Fix freebsdNN compat for [u]intptr_t syscall arguments

[bsdjhb]

• freebsd64_fcntl: Remove unnecessary cast
• freebsd32/64: Map [u]intptr_t syscall arguments to [u]intXX_t
• sysent: Regen

[brooksdavis]

I'm now feeling uncomfortable about fcntl being intptr_t on 32-bit architectures with >31-bit userspace addresses. I think that may be making kernel addresses.

[bsdjhb]

I'm now feeling uncomfortable about fcntl being intptr_t on 32-bit architectures with >31-bit userspace addresses. I think that may be making kernel addresses.

Yes, I wish it was a uintptr_t arg instead (I think)

[bsdjhb]

I'm now feeling uncomfortable about fcntl being intptr_t on 32-bit architectures with >31-bit userspace addresses. I think that may be making kernel addresses.

Yes, I wish it was a uintptr_t arg instead (I think)

freebsd32_fcntl() goes out of its way to use an unsigned cast to avoid sign extension for cases where it is either a pointer or flags field.

[bsdjhb]

I've changed this to add synthetic [u]intptrXX_t types that are used in place of [u]intptr_t. This avoids the false positive matches from isptrtype() in makesyscalls.lua and trims much of the noise from the diff.

[jrtc27]

I don't see what benefit the extra type gives over just saying int32_t?

[bsdjhb]

It avoids all the extra gunk due to isptrtype() matching on non-pointer instances. I didn't see an easy way to fix it in makesyscalls.lua.

[jrtc27]

It's a bit weird to have this on ILP32, and a bit weird for these to exist outside of compat code

[bsdjhb]

I originally wanted to put these in <compat/freebsdNN/freebsdNN.h>, but many places include freebsdNN_proto.h without freebsdNN.h and so failed to compile. One of those places is the generated freebsdNN_sysent.c, so I tried to find a header that was already included by various places. Given that it is kernel only I could guard the typedefs with #ifdef COMPAT_FREEBSDnn.