Bisected: The Unfortunate Reason Linux 4.20 Is Running Slower

[ryao]

That is literally the opposite of what you said in the comment I replied to.

... and doesn't change the fact that the no viable attacks have been demonstrated.

-- and if you bothered reading what you link, AMD are advising against using STIBP. They are only recommending one of the three mechanisms.

If you read my other 2 comments on the topic, you would know that I stated that AMD discourages stibp as a Spectre v2 mitigation. Furthermore, you are missing the point that they clearly state that they recommend a Spectre v2 mitigation.

The change in mainline was made because someone felt that the current measures do not go far enough. I have neither have studied this in enough detail to know one way or another nor have access to the likely confidential information that the mainline developers are using to make their decisions to evaluate them, but I doubt that you are in any better position than I am to second guess their decisions this. Anyway, here is what I know:

1. AMD is vulnerable to Spectre v2.
2. This patch is written in a way that should apply this mitigation to AMD hardware, whether it needs further mitigations or not.
3. Reports contradict #2.

If I had access to an AMD Zen machine, I could probably figure out the contradiction.

[ryao]

Yes, because AMD CPUs are invulnerable to those attacks. No wait, they aren't...

They are invulnerable to a handful of them, such as meltdown.

[juanrga]

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.

There is no way you can clock processors to 7 or 8 GHz for compensating for elimination of speculation. Engineers hit a frequency wall at around 4--5 GHz

[Weasel]

It is linear, not exponential. Voltage increases are quadratic. The unfortunate reality that higher clock speeds are dependent on voltage increases makes it cubic-ish.

Damn that's even worse than I thought then.

[juanrga]

All of my AMD systems were tested with their latest BIOS releases.

That doesn't imply that all the mitigations were enabled for them.

[juanrga]

They are invulnerable to a handful of them, such as meltdown.

I guess you missed recent research and my former post:

We also debunk implicit assumptions including that AMD processors are immune to Meltdown-type effects

[duby229]

I guess you missed recent research and my former post:

They would have to demonstrate an exploit in order to debunk AMD's supposed immunity, which they did not do. No exploit in the literal sense means no exploit.

[hotaru]

I guess you missed recent research and my former post:

that's a bit disingenuous, considering that neither of the two meltdown-type vulnerabilities that affect AMD processors can cross privilege boundaries, while meltdown itself can and three other meltdown-type vulnerabilities that can cross privilege boundaries affect Intel processors, but not AMD.

[ryao]

I guess you missed recent research and my former post:

I did not miss it. However, I only refer to Spectre v3 as Meltdown and AMD processors are reportedly immune to it,

[carewolf]

1. AMD is vulnerable to Spectre v2.

Only theoretically, not practically.. Which makes a huge difference in what kind of mitigations are warranted

2. This patch is written in a way that should apply this mitigation to AMD hardware, whether it needs further mitigations or not.

No, it it is written in a way where it COULD be applied to AMD hardware(*), but shouldn't necessarily, and isn't.

3. Reports contradict #2.

Not sure what you are talking about. You are contradicting yourself so much it is hard to keep track.

* Note, when I repeat your term "AMD hardware" here, I mean Zen. Older AMD chips have indirect branch predicators more similar to Intels and are just as vulnerable. Only the latest Zen architectures have a newer design, that is effectively immune unless someone sets up specific circumstances, or a new trick is discovered.