> As far as I know, it is an old (and documented) flaw of gcc: it needs 
   > at
   > least -O in order to get inline assembly correctly (the binding of 
   > local
   > variables to register is not handled properly by the dummy register
   > allocator used when compiling without optimisations). And the kernel
   > uses inline assembly to talk to the hardware.
   >
   I guess this actually enough of a problem to warrant a pr - and I just 
   did a quick search and saw toolchain/18270.  From what I can see, COPTS 
   is reset even if "+=" is used.  I'll see if I can just append my 
   observations to the PR.  At the very least, it's a documentation bug in 
   mk.conf(5)...


this is exactly why CPUFLAGS was invented...