make buildworld fails on 4.8-release

Chuck Swiger cswiger at
Sun May 4 10:42:54 PDT 2003

Jeremy Gaddis wrote:
>>> CPUTYPE=i686
>>> NOPROFILE=true
[ ... ]
> Still don't know why those two lines would cause it
> to fail, especially since I took them straight out of
> some {documentation|HOWTO|handbook|something}.

Sometimes the compiler has problems with higher optimization levels, 
some of which are implied by the CPU-specific scheduling and such the 
"-mcpu=i686" ("-march=i686"?) your CPUTYPE setting produces.

If the compilation terminates, it's either a bug with gcc itself, or 
possibly an interaction with the source code explicitly trying to do 
something else.  For a possible example, OpenSSL's hand-tuned assembly 
and specific -m flags it wants instead.

If the system compiles but then panics or simply encounters very odd, 
intermittant bugs, because the kernel was optimized too highly, that's 
harder to diagnose.  Perhaps some critical reference to a volatile 
memory location got optimized away because it didn't look like it was 
doing anything, or the compiler failed to understand pointer aliasing 
when the kernel is adjusting the VM page tables, or god knows what, 
well, it'd be really nice to identify and solve those bugs, but it's 
also hard to reproduce them and track down the causes.  Please do try, 
however.  :-)

"Warranty: If this software breaks, you get to keep both pieces."


More information about the freebsd-questions mailing list