Bisected: The Unfortunate Reason Linux 4.20 Is Running Slower

[gmturner]

Pretty sure insider-types have been making ominous statements to the effect that they expect many more performance-harming exploits/mitigations to appear... Scary to think there may well be more bad news of this nature in the pipeline. Also to anyone saying how great this is for AMD, consider this: if you're a security researcher that's uncovered a new class of exploit in, let's say, mid 2017, the obvious platform to code your proof-of-concept exploit for is Intel. This is like saying that linux is more secure because it has less desktop malware in the wild... basically your status as a minority platform provides a relative disincentive to attackers. It's pretty clear that Intel's market-share will not continue as it has in the past, and not only because of the these security issues. IME it's a safe bet that similar exploits will appear for Zen targets sooner or later, and AMD has been wise not to seize this window of opportunity to make sweeping claims about their hardware being free from security flaws.

Finally, to whomever said Intel was cheating, nobody is saying that so far as I know (at least, not about Spectre/Meltdown). This isn't purported to be about cheating, but about performance enhancing features of modern desktop/server platforms that leaked information in ways so subtle that nobody thought of it until now. It would be somewhat scammy if Intel/AMD knew about these issues all along, but chose to suppress that information, rather than develop mitigations. I have not heard any allegations to that effect, and even if such accusations emerged, so long as the exploits in question remained theoretical, I would still take them with a grain of salt.

[duby229]

With all security mitigations enabled Skylake has lower IPC than Bulldozer... And they called Bulldozer a "failure". Sure, when you don't care about security at all you can get better IPC... How come AMD didn't think of this?

And Bulldozer only had 2 integer units available per thread while in single threading Skylake could use3 of them..... So yeah it speaks volumes.

[duby229]

Intel could bring back Itanium as the solution. It was designed to avoid speculative execution from the start.

Yeah, by executing every branch, including every branch not taken. That's why sucked so hard at emulating x86 and as a general purpose processor in general. You can write specialized algorithms to take advantage of how it parallelizes branches, but it's just not how most loads work.

[dungeon]

Also to anyone saying how great this is for AMD...

Well, to quote finalzone post stats on AMD Ryzen says this:

* CPU vulnerability to the speculative execution attack variants
* Vulnerable to CVE-2017-5753 (Spectre Variant 1, bounds check bypass): YES
* Vulnerable to CVE-2017-5715 (Spectre Variant 2, branch target injection): YES
* Vulnerable to CVE-2017-5754 (Variant 3, Meltdown, rogue data cache load): NO
* Vulnerable to CVE-2018-3640 (Variant 3a, rogue system register read): NO
* Vulnerable to CVE-2018-3639 (Variant 4, speculative store bypass): YES
* Vulnerable to CVE-2018-3615 (Foreshadow (SGX), L1 terminal fault): NO
* Vulnerable to CVE-2018-3620 (Foreshadow-NG (OS), L1 terminal fault): NO
* Vulnerable to CVE-2018-3646 (Foreshadow-NG (VMM), L1 terminal fault): NO

Older Intel CPUs have YES to all and that is from where AMD users comments coming Intel became nearly equal to AMD since only newest 9xxx, also since 9800K have HT, this probably making non HT 9700K sexier since these have also fixed Meltdown/Foreshadow

It is great for existing AMD users as they are not so affected and that is valid. Software mitigations will eat some performance but is correct thing to do, of course by the time vendors will fix all these on new hardware.

At the end if you don't have security concerns then disable all mitigations and fly

[NateHubbard]

Microsoft kernel's mitigation for these hyperthreading security problems only applies between threads belonging to different processes running on the same physical core. Why not implement this in Linux?

I'd rather we didn't start taking security advice from Microsoft.

[RealNC]

So disable SMT.

Good thing Intel removed HT in new i7 CPUs. At first I thought this was just Intel being Intel (== charging more money while offering less in return), but it turns out they were pro-consumer. HT makes your CPU slower now, so they were thinking ahead.

* applause *

[dungeon]

Yeah world will be safer if CPU vendors don't use HT anymore and if nVidia does not release Ti cards anymore

PlayStation is much safer anyway, it caps everything so you can't go crazy

[abott]

Some people here are highly moronic.

Intel beats AMD in single thread, but it's not from "cheating". They have a flaw which reduces performance to fix it currently, but in-hardware the fix would be no overhead at all. Moronic to call it "cheating" and it's really pathetic to see that here.

Hyperthreading is useful for running multi-threaded programs that only need two threads. It doesn't have to hit another cache for the data it needs, it can stay on-core for it, and it's not a bad feature. It also can block it's self, and not all cores for data because it never leaves the core. It will never go away, either, until we have another architecture bigger than x86/x64. It's also moronic to suspect that.

The amount of FUD here is pathetic, it is painful to read quite a few of these posts. Take off your tinfoil hats, you pathetic morons.

I just can't wait until Zen 2/Intel 10nm hardware to come out to fix them and be secure, that's all there is to it really. Give them time, they'll fix it, and we'll find another thing worse in 10 years again.

[mayankleoboy1]

can someone take these findings and post them on Reddit or Anandtech? I want this to get viral and make shintel even more uncomfortable

[ssokolow]

Seven more speculative execution vulnerabilities found. An absolute disaster for CPU design. I guess the answer is remove Hyper Threading from mainstream processors and bump up the core count and remove speculative decision making from the architecture and offset the losses with clock speed increases. 7 to 8 GHz would be good? I guess smarter and more efficient processor design means worse security.

Then you'd have a different kind of meltdown. The big thing that's holding back clock speeds is, once you get up to around 4GHz, it gets exponentially harder to increase clock speed without causing waste heat to shoot through the roof.