corinne has quit [Remote host closed the connection]
corinne has joined #osdev
Matt|home has quit [Quit: Matt|home]
Pixi has joined #osdev
Pixi` has quit [Ping timeout: 248 seconds]
goliath has quit [Quit: SIGSEGV]
edr has quit [Quit: Leaving]
netbsduser`` has quit [Ping timeout: 276 seconds]
corinne has quit [Remote host closed the connection]
corinne has joined #osdev
jcea has quit [Ping timeout: 252 seconds]
corinne has quit [Remote host closed the connection]
corinne has joined #osdev
corinne has quit [Remote host closed the connection]
corinne has joined #osdev
n00by_ has joined #osdev
karenw has quit [Ping timeout: 276 seconds]
n00by_ has quit [Ping timeout: 260 seconds]
n00by_ has joined #osdev
n00by has joined #osdev
n00by has quit [Client Quit]
n00by_ has quit [Client Quit]
n00by_ has joined #osdev
n00by has joined #osdev
n00by has quit [Client Quit]
arminweigl_ has joined #osdev
eluks has quit [Remote host closed the connection]
arminweigl has quit [Ping timeout: 252 seconds]
arminweigl_ is now known as arminweigl
eluks has joined #osdev
arminweigl has quit [Ping timeout: 252 seconds]
arminweigl has joined #osdev
n00by has joined #osdev
n00by_ has quit [Quit: Quit]
corinne has quit [Remote host closed the connection]
corinne has joined #osdev
arminweigl_ has joined #osdev
arminweigl has quit [Ping timeout: 272 seconds]
arminweigl_ is now known as arminweigl
arminweigl has quit [Ping timeout: 268 seconds]
arminweigl has joined #osdev
n00by has quit [Quit: leaving]
n00by has joined #osdev
dormito has quit [Ping timeout: 248 seconds]
dormito has joined #osdev
Fingel has quit [Ping timeout: 248 seconds]
Fingel has joined #osdev
goliath has joined #osdev
Lucretia has joined #osdev
xvmt has quit [Ping timeout: 244 seconds]
chiselfuse has quit [Remote host closed the connection]
chiselfuse has joined #osdev
ZipCPU has quit [Ping timeout: 252 seconds]
the_oz has quit [Remote host closed the connection]
the_oz has joined #osdev
ZipCPU has joined #osdev
GeDaMo has joined #osdev
netbsduser`` has joined #osdev
pebble has joined #osdev
corinne has quit [Remote host closed the connection]
corinne has joined #osdev
netbsduser`` has quit [Ping timeout: 272 seconds]
corinne has quit [Remote host closed the connection]
corinne has joined #osdev
netbsduser`` has joined #osdev
steelswords94361 has quit [Read error: Connection reset by peer]
steelswords94361 has joined #osdev
edr has joined #osdev
jcea has joined #osdev
corinne has quit [Remote host closed the connection]
corinne has joined #osdev
pebble has quit []
netbsduser`` has quit [Ping timeout: 260 seconds]
antranigv has quit [Ping timeout: 248 seconds]
antranigv has joined #osdev
the_oz_ has joined #osdev
the_oz has quit [Quit: Leaving]
Left_Turn has joined #osdev
trillion_exabyte has quit [Ping timeout: 268 seconds]
trillion_exabyte has joined #osdev
corinne has quit [Remote host closed the connection]
corinne has joined #osdev
Turn_Left has joined #osdev
Left_Turn has quit [Ping timeout: 245 seconds]
xvmt has joined #osdev
Left_Turn has joined #osdev
Turn_Left has quit [Ping timeout: 245 seconds]
k0valski18891621 has quit [Ping timeout: 272 seconds]
corinne has quit [Remote host closed the connection]
corinne has joined #osdev
basil has quit [Quit: Ping timeout (120 seconds)]
marv7000 has quit [Quit: Ping timeout (120 seconds)]
basil has joined #osdev
alexander has quit [Quit: ZNC 1.8.2+deb3.1+deb12u1 - https://znc.in]
marv7000 has joined #osdev
alexander has joined #osdev
DragonMaus has quit [Ping timeout: 252 seconds]
JTL has quit [Ping timeout: 276 seconds]
DragonMaus has joined #osdev
JTL has joined #osdev
Turn_Left has joined #osdev
Left_Turn has quit [Ping timeout: 272 seconds]
the_oz_ has quit [Ping timeout: 260 seconds]
the_oz_ has joined #osdev
CarbonWriter has quit [Quit: ZNC 1.8.2+deb2+deb11u1 - https://znc.in]
corinne has quit [Remote host closed the connection]
corinne has joined #osdev
sprock has quit [Ping timeout: 276 seconds]
sprock has joined #osdev
<zid>
heat did you check your underwear draw for an eeeeeeepc
<heat>
i don't have one
<zid>
drawer*
<zid>
fuck
<zid>
I CANNOT ENGLICH
<the_oz_>
VERDAMMT
netbsduser`` has joined #osdev
netbsduser`` has quit [Ping timeout: 260 seconds]
GeDaMo has quit [Quit: 0wt 0f v0w3ls.]
