You answered your own question. Linux uses an ELF 
file format. When the ELF is loaded, the variables loaded
into the DATA segment are marked as read-only pages.
This, of course, will cause a write to such pages to 
manifest a SIGSEGV. 

Plan 9, however, makes no effort to do this. Objects
marked into the DATA segment of the Plan 9 file format
are loaded into pages that are actually RWX. So, it's really
a two-pronged issue. 
	1) The kernel loader doesn't handle the management
	    of page protection in the DATA, Stack, BSS, or even
	    the TEXT. 
	2) The current binary format doesn't allow for specification
	    of whether or not the DATA segment should *be* ~W.

There is one main reason why you would want a W DATA. 
Variables that end up initialized to a value other than nil must 
be put in the DATA segment. Otherwise, they end up in the 
BSS, which is initialized to nil.

Thus, global pointers to objects in the DATA (not const?) must
be R|W, while the object itself (raw data) must be ~W. This is
the ideal solution. 

This is being implemented in my work on the SPARC/SPARCv9
and AMD64 ports. I've discussed this with other members of
the AMD64 port, and there've been no nay-sayers. There will
be other improvements, as well. However, they are to be seen
later. 

FYI, I'm writing the toolchain(s) from scratch, so the new binary
format for these archs will look only slightly different from the
current. 

Don (north_)