Clear Linux Rolls Out KPTI Page Isolation & Retpoline Support

[starshipeleven]

AMD is not so clever either. They provided the wrong documentation to Microsoft, which resulted in bootloops: https://www.ghacks.net/2018/01/09/mi...t-amd-devices/

I'll take that statement with a grain of salt as it's still coming from MS themselves.

Not that I think AMD is innocent, afaik their chipsets are still made/designed by Asmedia so I'm sure they are full of weird quirks not even the designers are fully aware of.

Although I don't know what MS means with "chipsets" in their statement.

[numacross]

AMD is not so clever either. They provided the wrong documentation to Microsoft, which resulted in bootloops: https://www.ghacks.net/2018/01/09/mi...t-amd-devices/

And if Microsoft didn't outsource their QA to the general public they would've caught it earlier in testing

[Vistaus]

Yet those architecture changes to speed things up are now a threat vector and the changes both software, firmware and architecturally negate a lot of those speed ups. Once again....AMD at least for Meltdown and Ryzen made architecture choices that hurt their performance against Intel but they saw the potential threat vector and decided against doing that just to score higher in perf or for marketing purposes. Intel...well....

AMD is not so clever either. They provided the wrong documentation to Microsoft, which resulted in bootloops: https://www.ghacks.net/2018/01/09/mi...t-amd-devices/

[starshipeleven]

x86 is beginning to feel like a legacy shit architecture.

It's been a legacy shit architecture for at least 15 years.

[zamadatix]

What's your definition of a "high performance CPU"?
Old Pentium 4 CPUs are affected. I'd hardly call these high performance. They weren't even high performance when they came out.

I'm not looking at "Only the CPUs with the highest prim95 score at launch" I'm talking about "any modern CPU design that actually has speculative prefetching and similar optimizations". Hell, the Pentium Pro from '95 had speculative prefetching but current RISC-V designs haven't gotten there.

RISC-V does have an implementation. Namely, RISC-V Rocket. That one was not affected by Spectre or Meltdown.

Because the RISC-V Rocket doesn't support speculative prefetching.... as I said, RISC-V isn't done yet and that is the only reason it wasn't affected. As a result of that and other design limitations the Rocket's relative performance is shit compared to other recent cores; ARM cores get 2x-3x the dmips/core/MHz of the higher end BOOM configuration and I don't want to imagine how far behind POWER/x86 that is.

Doesn't matter that there aren't commercial implementations, yet. Clearly, the architecture itself can be considered a fairly modern ISA and was not affected.

Modern as in "new design" not modern as in "high performance design". Besides, just because the vuln was found before they ever made a CPU with speculative prefectching (which is planned) does not mean RISC-V did anything right in their design it means they hadn't got that far. Like I said, the 2024 Mustang has yet to have a single engine failure!

Don't get me wrong RISC-V is neat as hell and on a great track towards being a high performance open CPU but it is by no means that right now and not being affected by the current vuln because they hadn't designed the feature yet is in no way a win for it to to be touting around.

[unixfan2001]

Meanwhile the 2024 mustang hasn't had an engine failure yet!*

*Please disregard that the engine hasn't been designed or built.


But seriously, the only reason RISC-V wasn't affected was they haven't released a high performance CPU yet and only high performance CPUs were affected. It has nothing to do with the ISA being new or open source, simply not complete yet.

What's your definition of a "high performance CPU"?
Old Pentium 4 CPUs are affected. I'd hardly call these high performance. They weren't even high performance when they came out.
Merely an alternative to electric heating devices.

RISC-V does have an implementation. Namely, RISC-V Rocket. That one was not affected by Spectre or Meltdown.
Doesn't matter that there aren't commercial implementations, yet. Clearly, the architecture itself can be considered a fairly modern ISA and was not affected.

[ptyerman]

At this rate it won't be long before my Amiga's are outperforming recent platforms.

[L_A_G]

x86 is beginning to feel like a legacy shit architecture.

Looks like somebody's a bit late to the party... People have been talking about the potential (and actual) exploitability of x86 hardware for years already. Here's a pretty good talk on the subject from the 2016 Chaos Computer Congress.

But seriously, when you've extended an architecture as far as x86, an architecture originally designed in the mid 1970s, has been extended you're more or less guaranteed to see all kinds of "interesting" side effects of this.

[DanL]

The level of ignorance and FUD being spread right now is inexcusable.

Oh, good. I was wondering what the resident Intel shill thought of the matter.

[numacross]

It's hilarious that you would rant about those that are "ignorant" and spread FUD yet make the above statement. I'll let you in on a little secret, but don't tell anyone, ARM DOES NOT produce it's own cpu's, the license the ISA to anyone that wants to make a cpu based on it. There is no such thing as a cpu sold by ARM.

That is not strictly true. There are multiple types of ARM licenses, one can buy a design for an entire core (Cortex-A53 for example, which is a complete processing core) or have an architectural license that allows them to design their own cores (like Apple, Samsung or Qualcomm). Compute cores, GPUs and interconnects are among the primary products of ARM. They are sold as complete parts to make a SoC by the licensee. Though we might never know for sure but I'd wager that if you buy a certain core from ARM you're not permitted to modify beyond what's needed in order to integrate them into your product.

The fact that ARM themselves are not physically making the silicon doesn't change much here. It's still their design that's largely immutable. Some of them are vulnerable to at least Spectre and some are not (because they lack the speculative execution among other things).

I also think the timeline of all this is pretty telling, first it's revealed that all Intel cpu's run a secret low level OS, based on Minix, as part of it's ironically named "Trusted" Execution Engine and Management Engine, then we find out that Google's engineers are looking at ways to remove that crap from the Intel cpu's they are using and then it's revealed that the design of the cpu's at the hardware level is susceptible to attacks.

It's been known since the C2D era what Intel ME/AMT are. Not only there's documentation from Intel about its capabilities but also when you look at what AMT can actually do it's pretty obvious how powerful the ME must be in comparison to the host CPU.

And it hasn't been Minix always - it's only since Skylake that it's running on a 3-core Quark-like x86. Previous versions were using derivatives of the ARC architecture with a Java-based software stack.

Clearly this is done intentionally and I think we, as consumers, need to send a strong message to Intel, ARM, AMD and NVIDIA by boycotting the next generation of their products, not buying any of the new refreshes/releases that are coming out in 2018, let them feel the sting in their bottom line for a year or two and then maybe they'll get their act together.

But if anyone buys a new Intel, AMD or NVIDIA product after what we know now, then they have only themselves to blame if this shit keeps happening.

Of course it's intentional. There is a market for having your fleet of computers remotely manageable

The "flaws" in that product and their origins are another matter though...