dude12312414 has quit [Quit: THE RAM IS TOO DAMN HIGH]
netbsduser`` has joined #osdev
the_oz_ has quit [Remote host closed the connection]
the_oz_ has joined #osdev
Left_Turn has joined #osdev
Turn_Left has quit [Ping timeout: 245 seconds]
air has quit [Ping timeout: 248 seconds]
netbsduser`` has quit [Ping timeout: 260 seconds]
<clever>
zid: i have a pair of 701's
dude12312414 has joined #osdev
dude12312414 has quit [Remote host closed the connection]
karenw has joined #osdev
<zid>
Perfect, you can send one to heat
<heat>
i've been thinking about buying a pi 5
<zid>
that's not en eeeeeeeeeeepc though
<heat>
apparently USB is much faster and it now has a PCIe slot
<heat>
and that's cool
<heat>
i don't know if the ultimate pi5 storage solution is like a SATA controller card sitting on the PCIe bus
<zid>
nvme drive
<zid>
surely
<clever>
heat: pi5 has 4 lanes of pcie gen2 going off to a southbridge, which drives gbit ethernet, 2xusb2, 2xusb3, the csi/dsi ports, and all of the gpio
<clever>
plus a spare pcie lane, gen2 certified (but gen3 capable) for user addons
<zid>
remove the superio, replace with a 4x nvme
<clever>
zid: the firmware refuses to boot if its missing
<zid>
replace the firmware while you're at it then
<heat>
firmware? more like flimsyware
<clever>
the firmware is signed, and i have yet to crack it
<clever>
its more locked down then past models
<zid>
include a superio emulator in your nvme firmware
<zid>
so that it boots correctly
<heat>
ok say in theory i wanted to add an NVMe drive to my pi5
<heat>
because i'm an idiot
<heat>
would that be possible
<clever>
part of it, is that the firmware talks to the southbridge over i2c
<kof673>
Presently, they [software and hardware] met Firmware, who was dressed in tattered rags and hobbled along propped on a thorny stick --- tao of programming. firmware just hobbles along
<clever>
that pdf goes over a dozen sata ssd's, subjects them to sustained writes, then yanks the power rail (mosfets on the sata power port)
<clever>
and then looks at what mess it left behind
<heat>
zid: 80 final versions is kind of insane
<clever>
3 drives suffered bit corruption
mahk has quit [Ping timeout: 252 seconds]
<clever>
3 drives had half a sector worth of data written
<zid>
Someoen tripped the breaker in my house
<zid>
it RUINED my ntfs on ssd :(
<clever>
8 sata ssd's, and 1 mechanical drive, had unserializable writes
<zid>
good intel enterprize ssds had big caps
<clever>
1 drive suffered metadata corruption, i think it was fine if you read the first 20gig, and if you read past that, *dead*
<zid>
it was a selling point
<clever>
complete firmware crash
<clever>
and one ssd, suffered total failure, just by cutting power
<clever>
zid: oh, ive heard that some high quality nvme, wont write the wear leveling translation tables back to flash, for speed
<clever>
and on power-loss, there will be a mad super-cap fueled race, to commit them, lol
<heat>
this is why enterprise buys enterprise nvmes
<heat>
versus consumer garbage
<clever>
Model Number: INTEL SSDPE2KX020T8
netbsduser`` has quit [Ping timeout: 245 seconds]
<clever>
i'm guessing these are enterprise grade?
