Massive security hole in CPU's incoming?Official Meltdown/Spectre Discussion Thread

[Topweasel]

I think he was suggesting that the exploits were "fixed" through no intention of Intel, just through design changes. I think an exploit and a bug can be "fixed" if it isn't done so intentionally--it just means that the exploit is no longer there due to design change.

....I do find this hard to believe, however, as the new architecture coming out this year is already 1+ years old, right? Early on in this debacle, posters were saying that it really takes up to 5 years from design to release a new architecture, right? Or are these minor tweaks within the current design, for this upcoming generation, that fortuitously managed to fix the exploits?

There is a lot of work that goes into producing a chip. In theory anything AMD or Intel releases this year should be finalized designs. Any work going on now would be about working through the process to get yields up high enough.

Intel does have the finances and the tech to hold up a launch work implement the fix have the Fabs pop up silicon and just triple the workforce on validation. But that would still mean that they at least started on that when they got the word and not recently like they suggested. Which is fine. What the Government is calling everyone out for is the background dealings of all these companies when dealing with a security issue months before most companies and people know. But honestly what would have them do, publish the information of a possible exploit that no one else knows about months before security firms and in this case hardware developers can find solutions. This was why it was such a big issue when Vista was getting updates that XP wasn't when XP's support was dropped. The first vulnerabilities that were patched on Vista post XP support would give hackers inside information on exploits that would work on at the time 60+% of the systems out there. This is the same all of a sudden you would have an exploit affecting 100's of millions of machines in the last decade, with the information on what the exploit is, blasted to the hackers out there, months before there was any security solution.

 

[darkswordsman17]

I hope AMD keeps their relatively low profile on this, as it could turn around and bite them. By that I mean, don't go on a big marketing push to disparage Intel. Focus on your own stuff, making sure its secure and stable and keep working to improve performance. Now, sure make a point to have reviewers test with all the proper patches (so if they want to include Intel results for comparison they have to be fully patched), and if there's areas you can point to (like streaming while gaming) and things that will exacerbate issues with Intel's performance after patching, then do that. But don't run ads going "and our chips don't Meltdown" or some such nonsense.

I won't be surprised to find out that half of Intel's work is trying to find similar exploits of AMD and ARM designs so they can go "see it wasn't just us!" because that seems about how they operate.

I think I read somewhere that it's about as difficult to take finished silicon and go in and tweak it, as it is to make a new chip?

Anyway, Intel has made it's bed. It must now sleep in it.

Not really, that's more just saying that tweaking the silicon would still involve the remasking and other stuff as part of the validation and production part of things, that a new design would also need. But a new design is absolutely more of an undertaking. It has benefits that can make it worth it though. But it comes with risks (as we've seen, every company has had issues, Intel with some of their bugs and then just the general architecture like with Netburst; AMD with bugs like Phenom, and architecture like Bulldozer; even ARM designs aren't immune as seen with the Snapdragon 810).

Simply put, I think Intel is just going to ship chips with updated microcode and they'll claim they "fixed the silicon" even though they really didn't in the sense that us enthusiasts would say, but because the microcode controls how the silicon operates it would be legally defensible. They'll probably rely on clock speeds/binning to push a bit higher so they can claim no loss in performance or something.

They're supposed to be bringing a significant change in their architecture sometime (which perhaps has the fix, although it seems like there's enough people doubting that might be the case, after all they had significant architecture changes before and didn't resolve it then), but I thought that's been delayed or something? Their roadmaps the past couple of years has gotten so stupid and indecipherable (I think intentionally so, they're trying to hide that they've not really been updating their architectures, and because they're not moving to new processes they're not getting gains from tock like in the past; but they have to keep pushing a new generation, adding some platform features here and there and eking out what extra they can offer; I think the last what 2 gens was mostly GPU upgrades and then pushing clock speeds a bit more and adding cores and cache to higher models, hasn't it?), that I'm not even sure Intel knows what its doing, and this I'm sure muddies things substantially.

 

[LTC8K6]

Simply put, I think Intel is just going to ship chips with updated microcode and they'll claim they "fixed the silicon" even though they really didn't in the sense that us enthusiasts would say, but because the microcode controls how the silicon operates it would be legally defensible. They'll probably rely on clock speeds/binning to push a bit higher so they can claim no loss in performance or something.

That isn't going to work at all, imo.

And as someone else said, what about all the chips that are already out there?

Maybe Intel could slide out of End of Life chips, but there are a heck of a lot of chips out there that are not EOL that are affected.

I think most Haswell chips are EOL now, but after that they should still be supported.

 

[Kenmitch]

That isn't going to work at all, imo.

And as someone else said, what about all the chips that are already out there?

Maybe Intel could slide out of End of Life chips, but there are a heck of a lot of chips out there that are not EOL that are affected.

I think most Haswell chips are EOL now, but after that they should still be supported.

Anything purchased after the discovery should be eligible for a refund at Intels expense.

 

[maddie]

...............................................

