the magic
and then stage into the real ramdisk
for windows isos, fake up netboot
on linux maybe use the ramdisk stuff to copy the iso to there
or just inject a driver that autoprobes itself and always declares there to be the right iso in memory
and claims to be a block device
or fake some sort of device
somehow, ideally with aml
ok cursed idea
choose a device with a doorbell
point doorbell at lapic self-ipi
to the acpi event vector
simulate a disk in aml
fleet osprey
How many machines do you have holy shit
to quote my notes from testing Limine USB boot (where I tested Astral as well as the C template across various boot methods), mostly because I'm too lazy to retype it
[2] - I have the tested all cases possible based on following list of hardware: - HP ProBook 4540s [UEFI (known buggy), UEFI CSM) - Dell OptiPlex 755 [Legacy BIOS] - Lenovo ThinkPad T14 Gen 1 (AMD) [UEFI only] - Custom PC, ASUS PRIME H610M-E D4, Intel Core i3-12100f [UEFI] - Dell Precision T1650 [UEFI, UEFI CSM] - QEMU, see note 3! [UEFI, Legacy BIOS] I planned to also test on the following, but couldn't due to one reason or another, unrelated to Limine - Dell Latitude E5250 [UEFI, UEFI CSM]
oh and if you would've asked that a month ago, I would also add a ThinkPad T480, and a Latitude 5491
I sold those 2 sadly
grand shadow
to quote my notes from testing Limine USB boot (where I tested Astral as well as...
Damn
I think after I get qemu working I will go and do some more general improvements along the kernel
Like making pci check for drivers in a section
Same thing for other things which need 192829378 blahblah_init
To clean up the code
grand shadow
Goal is to have as little driver init code in main as possible
fleet osprey
good goal
supple hatch
techflash you can ever try nyaux on ur optimux prime pc 
?
mayybeeee
i mean do in ur own time but it would be quite helpful when i get to writing drivers for the OS, currently trying to fix uacpi atm
fleet osprey
Like making pci check for drivers in a section
just a note on this though - I would probably have a global driver_init or similar, rather doing it in PCI init
that way it's set up to handle non-PCI backed devices when support for those are ready
like probe for devices on all available devices, like PCI, USB, (maybe weird shit like i2c or whatever?), then based on that list of all devices found, init some drivers
supple hatch
like i plan to support things like smp and multiple ioapics in this rewrite
fleet osprey
techflash you can ever try nyaux on ur optimux prime pc <:memedown:1267962938132...
yea, when I get home
sure
fleet osprey
if you would've asked an hour ago I could've done it immediately
but now you gotta wait until I get out of school :(
supple hatch
no need i still need to fix uacpi
fleet osprey
ah
fleet osprey
k
grand shadow
just a note on this though - I would probably have a global `driver_init` or sim...
Ikik I meant like pci_init automatically calling the drivers for any pci devices it finds
Same thing for acpi and usb and etc
fleet osprey
ah
grand shadow
why did init die
the qemu window did open though
so thats progress
signaled so hard I killed init
grand shadow
ok so
does seem like some thread is stuck in kernel mode
nvm?
I think something is racey
oh defo
fleet osprey
uh oh
grand shadow
seems like something deadlocks the vmm
the kernel vmm space lock in particular
pulled out gdb and looking into this rn
grand shadow
the mutex has several threads waiting on it without a thread holding it which wtf
either my wakeup is broken or my semaphores are broken
grand shadow
seems like the thread just doesnt get woken up?
grand shadow
welp that was fucking stupid
fleet osprey
oh?
