cvs commit: src/sys/vm vm_map.c

[Bruce Evans]

On Tue, 29 Jun 2004, Andrew Gallatin wrote:

> Bruce Evans writes:
>  > On Mon, 28 Jun 2004, Andrew Gallatin wrote:
>  > >  > > >   Log:
>  > >  > > >   Fix alpha - the use of min() on longs was loosing the high bits and
>  > >  > > >   returning wrong answers, leading to strange values vm2->vm_{s,t,d}size.
>  >
>  > MIN() and MAX() should not be used in the kernel.  4.4BSD removed them in
>  > the kernel, but FreeBSD broke this in rev.1.141 of sys/param.h.  They
>  > remain removed in RELENG_4.
>
> OK.  Then what's the correct fix? ulmin()?

Fixing min() to handle all unsigned types is probably best.  Some not-quite
correct fixes:

1. Just use ulmin().  It would work on all supported machines because
   vm_offset_t is smaller than u_long (actually the same).  This is fragile.
2. (1) plus a compile-time assertion that
   sizeof(vm_offset_t) <= sizeof(u_long).  This is quite practical.  There
   are only a few min()'s in all of vm, and only the 2 that you fixed seem
   to need special handling (the others are mainly for page counts, and
   2^32 pages should be enough for anyone).
3. (1) plus a runtime assertion...
4. (3) with calls to uqmin() instead of failing assertions for larger
   vm_offset_t's.
5. (4) plus support for uintmax_t's.
6. Support for uintmax_t's, and just use uimax_min().
7. (6), plus wait for gcc to optimize uimax_min().  gcc currently always
   widens the args before comparing them, even with -O3.

Bruce