Re: [Qemu-devel] [PATCH] pci: add standard bridge device

[Avi Kivity]

On 09/04/2011 04:01 PM, Michael S. Tsirkin wrote:
> On Sun, Sep 04, 2011 at 03:40:58PM +0300, Avi Kivity wrote:
> >  On 09/04/2011 03:30 PM, Michael S. Tsirkin wrote:
> >  >>
> >  >>   Create a memory region for the bridge's address space.  This region
> >  >>   is not directly added to system_memory or its descendants.
> >  >
> >  >I do this for each bridge in the hierarchy, right?
> >
> >  Each bridge does this independently (so yes).
> >
> >  >>   fx440 does exactly this, with the following cosmetic changes:
> >  >>
> >  >>   - the windows are different (vga, pci hole, 64-bit pci area, PAMx, 
> SMRAM)
> >  >>   - instead of mapping them to the parent bridge's
> >  >>   pci_address_space(), we map them to get_system_memory()
> >  >
> >  >Hmm, what ties the windows of a child bridge
> >  >to be within the windows of a parent?
> >  >
> >
> >  system_memory
> >    |
> >    +--- pci0_alias0 (aliases part of pci0)
> >
> >  pci0
> >    |
> >    +--- pci1_alias0 (a bridge)
> >
> >  pci1
> >    |
> >    +--- pci2_alias0 (another bridge)
> >
> >  pci2
> >    |
> >    +--- BAR
> >
> >  When rendering the memory hierarchy, the only parts of BAR which are
> >  visible are those that fit the clipping regions pci0_alias0 ^
> >  pci1_alias0 ^ pci2_alias0.  If there are multiple aliases (like the
> >  low and high PCI holes, and PAM, it becomes (pci0_alias0 v
> >  pci0_alias1) ^ (pci1_alias0v pci1_alias1) ^ (pci2_alias0 v
> >  pci2_alias1). ( "^" == intersection, "v" == union )
>
> What about BAR directly behind pci0?
> That should be unaffected by bridges pci1 and pci2
> since the device is not behind that.

It follows naturally:

system_memory
  |
  +--- pci0_alias0 (aliases part of pci0)
               |
pci0 <-+
  |
  +--- pci1_alias0 (a bridge)
  |            |
  +--- BAR0.0  |
               |
pci1 <-+
  |
  +--- pci2_alias0 (another bridge)
  |            |
  +--- BAR1.0  |
               |
pci2 <-+
  |
  +--- BAR2.0


BAR0.0 is only filtered by pci0_alias*, BAR1.0 is filtered by 
pci[01]_alias*.  Since pci_register_bar() adds the BAR as a subregion of 
the bridge's pci_address_space(), it works without any global knowledge, 
with this topology, or if pci1 and pci2 are siblings instead of parent 
and child.

--
error compiling committee.c: too many arguments to function