grand shadow
it was an issue in my yield -> sleep -> get interrupted by a signal instantly instead of sleeping path
thread_t *next = runqueuenext(sleeping ? 0x0fffffff : thread->priority);
bool gotsignal = false;
for (int i = 0; i < NSIG && thread->proc; ++i) {
void *action = thread->proc->signals.actions[i].address;
if (SIGNAL_GET(&thread->signals.urgent, i)) {
gotsignal = true;
break;
}
if (action == SIG_IGN || (action == SIG_DFL && signal_defaultactions[i] == SIG_ACTION_IGN) || SIGNAL_GET(&thread->signals.mask, i))
continue;
if (SIGNAL_GET(&thread->signals.pending, i) || SIGNAL_GET(&thread->proc->signals.pending, i)) {
gotsignal = true;
break;
}
}
if (sleeping && (thread->shouldexit || gotsignal) && (thread->flags & SCHED_THREAD_FLAGS_INTERRUPTIBLE)) {
sleeping = false;
next = NULL;
thread->flags &= ~(SCHED_THREAD_FLAGS_SLEEP | SCHED_THREAD_FLAGS_INTERRUPTIBLE);
thread->wakeupreason = SCHED_WAKEUP_REASON_INTERRUPTED;
spinlock_release(&thread->sleeplock);
}
it gets the thread at runqueuenext but if it gets interrupted it doesnt put it back there
which means the thread gets lost in the limbos of the kernel
fleet osprey
ah
grand shadow
only happened now because the way qemu does coroutines forces this to happen
fleet osprey
you may want to implement some form of locked up thread detection
I've found that immensely useful when poking at Linux drivers
and that would've caught this
grand shadow
this is the thread effectively being lost by the scheduler and never found again 💀
2 line fix anyway
fleet osprey
ah... yeah may not have solved that, I misunderstood what you meant ig
I thought you meant that the thread technically still existed, but the scheduler just didn't actually give it any time
grand shadow
anyways
still has some rough edges (wtf is causing it to segfault? and only one thread?)
but it does work
gonna see if some more complex iso works
what should I try
not too big so I dont have to wait for the download too much
fleet osprey
boot Astral in Astral 
grand shadow
hmmmm
I think I have a minimal iso in my server
100 mb
damn
I'll just toss it into the hard disk image
fleet osprey
I wonder how many layers deep you could theoretically get before something blows up
maybe you could get Astral in Astral in Astral before any of the QEMU processes segfault, if you're fast enough
fleet osprey
oh yeah that doesn't work with how you make your ISO for some reason
that's an issue I was looking at investigating eventually, because other Limine ISOs (particularly, I tested with the C Template) do work with that setup
grand shadow
yeah qemu is super unstable
I need to see what it complains about
also I wonder what 75 is
yeah I need to see WTF its trying to do with memory
but thats something for tomorrow
oop almost got to the limine screen
signal stack keeps being bad
which idk if is my sigaltstack issue or qemu's
I think it trying to load from disk is causing it to be slow
ohhh yeah the background
almost there
fleet osprey
damn
even if it can't do it yet, it's still impressive as hell that it's even to the point of trying
grand shadow
I just need to unfuck whatever qemu is trying to do with memory and make it faster
yeah its too unstable
it didnt load the kernel
fleet osprey
tbf, QEMU w/ TCG is just slow :P
fleet osprey
hmmm, wonder why
does it suck trying to load from disk perhaps?
like you're thinking?
grand shadow
that is probably one of the things
its still very slow regardless
but yeah it taking really long to load the disk is probably just astral's disk drivers sucking
I can just run qemuprof.sh from mint too
tomorrow tho jesus christ thats enough osdev for me today
grand shadow
@craggy tide look recursive limine
alright tomorrow I will make qemu stable and then make it fast (as fast as I can anyways)
grand shadow
true
grand shadow
something is telling me my signal alternative stack is not right
addr2line says the } after return ret segfaults
which is more like the stack is bad
and its not a protection issue, there is just no range
I should log to see what signal stack it wants to set
I hate posix signals
obsidian pivot
something is telling me my signal alternative stack is not right
I wonder how that'd even happen 
grand shadow
well that was idiotic
I think I fixed it
time to see if I can load astral for real this time
So far so good
Go into limine promot
Aaand pressing enter made it kill itself
Nice
obsidian pivot
Aaand pressing enter made it kill itself
sounds like something my ps/2 driver would do (if I had one)
you click enter and it triple faults
bcuz curse
grand shadow
Qemu passed an invalid sigstack when I pressed enter for some reason
obsidian pivot
but ofc having qemu is a big W
grand shadow
sounds like something my ps/2 driver would do (if I had one)
Hey thats the mathewnd curse
obsidian pivot
no it's the triple fault part
since I got an astral iso I'm going to test it
on my hw
grand shadow
Qemu passed an invalid sigstack when I pressed enter for some reason
it wasn't even like memory that wasn't mapped it was just mapped read only????