<heat>
i hope so
<heat>
i would be very upset if my enterprise SSDs weren't fault resillient
<heat>
because all of these issues are very observable at scale
<zid>
tbh
<zid>
if a data center hard-loses power
<zid>
that's a fucking disaster regardless
<zid>
reimaging some ssds is probably the least of their worries
<zid>
the big fire in the turbine hall slightly more pressing :p
<heat>
yeah but database
<zid>
They have 40 different sets of corruption they can average them, np
<zid>
replicate from US EAST
<heat>
you're assuming this is google
<zid>
people exist who aren't google?
<heat>
if you work at google, the answer is no
<zid>
exactly
<zid>
and google buys all the hw
<zid>
qed ergo cogito sum ssds dying is a fire damage issue
<heat>
though i would love to know the frequency of observed disk/IO failures over google's fleet of servers
<heat>
i know they have a daemon to collect EIO's
<zid>
my friend's job at google
<zid>
is to count ecc errors
<gog>
E I E I O
<heat>
kick RAM right in the teeth
<heat>
until it fails
<clever>
i saw a talk before, about bit squating
<zid>
He told me a fun story about an argument with a customer about a flipped bit
<clever>
register dns domains similar to the target, but with single bit errors
<zid>
He determined it was in the customer's NIC
<clever>
then get ssl certs for those domains
<zid>
given it was checksummed etc everywhere else
<zid>
it must have been
<clever>
the guy wound up getting http requests from legit google servers
<heat>
clever: lol jesus
<zid>
and then my friend had to go around making sure the NICs were ecc'd :P
<clever>
and the little bit of JS he slipped into the response (as a test), got cached by google, and served to 1000's of clients
<heat>
zid: the checksum didn't make it drop the packet?
<zid>
also I believe it's now google policy that ERR_OK and ERR_BAD or whatever, are not 0 and 1, but 0x5555555 and 0xAAAAAA or such
<clever>
SSL doesnt do jack, if the bit flips before openssl gets the domain
<heat>
i guess you can technically have, like, a bitflip after csum validation
<zid>
after someone randomly found themselves to be admin on gmail
<clever>
heat: my nas is using zfs, twice now ive recorded bit-flips, that the zfs checksums dont catch
<clever>
heat: one bit flip turned a 3 in the filename into a #
<heat>
huh
<clever>
all of the tooling reports that the file exists, yet you cant read or delete it, because the filename is invalid
<heat>
what is it using for csums?
<heat>
ext4 does crc32c
<clever>
basically, 123.txt turned into 12#.txt
<clever>
let me look
<nikolar>
there are several options
<nikolar>
fletcher4 by default i think
<heat>
inet csums are pretty naive, just a big-ass sum of the whole packet, but even then a single bit flip having it still valid is a pretty big coincidence
<heat>
even worse when you factor in encryption/decryption of TLS packets
<zid>
heat: it was detected, that was the problem, just not by the NIC
<zid>
or rather, it was detected, maybe even by the NIC
<nikolar>
clever: so you had bit flips that weren't caught by checksums
<nikolar>
how did that happen
<zid>
but you don't know *where* the bit flip happened
<clever>
nikolar: it flipped before the checksum was computed
<nikolar>
ah makes sense
<nikolar>
yeah literally nothing you can do about that
<nikolar>
heat: what the heck are you supposed to do when a bit flips before it even gets to you
<heat>
i would flip z into a different character for a joke
<clever>
and this, is a bit flip in a .mkv file i downloaded
<heat>
but i can't be arsed to look up binary versions of characters
<heat>
so
<clever>
it was the same bit as the filename, and in the same direction (clearing a bit)
<heat>
<bitflipped z into some other letter>fs
<zid>
Zfs
<zid>
:P
<zid>
flipped the 32 valued bit for you
<nikolar>
kek that works
<heat>
lol
<heat>
i mean yeah bitflips are fucked
<heat>
but also why aren't filenames properly csummed?
<zid>
oh right, we're post 23:00, let's check my EBOOKS
<heat>
this is kind of recoverable
<nikolar>
bookies
<zid>
nothing, rip
<clever>
heat: the filenames are in a data block that is checksummed
<heat>
and it didn't find the bitflip? very weird
<clever>
it flipped before the checksum was computed
<nikolar>
because it was flipped before it was checksummed i imagine
<zid>
depdends where the flip is
<nikolar>
there's literally nothing you can do in that case
<zid>
you write the data out, you write the metadata out
<zid>
both are checksummed
<nikolar>
like, have raid ram or something lol
<zid>
but nothing exists to *correlate* those checksums
<heat>
nikolar: you would like CXL
<nikolar>
what dat
<heat>
it's a weird new interconnect pushed by intel and other big corps
<zid>
cocks x lampposts
<nikolar>
oh i vaguely remember it
<heat>
where they want to do stuff like memory over CXL
<clever>
zid: in a directory, you have filename->inode pairs in a big list, that list is broken up into 16kb chunks, and then it uses a prefix of the hash to pick which 16kb chunk to put a filename into
<heat>
where that memory is suuuuuuuuuper latent
<heat>
because, of course
<nikolar>
more than it is now?
