Re: Capabilities, Anybody?

On 2024-03-08 9:25 p.m., Lawrence D'Oliveiro wrote:

On Fri, 8 Mar 2024 21:15:29 -0500, Robert Finch wrote:
 

I gather that capabilities are generally fine-grained, and capability
pointers would be generated and handed out by the OS. What happens when
a pointer is incremented?

 Each capability is a descriptor, describing a range of memory, not
(necessarily) just one address. So it is valid to use that to address any
area within the range.

Yeah, I got that after a little more reading. It looks a bit like a segmentation system. It adds quite a bit to the instruction set, doubling up on the load and store instructions to use capabilities and to use ordinary pointers. It also needs wider registers. It is something else to experiment with. Tags can be incorporated into the current project for instructions relatively easily. But adding the extra tag bit for data is trickier. It is desirable not to need to read two different memory locations on every access. I suppose "block" capabilities could be used instead of fine-grained. In a sense with the compressed format for the bounds things are blocked.