Linus Torvalds Bashes Intel's LAM - Rejected For Linux 6.2

[phoronix]

Phoronix: Linus Torvalds Bashes Intel's LAM Feature - Rejected For Linux 6.2

Linus Torvalds can be known for his hardware commentary at times like hoping AVX-512 "dies a painful death", Intel's "bad policies" around ECC memory, and giving NVIDIA the finger. The latest colorful commentary by the Linux creator is around Intel's new Linear Address Masking (LAM) feature that aimed to land in Linux 6.2 but is now delayed until the code can be reworked...

https://www.phoronix.com/news/Torvalds-Bashes-Intel-LAM

[Danny3]

I wish he did the same with Intel's planned obsolescence like unlocking some parts of the hardware with software instead of having all the hardware unlocked when you buy it!

[evert_mouw]

Ah Linus looks like he is finally recovering from previous psychosocial attacks and becoming himself again
The future for Linux is bright again

[user556]

Hehe, I bet Linus isn't so anti-AVX-512 now that AMD has demo'd it working surprisingly effective without nerfing the clock ... and adding it as a equal citizen extension too. Fingers crossed even the APUs will get AVX-512.

[coder]

Hehe, I bet Linus isn't so anti-AVX-512 now that AMD has demo'd it working surprisingly effective without nerfing the clock ...

I expect he feels it's still a waste of silicon and context, mostly for the sake of workloads he doesn't care about.

Not saying I agree with him, but if we're to subtract the issues AMD solved from his original litany of complaints, that's what we'd be left with.

Even though AMD didn't implement it at full width, they can't avoid the register & context bloat. And register bloat is starting to look worse, these days, because:

https://fuse.wikichip.org/news/7343/...death-of-sram/

[Draget]

I strongly agree, that making up new acronyms has a hidden, really awful cost of some time down the line, causing subtle confusions and bugs or nudge the complexity of some system just past that point, where someone who wanted to contribute to Linux will just not do it.

I fully agree, that acronyms, especially three letter acronyms, should be introduced *really* carefully. If avoidable, not at all. Or if there is something already that one could use. The 700ms you save, compared to when the word is just typed out, is nothing compared to the tax paid in complexity, confusion and increased learning-curve to a system.

it is a general rule for me in software development and I see it way too often ignored in many projects. :-(
(Also see 'ASS rule from SpaceX')

[Vorpal]

Except for this making me laugh, I would also like if someone could quickly explain to me what is a "mm", so that I can figure out what is actually going on. Also, note that there are only 676 available two-letter acronyms in US-ASCII, so we should use those sparingly.

(edit) NIH is supposed to be "Not Invented Here", I guess.

(edit2) "mm" = managed machine?

mm == memory management. It is a subsystem in the Linux kernel.

[Dar13]

OK, but then what does the expression "per-mm" mean?

I'm kind-of getting it: if this bit-masking can be done per-thread, then such might be a better solution (more granular); per-thread bit-masking can also be easily made to emulate "per-mm" bit-masking if necessary.

In this specific context I'm fairly certain it means 'memory manager' and each process has its own 'memory manager' instance. Making it 'per-mm' means it's a per-process setting instead of per-thread.

[Alliancemd]

The strongest word he used was "Christ". He made suggestions with real-world examples, it's not bashing.

[jeisom]

Except for this making me laugh, I would also like if someone could quickly explain to me what is a "mm", so that I can figure out what is actually going on. Also, note that there are only 676 available two-letter acronyms in US-ASCII, so we should use those sparingly.

(edit) NIH is supposed to be "Not Invented Here", I guess.

(edit2) "mm" = managed machine?

I thought NIH was "not important here". I wonder which of us is right?