#EvalynOS
1 messages Β· Page 25 of 1
Kinda
Though no, I'd have to self ipi to userspace as well
I've been thinking about it the other day
You can get preempted in the kernel
I mean clanker said it to me as well
The handler has to get its own stack tho
#MCE is an NMI
100% cpu maxing discord on pentium-II
So you should use a TSS
I've got to log it and yeet out of the kernel
there should be 2 noncontiguous sticks of memory bad
jesus i have like 15 second animation delay lmao
Is this Evalyn's kernel?
yep
MFW
cuz one stick doesn't have labeling
well i did put 2 broken sticks and one normal
Yes
Let me take a shit and see how large the reserved regions are
yo but the laptop is nice and toasty
warming up my thighs quite nicely :3
You can try my kernel and see what happens 
send me a functioning x86-64v2 iso image or smth
not from the iso
i had to limine bios-install /dev/sda
Is my thing still running?
yep
memset round 420
nice
the memory is btw 4 (broken) 4 (broken) 8 (ok)
memset round 550
i don't think we'll get anywhere interesting
So I donβt think it managed to pick it up

Because it only detected 8gb as usable
Maybe if we get the physical address of the bad regions I can just force it to be used
Yeah maybe the firmware just saw bad memory and decided not to report it
so wanna swap one of the bad sticks for a good stick of equal size?
Yes
I'll ask copilot
choose one:
cool one with a black heatsink
garbanzo green IBM OEM one
Yeah
[MCE] CPU has 25 MCE Banks
[MCE] MCE+MCA Exception MSRs Setup
Nice to see this working though on many machines
you're getting the TridentZ royal treatment of the 2012's IBM
awww
my plastic thingymagic got cracked :c
100% fault of evalynos, trust
Aw
207-Memory initialization error on Processor 1 Socket 12. The operating system
may not have access to all of the memory installed in the system.
207-Memory initialization error on Processor 1 Socket 12. The operating system
may not have access to all of the memory installed in the system.
Advanced Memory Protection Mode: Advanced ECC Support
the error log from ILO btw
and it triplefaulted and restarted
well
idk abt triple fault
but it restarted
with mapping memory being the last message
did it get bad memory into page tables?
Undervolt the RAM sticks while the OS is running somehow and that should get the job done i think 
283-Memory Address/Command Parity Error Detected - Processor 1, Channel 1.
Action: Reseat the DIMMs. If the issue persists, contact HP service.
oh look
it doesn't boot now
π
YOU BROKE IT π’
π
oh shit!
it's brokeeeeeennnnn π’π
this did detect the bad memory as part of the map
wait it legit is boot looping even before the system rom
He used memory so trash πποΈ even the BIOS crashed
π
again
iirc this is proper as per SDM
i feel like some bad pages are getting into the page tables maybe?
i guess?
idk you're the evalynos kernel dev
if we can get down the paddr's of good memory
and make sure the loaders dont use it
it may work
want to swap another stick?
if i hard code it to only use bad mem to be mapped there
the seconf bad one
ehhhhh
detailll
this is the diff
It seems like the bad memory is at the outer end of that region?
Yeah you can disable that
Or I can compile it without PIC
the bad memory should be 4gb offset from that last region
in that config
i'm using channels
1A 2A 3A
so if you run without KASLR
with the broken one being 3A
broken one i replaced*
so kaslr: no or completely remove?
kaslr: no should work
okies
im glad my serial driver is workin tho
i mean i would've fallen back on sending pics/videos
but the serial is very nice
Iβm very proud of that driver
https://git.evalyngoemer.com/evalynOS/evalynOS/src/branch/main/kernel/src/arch/x86_64/drivers/16550uart.c
https://git.evalyngoemer.com/evalynOS/evalynOS/src/branch/main/kernel/src/arch/x86_64/drivers/16550uart.h
i should do something more robust and not just ramming tty stuff into the io port
though i myself am proud of my tty implementation
i implemented everything (other than control stuff and baud) that posix mandates
Check my old repo for one that does interrupts
Though itβs a bit messier
This one only does output so I never implemented it lol
And I need proper device abstractions for my rewrite
the shiz is bootlooping again π₯
i'm glad i have the 198471875th ILO IPv6 address
but i'd like a boot menu
thank you very much
bruh, my kernel is trolling
this is actually S-Tier ragebait
who could've known that running unstable (unstableos reference? π) ram leads to unstable behavior
not fair
dad entered the room and asked about the "jets taking off" sound π
lmao
Lmfaooo
Okay let me make the changes
So paging tables donβt get fucked
π
it should have just been fine and delivered an #MCE but???
