Re: [PATCH-RFC 07/10] parisc: switch to GENERIC_PCI_IOMAP

["Michael S. Tsirkin" <mst@redhat.com>]

On Mon, Dec 05, 2011 at 10:23:17AM -0800, Grant Grundler wrote:
> On Sun, Dec 4, 2011 at 6:30 AM, Michael S. Tsirkin <mst@redhat.com> wrote:
> ...
> > AFAIK, on PCI a resource is either IO or MEM, but never both.
> > What am I missing?
> 
> Yes, a particular PCI BAR (Base Address Register) is either IO or MEM
> (not both) address space resource.
> 
> James is observing many older devices have the same set of registers
> mapped in both address spaces using two BARs. As you already know,
> MMIO is "lighter weight" (CPU cost to generate transactions on PCI
> bus). I think this is orthogonal to the issue you are addressing.
> 
> Regarding handling of IO Port space cookies:
> > -     if (flags & IORESOURCE_IO)
> > -             return ioport_map(start, len);
> 
> I don't know who all the consumers of pci_iomap() are. If user space
> can somehow call this, it should fail for IO port space "mappings"
> since unlike x86, parisc has no INB/OUTB instructions. Generating IO
> Port space requires frobbing PCI Bus controller registers that I don't
> think we want to expose to user space.
> 
> I'm going to assume this is NOT a problem since any code that
> generates anything that tries to look like an INB/OUTB just won't
> compile on parisc.  I think removing the parisc specific definition
> and using the generic one instead looks fine to me.  But I'm not
> confident enough to offer a formal ACKed-by line. Sorry. :(
> 
> cheers,
> grant

How about a Tested-by line? I presume you have the hardware to
try this out? You can also get the patches from my tree:
git://git.kernel.org/pub/scm/linux/kernel/git/mst/vhost.git linux-next
or from linux-next.

-- 
MST