obsidian pivot
currently at af92e8eb871e3e6317e02b55637bc24a64eeed07
grand shadow
grand shadow
sadly it doesnt
obsidian pivot
which reminds me I have my disk controller on ide emu
because I was tryna run DOS
and thought that would matter for some reason
obsidian pivot
Was that vmm bug from techflash fixed
grand shadow
and doing it wrong I think
grand shadow
Was that vmm bug from techflash fixed
wasnt a vmm bug it was firmware being fucky I think
obsidian pivot
Guess my firmware is fucky then
OBOS almost boots fine on my hw
It only gpfs sometime during kernel init
grand shadow
Guess my firmware is fucky then
did it happen
obsidian pivot
Yes
grand shadow
do you have it built
obsidian pivot
Yes
grand shadow
can you do this
obsidian pivot
Sure
grand shadow
this time it'll be fr
oh god its gonna take so long to load
I think I'm just loading the kernel for now 💀
actually waitttt
I might just run from tmpfs this one time
obsidian pivot
my camera is dogshit
But
It had an unaligned address before crashing
E057e400
Type 5
grand shadow
type 5 is bootloader reclaimable I think
grand shadow
which needs to be 4096 aligned
bruh
@fleet osprey what ended up being the issue with the unaligned bootloader reclaimable
fleet osprey
<@632725314111275018> what ended up being the issue with the unaligned bootloade...
compiler issues with Limine
obsidian pivot
This length looks sus
fleet osprey
official builds use clang, Astral's uses GCC, which had a bug, it was fixed now tho
obsidian pivot
0xA5433cc00bytes of reclaimable bootloader ram?
fleet osprey
if you want to bump to 8.0.10, it's fixed there
specifically, this commit fixed it https://github.com/limine-bootloader/limine/commit/8382cd1794b32224f9f37f3571672cc651bf938a
grand shadow
ohhhhh
yeah that sounds like a good idea
@obsidian pivot it wasnt astral vmm bug nor firmware bug it was gcc bug
fleet osprey
it was a GCC bug that created a Limine bug that makes Astral explode
obsidian pivot
Ignore thatmsg
fleet osprey
such a chain of events lol
obsidian pivot
so how exactly
do I bump limine
I have no idea how astral's code structure works
fleet osprey
Look at what I did in my pr
Just don't touch the config like I did (was only necessary due to major revision bump, 7.x.y to 8.x.y)
Only do the version bump
Hell I might just do it from the GitHub web editor on my phone
grand shadow
you still need the checksum of the package
fleet osprey
Bold of you to assume I can't obtain a sha512 on Android
obsidian pivot
it'd be funny if astral crashes on my hw because of an allocator bug
grand shadow
anyways time for the truth
obsidian pivot
during uacpi init
obsidian pivot
anyways time for the truth
astral is secretly a fork of vinix?
grand shadow
it'd be funny if astral crashes on my hw because of an allocator bug
theres a chance I dont have a slab of big enough size in alloc() for uacpi memes
grand shadow
astral is secretly a fork of vinix?
no if qemu will work
oh god its still slow
a bit faster though
obsidian pivot
bruh all I did was do the thing
git pull
why is it doing all this
and what does libotolize do
and why is it taking so logn
grand shadow
poor oberrow
I did a fuckton of changes to packages youre gonna be in there for a while
do a make clean and change DISTROTYPE to minimal
obsidian pivot
fleet osprey
you still need the checksum of the package
Anything is possible if you can get a shell :)
grand shadow
just ssh into astral 5head
obsidian pivot
it's building GCC again
I've had enough of building GCC
(ptsd from m68k port)
((and slightly cursed gcc))
(((and binutils)))
((((it didn't like my ld script))))
(((((only compiling with gcc though)))))
((((((clang worked fine))))))
(((((((common GCC L)))))))
obsidian pivot
sounds about right
they gave me free storage for my computer backup
(they don't know it)
((hehe))
dam that was fast
already on mlibc build
b r u h
t h a t
w a s
b i n u t i l s
fleet osprey
There we go
Created, committed, and made a PR, all from my phone, sitting on the toilet
Quite possibly the funniest thing I have done in my entire life
fleet osprey
@grand shadow you should merge it, for the funnies, lol
obsidian pivot
good
grand shadow
@fleet osprey if it doesnt work I will come out of the toilet and get you
grand shadow
cant test atm cuz qemu is busy waiting for qemu waiting for limine to load an initird
fleet osprey
No reason it should break tho, it's a simple version bump
obsidian pivot
well guess I'm testing it then
fleet osprey
👀 whar
obsidian pivot
ban him
grand shadow
pushed the fix
fleet osprey
I missed the quote?