slightly out of sync
msg so this stays pinned in my channel list
lemme first test with this working stick
if it page faults or something
i'm so lost atp
WAIT
FWAIT?
im mapping the phys addresses qrong
i put the alr HHDM'd ones
π
which is way past PLEN max
so ofc it will PF
imma help you
the issue is somewhere either on a piece of silicon or copper :3
and actually now that i'm this helpful
i might've spotted a JTAG on the mb
if i find my jtag convertor and i solder pins it might work and help
i highly doubt it
qemu a bit confused
poor bird
RIP=ffffffff80003a13
/home/evalyn/Documents/Programming/evalynOS/evalynOS-new/kernel/src/libc/string.c:13
yeah about what i expected
no fault tho
just locked up
welp this should be valid?

i'm sorry it's dragging this long on
it's 11:04 PM here
5PM for me
i do wonder how longer i can keep doing this before i get raided to stfu
π
I have the hyper image, but it's tripple faulting in early boot
In qemu
wtf
if i never write anything again here, assume the worst
i've been abducted by the HPE gods and forced to repent for my bad ram diddling sins
Oh also Iβve never been able to get a #MCE to trigger
Even using some super strange AMD spesific MSRs
ok so i've figured out a pattern
it boot loops 4 times
then tries PXE
subsequently fails
and then after 2 minutes gives me the option to boot from USB
if i disable PXE it bootloops forever ππ₯
That have a fancy watchdog where if an opcode takes too long it dies
But even on the minimum setting
It never triggered
π
And I legit tried to slow down the hot paths as much as I could
fucking reloading page tables on every iretq
π
Well
and instatly rebooted

Well
Now we know it is diddling the ram
I wonder if the ram is like too fucked?
To where it goes #MCE then #MCE before it can clear the flag?
um
i can try and reseat the one throwing parity errors and put the broken one to that offset
Maybe?
Just keep the one slot with bad memory the same
As we know that gets us in with thoes phys addresses
okies
i'll send the log after wrangling with the pixies
Lmfao
If only Intel and AMD were smart and had an MSR to trigger an MCE in x cycles for debugging
π
Or something else
But noooo we have to suffer with this fuckass ram
i mean
do a debug trap idt into undefined memory
Hm?
that'd be double fault prolly
π
so progress
we're on our 4th bootloop
so it's pxe and usb boot time π₯
ts sucks
this is how i handle them
Hm
the stick always gives off parity errors
so we officially have a reliable way to get parity errors
thats good
or well
as reliable as "random parity errors" can be
niceeeeeee
it's so fucking funny that in the attempts to run broken ram, we've broken a ram stick
Put cause of death
βTesting EvalynOS #MCE Handlingβ
giant wooden cross with that scribbled on it with blood
I'M NOT FUCKING KIDDING RN
ONE MEMORY STICK UNBROKE ITSELF
Or you can sell the ram on eBay
HELOOO??????????

???????
HOW DO YOU UNBREAK A RAM STICK
This operating system has quite the history of shit like this tbh
π
So many strange fucking errors
My code is fucking magic
For better or for worse
watch as magically no MCE is raised but somehow a coprocessor segment overrun
to spite us for trying actual x64 features
what if all reserved exceptions are more MCEs
okies
so for good measure
it doesn't bootloop anymore
I just have a blank black screen
and the ILO here blinking every led in a pattern i cannot decipher

Lmfao
Well
How good is that fixed ram btw
Maybe we can make this into a business
And fix peopleβs ram
um
is fixed?
idk?
hasn't booted since π
With my fucking magic code of wizardry and witchcraft
Not even I know how it did that
π
by the booting (lack thereof) i'd say fixed a ram at a cost of a ddr3 memory channel controller π
which was the broken one π
Flip a coin
i went to my ddr3 shelf and picked a random stick
peak strategy
lose track of the broken ones, so that future foxerity has something to do in his free time
lol
i just saw a text message
evalynOS-potential-magic-ram-fixer
no attachment
iLO 4 Server and iLO 4 Diagnostics page.
fuck off no way
π
okay
so what do we know
we have 2 completely dead (ilo says the dimm slot error) sticks
one was throwing parity errors
maybe dead ddr3 channel C memory controller
dying ilo 4 chip
BUT
a fixed 4 gb ram stick :3
i'd call this an absolute success
there has to be enough broken components to raise an #MCE
PLEASEE
and the ram stopped throwing parity errorrrrssss π π

it's the magic iso π
currently 3 working + the one that was throwing parity errors
π
they're getting shorted
cpu pins AA-ZZ
then we'll see
what if i just straightup stab the memory chip with like a knife or smth
Maybe?
