Researcher Talks Of Possible CPU Security Mitigations With No Performance Cost

[phoronix]

Phoronix: Researcher Talks Of Possible CPU Security Mitigations With No Performance Cost

A security researcher presented at last weekend's Free and Open source Software Developers' European Meeting (FOSDEM) conference around mitigating processor vulnerabilities like Spectre and Meltdown but without negligible performance cost...

https://www.phoronix.com/scan.php?page=news_item&px=Specmelt-No-Cost-Research

[CochainComplex]

this would be a great thing ...we would get - at least for some cpus - a few percentage of performance back. hopefully this will be realised.

[paulpach]

but without negligible performance cost

Should be:
"but with negligible performance cost"

[LightBit]

Similar idea: https://lore.kernel.org/lkml/[email protected]/

[direc85]

The low-hanging fruit is sometimes the sweetest. I once found out that nobody in a software project had added all necessary indexes to the database.

Let's hope this mitigation is researched and implemented!

[Amano]

This similar idea (process-local memory within the virtual address space) seems closer to be usable in the mid term. At least the code is there already. But the discussion died mid 2019 and it was mentioned that Linus didn't like that approach.

[bug77]

Keep in mind the claimed impact of the current mitigations is NOT confirmed in benchmarks Michael ran.
But this new concept sounds promising. Probably because it's so simple, it makes you wonder why it isn't the default implementation already.

[waxhead]

All good, and sounds like the right way forward, but I doubt there will be a noticeable performance increase on professors already patched with microcode mitigations.

[tildearrow]

Somebody tried to do this 4 years ago, but sadly nobody cared and the project didn't go any further.

[maurossi]

Similar idea: https://lore.kernel.org/lkml/[email protected]/

Hello, is someone knowing why Linus Torvald didn't like the idea?
Thanks