3 ms·
This is a good primer for understanding the Spectre[0] issue. In a nutshell, Spectre is caused by the CPU keeping the branches in an unencrypted cache that is
by jedberg 7y ago
This is a good primer for understanding the Spectre[0] issue. In a nutshell, Spectre is caused by the CPU keeping the branches in an unencrypted cache that is accessible from programs other than the one that generated the branch.
Their fix was to add an extra kernel check for unauthorized access, which is why it causes a huge performance hit. The real fix will be in the next generation of chips, which will either encrypt or at the least protect the branch cache from unauthorized access from other programs.
[0] https://en.wikipedia.org/wiki/Spectre_(security_vulnerability) https://en.wikipedia.org/wiki/Spectre_(security_vulnerabilit...
- saagarjha 7y ago> In a nutshell, Spectre is caused by the CPU keeping the branches in an unencrypted cache that is accessible from programs other than the one that generated the branch. I would instead say that the branch prediction has observable side effects even when it’s thrown away; say, for example, the branch performs a memory access: it may leave things in the cache even if the processor doesn’t take the branch and hence it may be possible to recover information by running a timing attack on how long it takes to access certain things.