There is a serious security flaw in billions of Intel CPUs that can let attackers steal confidential data like passwords and encryption keys. Firmware updates can fix it, but at a potential significant performance loss.
I just skimmed through the article and it seems like this vulnerability is only really meaningful on multi-user systems. It allows one user to access memory dedicated to other users, letting them read stuff they shouldn’t. I would expect that most consumer gaming computers are single-user machines, or only have user accounts for trusted family members and whatnot, so if this mitigation causes too much of a performance hit I expect it won’t be a big risk to turn it off for those particular computers.
It would indeed imply that which is why this vulnerability is also serious for single user contexts.
The vulnerability is caused by memory optimization features in Intel processors that unintentionally reveal internal hardware registers to software. This allows untrusted software to access data stored by other programs, which should not normally be accessible.
All these kind of CPU level vulnerabilities are the same, they are only really “risky” if there is malicious software running in the computer in the first place.
The real problem is that these CPU-level vulnerabilities all break one of the core concepts of computers, which is process separation and virtual memory. If process separation is broken then all other levels of security become pointless.
While for desktops this isn’t a huge problem (except when sometimes vulnerabilities might even be able to be exploited though browsers), this is a huge problem for servers, where the modern cloud usually has multiple users in virtual machines in a single server and a malicious user could steal information across virtual machines.
Your first paragraph isn’t quite right.
Modern hacks/cracks aren’t a “do this and suddenly you are in” type deal.
It’s a cascade chain of failures of non-malicious software.
Saying “don’t have a virus” is absolutely correct, however that’s not the concern here.
The concern is about the broadening of the attack surface.
A hacker gets minor access to a system. Leverages some CVE to get a bit more access, and keeps poking around and trying CVEs (known or unknown) until they get enough access to run this CVE.
And then they can escape the VM onto the host or other VMs on the same system, which might then give them access to a VM on another host, and they can escape that VM to get access to another VM, and on and on.
Very quickly, there is a fleet of VMs that are compromised. And the only sign of someone poking around is on the first VM the hacker broke into.
All other VMs would be accessed using trusted credentials.
ETA:
Infact, it doesn’t even need to be a hacker.
It could be someone uploading a CI/CD task using their own account. It extracts all API keys, usernames and passwords it can find.
Suddenly, you have access to a whole bunch of repositories and APIs.
Then you can sneak in some malicious code to the git repo, and suddenly your malicious code is being shipped within legit software that gets properly signed and everything.
Processes that run on the same system can run as different users (including kernel) which is used for privilege separation. This can still allow a program in userland to peer into otherwise restricted system processes or the kernel. Every system is a “multi-user” system, even if there is only a single human user.
Yes, but all the data that I care about is in my single human user’s account already. If I install malicious software then I’m already hooped regardless.
Look, I’m not saying this is no biggie. There are plenty of systems out there that will have to install this patch. Single-user computers probably should too. The situation I’m addressing is the case where a gaming computer has its performance as a gaming measurably harmed by the patch’s overhead, which is reportedly significant in some cases. In those cases it’s reasonable to weigh the merits and decide that this vulnerability isn’t all that big a problem.
this vulnerability is only really meaningful on multi-user systems
Well, that says it all. CPU manufacturers have no incentive at all to secure the computations of multiple users on a single CPU (or cores on the same die)… why would they? They make more cash if everyone has to buy their own complete unit, and they can outsource security issues to ‘the network’ or ‘the cloud’…
Years ago when I was in University this would have been a deathblow to the entire product line, as multi-user systems were the norm. Students logged into the same machines to do their assignments, employees logged into the same company servers for daily tasks.
I guess that isn’t such a thing any more. But wow, what a sh*tshow modern CPU architecture has become, if concern for performance has completely overridden proper process isolation and security. We can’t even trust that a few different users on the same machine can be separated properly due to the design of the CPU itself?
Fair enough, probably was hyperbole :) But performance does seem to be a higher priority than security; they can always spin PR after the next exploit, after all, users already have the CPU in their system, they’ve made their money; what are users really gonna do if an issue comes up after they’ve bought their box?
Are you aware that the majority of cpus sold today go to cloud computing? Believe it or not, but that is an application space with multiple users on the same machine.
Even on a single user machine, multiple users are very much a thing. Even Apple has left behind the DOS-like architecture where everything runs with the same rights. All current systems run with multiple concurrent users, notably root (or the Windows equivalent) and the keyboard operator (as well as dedicated ones for the various services, although that’s maybe more a thing in Unix/Linux than Windows).
Good point. But I think performance is still a greater priority for those who make purchasing decisions, rather than basic security, and that’s the problem.
Not at the enterprise level.
Security means compliance, which means getting/keeping contracts and not getting sued.
And they care more about performance-per-watt and density.
Processor manufacturers target their devices and sales towards cloud computing so they have a huge incentive to avoid having issues like these. It’s ridiculous to suggest otherwise.
You’re reading an awful lot into what I said that wasn’t put in there.
There’s nothing wrong with multi-user systems existing, there’s plenty of use for such things. This bug is really bad for those sorts of things. I was explicitly and specifically talking about consumer gaming computers, which are generally single-user machines. Concern for performance is a very real and normal thing on a gaming computer, it’s not some kind of weird plot. An actual multi-user system would obviously need to be patched.
