Linus Torvalds Lands A 2.6% Performance Improvement With Minor Linux Kernel Patch

[asmboid]

So funny that people praise O Mighty Linus so much when he only took the idea of Josh Poimboeuf and put his name on it. At least phoronix mentioned the name of the original author, sometimes articles miss even that.

[marios]

More secure critical systems - oh, the humanity!

More undefined critical systems, I would say, since rust is undefined (there is no definition on how the binary generated by Rust code should behave).
I had no intention to flame Rust once more, but you asked for it...

Enlighten us, o Nostradamus of Phoronix - what are these known reasons? And surely the lines with the same mitigation in the lib/usercopy.c were not related?

lol, looks like someone does not know the meaning of "I am not certain it is true" and "if".

More seriously - Linux kernel code is messy. It's a wonderful piece of software, but the code quality in the kernel... highly varies, to say the least. Also, ultimately, all large software is bound to have some code smells and bugs. This is a mix of that, plus hardware issues with CPU speculation. See this post for seemingly the context of these changes: https://www.phoronix.com/news/Torval...rated-Buggy-HW

I agree with that.

[Wuzado]

I am not certain it is true. If Rust indeed causes 2.6% (at least) to C code though, even if it is for special cases, it is reason enough to kick it out of the kernel asap. Sadly it is not going to happen, because of the known reasons.

More secure critical systems - oh, the humanity!

Enlighten us, o Nostradamus of Phoronix - what are these known reasons? And surely the lines with the same mitigation in the lib/usercopy.c were not related?

More seriously - Linux kernel code is messy. It's a wonderful piece of software, but the code quality in the kernel... highly varies, to say the least. Also, ultimately, all large software is bound to have some code smells and bugs. This is a mix of that, plus hardware issues with CPU speculation. See this post for seemingly the context of these changes: https://www.phoronix.com/news/Torval...rated-Buggy-HW

[xhustler]

2.6% improvement, come on Linus - you can do better. From you, we expect 1,589% Axboe type improvements.

[mrg666]

Wow! This is the stroke of genius. Thanks, boss!

BTW, 6.11.6 was just updated with these! I am already running it

[marios]

Just because it mentions speculation doesn't mean that it's related to a mitigation. digging through the history of this code it seems like the adding the barrier to this function it seems like it was added to 6.11, because of something rust linux needed for googles effort to move binder to rust:

https://lore.kernel.org/all/[email protected]/

Speculation is just a feature of modern cpus, and needs to be dealt with at some points, but a barrier is the heaviest weight way deal with it. although I'm not a kernel guy and could be way off.

I am not certain it is true. If Rust indeed causes 2.6% (at least) to C code though, even if it is for special cases, it is reason enough to kick it out of the kernel asap. Sadly it is not going to happen, because of the known reasons.

[quaz0r]

Yeah, 'thanks' for being corporate bitch with 12 years old mentality.

LMAO thanks for the chuckle friend I actually laughed out loud for a hot minute even
meanwhile everyone else cant wait to cry
win/win

[maxmousee]

Love to see it, specially in the next release.
Linus (and all the kernel devs) have been making great improvements lately.

[tandemcykeln]

Yeah, 'thanks' for being corporate bitch with 12 years old mentality.

Linus is the best. Thank you very much.

[fitzie]

If i'm reading this right, this doesn't really help you if you run with "mitigations=off" right?

Just because it mentions speculation doesn't mean that it's related to a mitigation. digging through the history of this code it seems like the adding the barrier to this function it seems like it was added to 6.11, because of something rust linux needed for googles effort to move binder to rust:

https://lore.kernel.org/all/[email protected]/

Speculation is just a feature of modern cpus, and needs to be dealt with at some points, but a barrier is the heaviest weight way deal with it. although I'm not a kernel guy and could be way off.