Idk if my code could fix that tho
evalynOS-potential-magic-ram-fixer
fixes everything
except integrated management controllers apparently
this is fun
i'm surprisingly having fun
Oh this is quite entertaining
ok
283-Memory Address/Command Parity Error Detected - Processor 1, Channel 2.
Action: Reseat the DIMMs. If the issue persists, contact HP service.
Redundant ROM Detected - This system contains a valid backup System ROM.
Inlet Ambient Temperature: 22C/71F
207-Memory initialization error on Processor 1 Socket 1. The operating system
may not have access to all of the memory installed in the system.
207-Memory initialization error on Processor 1 Socket 9. The operating system
may not have access to all of the memory installed in the system.
207-Memory initialization error on Processor 1 Socket 9. The operating system
may not have access to all of the memory installed in the system.
Advanced Memory Protection Mode: Advanced ECC Support
WARNING: The iLO Security switch is set to the ON position.
Platform security is DISABLED.
got this
this is the current state
i'll
send you the evalynos map
once/if it boots up
why is IXOFF not workingggggggg
this is working fine so far
evalynOS-potential-magic-ram-fixer
no fucking shot lmao
gg
ig evalyn just knows how to diddle that ram
π
how the fuck did you do it
it cost me my management controller :c
But we have found fucking gold in this economy
you owe me
uhhh
idk
somethin
a funny gif
yea that
memtest running a second time bc why not
not madoka magica π
Itβs just gonna get fixed
try
i'll run the parity broken stick
wait
what's channel 2????
slot 2 is empty
and i have channel ABCD
Across computers
dw if the os makes the PC reboot
I wonder if @devout path knows anyone?
dw i can break my collection of 120 ram sticks for the cause
I have hyper image
memtest #2 50% passed
π
We need to test more machines too tbh
it's still going btw
i have hyper depression from neteXtreme PXE rom π₯
But I don't handle MCEs
We have discovered a magic way to revive ram sticks
π
how
idk
My ISO
it unfucked my ram
i have a zen2 system
zen4
xeon e5 gen 2, gen 3, gen 4
xeon e3 somewhere
xeon x5650
xeon scalable 3rd gen
some random celeron of who knows origin
intel celeron (M, slot 1 socket)
XD
Itβs still not proven this works
@hazy saddle add a quote to your os: Also called the "RAM Unfucker 2026". XD
Well itβs only this one ISO
This ISO has patches that diddle a spesific set of phys memory
what if it will be un the next versions too
and we're back into brokenville, population: 12 GBπ£οΈ
mine reached this point too
then it rebooted after 2 messages XD
well
"memset round 0" is strictly speaking "a message"
after that i saw 2 messages that started with [MSG] ...
and then it died
close enough
booted via ventoy btw
got the message accross π π
HP Power Profile Mode: Custom
Power Regulator Mode: Dynamic Power Savings
283-Memory Address/Command Parity Error Detected - Processor 1, Channel 2.
Action: Reseat the DIMMs. If the issue persists, contact HP service.
Uncorrectable Processor Machine Check Exception(s) were detected
and are logged in the Integrated Management Log (IML.)
Redundant ROM Detected - This system contains a valid backup System ROM.
Inlet Ambient Temperature: 22C/71F
207-Memory initialization error on Processor 1 Socket 1. The operating system
may not have access to all of the memory installed in the system.
207-Memory initialization error on Processor 1 Socket 9. The operating system
may not have access to all of the memory installed in the system.
207-Memory initialization error on Processor 1 Socket 9. The operating system
may not have access to all of the memory installed in the system.
Advanced Memory Protection Mode: Advanced ECC Support
WARNING: The iLO Security switch is set to the ON position.
Platform security is DISABLED.