Bruh, GitHub default editor moment
obsidian pivot
no you missed the tarball_sha512="
fleet osprey
Oh shit it nuked the entire var?
obsidian pivot
fleet osprey
Wtf??? I'm ashamed in GitHub now
grand shadow
damn limine is still loading the initrd
its gonna be beuatiful when it works
or I could have just
passed a minimal nvme disk to it
oh
well too late now sunken cost time
ah fuck it
grand shadow
btw do git pull again when youre done I removed a funny debug print I forgor about
(itll just rebuild the kernel if you run make)
aaaaa more sigaltstack memes I think
obsidian pivot
IT'S ALMOST DONE
nvm
WHY
TH
FUCK
IS IT
is that
coreutils
or the other thing
gcc
grand shadow
probably coreutils since thats in the minimal iso
if you remembered to change it to DISTROTYPE=minimal in the makefile lol
obsidian pivot
I did
obsidian pivot
ye its coreutils
grand shadow
aaaaa more sigaltstack memes I think
nope I think its actually just my vmm complaining during a safe user op
grand shadow
damn rip
obsidian pivot
Don't wanna deal with this shit
grand shadow
did it run with the new limine
grand shadow
damn
@wild marsh trying to run with an nvme disk in qemu results in this
Was that what yall had run into?
gonna try this in real hardware then go to sleep
grand shadow
Cool probably some mprotect memes?
Too lazy to find out
(note the kernel didnt panic I just printed the context of the segfault)
fleet osprey
i3 my beloved :)
grand shadow
I like how fvwm looks but maybe I can add another option other than twm and fvwm
grand shadow
love the kerneç
ç is right next to l on the keyboard and I didnt notice it for a while and now itll stay there until I eventually remove the printf
neon crane
ç is right next to l on the keyboard and I didnt notice it for a while and now i...
I thought it was some kind of corruption lmao
grand shadow
Nah its just my brazilian ass doing a typo
wild marsh
damn
i don't recall seeing any mlibc errors, just grub failing to load itself
but i can check again in a sec https://uwu.foundation/5JSBxyG.png
grand shadow
I mean its just a missing sysdep its not that hard
Its like a pread but readv
Though yeah checking it again would be nice, I assume it mightve been fixed already
(since limine loads the kernel just fine)
Or maybe its a grub only issue
wild marsh
link to astral iso?
supple hatch
i don't recall seeing any mlibc errors, just grub failing to load itself
qookie will i use xbstrap this time for nyaux
or nah
grand shadow
link to astral iso?
Theres one at https://avalios.dev/bootdisk.iso
Its a bit old but should work
wild marsh
lemme try that then
wild marsh
Theres one at https://avalios.dev/bootdisk.iso
seabios fails to even boot into the cd'
Boot failed: Could not read from CDROM (code 0005)
and i don't see it calling preadv
only some warnings about mincore
grand shadow
and i don't see it calling `preadv`
Preadv was nvme
grand shadow
`Boot failed: Could not read from CDROM (code 0005)`
Wtf
wild marsh
but i don't think those are from qemu (maybe mesa? idr)
wild marsh
i have absolutely no idea why the reads seem to fail
grand shadow
Maybe managarm memes
robust geyser
qemu-system-x86_64 -cdrom bootdisk.iso boots into limine
probably
or maybe old qemu/seabios
maybe a bugged seabios version
wild marsh
qemu 8.1.4 with seabios 1.16.2
robust geyser
1.16.3 here
and qemu 9.0.2 or whatever
so yeah i doubt
what command line did you use?
robust geyser
can you sha256sum the bootdisk?
