Extend the CPU ID types in ULE, struct thread and struct kproc_user to be an int instead of a char.

[kib]

Summary

The CPU ID fields in ULE, kproc_user and thread are all char's, which
limit the number of CPUs to 254.

So:

• ule: char -> int;
• thread: char -> int;
• kproc: add new int fields; rename older fields to _old;
• redefine NOCPU from 255 to -1
• freebsd32: add new int fields; rename older fields to _old
• freebsd32: add new int fields; rename older fields to _old.

[kib]

sys/compat/freebsd32/freebsd32.h

347

This breaks ABI. Why did you put new fields into arbitrary place inside the structure ? It must consume the same spare ints as the struct kinfo_proc on 32bit arches.

sys/kern/kern_proc.c

991

This is weird. If CPU index is greater than 255, it is taken mod 256 for old fields.

I see two reasonable approaches to fill the ki_oncpu_old and ki_lastcpu_old. Either set old fields to 255 when CPU index is greater than 255, which means 'no cpu'.
Or (which, IMO, is much better) dedicate some currently unreacheable index, e.g.
254, to all CPUs >= 254. There will be a mis-reportning and bias for huge-cpu machines, but old binaries will be working somehow.

1013

You missed update to freebsd32_kinfo_proc_out().

sys/sys/proc.h

604–605

This must be
#define NOCPU (-1)

[adrian]

sys/compat/freebsd32/freebsd32.h

347

How is it not the same as kinfo_proc?

sys/kern/kern_proc.c

1013

Fixed in my next patch.

sys/sys/proc.h

604–605

Ok, fixed in the next patch.

[adrian]

Update patch with kib@'s comments.

[kib]

Two small issues aside, I am fine with the change. Did you tested e.g. top or ps 32bit binaries on amd64 kernel ?

sys/kern/kern_proc.c

994

The 254 magic value ought its own define, IMO.

1187

This comment is not needed. The handling of NOCPU and wrapping in the native code should be enough.

[emaste]

What's next here?

[kib]

r273266