[parisc-linux-cvs] Re: U bit and memory mapped IO

[Thomas Bogendoerfer]

On Mon, Mar 04, 2002 at 04:40:54AM -0700, John Marvin wrote:
> > then the PA8000 in my C160 is sort of a strange CPU ? Without the U bit
> > set it HPMCs, when accessing the framebuffer from userspace. I read the
> > PA2.0 docs and there it is noted, that the behaviour of an unset U-bit for
> > IO memory space is undefined.
> 
> Well, I can't explain what you are seeing. It might be interesting to
> actually analyze the HPMC you are seeing.

ok, will post that later.

> to find out the actual implementation.  Here is what the PCXU (PA8000) cpu

thanks for looking that up. So basically I'm seeing something which shouldn't
be influenced by the U-bit. Strange.

Another issue on which the PA2.0 docs aren't clear and you might even
know. If our kernel is running 32bit (W = 0), how must TLB entries
for IO space addresses look like ? As I read the PA2.0 we would need
to supply 64bit TLB entries for it, which we don't do at the moment. 
That would explain the framebuffer mapping problems on 32bit a lot of
people see at the moment. Unfortunately my assembler hacks either are
wrong, because the assumption is wrond or just plain broken. So do we
need 64bit TLBs entries for IO space mappings ?

Thomas.