OH
MCE
jesus christ coming through the screen
Can you test this? π₯
the first one who figures out why it doesn't work gets a ":3" from me
@hazy saddle 20 mins and no crash + memtest 2 passed
lol
tmr i'm gonna leave it running for some hours
speaking of servers
have you ever been bitten by a server?
i have
a supermicro server bit me
had to go to the er to get stitches
a fan bit me
still have a scar
kinda fun
also in the official medical records is "a supermicro fan" listed as the cause
which is immensly funny
lee
mme actually look at the IML
how do you specify an IPv6 linklocal address interface in firefox?
cuz [fe80::9eb6:54ff:fe09:75c8%enp2s14] is not it
It should just be the IP itβs self no?
Real
fucking network manager
who sat down
and decided 30 seconds of no internet = hahaha interface goes down brrrrr
memtes 3 passed
so assuming the laptop didn't explode
this is in the IML
i keep bootlooping evelynos
in case something magical β comes up
@hazy saddle can I help somehow?
like anything that i can actively be doing to help?
i wanna help :3
You could look at my MCE code and see if somthing is fucked with it maybe?
These exceptions should be getting caught by me
i actually don't know whether those are from the firmware or from you
but i assume that i'd be more that probable that you'd catch at least one of those
Yeah

The Intel manual has a good bit of info on these things
Also I wonder what happens if you donβt touch the bad ram
And if it just spuriously does the thing
Yeah
hanging on that "press a key to reboot" no bootable device prompt and watch the iml
Il make an ISO that just does some basic loop things
And triggers a software interrupt
kkk
That I use to test the IDT
And maybe @tender mantle can find that spalloc test ISO that may do some things
yea maybe make it do something at least somewhat visual so we can see whether it isn't just hanging hanging
something like "\r%c", 'a' + counter++%26
or idk
maybe send me the elf file so i don't have to dissable the iso again
can't type today
disassemble*
spalloc being my single page / split page allocator (slab but specialized for 32b+ objects in a single 4kb chunk with bitmaps to be fancy)
im pretty proud of this
fuck me it autoreboots
I have to go into the bios so it doesn't reboot itself
considering it's stable after 3-4 days and you wrote it yourself you should be :3
spalloc hasn't died yet π€
i'm in the rbsu
no clue what that is
but yea
this is a diff thing
testing bad memory
for #MCE's
we have genuine bad ECC that gives off errors
we're intentionally breaking and magically fixing my server
:O
https://www.youtube.com/watch?v=yModCU1OVHY
this song should give extra luck btw @jade olive
of course evalyn of all people wants to play with bad ecc memory
π
okay fine
im a protogen
i eat ram
I ALSO HAPPEN TO WANA TEST MACHINE CHECKS
:3
:3
:3
okies
it doesn't raise MCE on its own
it's definitely something your kernel is doing
or at least the RBSU shell isn't raising them
so now that elf
well in any case, we know the MCE has to be caused somewhere in your kernel
i expect the management controller to at least know whether MCE happens
you may wana pull down my repo so you can start making changes
i don't know what i'd be doing tho
i do have copies of the intel sdm so might take a look at that
.!t sdm
grab a fresh one
and look at the #MCE code maybe
my kernel should be easy to build w/ clang22
π₯
i plugged in the USB and the server immediately rebooted itself lol
no MCEs yet
almost booted
POST Information: Processor 1, DIMM 12 could not be authenticated as genuine HP SmartMemory. Enhanced and extended HP SmartMemory features will not be active. blah blah blah
bootlooping again....
it's too broken ig
need to remove something
huh? what are lenovo engineers doing
bruh
i may have accidentally fixed the parity
we'll see
yk how hard it is to gaslight the ILO into thinking the ram is ok?
why is it so good at its job
IML still empty
i can go diddling with the flaky parity ram stick
so far nothign
btw how are my logs?

i don't think it's just gonna appear
i can go slowly unhook the ram until i get the MCE
i don't have that strong magnets
WHAT
the ram is completely unhooked
the only detected ram is unhooked completely on one side
and it's just goin on
fuck
so in case you were wondering
if all memory is badd
you'll get
No Usable Memory Found - System Halted!
and an insanely annoying beeping sound
anyone wanna mod the firmware to accept broken ram?
