Change representation of the SeqSet #568
Conversation
This changes the representation of SeqSet to be doubly linked. This is required to enable tracking fully used slabs. * Currently debug_check_empty can say empty when it isn't as a slab is fully used. * Implementing a heap walk we will also need to track fully used slabs. Future PR will use this to track fully used slabs.
There was a problem hiding this comment.
LGTM but I'm sorry you had to go through all that. I suspect moving back to an AuthMap design will simplify some of this and the unfortunate interaction with the StrictProvenance arenas. I did not foresee this (and other fun) when #402 was being developed.
| NextPtr* end{&head}; | ||
| }; | ||
| // We could use -static_cast<ptrdiff_t>(offsetof(T, node)) here but CHERI | ||
| // compiler complains. So we restrict to first entries only. |
There was a problem hiding this comment.
It's not really the CHERI compiler's fault per se, it's just that StrictProvenanceSlabMetadataMixin defines members atop the FrontendSlabMetadata parameter and so makes the Pagemap's SlabMetadata parameter not a Standard Layout type, and so offsetof is not defined.
I'm kind of surprised it works in the static_assert below.... oh, because T here is the superclass and the caller's done the non-Standard-Layout shift for us. It looks like we never need to perform the downcast?
There was a problem hiding this comment.
I think T is StrictProvenanceSlabMetadata for CHERI here. That class ties the parameter back in using F-Bounded polymorphism.
This changes the representation of SeqSet to be doubly linked. This is required to enable tracking fully used slabs.
Future PR will use this to track fully used slabs.