On 28/03/19 17:37, Jim Mattson wrote: >> We store it in SMRAM and not do this part: >> >> IF VMXE = 1 in CR4 image in SMRAM >> THEN fail and enter shutdown state; >> >> This is only for historical reasons; we got it wrong when first adding >> support for nested+SMM, it remained like that ever since, and it would >> be hard to change it without potentially breaking KVM_GET/SET_NESTED_STATE. > It seems like there are inherent flaws in a system design that > discourages bug fixes. The very fact that you might be asked to load a VM that you've snapshotted 10 years ago discourages bug fixes. In this case, it's broken enough that we can ignore compatibility, as Sean pointed out; for other cases there is the quirks mechanism. Either way this still requires someone to find the time to: - peruse the SDM until they notice the discrepancy - fix the bug - write a unit test, if not a unit test framework for the bug - test the fix on real-world code Paolo