i do have a chip flasher
idk anybody
surely
okies
got it to produce more uncorrectable errors
this is legit just relaunching a heavily modded game and hoping it actually turns on
the interrupt loop is running again
i don't want to touch that ram
it's exactly the right temperature, unhooked enough, and the right vibe to just barely throw parity errors
i hope it dosnt hit it within a spinlock
i could disable the FB and the lock if happens tho
the error light for DIMM12 (the flaky one) just lit up
at random
like it's still running
grep for Got a non fatal #MCE
i have the terminal opened
i'm staring at it
like intesively
have to assert dominance
lol
nothing in the IML
the iml is in UTC time for some reason
true
well i think the last MCE from the IML is still the current memory setup
you could poke an address in that range
and like have some sort of a progress bar
here you have some
of the memory map
i don't have that far of a scrollback
i'll reboot then
it should be the entire range 0x200000000 -0x3FFFFFFFF
towards the kernel, the ram, the server, me, or our patience?
the ram and kernel
it should like uhh
make less crazy errors
than memsetting shit
page fault
:3
it's also blinking
i guess the ram moved?
possible i guess
does it display right in the serial term btw?
try this
i moved it to match the mmap
blinking in serial
i'll wait then
the current DIMM is in slot 12, which is A (ram going A-B-C-D-E-F-G-H-I-J-K-L)
then first broken is in B
then in C
so assuming the manual isn't lying about how this works, they should be contiguous
kkkkk
well scrollback took the map
but i assume that's ok
so far at offset 2000 and it's ok
at the speed it's going i don't know how long it's gonna take assuming this entry somewhat corresponds to real addresses
| 00023 | 0xffff800100000000 | 0xffff80013ffff000 | 0x000000003ffff000 | Usable |
are you sure about this?
we're at 0xffff800100008c00
that region is HHDM mapped
to the phys region of things
so it should? be correct
and it will go up to 120000000
we're going like 400 bytes/s
that's gonna take ages
i don't mind and will gladly give you the """compute"""
just saying
it is only one core rn 
i could also chunk messages
or disable FB
which should speed it up
maybe write out a message per like 1 page? like 4096 bytes?
we're at 10002C800 btw
give or take a few bytes
like this or so?
like
the server hasn't yet spontaniously combusted
and are you sure those are the valid ranges?
0xffff800100000000 to 0xffff80013ffff000 yes
if its divisible by 0xffff80013ffff000 it will reset to 0
etc
that's the good memory though
or well the somewhat flaky that doesn't flaky anymore
gg
i need to map it
completely
like completely offtopic
but why does the router vodafone gave us
NOT SUPPORT STATIC ROUTES???????????
THERE'S NOWHERE TO INPUT THEM
f this
mikrotik all the way
i'm gonna write the most convoluted dhcp server and relay config you have ever seen
just so i can ssh into the pentium laptop and watch the serial log on my main pc
okay
assuming my internet works?
attempt #3
thank you
finally
reboot after the last message
yooo have to rewatch fantastic mr fox
yep
MCE
121 CPU 05/15/2026 00:10 05/15/2026 00:10 1 Uncorrectable Machine Check Exception (Board 0, Processor 1, APIC ID 0x00000000, Bank 0x00000018, Status 0xFE200000'0003110A, Address 0x00000002'00000300, Misc 0xD0000838'80801086)
120 CPU 05/15/2026 00:10 05/15/2026 00:10 1 Uncorrectable Machine Check Exception (Board 0, Processor 1, APIC ID 0x00000000, Bank 0x00000017, Status 0xFE200000'0003110A, Address 0x00000002'00000140, Misc 0x94000838'80801086)
119 CPU 05/15/2026 00:10 05/15/2026 00:10 1 Uncorrectable Machine Check Exception (Board 0, Processor 1, APIC ID 0x00000000, Bank 0x00000016, Status 0xFE200000'0003110A, Address 0x00000002'00000180, Misc 0x58000838'80801086)
118 CPU 05/15/2026 00:10 05/15/2026 00:10 1 Uncorrectable Machine Check Exception (Board 0, Processor 1, APIC ID 0x00000000, Bank 0x00000014, Status 0xFE200000'0003110A, Address 0x00000002'00000000, Misc 0xD4000838'80801086)
117 CPU 05/15/2026 00:10 05/15/2026 00:10 1 Uncorrectable Machine Check Exception (Board 0, Processor 1, APIC ID 0x00000000, Bank 0x00000012, Status 0xFE200000'0003110A, Address 0x00000002'000006C0, Misc 0x5C000838'80801086)
116 CPU 05/15/2026 00:10 05/15/2026 00:10 1 Uncorrectable Machine Check Exception (Board 0, Processor 1, APIC ID 0x00000000, Bank 0x00000011, Status 0xFE200000'0003110A, Address 0x00000002'000000C0, Misc 0x20000838'80801086)
yep
the addresses are a bit off
they look right
its giving phys addr
i log virt
well i get that with the 0xFFFFF stuf
but 0xC0? when we passed through 0x0000 to 0x1000?
well precisely
then why MCE at 0xC0
then i'm slightly lost
the ILO clock is behind about 4-7 minutes (not sure how much that's the insane networking setup and caching firefox does)
can we disable the IML?
