The Elite Intel Team Still Fighting Meltdown and Spectre (wired.com)

Throughout 2018, researchers inside and outside Intel continued to find exploitable weaknesses related to Meltdown and Spectre class of "speculative execution" vulnerabilities. Fixing many of them takes not just software patches, but conceptually rethinking how processors are made. From a report: At the center of these efforts for Intel is STORM, the company's strategic offensive research and mitigation group, a team of hackers from around the world tasked with heading off next-generation security threats. Reacting to speculative execution vulnerabilities in particular has taken extensive collaboration among product development teams, legacy architecture groups, outreach and communications departments to coordinate response, and security-focused research groups at Intel. STORM has been at the heart of the technical side. "With Meltdown and Spectre we were very aggressive with how we approached this problem," says Dhinesh Manoharan, who heads Intel's offensive security research division, which includes STORM. "The amount of products that we needed to deal with and address and the pace in which we did this -- we set a really high bar."

Intel's offensive security research team comprises about 60 people who focus on proactive security testing and in-depth investigations. STORM is a subset, about a dozen people who specifically work on prototyping exploits to show their practical impact. They help shed light on how far a vulnerability really extends, while also pointing to potential mitigations. The strategy helped them catch as many variants as possible of the speculative execution vulnerabilities that emerged in a slow trickle throughout 2018. "Every time a new state of the art capability or attack is discovered we need to keep tracking it, doing work on it, and making sure that our technologies are still resilient," says Rodrigo Branco, who heads STORM. "It was no different for Spectre and Meltdown. The only difference in that case is the size, because it also affected other companies and the industry as a whole."

Comments

Intel shits bed massively, hires cleaning crew

[Anonymous Coward]

To clean their massive shitty bed.

Better solution: less Intel

[xxxJonBoyxxx]

It's great to see ARM chips take off in popularity. In ten years, Intel will be like Sun is now. ("Intel, like the old chipmaker gramps?")

Why Intel has this issue and AMD does not

[Anonymous Coward]

Go back many years to the putrid Intel Netburst architecture. Single core, very long pipeline, massive caches and the goal of 10GHz. It was the post Pentium 3 design and it was DREADFUL. But Intel paid all major tech outlets, including this one, to sing its praises.

Then AMD invented AMD64 (now called x64) and true mutli-core x64 chips. AMD's tech lead over Intel was massive- even if sites like this one still shilled Intel Netburst.

But AMDand Intel had a cross patent agreement. Intel took the best of AMD's new tech, crossed it with the older Pentium 3 design, and invented Core 2 - which was then used for Intel's much later true dual core parts. And here arose the issue.

For the first time an Intel chip would have TWO threads running on the same chip at the same time, sharing many on-chip and off-chip memory resources. The ONLY way to do this properly is called 'lock and key'. Every thread has a 'key' (unique id) and each access of shared memory must use that 'key' to 'unlock' access to a resourse intended for that thread alone. But 'lock and key' is complex to design. Uses a LOT of transistors. Uses power. And introduces sinificant memory latencies. And makes it harder for the NSA to hack into the chip. So Intel NEVER implemented 'lock and key'. Instead Intel worked with another NSA partner, Microsoft, to use an OS 'solution' that the NSA could easily bypass.

For 10+ years all tech sites conspired to lie and state the OS thread system could provide thread security. It could not. Then the bubble burst, and Spectre and Meltdown finally revealed the atrocious state of ALL Intel CPUs.

Meanwhile AMD had always implemented 'lock and key' in CPU hardware. As a result AMD's current fantastic Ryzen Zen parts cannot hit Intel speeds and have higher memory latencies than Intel- leading to worse gaming performance for gamers wanting >120 Hz refresh. But Intel's clock and memory latency win is only possible cos Intel chips all fail to implement thread security. So intel CHEATS and pays sites like this one to hide this fact.

Zen 2- announced in a few days time, uses superior engineering and TSMC's leading 7nm process to finally close the gap with Intel. A gap already made irrelevant when using decently coded programs that are properly multi-threaded because of AMD's core advantage (at any given price point).

Intel is curently paying tech sites to benchmark using decades old CAD programs that are single threaded and use the long obsolete x87 FPU instructions cos Intel shows a big win here. Intel pays sites like this one to spread the FUD that Intel is fixing their problem (they cannot) and that anyway AMD has the same issue (totally untrue). Intel's real fix happens when they totally redesign their CPU (which will take at least FIVE years) and even then the redesign will massively crater Intel's performance.

Today the ONLY way to safely use an Intel CPU is to only run one thread at a time on the chip, and do a complete state flush on the CPU between multi-tasking thread swaps. A modern coffee-lake six core Intel CPU would see its performance drop by 90%+ if this fix were implemented tho- so you can see why Intel is desperate to pay for lies suggesting the fix is not needed.

Anyone using Intel CPUs today is a complete fool.

Re:Why Intel has this issue and AMD does not

[nuckfuts]

Anyone using Intel CPUs today is a complete fool.