793f17469b0518fd92dcfb4ccd1dc6abcb7374cf735cab383f59582fe3192126
wild marsh
https://uwu.foundation/2qJMEEo.png seems to match
robust geyser
also try qemu-img info bootdisk.iso
maybe qemu is fucking something up
image: bootdisk.iso
file format: raw
virtual size: 101 MiB (105549824 bytes)
disk size: 101 MiB
Child node '/file':
filename: bootdisk.iso
protocol type: file
file length: 101 MiB (105549824 bytes)
disk size: 101 MiB```thats the result for me
wild marsh
prints the same thing
wild marsh
i mean it reads fine for a bit and then fails later on
robust geyser
odd
wild marsh
from seabios code (return value is the code that's shown) ```
// Validity checks
if (buffer[0])
return 4;
if (strcmp((char*)&buffer[1], "CD001\001EL TORITO SPECIFICATION") != 0)
return 5;
inb4 off by one error in pread or something
hm i wonder if it behaves any better if i put the iso on tmpfs
okay it works better if i put the iso in /tmp
so it's a bug in libblockfs
now the kernel panics
grand shadow
Damn assertos moment
errant zephyr
lmfao
grand shadow
https://uwu.foundation/3dDotUJ.png
Also wtf does sdl on managarm not have window decorations?
wild marsh
Also wtf does sdl on managarm not have window decorations?
thats a wayland and weston thing
weston doesn't draw any server side decorations
sdl supports libdecor though which provides client side decorations for it so we could get that
wild marsh
weston doesn't draw any server side decorations
to be clear, if you have sdl without libdecor on linux it'd have no window decorations on weston as well
grand shadow
Just run it on xwayland ez
wild marsh
i don't think it makes a difference for sdl?
carmine swallow
btw @grand shadow what is the codec model on either one of the pc's/laptop that you tested the hda driver on? iirc you can find that in device manager when going to the properties of the soundcard (or alternatively you could just send the string in the bios path property or whatever its called that contains the pci vendor/device id)
carmine swallow
thanks, the reason that I wanted them was that I want to make sure that there is not some random weird gpio stuff that would have to be done for those specific cards because if its something like that then there is no way I can fix the issue by just guessing what the problem is
grand shadow
thanks, the reason that I wanted them was that I want to make sure that there is...
HDAUDIO\FUNC_01&VEN_10EC&DEV_0887&SUBSYS_104386C7&REV_1003\4&2000AE52&0&0001 something like this?
this is my desktop
my desktop has the nvidia gpu one too but I sent the mobo one
INTELAUDIO\FUNC_01&VEN_10EC&DEV_0295&SUBSYS_1025143F&REV_1000\5&42ED54C&0&0001
my laptop
carmine swallow
looks like they are alc887 for the desktop and alc1220 for the laptop
or maybe alc295 for the laptop, idk there seems to be some conflicting info but its good enough anyway
grand shadow
Hopefully you can figure out whats wrong that makes them not work
carmine swallow
at least for the desktop one doesn't look like there are any other quirks than some pll hw bug (I highly doubt that it would cause the issue you are having tho)
but ig I can add the fix for it
grand shadow
What is this pll bug
carmine swallow
idk it doesn't explain what it causes ```c
/*
- Fix hardware PLL issue
- On some codecs, the analog PLL gating control must be off while
- the default value is 1.
*/
https://en.wikipedia.org/wiki/Phase-locked_loop this thing?
carmine swallow
yes
though the thing with hda in general is that there are a lot of quirks (there is a 10k line file in linux for the quirks in realtek codecs, though tbh that's like the majority of all of them anyway)
idk how likely do you run into them tho
especially with a playback only driver that doesn't do any jack detection or anything more advanced like that
analog berry
though the thing with hda in general is that there are a lot of quirks (there is...
lmao
what a pile of crap
winged basin
though the thing with hda in general is that there are a lot of quirks (there is...
i remember having to hand craft them to get reliable audio on very new laptops with netbsd or freebsd
horrible
carmine swallow
yeah that's bad
carmine swallow
I added the quirk, idk if its worth testing again yet tho (and not rn anyway because I am rebuilding astral)
if only I would have more machines myself that I could test this stuff on 
I only basically have the tigerlake laptop and then my desktop that has alc897
- qemu and vbox ofc but pmuch any driver will work on them no matter how garbage it is
obsidian pivot
@grand shadow
I have pulled
Astral
and my limine is a version behind
I run the jinx rebuild command twice for limine
fleet osprey
yeah, if Limine isn't built, how do you expect to be able to build an image using it's files 😅
fleet osprey
that probably should've rebuilt Limine, but just run that command for me
just to verify
because if it doesn't exist or is empty, well there's your problem
grand shadow
I added the quirk, idk if its worth testing again yet tho (and not rn anyway bec...