IML is just logging
maybe its stealing my MCE?
its still going through SMM
or some crap?
lets try this
idk how to check that
but regardless it's impossible to truly turn off ILO
and even if possible
i do remember linux throwing some ECC errors on this machine a LONG time ago
so i assume it's not completely stolen
rahhhhh π₯π₯
so the clock is 6 minutes behind
give or take
so the MCE reports come in after your kernel crashes
so i assume the IML just gathers the reboot reason after it happens
maybe
if i reply the MCE in QEMU it goes off aswell too @jade olive
i wonder
what if the MCEs are so fast
that my recurssive lock preventor kills it?
i need to make it iretq and not hcf()?
thr rahhh will test that
nothing happened
again hung at read from memory addr ffff800200001000 and got 255
no printf
why lock when you can unlock
or smth like that
i meanwhile was so recordbreakingly productive that i fixed the time
fyi
the rom based setup utility sees 12288 MB of ram
(it's literally called the "rom based setup utility" (rbsu))
lol
also if you want
this server has an NMI button
so i can manually raise an NMI
If we canβt get MCEs after this test
My code may be cucked π
We can test NMIs then
Hey i dreamed of all of this shit happening 5 months ago
fair enough π
deja vu?
wait actually no
deva vu is something different
Do we ever get my kernel to get an #MCE
Not really
Idk i remember up until the part diamantino says it unbroke their laptop's ram or something like that
i also sometimes have dreams like that
I wish I had dreams lowkey
aww
[MEMORY] Paging Init [OK]
read from memory addr ffff800200000000 and got 255
read from memory addr ffff800200001000 and got 255
read from memory addr ffff800200002000 and got 255
read from memory addr ffff800200003000 and got 255
read from memory addr ffff800200004000 and got 255
read from memory addr ffff800200005000 and got 255
read from memory addr ffff800200006000 and got 255
read from memory addr ffff800200007000 and got 255
read from memory addr ffff800200008000 and got 255
read from memory addr ffff800200009000 and got 255
read from memory addr ffff80020000a000 and got 255
read from memory addr ffff80020000b000 and got 255
read from memory addr ffff80020000c000 and got 255
read from memory addr ffff80020000d000 and got 255
read from memory addr ffff80020000e000 and got 255
read from memory addr ffff80020000f000 and got 255
read from memory addr ffff800200010000 and got 255
read from memory addr ffff800200011000 and got 255
read from memory addr ffff800200012000 and got 255
read from memory addr ffff800200013000 and got 255
read from memory addr ffff800200014000 and got 255
read from memory addr ffff800200015000 and got 255
read from memory addr ffff800200016000 and got 255
read from memory addr ffff800200017000 and got 255
read from memory addr ffff800200018000 and got 255
read from memory addr fff
not badly copied
died mid print
Huh wtf
peak 2:40 AM activities
I diddnt remove the panic lock
But
Iβd expect some print
wtf
Iβm too sleepy for ts
π
Go brrr with the NMIS
gimme an infinite loop and i'll trigger an nmi
You got the old elf right?
can't remember if it was kernel 5.elf or kernel 8.elf
Should have one that does it
Maybe this machine just uses NMIs or something for memory errors
.