This kind of blanket statement show a complete lack of pragmatism. You are clearly an AMD fanboi in the same vein as anti-MS fanbois who would chant "anyone using a Windows operating system today is a complete fool".

The world of computing is not so black and white, and there are myriad reasons why one might choose a particular architecture or OS over another one. Don't presume to know everyone's usage cases better than they know them themselves.

More to the point, however, is the fact that in your entire rant against Intel, you only talk about "lock and key". You don't even mention the topic of speculative execution, which is the basis of these vulnerabilities.

Speculative execution is a class of vulnerabilities, not a specific implementation flaw, which is what makes it difficult to mitigate. If you stop using any speculative execution, you will take performance hit. So it becomes a question of risk vs benefit. Again, it's not simply black and white. The team at Intel is trying to figure out how to retain some of the benefits while mitigating the risks. Nobody wants to throw out the baby with the bathwater.

Perhaps in your world it's as simple as "just use only AMD", but I can assure you if everyone actually followed that advice, it would simply require the bad guys to focus all their attention on new targets, and inevitably, new vulnerabilities would be found. There is no perfect technology that can't be exploited, and probably never will be.

Re:Why Intel has this issue and AMD does not

[Anonymous Coward]

While speculative execution could create a new class of vulnerabilities, Intel's problem is /solely/ Intel's problem. You see, Intel released a processor architecture called the pentium (i586) in 1993. It supported the same privilege system as 386/486, but done using super-scalar design, improved cache, and improved addressing. It was a huge leap in performance. It also is not vulnerable to Meltdown/Spectre. Why? It's not because it simply lacked the execution attacks caused by speculative execution. It's that there are no way to blow holes into the privilege system. Why is this known to be true? Because the Atom architecture is a modernized form of the pentium, and is completely immune.

So then come 1995. There was a company called Cyrix. Note this isn't a matter of being and AMD fanboi as Cyrix was a completely separate non-licensee of Intel x86 that was very successful at doing x86 very well, arguably better, before AMD even adopted that strategy. They were already eating into Intel market share massively during around time the pentium was launched. Rumor has it they were faster at executing x86, and also providing higher clock speeds on common socket architecture, keeping things like the 386 and 486 class hardware alive when Intel wanted to get people to upgrade. They were insanely less expensive than Intel. So Intel was getting a little worried about them, even thinking about litigation because how dare they clean room reverse engineer and then optimize the x86 architecture without paying them! So one of Intel's project was to beat them on the performance punch by rapidly releasing another next gen architecture - pentium pro (i686), with speculative execution. But you see, it wasn't enough just to beat them on implementation done the right way. They decided to break the privilege layers when executing to go FASTER. This might have made sense in 1994-95 before the Internet really took off, but it certainly was pretty foolish.

So Cyrix in 95 was readying their pentium clone, which not too bad, then they had to respond to the i686 platform, which took them by surprise, and they had probably a harder time duplicating the super-scalar methods, because most likely they were trying to do it the right way, and Intel's performance was because they took as many short cuts as possible. Evidence that Cyrix and AMD were struggling getting it "right" is with how long Cyrix took on their i686 clone, and how much they were hammered by poor benchmarks. Cloning i686 literally did Cyrix in. AMD's version of i686 is not vulnerable to meltdown, which clearly shows that the designers knew the right method, but of course even AMD performs badly on unprotected execution of a benchmarked Intel.

But the kicker today, is that with the fix, Intel's chips are much slower since speculative execution has to be disabled. It was fundamentally broken. Compare that with AMD where speculative execution does work, now AMD is and was always the performance king.

So Intel lied to everyone and put out a broken product for years. It is totally in their implementation, and you can see the difference from various competitors to this day. So you can try to say that this was just pros or cons in different offerings, but pragmatism really does show when you know what happened is that Intel cheated to kill their competitor, and now they are reaping what they sowed.

Re:Why Intel has this issue and AMD does not

[drinkypoo]

Rumor has it they were faster at executing x86, and also providing higher clock speeds on common socket architecture,

Rumor is false. As a cheap bastard, I tried the cheapass architectures, which sucked until the k7 came out. Cyrix chips were stunningly cheap, but also staggeringly slow. AMD chips were in the middle in both regards, with Intel by far fastest and most expensive. But then Athlon showed up and knocked Intel's socks directly off for a time, especially in multiprocessor systems where Hypertransport was a drastic improvement.

However, what is often missed in the analysis is that until recently, Intel was the world leader in process technology for years. Even without failing at security design, that gave them a part of their performance advantage. But those days are now gone, so what's Intel got over AMD? Answer, nothing.

Need security? Don't use JIT!

[Gravis Zero]

Meltdown is an Intel problem. Spectre is only a problem if you use Just-In-Time compilation on your system. The obviously solution is to simply not use JIT in the first place. Nothing fundamentally needs it, it simply makes the execution of unverified code faster. Nobody writing applications needs to worry about Spectre... unless you are writing a JIT compiler. This is a very small number of applications and they can still run unverified code using an interpreter engine, it's just a bit slower.

The solution is simple: dump JIT.