I mean you could add a debunch of debug prints and ask me to test it
fleet osprey
I'm already trying it from a clean build just in case
but I don't see any reason it wouldn't work unless host-pkgs/limine just flat out doesn't exist, or is empty
fleet osprey
obsidian pivot
PS2 bugged
grand shadow
the mathewnd curse
grand shadow
do you have anything in that port/emulation turned on?
and does power button work? (for infy)
obsidian pivot
Power button works
grand shadow
based
obsidian pivot
But never shuts down
obsidian pivot
It said powering off
grand shadow
oh bruh
obsidian pivot
grand shadow
well its still safe to turn off your machine
grand shadow
@analog berry
grand shadow
bruh
grand shadow
do you have anything in that port/emulation turned on?
@obsidian pivot
obsidian pivot
I'll check
Doesn't look like anything with ps2 in bios settings was touched
And my older obos rewrites worked with ps2 emulation on that machine
Tbf they didn't support multiple ports
grand shadow
did they do the whole ps2 reset stuff
grand shadow
hm
analog berry
OBOS can shutdown fine
thank god
astral skill issue then
i doubt it even reached uacpi there
craggy tide
seems like whatever make does doesnt make {,} expand
that's a bashism
analog berry
but u can try with a different log level
grand shadow
yeah I figured
grand shadow
i doubt it even reached uacpi there
I mean after that last print all it does is write a letter to /dev/acpi which literally just has a swtich() in the device write lol
but he can see
analog berry
add printfs to kernel i guess
grand shadow
With native ps2
pls
grand shadow
my guess would be obos does some more initialization or smth I dont do
obsidian pivot
For what
grand shadow
acpi
obsidian pivot
Well I just initialize uacpi as I did in my 3rd kernel
Which was taken from the test runner
In February
Or January maybe
And I also added the gpe finalization thing
I need to fix gpes in obos
carmine swallow
I mean you could add a debunch of debug prints and ask me to test it
I made the hdaprints go to the screen and added a print to the stream irq handler that it prints at max 6 times, I am not really sure where else could I add prints that would be useful without them being spammed all over the place (and the test file is now a lot shorter and called test.wav, I just dd'd a part of it)
analog berry
And I also added the gpe finalization thing
wait i dont think i even initialize GPEs in astral lol
not that it should matter
but maybe it does on that hw
anyway we need logs to see what happens
analog berry
by logs i mean in astral kernel
perhaps new printf statements in the write handler for /dev/acpi
analog berry
I thought you meant to see if gpes are finalized
i know its not because i never call uacpi_finalize_gpe_initialization in astral
analog berry
lmao
analog berry
i know its not because i never call `uacpi_finalize_gpe_initialization` in astra...
i forgor to do that 💀
obsidian pivot
My panic has no stack trace
So I have no clue how this could've happened
All I get is an rip
obsidian pivot
My panic has no stack trace
TODO: This
fleet osprey
also
btw, just did a fully clean build and have no issues with installing Limine like this
sounds like you did something wrong on your end
craggy tide
his /bin/sh is not bash, that's why he had the issue
it was fixed anyways in latest Astral commit
fleet osprey
ah, alright
grand shadow
doesnt bash when ran on /bin/sh do a compat mode of sorts or smth
I could've sworn it did
but maybe not
on my system it seems like /bin/sh is dash
fleet osprey
yes-ish iirc
it does run with some compatibility stuff, but not all
you have to do extra steps within the script itself to ensure that you're running 100% compliant to POSIX with no extra extensions
which is stupid, but iirc is just how it is
grand shadow
downloading qwinci's iso to test
@carmine swallow WTF
It bugs out for a few seconds then just works????
wonder if it could be playing unitialized memory for a bit until it starts running the actual song?
analog berry
dont forget theres also the header for this file format
grand shadow
ohhhh true