<clever>
zid: each 16kb chunk is then hashed with edonr
<heat>
for unclear benefit
<nikolar>
why would they want
<heat>
yeah, more than it is now
<nikolar>
like it's already stupid slow
<nikolar>
no one would notice if it's slower :P
<nikolar>
that's the only reasoning i can imagien
<clever>
and then because its raidz1, it adds 8kb of parity data, and writes the 16kb+8kb to disk
<heat>
accessing CXL memory is the equivalent of accessing memory that's like 10 sockets over
<nikolar>
which is good, for some reason
<zid>
clever: grats?
<zid>
no idea why you said any of that
<clever>
the disk location, and hash, of each block, is then put into an L1 block, which itself is then edonr'd again
<heat>
yeah i think it allows them to do $unclear fancy stuff
<heat>
including adding a bunch of shit over the CXL bus
<heat>
for $workloads
<nikolar>
like 10x the latency is definitely going to be noticable
<heat>
of course since it's 2025 it's AI
<nikolar>
whenever you dip into ram
<nikolar>
which is often
<clever>
also, since this is "critical" metadata, zfs stores 2 copies of every block
<nikolar>
i wonder, how often do you get cache misses if you have like half a gig of l3
<zid>
depends what you're doing
<zid>
I'll give you a hint, if your data set is a lot bigger than half a gig, many many misses
<nikolar>
sure, of course
<zid>
if it is a lot smaller, no misses
<nikolar>
wasn't there some amd server cpu with thereabouts of l3
<zid>
768MB
<zid>
is the most I remember
<zid>
x3d baby
<nikolar>
oh yeah sometihng like that
<heat>
as far as I understand of linux-mm shitfests about this sort of stuff, their idea currently is to use all of this CXL memory as some sort of slow-tier
<nikolar>
got to love it
<heat>
where you could in theory slowly migrate fast page cache to slow page cache
<nikolar>
so you have normal ram and slow ram?
<heat>
then promote it back up eventually
<heat>
yeah some sort of that
<nikolar>
right that's not nearly as bad
<nikolar>
it's kind of a software managed cache i guess
<kof673>
primary storage, secondary storage, and 1.5 storage
<heat>
but a very significant group of people think this is very very very stupid and insane
<nikolar>
i didn't say it was good
<nikolar>
it's just not as bad as i initially thought
<heat>
(disclaimer: i am included in that group of people)
<heat>
then there are BASED AND CXLPILLED
<zid>
It's going to be
<zid>
for 'some workload'
<zid>
the best thing ever
<nikolar>
i guess the idea is that you use the same bus for all "peripherals" like the memory and the interupt contoller and timers and stuff
<zid>
but it won'tbe us
<heat>
yeah AI bro
<heat>
it's always AI
<nikolar>
which would also inclue the stupid npus they've started putting in now
<nikolar>
i hate that they are even calling them npus
<zid>
"confused numpties who have little idea of what they're talking about"
<zid>
see, fits fine
Gooberpatrol66 has quit [Remote host closed the connection]
<heat>
anyway i'm going to bed because of
<heat>
*checks notes*
<zid>
footballers legs
<heat>
AI
<zid>
won't ogle themselves
<heat>
footballer legs don't increase shareholder value
<heat>
well, i guess they do for like, sky sports
<heat>
but fuck sky sports
<zid>
and onlyfans
Gooberpatrol66 has joined #osdev
vdamewood has quit [Quit: Life beckons]
Turn_Left has quit [Read error: Connection reset by peer]
<zid>
gog did you beat ffx
<gog>
no
<gog>
i haven't been playing
* zid
updates christmas list
<zid>
chat how do you spell 'norty'
* kof673
points at bill goldberg, santa's slay movie, hourglass nebula by sagittarius old man saturn rules around there "christmas isn't over until i say it's over!"