970
this post was submitted on 13 Aug 2023
970 points (99.0% liked)
Technology
57432 readers
4465 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related content.
- Be excellent to each another!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, to ask if your bot can be added please contact us.
- Check for duplicates before posting, duplicates may be removed
Approved Bots
founded 1 year ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
ELI5, or ELIAFYCSS (Explain like I'm a first year CS student): modern x86 CPUs have lots of optimized instructions for specific functionality. One of these is "vector instructions", where the instruction is optimized for running the same function (e.g. matrix multiply add) on lots of data (e.g. 32 rows or 512 rows). These instructions were slowly added over time, so there are multiple "sets" of vector instructions like MMX, AVX, AVX-2, AVX-512, AMX...
While the names all sound different, the way how all these vector instructions work is similar: they store internal state in hidden registers that the programmer cannot access. So to the user (application programmer or compiler designer) it looks like a simple function that does what you need without having to micromanage registers. Neat, right?
Well, problem is somewhere along the lines someone found a bug: when using instructions from the AVX-2/AVX-512 sets, if you combine it with an incorrect ordering of branch instructions (aka JX, basically the if/else of assembly) you get to see what's inside these hidden registers, including from different programs. Oops. So Charlie's "Up, Up, Down, Down, Left, Right, Left, Right, B, B, A, A" using AVX/JX allows him to see what Alice's "encrypt this zip file with this password" program is doing. Uh oh.
So, that sounds bad. But lets take a step back: how bad would this affect existing consumer devices (e.g. Non-Xeon, non-Epyc CPUs)?
Well good news: AVX-512 is not available on most Intel/AMD consumer CPUs until recently (13th gen/zen 4, and zen 4 isn't affected). So 1) your CPU most likely doesn't support it and 2) even if your CPU supports it most pre-compiled programs won't use it because the program would crash on everyone else's computer that doesn't have AVX-512. AVX-512 is a non-issue unless you're running Finite Element Analysis programs (LS-DYNA) for fun.
AVX-2 has a similar problem: while released in 2013, some low end CPUs (e.g. Intel Atom) didn't have them for a long time (this year I think?). So most compiled programs wouldn't compile with AVX-2 enabled. This means whatever game you are running now, you probably won't see a performance drop after patching since your computer/program was never using the optimized vector instructions in the first place.
So, the affect on consumer devices is minimal. But what do you need to do to ensure that your PC is secure?
Three different ideas off the top of my head:
BIOS update. The CPU has a some low level firmware code called microcode which is included in the BIOS. The new patched version adds additional checks to ensure no data is leaked.
Update the microcode package in Linux. The microcode can also be loaded from the OS. If you have an up-to-date version of Intel-microcode here this would achieve the same as (1)
Re-compile everything without AVX-2/AVX-512. If you're running something like Gentoo, you can simply tell GCC to not use AVX-2/AVX-512 regardless of whether your CPU supports it. As mentioned earlier the performance loss is probably going to be fine unless you're doing some serious math (FEA/AI/etc) on your machine.