mapping "random" physical memory into kernel
Scott Long
scottl at samsco.org
Thu Feb 2 14:39:35 PST 2006
Andrew Gallatin wrote:
> Scott Long writes:
> > Andrew Gallatin wrote:
> > > Can somebody please remind me what the supported way to establish a
> > > kernel virtual address for a "random" chunk of physical memory is?
> > > I'm currently using pmap_mapdev() directly, but that just feels
> > > dirty...
> > >
> > > I need to twiddle the settings of an Nvidia NF4 bridge in extended
> > > PCI config space (offset 0x178). These offsets are not accessible
> > > via normal pci config space writes, but are doable when you use the
> > > 0xe000000 mapping.
> >
> > Extended pci config space is supported by default on i386, but I
> > haven't gotten around to figuring out the right strategy for
> > amd64. I'd like to be able to map the entire 256MB range instead
> > of the on-demand approach with i386, but I need to learn more about
> > the various VM maps on amd64 first.
>
> Cool. It would be great if it worked; I'd love to remove
> a bunch of grotty code ;)
>
If you, Peter, Alan, etc, have any suggestions, I'm all ears.
> > >
> > > The problem is that this memory is not really associated with
> > > anything, so a normal bus_alloc_resource() allocation doesn't
> > > seem like it would work.
> >
> > pmap_mapdev is the correct interface for this.
>
> Except it doesn't appear to exist on sparc64. But I doubt
> we'd ever see an NF4 bridge there, so I suppose I can ifdef
> it for i386/amd64.
>
> Drew
>
That's quite reasonable.
Scott
More information about the freebsd-current
mailing list