Benchmarking Linux With The Retpoline Patches For Spectre

[MWisBest]

Just an FYI, AMD systems are getting a more optimized patch than the generic retpoline that this was likely tested with. Note the last sentence on the first page of the article:

As the Linux Retpoline support matures, of course, there will be new tests on Phoronix.

This is far from finalized.

[Spacefish]

So we might be looking at a up to 60%+ performance penality for Redis with KPTI + Retpoline + new Microcode.. Damn Intel PR that seems to cover this up and probably will announce "so much better peformance in redis" for their 10th generation Ice Lake CPUs...

[juanrga]

Thx for the, ..........ahem........, "clarification"? And, of course, the name-calling.

Well, others asked for clarification, although this KPTI issue was clearly stated in the article. You, instead, wrote a "Please re-run the AMD tests without KPTI."

[juanrga]

Your anger is misplaced. Spectre is a hardware flaw, but can be mitigated via software at the cost of performance. AMD had the foresight to prevent Meltdown, while Intel did not. Spectre affects all of them, so until new hardware design comes out, Spectre has to be addressed via retpoline.

https://meltdownattack.com/

"At the moment, it is unclear whether AMD processors are also affected by Meltdown."

[juanrga]

The "negligible performance impact" is related to variant 1. The retpoline patchset fixes variant 2, which AMD deemed is impossible on their CPUs.

Not sure what is the plan forward. Either retpoline won't be enabled on AMD CPUs (since, as of now, it is not needed) or it will (just in case someone reproduces variant 2 on AMD CPUs). Since variant 2 depends on branch prediction inner workings, it is heavily dependent on CPU microarchitecture. The fact it was not done on AMD CPUs doesn't mean it's impossible. For now, without retpoline, we have something like "security by obscurity" - AMD CPUs are safe since their branch predictor was not reverse-engineered yet.

When AMD PR shared that slide about the near zero risk, AMD engineers were already working with kernel developers to patch AMD hardware

https://www.suse.com/es-es/support/u...su-20180011-1/

Finally AMD PR already starts admitting that its CPUs aren't so invulnerable like they pretended

https://www.cnet.com/news/amd-spectr...ips-intel-arm/
https://www.theverge.com/2018/1/11/1...tes-ryzen-epyc
https://seekingalpha.com/news/332250...y-spectre-flaw

But we have known that AMD CPUs are vulnerable since the first minute

"Which systems are affected by Spectre?

Almost every system is affected by Spectre: Desktops, Laptops, Cloud Servers, as well as Smartphones. More specifically, all modern processors capable of keeping many instructions in flight are potentially vulnerable. In particular, we have verified Spectre on Intel, AMD, and ARM processors."

https://security.googleblog.com/2018...-need.html?m=1

"These vulnerabilities affect many CPUs, including those from AMD, ARM, and Intel, as well as the devices and operating systems running on them."

[linner]

Personally I'd like some benchmarks of the effects against virtual machines (in particular qemu/kvm and xen) because this is another area where servers could take a hit.

[Rallos Zek]

The IBM POWER6 showed it's possible to design a high performance pipe-lined CPU with in-order-execution by utilising SMT amongst other design decisions, this allowed more of the die to be used for other things.

Is that why IBM admitted that Power5 was not all the up to snuff with older software and quickly returned to a OoO design with the power7? In Order designs have no place for general propose cpus, even mighty Intel gave up on In Order with the Atom and realized that it could only work by being OoO. In order is a failure outside of DSP's and pie in the sky navel gazers who drool over vaporware like the Mill.

[s_j_newbury]

Is that why IBM admitted that Power5 was not all the up to snuff with older software and quickly returned to a OoO design with the power7? In Order designs have no place for general propose cpus, even mighty Intel gave up on In Order with the Atom and realized that it could only work by being OoO. In order is a failure outside of DSP's and pie in the sky navel gazers who drool over vaporware like the Mill.

Except we now know the assumptions (at least historically) made in OoO design are invalid. POWER7 and later are susceptible to Spectre along with the rest of the industry. At least they didn't make the assumptions Intel did.