Benchmarks Of JCC Erratum: A New Intel CPU Bug With Performance Implications On Skylake Through Cascade Lake

[lowlands]

Well, well, well, if that isn't the consequences of my corner-cutting actions...

[milkylainen]

So now we get another round of "fixes" to fsck core kernel stuff for years to come.
Awesome. Just awesome..

[JPFSanders]

This is the neverending story. I wonder how many more are to come, it feels at this point that whatever performance-enhancing mechanisms you look at in modern Intel CPU's (AMD, ARM to some degree too) there is a huge security hole waiting to be discovered.

On the other hand, me being a cynical bstrd, can't stop thinking that this is some kind of scam to force people to buy new CPU's built around a brand new secure implemented architecture that returns to us the lost speed...

[SofS]

One thing that is not clear is how the newly generated code will distinguish between affect and unaffected x86 processors. Fat binaries and architecture specific code only? This overhead better be transparently managed if it's use is to be widespread.

[CochainComplex]

https://www.phoronix.com/vr.php?view=28453 Michael so much work ...and now you have to bench again

[Scellow]

sooo INTEL CPU were designed for spying, nice

[xorbe]

I didn't see it spelled out in the article, but the conclusion I would draw is that these particularly aligned JCC instructions, if the stars line up with other activity within the cpu, cause either incorrect bytes to enter the decoder, or an incorrect branch target, or branch taken wrong way. The security implications are interesting if you could hammer sensitive offending code depending on which one of these it is.

[plast0000]

......not to start a fanboy war but I think I won't be buying anymore intels from now on. I'm still on my 5yo Haswell PC

[Brane215]

Intel's engineers were kind enough to...

What a load of crap.

This is not some obscure bug that might happen under some circumstances.
This affects fundamental functionality of the product and if it can't be addressed by a patch, product should be replaced.

[archsway]

Typo:

end on a 32-bit boundary.