Re: [Qemu-devel] [PATCH 1/6] q35: make pci window address/size match guest cfg

[Michael S. Tsirkin]

On Tue, Sep 10, 2013 at 04:09:21PM +0200, Igor Mammedov wrote:
> On Tue, 10 Sep 2013 16:53:58 +0300
> "Michael S. Tsirkin" <address@hidden> wrote:
> 
> > On Tue, Sep 10, 2013 at 03:46:13PM +0200, Igor Mammedov wrote:
> > > On Tue, 10 Sep 2013 16:46:36 +0300
> > > "Michael S. Tsirkin" <address@hidden> wrote:
> > > 
> > > > On Tue, Sep 10, 2013 at 03:37:12PM +0200, Igor Mammedov wrote:
> > > > > On Wed, 4 Sep 2013 13:48:29 +0300
> > > > > "Michael S. Tsirkin" <address@hidden> wrote:
> > > > > 
> > > > > > For Q35, MMCFG address and size are guest configurable.
> > > > > > Update w32 property to make it behave accordingly.
> > > > > >
> > > > > > Signed-off-by: Michael S. Tsirkin <address@hidden>
> > > > > > ---
> > > > > >  hw/pci-host/q35.c | 10 ++++++++++
> > > > > >  1 file changed, 10 insertions(+)
> > > > > > 
> > > > > > diff --git a/hw/pci-host/q35.c b/hw/pci-host/q35.c
> > > > > > index 4febd24..3f1d447 100644
> > > > > > --- a/hw/pci-host/q35.c
> > > > > > +++ b/hw/pci-host/q35.c
> > > > > > @@ -214,6 +214,16 @@ static void mch_update_pciexbar(MCHPCIState 
> > > > > > *mch)
> > > > > >      }
> > > > > >      addr = pciexbar & addr_mask;
> > > > > >      pcie_host_mmcfg_update(pehb, enable, addr, length);
> > > > > > +    /* Leave enough space for the MCFG BAR */
> > > > > > +    /*
> > > > > > +     * TODO: this matches current bios behaviour, but it's not a 
> > > > > > power of two,
> > > > > > +     * which means an MTRR can't cover it exactly.
> > > > > > +     */
> > > > > > +    if (enable) {
> > > > > > +        mch->pci_info.w32.begin = addr + length;
> > > > > > +    } else {
> > > > > > +        mch->pci_info.w32.begin = MCH_HOST_BRIDGE_PCIEXBAR_DEFAULT;
> > > > > > +    }
> > > > > >  }
> > > > > I probably miss something but where is remapping in system address 
> > > > > space?
> > > > > If there is none then, then updated w32 might mismatch 
> > > > > actually/initially mapped alias.
> > > > > 
> > > > > >  /* PAM */
> > > > 
> > > > You mean mmcfg?
> > > > The re-mapping is in hw/pci/pcie_host.c
> > > no, I mean 32-bit PCI hole
> > 
> > It works differently - see mch_init.
> > 32-bit PCI hole is not remapped - instead mmcfg overlaps it
> > and shadows a chunk of it.
> > 
> it probably won't cause harm, but actual pci_hole alias doesn't match
> its window reported via properties anymore (it's just inconsistent).

That's exactly what happens at the moment too:
for q35 pci hole has a chunk that is not reported in w32.

Do you imply that w32 should just report the PCI hole,
and make users (e.g. acpi) take mmcfg out of that?


-- 
MST