Re: Ada's Slide To Oblivion ...

[Ian Wild <ian@cfmu.eurocontrol.be>]

Ole-Hjalmar Kristensen wrote:
>  
> ??? This is pure fantasy. How do you think the code got into the code
> space in the first place?

(sigh)

Let's take the example of a 68000 'cos I've got the data sheet
readily available.  There are three lines coming out of the
processor which distinguish various spaces.  Of interest are:

  supervisor code
  supervisor data
  user code
  user data

If the OS is clever, and many of them were, it's set up the MMU
tables such that 'user code' exists at some point in 'supervisor
data', at least long enough to load the executable from disc.  At
some later point it may re-map things so that 'user data' exists
at that same address in 'supervisor data', letting it initialise
the bss.  At some late stage it initialises the MMU so that
'user code' starts at address 0, 'user data' starts at address 0
(but with the first page unmapped to catch NULL pointers), and
does a mode switch.  There is then NO WAY from user mode to write
to the 'user code' space, and only limited (ie PC relative) access
for reading.

Oh - before anyone suggest it, no, the MMU is not visible from
user mode either.

Is this really rocket science?  I would've though the above
sufficiently obvious that I could have got a patent on it.