I just skimmed through the article and it seems like this vulnerability is only really meaningful on multi-user systems. It allows one user to access memory dedicated to other users, letting them read stuff they shouldn’t. I would expect that most consumer gaming computers are single-user machines, or only have user accounts for trusted family members and whatnot, so if this mitigation causes too much of a performance hit I expect it won’t be a big risk to turn it off for those particular computers.
Would it mean that a malicious application being run in non-admin mode by one user could see data/memory in use by an admin user?
It would indeed imply that which is why this vulnerability is also serious for single user contexts.
deleted by creator
This
All these kind of CPU level vulnerabilities are the same, they are only really “risky” if there is malicious software running in the computer in the first place.
The real problem is that these CPU-level vulnerabilities all break one of the core concepts of computers, which is process separation and virtual memory. If process separation is broken then all other levels of security become pointless.
While for desktops this isn’t a huge problem (except when sometimes vulnerabilities might even be able to be exploited though browsers), this is a huge problem for servers, where the modern cloud usually has multiple users in virtual machines in a single server and a malicious user could steal information across virtual machines.
Your first paragraph isn’t quite right.
Modern hacks/cracks aren’t a “do this and suddenly you are in” type deal.
It’s a cascade chain of failures of non-malicious software.
Saying “don’t have a virus” is absolutely correct, however that’s not the concern here.
The concern is about the broadening of the attack surface.
A hacker gets minor access to a system. Leverages some CVE to get a bit more access, and keeps poking around and trying CVEs (known or unknown) until they get enough access to run this CVE.
And then they can escape the VM onto the host or other VMs on the same system, which might then give them access to a VM on another host, and they can escape that VM to get access to another VM, and on and on.
Very quickly, there is a fleet of VMs that are compromised. And the only sign of someone poking around is on the first VM the hacker broke into.
All other VMs would be accessed using trusted credentials.
ETA:
Infact, it doesn’t even need to be a hacker.
It could be someone uploading a CI/CD task using their own account. It extracts all API keys, usernames and passwords it can find.
Suddenly, you have access to a whole bunch of repositories and APIs.
Then you can sneak in some malicious code to the git repo, and suddenly your malicious code is being shipped within legit software that gets properly signed and everything.
deleted by creator
It allows memory access across virtual machines as well, meaning the all cloud VMs are vulnerable.
The machines that are running cloud VMs should obviously be patched. I wasn’t talking about those.
Processes that run on the same system can run as different users (including kernel) which is used for privilege separation. This can still allow a program in userland to peer into otherwise restricted system processes or the kernel. Every system is a “multi-user” system, even if there is only a single human user.
Yes, but all the data that I care about is in my single human user’s account already. If I install malicious software then I’m already hooped regardless.
Look, I’m not saying this is no biggie. There are plenty of systems out there that will have to install this patch. Single-user computers probably should too. The situation I’m addressing is the case where a gaming computer has its performance as a gaming measurably harmed by the patch’s overhead, which is reportedly significant in some cases. In those cases it’s reasonable to weigh the merits and decide that this vulnerability isn’t all that big a problem.
Well, that says it all. CPU manufacturers have no incentive at all to secure the computations of multiple users on a single CPU (or cores on the same die)… why would they? They make more cash if everyone has to buy their own complete unit, and they can outsource security issues to ‘the network’ or ‘the cloud’…
Years ago when I was in University this would have been a deathblow to the entire product line, as multi-user systems were the norm. Students logged into the same machines to do their assignments, employees logged into the same company servers for daily tasks.
I guess that isn’t such a thing any more. But wow, what a sh*tshow modern CPU architecture has become, if concern for performance has completely overridden proper process isolation and security. We can’t even trust that a few different users on the same machine can be separated properly due to the design of the CPU itself?
I’m not happy with what’s happening and I know that corporations are money making evil machines.
But to say that chip makers have no incentive at all to secure their hardware is quite the hyperbole.
Fair enough, probably was hyperbole :) But performance does seem to be a higher priority than security; they can always spin PR after the next exploit, after all, users already have the CPU in their system, they’ve made their money; what are users really gonna do if an issue comes up after they’ve bought their box?
Are you aware that the majority of cpus sold today go to cloud computing? Believe it or not, but that is an application space with multiple users on the same machine.
Even on a single user machine, multiple users are very much a thing. Even Apple has left behind the DOS-like architecture where everything runs with the same rights. All current systems run with multiple concurrent users, notably root (or the Windows equivalent) and the keyboard operator (as well as dedicated ones for the various services, although that’s maybe more a thing in Unix/Linux than Windows).
Good point. But I think performance is still a greater priority for those who make purchasing decisions, rather than basic security, and that’s the problem.
Not at the enterprise level.
Security means compliance, which means getting/keeping contracts and not getting sued.
And they care more about performance-per-watt and density.
Processor manufacturers target their devices and sales towards cloud computing so they have a huge incentive to avoid having issues like these. It’s ridiculous to suggest otherwise.
I see the reasoning, fair enough. Just grumpy this evening I suppose :p.
You’re reading an awful lot into what I said that wasn’t put in there.
There’s nothing wrong with multi-user systems existing, there’s plenty of use for such things. This bug is really bad for those sorts of things. I was explicitly and specifically talking about consumer gaming computers, which are generally single-user machines. Concern for performance is a very real and normal thing on a gaming computer, it’s not some kind of weird plot. An actual multi-user system would obviously need to be patched.