Linux Fixing A "Hilarious/Revolting Performance Regression" Around Intel KVM Virtualization

[phoronix]

Phoronix: Linux Fixing A "Hilarious/Revolting Performance Regression" Around Intel KVM Virtualization

It's not too often that "fixes" to the Kernel-based Virtual Machine (KVM) are noteworthy but today is an interesting exception with among the KVM fixes sent in today ahead of the Linux 6.13-rc3 tagging is for beginning to deal with a "hilarious/revolting" performance regression affecting recent generations of Intel processors. This performance regression won't be fully worked around until Linux 6.14 but at least there is an interim step in place once the code is merged later today...

https://www.phoronix.com/news/Linux-Intel-KVM-Cache-CPUID

[[email protected]]

Thank you for finding the problem! Google

[sophisticles]

Makes me want to laugh and vomit at the same time:

Linux Fixing A "Hilarious/Revolting Performance Regression" Around Intel KVM Virtualization

[iggy]

Unless I'm reading this wrong, this only impacts nested virtualization which very few people use and even fewer people expect to perform well (especially on Intel CPUs). So it's unsurprising it has gone unfixed.

[Gonk]

Unless I'm reading this wrong, this only impacts nested virtualization ...

Yes and no. The CPUID performance problem is most obvious when using nested virtualization due how it multiplies the impact by executing it far more than necessary, but the problem is still there even when not using virtualization at all.

[mcloud]

They might as well improve the KVM code while at it, sheesh...

[Gonk]

They might as well improve the KVM code while at it, sheesh...

Uh, yeah, that's what they're doing. All the patches are to the kvm subsystem.

[FeRD_NYC]

Uh, yeah, that's what they're doing. All the patches are to the kvm subsystem.

Which could end up being "unnecessary", ultimately, if Intel come up with a µcode patch for these processors that fixes their insane CPUID latency at the hardware level. But, since it sounds like the KVM code was doing a lot of questionable faffing around in its emulation code, cleaning that up isn't the worst thing even if Intel eventually patch this away.

(It's like the whole climate-change thing: Worst-case scenario, what if it turns out that 97% of scientists are somehow wrong about human activity being the cause, and we end up making our industries more environmentally conscious, reducing our reliance on fossil fuels and our carbon output and greenhouse emissions, and just generally make the planet a cleaner, healthier, more pleasant place to live... "for nothing"? I guess we'll just have to take that chance.)

[FeRD_NYC]

Makes me want to laugh and vomit at the same time:

That's how you know it's working!

[carewolf]

Makes me want to laugh and vomit at the same time:

As a framework developer, that is my usual reaction to seeing user code.