#
#asahi · 2025-10-07 03:55
tpw_ruleshttps://asahilinux.org/docs/sw/broken-software/ some further context
#asahi · 2025-10-07 03:55
tpw_rulesbut it's unclear what part of the steps are actually needing jemalloc (whether it's the runtime or something in that particular dev environment), and i can't really help debug, sorry
#asahi · 2025-10-07 03:54
monthebiffriiiiiiighhhhht
#asahi · 2025-10-07 03:54
tpw_rules(other than hopefully use of it decreasing over time, thus reducing this problem)
#asahi · 2025-10-07 03:53
tpw_rulesjemalloc being dead is orthogonal
#asahi · 2025-10-07 03:53
tpw_ruleswell it's borked because jemalloc is designed poorly in particular for apple silicon machines
#asahi · 2025-10-07 03:52
monthebiffwait really? all i did was pull down the stock devcontainer feature via code, its borked because jemalloc is dead?
#asahi · 2025-10-07 03:52
tpw_rulesif you can modify the devcontainer you can replace jemalloc with a rebuilt version that is fine with 16/64k pages
#asahi · 2025-10-07 03:51
tpw_rules99.9% of machines use 4k pages and are fine. it was a poor design decision on jemalloc's part. but jemalloc is dead now anyway
#asahi · 2025-10-07 03:50
monthebiffreally, so its just jemalloc lagging behind upstream? wouldn't this break docker images across different machines though?
#asahi · 2025-10-07 03:49
tpw_rulesfixed*
#asahi · 2025-10-07 03:49
tpw_rulesmost programs are fized these days
#asahi · 2025-10-07 03:48
monthebiffthought that might be an answer, was just double checking there wasnt some known thing where vscode was being ran via fex/ muvm or something
#asahi · 2025-10-07 03:47
tpw_ruleshttps://github.com/jemalloc/jemalloc/issues/467 you have to rebuild whatever uses jemalloc
#asahi · 2025-10-07 03:46
monthebifffull context, attempting to use the docker out of docker devcontainer feature is throwing an error while building the devcontainer, offending step looks to be jemalloc (while being invoked by docker-compose).
#asahi · 2025-10-07 03:44
monthebiffafternoon all - been a while. have a page table related error from a docker container. is there a workflow around devcontainers/ asahi i should be aware of?
#asahi · 2025-10-05 15:31
leiochadmed: it's annoying when those dcp negotiated modes don't really match what your monitor can do because (apple) reasons :(
#asahi · 2025-10-05 15:18
sventhough i guess at those speeds most of link training just has to happen in hardware :D
#asahi · 2025-10-05 15:17
svenjust look at atcphy.c for how link training might otherwise looks like, and even that's already surprisingly simple and without much conditions or loops
#asahi · 2025-10-05 15:17
chadmedthe annoying stuff is mostly just that the firmware kinda sucks and has an interface that is annoying to work with
#asahi · 2025-10-05 15:16
bgtlover[mds]ahh ok, no wonder I was confused while passively reading chat
#asahi · 2025-10-05 15:16
chadmedall we need to do is ask it for the information
#asahi · 2025-10-05 15:16
chadmedthe dcp firmware already does all of this and even notifies us _when_ it's done it
#asahi · 2025-10-05 15:15
chadmedand not a likely one, for the reasons discussed
#asahi · 2025-10-05 15:15
chadmedits just a hypothetical
#asahi · 2025-10-05 15:15
chadmedno, we can and we do
#asahi · 2025-10-05 15:15
bgtlover[mds]ahh, gotcha! so y'all can't use apple's firmware?
#asahi · 2025-10-05 15:14
chadmedwhat we're discussing here is bypassing the firmware and for lack of a better term bit banging everything
#asahi · 2025-10-05 15:14
svenelectrons travels very slow actually (something like milimeters per second iirc), what you care about are the fields and the interference caused by the high speed switching between 0/1
#asahi · 2025-10-05 15:14
chadmedbgtlover[mds]: well yes, but someone has to write the firmware
#asahi · 2025-10-05 15:13
chadmedeven with all the cursed firmware versioning :p
#asahi · 2025-10-05 15:13
bgtlover[mds]although I thought that happens automatically without interference from the OS, I thought the gpu and the other side do the negotiations on a firmware level, but yeah, TIL
#asahi · 2025-10-05 15:13
chadmedeasier to just go "hello dcp what modes did you negotiate" and then forward that to userspace
#asahi · 2025-10-05 15:12
bgtlover[mds]but yeah, joking aside, it's very interesting that this stuff needs calibration, although achieving those speeds over thin wire must come with some tradeoffs because electrons can only travel so fast and in only so large a quantity before we get interference, causing information to degrade
#asahi · 2025-10-05 15:12
chadmedits not that the dp spec itself is extraordinarily bad or anything, but managing any high bandwidth data link is a PITA especially when we know the thing on the other end of the link is probably noncompliant in some way or another. virtually every monitor is broken
#asahi · 2025-10-05 15:12
svenML for both branch prediction and link training also makes perfect sense. it's a very well specified problem and you immediately know if you made a mistake and the cost of a mistake isn't all that high. on the other hand, getting something correct can be verified immediately and you directly gain something
#asahi · 2025-10-05 15:10
jannauno, it really didn't
#asahi · 2025-10-05 15:09
bgtlover[mds]the pun about llms :p
#asahi · 2025-10-05 15:09
svenno, it really didn't
#asahi · 2025-10-05 15:09
chaos_princesswhat
#asahi · 2025-10-05 15:08
bgtlover[mds]lol indeed, someone had to say that. After all, apparently ddr5 memory requires proprietary blobs to train and the training involves machine learning the same way the branch predictor does, so yeah, the pun had to be made
#asahi · 2025-10-05 15:07
svenusually they exchange some well-known pattern at the beginning and tune all their parameters until the other side can recover that pattern successfully
#asahi · 2025-10-05 15:05
svenit's usually clock recovery followed by a whole bunch of magic and adjusting filters and drive strengths and timings and whatever so that both sides can actually communicate at multi-Gbps speeds
#asahi · 2025-10-05 15:03
sven:/
#asahi · 2025-10-05 15:01
chaos_princesssigh
#asahi · 2025-10-05 15:01
bgtlover[mds]<chaos_princess> "link training is just what you..." <- link training? that sounds like something to do with llms :p I hope not
#asahi · 2025-10-05 14:57
jannauI think the HDMI spec is worse but since you can't access the HDMI (or implement it in open source (HDMI 2.1)) it can't hurt you
#asahi · 2025-10-05 14:53
chaos_princesslink training is just what you have to do on every high speed link, it is more physics hating you than the spec being obtuse on purpose
#asahi · 2025-10-05 14:52
bgtlover[mds]ha! that sounds as if the displayport spec is worse than hdmi
#asahi · 2025-10-05 14:50
chadmedid rather deal with clunky swaps than... *gestures at displayport spec*

plus ›