PCI mmap - How do I access a PCI board address space?

[gregory.menke at gsfc.nasa.gov]

Chris Caudle writes:
 > Eric Valette wrote:
 >   >Chris Caudle wrote:
 >   >> If using a typical personal computer style X86, I think that the
 >   >> machine BIOS should make sure that all of PCI memory space is
 >   >> non-cacheable for you.
 > 
 > > Not at all : How could the BIOS modyfy the way the CPU 
 > > access PCI address space?
 > 
 > The BIOS sets up the memory controller.  My experience is with
 > Compaq/HP Proliant servers, where I used RTEMS for testing new
 > ASICS.  The Proliant BIOS is pretty standard X86, nothing really
 > unusual, so I would expect that all Intel processor PC style
 > systems would behave the same way.  When the machine boots, the
 > BIOS finds how much physical memory is installed, and configures
 > the memory controller so that physical memory addresses are marked
 > as cacheable, all other address ranges are not cacheable.  On such
 > a system, you just access memory mapped PCI devices without
 > worrying about cache issues.  It just works.  Cache configuration
 > should be handled by the BSP, and in the case of pc386 that
 > configuration is effectively handled by the bootloader, so there is
 > nothing for the BSP to do.
 > 

Thats great until you don't have a PC BIOS to do it- and even if the
BIOS does configure some hardware, theres no reason why the bsp
shouldn't do whatever reconfiguration it wants to.  What happens when
you want your bsp to work on an x86 box with a broken or non-existant
bios?

And, nothing but nothing "Just Works".  If something seems to, you're
just not using it hard enough to reveal the problems.  A PC bios
hardware configuration may well be appropriate and bug-free for a
particular operating environment, but that doesn't make it universally
applicable.

Gregm