mmap的优缺点:

引用：http://lkml.indiana.edu/hypermail/linux/kernel/0004.0/0728.html

People love mmap() and other ways to play with the page tables to 
optimize away a copy operation, and sometimes it is worth it. 

HOWEVER, playing games with the virtual memory mapping is very expensive 
in itself. It has a number of quite real disadvantages that people tend 
to ignore because memory copying is seen as something very slow, and 
sometimes optimizing that copy away is seen as an obvious improvment. 

Downsides to mmap: 
 - quite noticeable setup and teardown costs. And I mean _noticeable_. 
   It's things like following the page tables to unmap everything 
   cleanly. It's the book-keeping for maintaining a list of all the 
   mappings. It's The TLB flush needed after unmapping stuff. 
 - page faulting is expensive. That's how the mapping gets populated, 
   and it's quite slow. 

Upsides of mmap: 
 - if the data gets re-used over and over again (within a single map 
   operation), or if you can avoid a lot of other logic by just mapping 
   something in, mmap() is just the greatest thing since sliced bread. 

   This may be a file that you go over many times (the binary image of 
   an executable is the obvious case here - the code jumps all around 
   the place), or a setup where it's just so convenient to map the whole 
   thing in without regard of the actual usage patterns that mmap() just 
   wins. You may have random access patterns, and use mmap() as a way 
   of keeping track of what data you actually needed. 

 - if the data is large, mmap() is a great way to let the system know 
   what it can do with the data-set. The kernel can forget pages as 
   memory pressure forces the system to page stuff out, and then just 
   automatically re-fetch them again. 

   And the automatic sharing is obviously a case of this.. 

But your test-suite (just copying the data once) is probably pessimal 
for mmap(). 

                Linus