[fpc-pascal] accessing hardware ports in FPC

[Daniël Mantione]

Op Tue, 22 May 2007, schreef Tomas Hajny:

> I don't know Linux, but: If ioperm needs to be called just once (without
> specifying a particular port number), this call could be added to unit
> initialization. If it needs to be called for particular port number or
> port range, we could add a function doing this to interface of unit ports
> (and document it as appropriate, of course), because it's better to
> provide common approach than to require users to fiddle with IFDEFs and
> platform specific API for this (such a function probably exists for most
> platforms and platforms not requiring such a thing may simply ignore it).

It would be a good idea to make even this platform independend, but only 
if you can provide similar semantics on all platforms. However, at least 
in the case of Linux, it such not entirely clear what this initialization 
should do.

There are two calls:
* ioperm, allows you to ports $0..$3ff, for each port individually.
* iopl, which moves the program into cpu ring 2. It can then access ports
  $0..$ffff and execute the cli/sti instructions.

A program can inherite port access, if the parent process has io port 
access, so has the child. So, perhaps the intention of the user could even 
be that no io permission is asked.
 
> Potentially, we could even add a boolean array to implementation part of
> unit ports maintaining the permission status for all the ports. Calling
> the "IOPerm" function would mark access to that port as permitted. Call to
> getter/setter would check the status first and call IOPerm if needed,
> otherwise skip the IOPerm call (that would allow both fast access and "no
> hassle" approach for our users).

This could be an elegant way to abstract the ioperm style functionality 
into a platform independend interface.

Daniël