In the last 10 years there has been a seemingly noteworthy uptick in hardware bugs in both intel and amd CPUs. Security researchers find and figure out potential attack vectors that rely on these bugs (ex. Specter/Meltdown). Then operating systems have to put workarounds in their kernel code to ensure that these hypothetical attack vectors are accounted for, at the cost of performance and more complicated code.
Linus is saying how annoyed he is with all this extra work they have to do, resulting in worse performance, all to plug vulnerabilities that we’ve never actually seen any real attackers use. He’s saying instead we should just write the code how it should be, and if the hardware is insecure, let it be the hardware company’s problem when customers don’t use the hardware.
The problem is, customers will continue to use the hardware and companies who need a secure OS (all of them) will opt to not use Linux if it doesn’t plug these holes.
Plus a lot of these bugs don’t get fixed, because they exist to allow the processors to “look ahead” for improved performance, at least on unmitigated benchmark tests.
Notably, that’s not what he says. He didn’t say in general. He said “for once, [after this already long discussion], let’s push back here”. (Literally “this time we push back”)
who need a secure OS (all of them) will opt to not use Linux if it doesn’t plug these holes
I’m not so sure about that. He’s making a fair assessment. These are very intricate attack vectors. Security assessment is risk assessment either way. Whether you’re weighing a significant performance loss against low risk potentially high impact attack vectors or assess the risk directly doesn’t make that much of a difference.
These are so intricate and unlikely to occur, with other firmware patches in line, or alternative hardware, that there’s alternative options and acceptable risk.
In the last 10 years there has been a seemingly noteworthy uptick in hardware bugs in both intel and amd CPUs. Security researchers find and figure out potential attack vectors that rely on these bugs (ex. Specter/Meltdown). Then operating systems have to put workarounds in their kernel code to ensure that these hypothetical attack vectors are accounted for, at the cost of performance and more complicated code.
Linus is saying how annoyed he is with all this extra work they have to do, resulting in worse performance, all to plug vulnerabilities that we’ve never actually seen any real attackers use. He’s saying instead we should just write the code how it should be, and if the hardware is insecure, let it be the hardware company’s problem when customers don’t use the hardware.
The problem is, customers will continue to use the hardware and companies who need a secure OS (all of them) will opt to not use Linux if it doesn’t plug these holes.
Plus a lot of these bugs don’t get fixed, because they exist to allow the processors to “look ahead” for improved performance, at least on unmitigated benchmark tests.
Notably, that’s not what he says. He didn’t say in general. He said “for once, [after this already long discussion], let’s push back here”. (Literally “this time we push back”)
I’m not so sure about that. He’s making a fair assessment. These are very intricate attack vectors. Security assessment is risk assessment either way. Whether you’re weighing a significant performance loss against low risk potentially high impact attack vectors or assess the risk directly doesn’t make that much of a difference.
These are so intricate and unlikely to occur, with other firmware patches in line, or alternative hardware, that there’s alternative options and acceptable risk.