Re: [PATCH v4 04/19] PCI/MSI: Add hooks to populate the msi_domain field

[Bjorn Helgaas <bhelgaas@google.com>]

On Wed, Jul 22, 2015 at 03:54:14PM +0100, Marc Zyngier wrote:
> On 22/07/15 15:48, Bjorn Helgaas wrote:
> > On Wed, Jul 22, 2015 at 9:42 AM, Marc Zyngier <marc.zyngier@arm.com> wrote:
> >> Hi Bjorn,
> >>
> >> On 21/07/15 22:26, Bjorn Helgaas wrote:
> >>> On Wed, Jul 15, 2015 at 01:16:38PM +0100, Marc Zyngier wrote:
> >>>> In order to be able to populate the device msi_domain field,
> >>>> add the necesary hooks to propagate the host bridge msi_domain
> >>>> across secondary busses to devices.
> >>>>
> >>>> So far, nobody populates the initial msi_domain.
> >>>>
> >>>> Signed-off-by: Marc Zyngier <marc.zyngier@arm.com>
> >>>> ---
> >>>>  drivers/pci/probe.c | 30 ++++++++++++++++++++++++++++++
> >>>>  include/linux/pci.h |  1 +
> >>>>  2 files changed, 31 insertions(+)
> >>>>
> >>>> diff --git a/drivers/pci/probe.c b/drivers/pci/probe.c
> >>>> index cefd636..376f6fa 100644
> >>>> --- a/drivers/pci/probe.c
> >>>> +++ b/drivers/pci/probe.c
> >>>> @@ -661,6 +661,20 @@ static void pci_set_bus_speed(struct pci_bus *bus)
> >>>>      }
> >>>>  }
> >>>>
> >>>> +void __weak pcibios_set_host_bridge_msi_domain(struct pci_bus *bus)
> >>>> +{
> >>>> +}
> >>>
> >>> I don't think there's anything in this series that requires this to be a
> >>> weak function, is there?  This is the only definition I see.
> >>
> >> It looks like all the pcibios_* functions so far have a weak attribute,
> >> and I've added it as a matter of consistency.
> > 
> > We've used pcibios_* names where we might need an arch-specific
> > implementation.  I'm not sure that's the case here -- do you envision
> > an implementation under arch/* someday?  If not, maybe it should just
> > be a pci_* function instead of pcibios_*.
> 
> I could definitely see non-OF driven architectures wanting to override this.
> 
> Or maybe we should turn it the other way around and make it call the
> various firmware interfaces (OF, ACPI...) until one of them succeeds. In
> which case your suggestion of making it a pci_* function makes a lot of
> sense.

Here's my advice, FWIW:

  - remove __weak

  - rename pcibios_set_host_bridge_msi_domain() to
    pci_host_bridge_msi_domain() and have it return a struct irq_domain *

  - pci_host_bridge_msi_domain() can call whatever firmware- or
    arch-specific code you need to look up the irq_domain

  - move the dev_set_msi_domain() call from pci_set_phb_of_msi_domain() to
    pci_set_bus_msi_domain()