Open asteinha opened 4 years ago
Not to pile too many hopes on this issue, but as we consider moving to a new mechanism and interface for fault-recovery it could be nice to keep an eye on supporting fault suppression using TSX as well, maybe through one of two swappable implementations.
Not to pile too many hopes on this issue, but as we consider moving to a new mechanism and interface for fault-recovery it could be nice to keep an eye on supporting fault suppression using TSX as well, maybe through one of two swappable implementations.
That's true, however I don't see it as a priority, because it is not a portable solution - does not work on AMD and on older Intel CPUs - and it is not neccessary in any of our current examples. I'll just check the Meltdown-OF whether it would not work better with Intel TSX on the newest Intel CPUs. That one performs much better on AMD.
In Meltdown examples we currently move the instruction pointer in a signal handler. If there were some register spills, the correctness of the C++ execution model can be violated. We should look for a solution that maintains current functionality of our Meltdown examples while always preserving correctness.