Add a signed version of ptraddr_t

[jrtc27]

This is sometimes needed as a useful way to express "signed machine word" (which ssize_t and ptrdiff_t are not required to be, even if in practice they are, nor does long suffice due to Windows's LLP64 which presumably would be LL64PC128 for CHERI) or as a way to deal with sentinel values like -1 similar to ssize_t, and also to be able to perform signed comparisons against addresses to distinguish between kernel and user addresses. The general consensus seems to be that sptraddr_t is the best name for it. See #8 for earlier discussions.

[arichardson]

Doesn't sptraddr_t suffer from the same theoretical issues as ssize_t/ptrdiff_t if you want it to be a "signed machine word"? In practise it will almost always be the same as the "signed machine word", but for something such as MIPSn32/x32/etc. a machine word is 64 bit whereas ptraddr_t would be 32 bits.

I'm not opposed to adding this type since the signed kernel address comparison is a valid use case (while you could also do unsigned comparisons that may require changing more code). Maybe we also need a "signed machine word" type?

[jrtc27]

Doesn't sptraddr_t suffer from the same theoretical issues as ssize_t/ptrdiff_t if you want it to be a "signed machine word"? In practise it will almost always be the same as the "signed machine word", but for something such as MIPSn32/x32/etc. a machine word is 64 bit whereas ptraddr_t would be 32 bits.

I'm not opposed to adding this type since the signed kernel address comparison is a valid use case, but maybe we also need a "signed machine word" type?

Well, "signed ABI word" then. The problem with ssize_t/ptrdiff_t is not n32/x32/ilp32 but that they're permitted to be only as big as the largest allocation you support, either because you set an arbitrary limit or because you have a segmented architecture (e.g. 32-bit segment number and 32-bit segment offset would be a valid pointer representation and give a 64-bit ptraddr_t/uintptr_t (without CHERI) but permit a 32-bit size_t/ssize_t/ptrdiff_t). In practice that's not true and size_t is always an "unsigned ABI word", but ssize_t isn't as helpful as the only negative number it's required to support is -1.