Conversation

Notices

1. Foone🏳️‍⚧️ (foone@digipres.club)'s status on Friday, 27-Oct-2023 04:07:10 JST Foone🏳️‍⚧️

   xfce4-session crashed with SIGSEGV in __strcat_avx2()

   that's probably not a good sign

   • Foone🏳️‍⚧️ (foone@digipres.club)'s status on Friday, 27-Oct-2023 04:12:53 JST Foone🏳️‍⚧️

     in reply to

     Actually it might just be a bad pointer bug, which is bad, obviously, but not as bad as "your cpu just randomly forgot how to do AVX2"

   • Foone🏳️‍⚧️ (foone@digipres.club)'s status on Friday, 27-Oct-2023 04:18:44 JST Foone🏳️‍⚧️

     in reply to

     it crashed while trying to execute "lock vpmovmskb %ymm6, %ecx" which is a completely reasonable bit of x86-64 machine code I'm sure

   • Foone🏳️‍⚧️ (foone@digipres.club)'s status on Friday, 27-Oct-2023 04:21:39 JST Foone🏳️‍⚧️

     in reply to

     strcatting 256 bits at a time seems scary. I'm sure that's not what broke, but it's still a lot.

     back in my day we moved AT MOST 4 bytes at once, never 32 of them all willy-nilly!

   • Foone🏳️‍⚧️ (foone@digipres.club)'s status on Friday, 27-Oct-2023 04:25:45 JST Foone🏳️‍⚧️

     in reply to

     the fun thing is that I have an Alder Lake processor, which means that SOME of my CPU cores have AVX-512 support, so I could in theory be moving 64 bytes at once.

     But in reality, because of the fact that Alder Lake is "an Abomination of Asymmetric CPU Design" (technical term), the AVX-512 support is disabled to prevent problems when my processes get migrated and suddenly lose AVX-512 support

   • Foone🏳️‍⚧️ (foone@digipres.club)'s status on Friday, 27-Oct-2023 04:27:54 JST Foone🏳️‍⚧️

     in reply to

     This particular CPU has 14 cores: 6 of them "performance" and 8 of them "efficiency" and they have different CPU instruction support. Since this would cause no end of problems, the CPUID is masked to only report the lowest common denominator of supported functions

   • Foone🏳️‍⚧️ (foone@digipres.club)'s status on Friday, 27-Oct-2023 04:33:09 JST Foone🏳️‍⚧️

     in reply to

     A fun thing about this is that Intel had to develop something called the Intel Thread Director, aka the Enhanced Hardware Feedback Interface.

     BASICLY it means the CPU will generate an interrupt to tell the OS which cores are a good idea to put threads on right now.

     https://docs.kernel.org/arch/x86/intel-hfi.html

   • Foone🏳️‍⚧️ (foone@digipres.club)'s status on Friday, 27-Oct-2023 04:34:29 JST Foone🏳️‍⚧️

     in reply to

     it can do this multiple times a second, and works by telling how much performance and energy efficiency each CPU core has at the moment, based on things like current temperature and battery/AC power.

   • Foone🏳️‍⚧️ (foone@digipres.club)'s status on Friday, 27-Oct-2023 04:45:17 JST Foone🏳️‍⚧️

     in reply to

     • Will

     @waffle_iron Atom is the single most important CPU design Intel ever made and they will never let it die

   • Will (waffle_iron@nyan.lol)'s status on Friday, 27-Oct-2023 04:45:18 JST Will

     in reply to

     @foone It amuses me to no end how Intel is happy to let people forget that efficiency cores are evolutions of Atom cores.

   • Foone🏳️‍⚧️ (foone@digipres.club)'s status on Friday, 27-Oct-2023 04:46:53 JST Foone🏳️‍⚧️

     in reply to

     don't worry though. by default, linux throttles Enhanced Hardware Feedback Interface updates to 250 a second

   • Foone🏳️‍⚧️ (foone@digipres.club)'s status on Friday, 27-Oct-2023 05:00:34 JST Foone🏳️‍⚧️

     in reply to

     anyway reportedly the big problem which made the split-CPUID scheme an issue was DRM:
     various programs monitor their CPUID so they can see if it ever changes, on the assumption that if it ever does, someone is doing something fucky with a VM or emulation, and refuse to work.

     So intel had to modify the microcode to make Alder Lake cores always return the same CPUID, to avoid it breaking DRM

   • Foone🏳️‍⚧️ (foone@digipres.club)'s status on Friday, 27-Oct-2023 05:03:17 JST Foone🏳️‍⚧️

     in reply to

     which is funny, because one of the other things they did with Adler Lake is INTENTIONALLY break DRM

   • Foone🏳️‍⚧️ (foone@digipres.club)'s status on Friday, 27-Oct-2023 05:06:04 JST Foone🏳️‍⚧️

     in reply to

     They dropped support for Intel Software Guard Extensions, which is a way to set up enclaves on the CPU where all information is encrypted on the fly, so even the OS or a hypervisor can't spy on it.

     and Intel SGX was required to playback Ultra HD Blu-ray discs on PCs

   • Foone🏳️‍⚧️ (foone@digipres.club)'s status on Friday, 27-Oct-2023 05:08:43 JST Foone🏳️‍⚧️

     in reply to

     • badsynthesis

     @badsynthesis it's supposed to be hidden from you by the OS and the CPU microcode. "supposed to" being the key words

   • badsynthesis (badsynthesis@infosec.exchange)'s status on Friday, 27-Oct-2023 05:08:44 JST badsynthesis

     in reply to

     @foone ...how is one supposed to develop for these abominations? CPU/core pinning so the instruction set doesn't change halfway for everything doesn't seem very scalable.

   • Foone🏳️‍⚧️ (foone@digipres.club)'s status on Friday, 27-Oct-2023 05:14:47 JST Foone🏳️‍⚧️

     in reply to

     • Glyph

     @glyph I believe so, yeah.

   • Glyph (glyph@mastodon.social)'s status on Friday, 27-Oct-2023 05:14:48 JST Glyph

     in reply to

     @foone …wait. does that mean that it's, like, physically impossible to play UHD discs on a mac

   • Foone🏳️‍⚧️ (foone@digipres.club)'s status on Friday, 27-Oct-2023 06:44:10 JST Foone🏳️‍⚧️

     in reply to

     • Kit Bashir

     @Unixbigot the last time Intel designed a good CPU, it couldn't divide properly

   • Kit Bashir (unixbigot@aus.social)'s status on Friday, 27-Oct-2023 06:44:11 JST Kit Bashir

     in reply to

     @foone some companies should not be allowed to have architecture

   • Foone🏳️‍⚧️ (foone@digipres.club)'s status on Saturday, 28-Oct-2023 05:18:59 JST Foone🏳️‍⚧️

     in reply to

     GAH it crashed again
     another segfault: it tried to read from address 0x10000 which wasn't readable.

     having your fucking session manager crash is no fun, it means everything you have open dies

   • Foone🏳️‍⚧️ (foone@digipres.club)'s status on Saturday, 28-Oct-2023 05:23:18 JST Foone🏳️‍⚧️

     in reply to

     I may need to memtest this machine