thx <3
Iβm not even getting an NMI
I know that on some (arguably rare) machines the SMM handles these things like MCE and NMI and you gotta diddle the chipset for them to pass through
Which is like π
yeah i think thats whats happening here
there is motherboard logs with the #MCEs
nothing happened π
and i dont got code to diddle the chipset
π
135 OS 05/15/2026 02:43 05/15/2026 02:43 1 User Remotely Initiated NMI Switch
134 CPU 05/15/2026 02:42 05/15/2026 02:42 1 Uncorrectable Machine Check Exception (Board 0, Processor 1, APIC ID 0x00000000, Bank 0x00000015, Status 0xFE200000'0003110A, Address 0x00000002'000003C0, Misc 0x1C000838'81801086)
133 CPU 05/15/2026 02:42 05/15/2026 02:42 1 Uncorrectable Machine Check Exception (Board 0, Processor 1, APIC ID 0x00000000, Bank 0x00000013, Status 0xFE200000'0003110A, Address 0x00000002'00000400, Misc 0x980008F8'81801086)
no clue how time works for that MCE
i really do think it's just post-incident collection
If you get whatever part the chipset is i might be able to find docs on it
I donβt think it can do that with MCA
I have a few sources for these things
Which cannot be indexed by google
before i start looking that up, it's the HPE ProLiant dl380p gen8
Since the banks get reset on CPU reset
maybe try the other server?
and download the repo
i gotta do irl crap for a minute
i can help compiling it
and then we'll get the 7 minute rebooting
it does do them
you can also find MCEs in the ILO

i remember when i bought it it had like 5000 MCEs from way back into 2009
Well
I guess Iβm gonna have to find out how to diddle the chipset
Hopefully you or @clear bison can find this out

And hopefully I donβt need to fucking fully implement PCIe
i bet the ILO interface must be very nice to work with
C600?
Tbf either way
We got alot of valuable information
Somehow my code can revive ram sticks
And we had fun while doing this
what about the IA32_MCG_CAP register bit 24?
do you read that?
cuz if i'm reading the SDM correctly, that indicates whether the MCE is even software handlable
more precisely, whether an uncorrectable MCE is software handlable
processor supports software error recovery (see Section 17.6), and IA32_MCi_STATUS MSR bits 56:55 are
used to report the signaling of uncorrected recoverable errors and whether software must take recovery
actions for uncorrected errors. Note that when MCG_TES_P is not set, bits 56:53 of the IA32_MCi_STATUS MSR
are model-specific. If MCG_TES_P is set but MCG_SER_P is not set, bits 56:55 are reserved.```
Would it ever report them as recoverable in the error code etc if thatβs the case?
And it still should log it?
If it got it at all
Maybe I can check for that??
Though the motherboard says itβs unrecoverable errros
sure
considering the parity error is "uncorrectable", then i assume that this being 0 means you won't get it
It wouldn't show you the uncorrectable errors
the chipset supports AMT
that surely has some mechanism to read this
Instead it would let the firmware handle it
So thatβs whatβs happening
Maybe
we don't know
depends on the value of that msr
Check those MSRs so you can be sure
So I need to check this and then also set a certain bit to handle them?
Or is it just straight up not at all in some cases
i guess straight up not at all?
and once again, i'd like to thank okular for it's immaculate resource hogging
not helping for a netbooter here
i need the cache π₯
I rarely even get above 16gb ram usage
i like to torture myself it seems
i'm running NFSv4 over 10 gig link as my netboot target
Torturing yourself is just apart of osdev
at least okular doesn't just stop paging the arm manual (yeah it's like 16k pages but still)
lord praise ZFS and its level 1 arc π
Thatβs fine for a server tbh
:3
Though with the three new Linux CVEs :x
what the fuck
:3
These are the docs for that but i don't see any mentions of MCEs
https://www.intel.com/content/dam/www/public/us/en/documents/datasheets/c600-series-chipset-datasheet.pdf
Yo reboot yo shi π
nahhhhh
@hazy saddle what about the IA32_MCG_EXT_CTL MSR? bit 0 Enable / Disable local machine check exception.
0x4D0
there should be one raspberry pi at grandparents' that has to have like 2000 days uptime atp
Does my code set that?
I wonder
and i know this one guy working for the national fibre grid here in czechia
he was lookin at some mikrotik router dashboard
3943 days uptime
i don't think so?
the wording of that entire MSR is about sending the MCE to a single LAPIC node?
not sure though
the pin is clear enough i'd say
yea
i see MSR_MCE_CTL which is IA32_MCG_CTL
so
missing the EXT_CTL completely at least
gg
well
the main thing is
that we got to the answer eventually
better get to it later than almost kill a management controller
oh wait π₯
nah dw it's fine
good thing
is that evalynos is still running lmao
chunk 1772
I donβt think itβs my OS that killed it
It probably was some random spurious shit
a random fly flew by flying above the fans being sucked in and it's fine mist flown all over the management controller

100% the fly's fault