Simply put, I think Intel is just going to ship chips with updated microcode and they'll claim they "fixed the silicon" even though they really didn't in the sense that us enthusiasts would say, but because the microcode controls how the silicon operates it would be legally defensible. They'll probably rely on clock speeds/binning to push a bit higher so they can claim no loss in performance or something.
..................................................................

This is what I believe also, as a first step to a proper redesign, but we'll have to wait and see.

 

[StinkyPinky]

I see my mobo manufacture has pulled the previous bios microcode update and released a new one. The point is my system is stable even with the pulled update...should i update the bios anyway?

What a mess.

 

[LTC8K6]

Anything purchased after the discovery should be eligible for a refund at Intels expense.

I think the warranty is 3 years.

Interestingly, my i3-4360 is not EOL, but the 4790K, and G3258 are.

My E3-1231 V3 is also not EOL.

 

[LTC8K6]

This is what I believe also, as a first step to a proper redesign, but we'll have to wait and see.

If they do that, on the day it's announced, I'll buy a Ryzen CPU and build a system with it.

I'll also get rid of the NIB i5-8400 sitting on my desk.

 

[StinkyPinky]

The thing about this warranty, is what are the legal requirements here? Would intel need to just supply a new CPU or also pay for the installation of it (since most people wold have no clue)

 

[LTC8K6]

The thing about this warranty, is what are the legal requirements here? Would intel need to just supply a new CPU or also pay for the installation of it (since most people wold have no clue)

Intel isn't going to do anything about old CPUs except supply microcode updates.

For most of those systems, the warranty is with the system mfg, not Intel.

If you buy a Dell and the CPU dies 3 months in, you go to Dell for the warranty.
Same for the hard drive or SSD.

 

[jpiniero]

Anything purchased after the discovery should be eligible for a refund at Intels expense.

I think you'd have a hard time convincing a court to do that, assuming Intel properly fixes the vulnerability. At least for consumers... servers/enterprise is another story.

 

[LTC8K6]

WCCFTECH link.

https://wccftech.com/amd-zen-2-cpus-fix-spectre-exploit/

AMD Zen 2 Architecture Will Address All Potential Spectre Exploits – On-Silicon Changes Made To Strengthen Security

So it looks like both Intel and AMD are claiming to have made silicon changes.

 

[SamMaster]

At first I thought Intel either knew about the vulnerabilities before Google, or that they rushed the fixes and we will have another launch like Coffee Lake, but if both claim the same, then could it be that BOTH knew about the vulnerability beforehand, or that we will see limited launches, or just maybe the fix is not that complex to begin with and can be done with little impact on availability?

 

[KompuKare]

WCCFTECH link.

https://wccftech.com/amd-zen-2-cpus-fix-spectre-exploit/


So it looks like both Intel and AMD are claiming to have made silicon changes.

I think the Intel statement remains a bit vague and undefined, but if what AMD say is true that must meant they at least must have known about this way before July 2017 as Zen2 is just meant to have taped out recently.

At first I thought Intel either knew about the vulnerabilities before Google, or that they rushed the fixes and we will have another launch like Coffee Lake, but if both claim the same, then could it be that BOTH knew about the vulnerability beforehand, or that we will see limited launches, or just maybe the fix is not that complex to begin with and can be done with little impact on availability?

Since AMD just said that Zen2 taped out, the design must have been finished ages ago (probably before the Ryzen launch).
As for both knowing about the vulnerability before Google told them, one explanation is that the Google zero day research was itself driven by some prior research, microarchitectural discussion etc., and both AMD and Intel saw the implications?

 

[William Gaatjes]

I think the Intel statement remains a bit vague and undefined, but if what AMD say is true that must meant they at least must have known about this way before July 2017 as Zen2 is just meant to have taped out recently.

Since AMD just said that Zen2 taped out, the design must have been finished ages ago (probably before the Ryzen launch).
As for both knowing about the vulnerability before Google told them, one explanation is that the Google zero day research was itself driven by some prior research, microarchitectural discussion etc., and both AMD and Intel saw the implications?

I think the whole of exploit flaws are part of a patent or two. A patent that has been licensed to all manufacturers of high performance cpu cores.
Intel/ AMD /Apple / Nvidia / ARM.
These cores that are vulnerable all look so much alike in how they function that it makes me wonder if exchanging the instruction decoder that is different for x86 or arm would still present a working core. Maybe not optimal but still functional.

edit:
Also, cpu designers often hop from one company to another and with them, their patent portfolio.

 

[LTC8K6]

I think the Intel statement remains a bit vague and undefined, but if what AMD say is true that must meant they at least must have known about this way before July 2017 as Zen2 is just meant to have taped out recently.

I can't see how this is "a bit vague" from Intel:

We’re working to incorporate silicon-based changes to future products that will directly address the Spectre and Meltdown threats in hardware. And those products will begin appearing later this year.

