[RFC 2/4] PCI: generic: Add support for ARM64 and MSI(x)

[Lorenzo Pieralisi]

On Fri, Jan 02, 2015 at 09:09:33PM +0000, Arnd Bergmann wrote:
> On Friday 02 January 2015 12:18:06 Suravee Suthikulanit wrote:
> > On 1/2/2015 5:55 AM, Lorenzo Pieralisi wrote:
> > > Hi Suravee,
> > >
> > > On Mon, Dec 29, 2014 at 07:32:44PM +0000, Suravee Suthikulpanit wrote:
> > >> >Hi,
> > >> >
> > >> >I am not sure if this thread is still alive. I'm trying to see what I
> > >> >can do to help clean up/convert to make the PCI GHC also works for arm64
> > >> >w/ zero or minimal ifdefs.
> > >> >
> > >> >Please let me know if someone is already working on this. I noticed that
> > >> >Lorenzo's patches has already been in 3.19-rc1, and in Bjorn's
> > >> >pci/domain branch. Otherwise, I'll try to continue the work based on the
> > >> >sample patch from Arnd here.
> > > If I am not mistaken, the only bit missing to remove pci_sys_data (and so
> > > having a generic host controller driver that works on ARM32/64) is generic
> > > MSI management.
> > 
> > Lorenzo,
> > 
> > Do you mean to remove pci_sys_data from pci-host-generic.c or removing 
> > it completely? I assume the former case.
> 
> Something inbetween: We should be able to remove pci_sys_data and
> pci_common_init_dev from all drivers in drivers/pci/host/, but keep them
> for all drivers in arch/arm/*/pci.c
> 
> > So, looking at the current code in the pci-host-generic.c, my 
> > understanding is that the:
> >      *gen_pci = pci_bus->sysdata->private_data
> > will be changed to:
> >      *gen_pci = pci_bus->sysdata
> > 
> > Then, we can simply just call pci_scan_root_bus() directly since we no 
> > longer need to declare hw_pci for calling pci_common_init_dev().
> 
> Right.
> 
> > > I know for certain Marc is working on it, and the solution is WIP,
> > > I think we should prevent adding more churn to pci_sys_data, since
> > > I managed to remove most of the dependencies (domain, mem_offset).
> > 
> > Thanks for cleaning up the domain and mem_offset.
> > 
> > I saw Marc's irq/msi_domain patch series 
> > (http://git.kernel.org/cgit/linux/kernel/git/maz/arm-platforms.git/log/?h=irq/msi_domain). 
> > 
> > 
> > My understanding is that deals with associating the newly introduced 
> > msi_domain to each device, which replaces the need for pci_bus->msi and 
> > hw_pci->msi_ctrl when configure with CONFIG_PCI_MSI_IRQ_DOMAIN (not sure 
> > if this would be the plan for all arm32).  For ARM32, if not define 
> > CONFIG_PCI_MSI_IRQ_DOMAIN, it would still fall back to using the 
> > [pci_sys_data|hw_pci]->msi_ctrl.
> 
> For all I can tell, we have two cases on ARM regarding MSI:
> 
> - arch/arm/mach-iop13xx/pci.c uses its own
>   arch_setup_msi_irq/arch_teardown_msi_irq implementation and does not
>   use pci_bus->msi.
> 
> - everything else that supports MSI has a modern driver with multiplatform
>   support and uses msi_controller. If any platform wants to support GICv2m,
>   we have to use CONFIG_PCI_MSI_IRQ_DOMAIN for all of them, and that
>   seems like the best way forward.

Yes, I think that's the current situation. With a hook (added by Marc's
code) in PCI core to set the msi domain (DT), all we need to do is convert
the generic host controller code to pci_scan_root_bus() and remove the
pcibios init calls, it is a mechanical change when all patches mentioned
above are merged.

Thanks,
Lorenzo