smi speedstep patch

[John Baldwin]

On Monday 16 July 2007 11:42:53 am Nate Lawson wrote:
> John Baldwin wrote:
> > On Saturday 16 June 2007 01:09:27 pm Nate Lawson wrote:
> >> Nate Lawson wrote:
> >>> Nate Lawson wrote:
> >>>> If you have a pentium 3 that works for speedstep, please try this 
patch.
> >>>>  It fixes the PAE case.  Compile-tested.
> >>> Hmm, I see it's no longer getting the physical address, just virtual.
> >>> So need to fix that part. 
> >> Attached is the updated patch.  It uses the low kernel map where P=V and
> >> fixes the callback.  I just need someone who is using smi-speedstep
> >> (440bx chipset with pentium 3) to make sure the device still attaches.
> >> Patch should work on 6.x and 7.x.
> > 
> > Why do you need a V == P page?
> 
> I don't.  I was just trying to choose a range where I was guaranteed
> there would be any mapping at all.  Aren't there some physical regions
> (VGA, ISA hole) that don't have a mapping?  Thus, it's best to ask for
> RAM at 1 MB+ unless the device has special requirements, right?
> 
> BTW, this is already committed to -current.

What do you mean by a range that doesn't have a "mapping"?  Do you mean you 
want bus_dma to give you memory not already in use?  Why would it do that? :)
Why can't you just allocate any buffer under 4GB?

-- 
John Baldwin