 

[moinmoin]

I think the Intel statement remains a bit vague and undefined, but if what AMD say is true that must meant they at least must have known about this way before July 2017 as Zen2 is just meant to have taped out recently.

Since AMD just said that Zen2 taped out, the design must have been finished ages ago (probably before the Ryzen launch).
As for both knowing about the vulnerability before Google told them, one explanation is that the Google zero day research was itself driven by some prior research, microarchitectural discussion etc., and both AMD and Intel saw the implications?

Zen 2 may also just build upon the work on SEV already in Zeppelin. Zeppelin is not affected by Meltdown as privileges are respected. Spectre v2 is also "highly unlikely" as it involves privilege escalation. Spectre v1 doesn't but allows snooping between same privilege processes. SEV (Secure Encrypted Virtualization), as an extension of SME (Secure Memory Encryption), already allows encrypting the memory of separate VMs to isolate them from each other. I'd expect AMD to work on extending that to work on a per process basis. This would make Spectre v1 and v2 fail as raw access would be still possible but decryption would not.

 

[Excessi0n]

I can't see how this is "a bit vague" from Intel:

"Silicon-based changes" is incredibly vague because it could refer to anything from an unchanged die with some functions physically crippled to a complete architectural overhaul. He also did not claim that they had actually fixed the problems; the statement was merely that they were making changes aimed at addressing the exploits.

Given how circumspect his wording was, I'm inclined to believe that any hardware changes in the next couple of years will be relatively minor alterations that mitigate the exploits but do not actually fix them.

 

[coercitiv]

Given how circumspect his wording was, I'm inclined to believe that any hardware changes in the next couple of years will be relatively minor alterations that mitigate the exploits but do not actually fix them.

Indeed, the wording became even more circumspect once Intel's upcoming hardware fixes were revealed to be opt-in and disabled by default, and it does seem like their hardware fix has an a performance cost:

The expectation here, at least on Torvald's part, is that a future chip addressing past flaws should include a flag or version number that tells the kernel it's not vulnerable, so no unneeded and potentially performance-killing mitigations need to be applied. In other words, the chip should indicate to the kernel that its hardware design has been revised to remove the Spectre vulnerability, and thus does not need any software mitigations or workarounds.

Intel's approach is backwards, making the fix opt-in. Processors can, when asked, reveal to the kernel that Spectre countermeasures are present but disabled by default, and these therefore need to be enabled by the operating system.

 

[LTC8K6]

"Silicon-based changes" is incredibly vague because it could refer to anything from an unchanged die with some functions physically crippled to a complete architectural overhaul. He also did not claim that they had actually fixed the problems; the statement was merely that they were making changes aimed at addressing the exploits.

Given how circumspect his wording was, I'm inclined to believe that any hardware changes in the next couple of years will be relatively minor alterations that mitigate the exploits but do not actually fix them.

AMD used the same terms as Intel as far as I can see.

 

[Fanatical Meat]

I’m no expert but couldn’t intels fix be something built into the motherboard chips or some kind of less aggressive speculation on the CPUs ? Assuming it’s this or a combination couldn’t intel just bump the clock speeds up a bit to cover the performance impact?

What about chips in you cable modem or router or anything else on the network like tablets or phones?

 

[Topweasel]

I’m no expert but couldn’t intels fix be something built into the motherboard chips or some kind of less aggressive speculation on the CPUs ? Assuming it’s this or a combination couldn’t intel just bump the clock speeds up a bit to cover the performance impact?

What about chips in you cable modem or router or anything else on the network like tablets or phones?

Bumping clock speeds means bumping power and that opens a whole nother ball of issues. The biggest hit. The most important hit. Is also where they can't simply bump up the clocks. The server and specially VM server solutions are hit nearly 30%. These are the most expensive CPU's Intel sells and they can't magically give them an extra 1 GHz.

 

[IEC]

A certain EHR vendor posted preliminary patch performance testing results on their customer documentation portal this past week. Last update: 1/30 for those of you who may have access...

Per terms, I am not allowed to screenshot or share the document, or even reproduce the text of the document here. However, if this preliminary data is borne out by further testing, this is going to be extremely painful. The previous estimated 30% is nowhere near the worst case using common customer configurations.

 

[moinmoin]

Fefe (a well know German IT blogger) did some simple web server throughput test with his Haswell under Linux 4.15 including Meltdown fixes, Retpoline, without SSD:

./bench -n 1000 -c 10 -k -K 5 http://127.0.0.1/testfile

bench: 164835678432 bytes in 24.7616 seconds.
bench: Throughput: 6.2GiB/sec
bench: Requests per second: 40

bench: 164870699232 bytes in 39.3921 seconds.
bench: Throughput: 3.9GiB/sec
bench: Requests per second: 25

He did a microcode update but that turned out to be from 2017, resulting in 4.5GiB/sec. So in this simple case performance loss appears to be between 27-37%.

 

[DrMrLordX]

Here they come:

http://www.securityweek.com/malware-exploiting-spectre-meltdown-flaws-emerges