Another nail in the coffin of x86?
These researchers shouldn't have had to reverse engineer this stuff; it only obscured the security problem. As with software, the more open the design, the better for security. This isn't rocket science.
Chip-busting boffins in America have devised yet another way to filch sensitive data by exploiting Intel's processor design choices. Doctoral student Riccardo Paccagnella, master's student Licheng Luo, and assistant professor Christopher Fletcher, all from the University of Illinois at Urbana-Champaign, delved into the way CPU …
They didn't test ARM. They didn't even test AMD, so we have no idea if this attack is cross platform. What they did, essentially, is what all attackers have done so far: attack the target with the largest footprint, e.g. Windows vs MacOS. It's not that MacOS / AMD is not 100% bulletproof, it was simply not "good" enough for them to target-test.
The future may hold different results, if they bother that is.
This is not a CPU core problem. It is a problem with the way Intel CPU cores share access to memory. You could create the same problem with ARM cores by connecting them to memory with a ring network like Intel uses in older generation chips. You could eliminate the flaw (or at least make it harder to exploit) by connecting cores to memory with a mesh network.
One of the amazing things about x86 is that an army of robots with nail guns have so far been insufficient to keep its coffin lid shut.
I didn't make my point well.
These cores from Intel, (and I assume AMD) are coming with a lot of extra stuff that's both poorly documented and proving to be a security headache. It seems to this layman that the ARM ecosystem is inherently more open because of the way things are licensed, thus allowing a lot of early scrutiny by independent people.
As with the previous post referring to "not good enough", it's worth mentioning that it appears a lot of reverse engineering at a really low level took place. They picked Intel, they showed the vulnerability. Point made.
Now it's up to them later, or somebody else, to figure out how to make this work on AMD, ARM, etc.
This post has been deleted by its author
"The exploits rely on you being able to run your own code on the machine. Xboxes & PS5s don't tend to allow this"
Xboxes and PS5's use AMD chips, whether these suffer the same we don't know. And, Xbox One X and Series X has a Developer Mode, so you could potentially run your own code using that. I'm not an expert on it though, just a possibility,
Correct about the second generation XBox using PowerPC, but I was talking about the original XBox which used a modified Pentium III processor:
GPU back then was green team rather than red team.
If you read the article it tells you that the researchers exploited a feature of Intel's ring interconnect. That is Intel technology. AMD x86 and other architectures do not use Intel ring interconnect (also said in the article) so are not vulnerable to this exact attack. However if one reverse engineered their interconnect technology then perhaps AMD x86 and other architectures like ARM would be vulnerable to a similar attack.
Reading comprehension FTW.
"However if one reverse engineered their interconnect technology then perhaps AMD x86..."
So far as this old brain can recall (corrections very welcome!), AMD published HyperTransport, and I think its successor, as open source. They did this to foster / enable novel co-processors based on FPGAs. So, perhaps no reverse engineering necessary.
My (barely educated) guess is that this kind of weakness might exist in any SMP-faked-on-top-of-NUMA architecture like modern x64 processors. In fact, I'm wondering if it would also be the case for true SMP designs like Intel's old fashioned Netburst (Pentium 4s et al) architecture. There you really would have delays to memory accesses caused by other code - there was just one memory bus...
I'm wondering if the real fix for this is no SMP whatsoever...
"I'm wondering if the real fix for this is no SMP whatsoever..."
Nah. The real fix for this is to run your own code on your own hardware. Despite decades of marketing hype, and some pretty serious efforts at encrypted computation, it remains true that if you lose physical security, you lose security.
Given the collapsing costs of actual hardware and the availability of free-as-in-beer operating systems and hypervisors, the movement to renting space in someone else's machine, alongside god-knows-who and at the far end of a long wire, is surely one of the more unexpected trends of recent years.
Biting the hand that feeds IT © 1998–2021