Phoronix already tested this: https://www.phoronix.com/scan.php?page=article&item=l1tf-ear... and https://www.phoronix.com/scan.php?page=article&item=l1tf-for....
The performance loss isn't that bad in most cases.
The performance loss isn't that bad in most cases.
I assume the microcode update could be either equal o slightly worse in terms of performance, as the CPU might need to flush more frequently.
Which is pretty sad, as the status of kernel+microcode updates is quite confusing already. Some mitigations can take advantage of new microcode updates, if the kernel is recent enough. How does the pure soft-workaround compare in terms of performance to the microcode-assisted one?
Note that, combining all workarounds for meltdown+spectre-v1/v2+l1tf can have a significant performance hit for some workloads which are not purely cpu-bound. On top of that HT is now looking like a bad idea to start with.
I'm pretty sure that for a server where there's a lot of I/O and virtualization going on, enabling all the patches and workarounds + disabling HT can take a massive cut in overall throughput.
Their security advisory says that
> Optimized L1 data cache flushing is available via intel-microcode updates. The updated kernels implement a software fallback cache flushing mechanism for processors that have not received microcode updates.
It looks like the kernel will say "VMX: conditional cache flushes" on the new microcode so according to the Phoronix screenshots, they were running the older version.
Maybe we'll see a new set of benchmarks.
Also there was significant reduced performance when hyperthreading was disabled as required to fully mitigate the threat of the vulnerabilities. If the microcode update changes the behavoir of hyperthreading in order to fully mitigate the vulnerabilities without having to completely disable it then there is a chance these changes effect the performance benefit of hyperthreading.
Given Intel's anti-benchmark license clause without evidence otherwise I assume it is not just a chance but a strong likelihood that the performance benefits are significantly impacted.