Re: [PATCH 5/5] powerpc/pci: Add config option for using all 256 PCI buses

["Pali Rohár" <pali@kernel.org>]

On Tuesday 26 July 2022 13:10:01 Pali Rohár wrote:
> On Tuesday 26 July 2022 21:02:22 Michael Ellerman wrote:
> > Pali Rohár <pali@kernel.org> writes:
> > > On Wednesday 06 July 2022 12:43:08 Pali Rohár wrote:
> > >> By default on PPC32 are PCI bus numbers unique across all PCI domains.
> > >> So system could have only 256 PCI buses independently of available
> > >> PCI domains.
> > >>
> > >> This is due to filling DT property pci-OF-bus-map which does not reflect
> > >> multi-domain setup.
> > >>
> > >> On all powerpc platforms except chrp and powermac there is no DT property
> > >> pci-OF-bus-map anymore and therefore it is possible on non-chrp/powermac
> > >> platforms to avoid this limitation of maximal number of 256 PCI buses in
> > >> system even on multi-domain setup.
> > >>
> > >> But avoiding this limitation would mean that all PCI and PCIe devices would
> > >> be present on completely different BDF addresses as every PCI domain starts
> > >> numbering PCI bueses from zero (instead of the last bus number of previous
> > >> enumerated PCI domain). Such change could break existing software which
> > >> expects fixed PCI bus numbers.
> > >>
> > >> So add a new config option CONFIG_PPC_PCI_BUS_NUM_DOMAIN_DEPENDENT which
> > >> enables this change. By default it is disabled. It cause that initial value
> > >> of hose->first_busno is zero.
> > >>
> > >> Signed-off-by: Pali Rohár <pali@kernel.org>
> > >> ---
> > >>  arch/powerpc/Kconfig         | 11 +++++++++++
> > >>  arch/powerpc/kernel/pci_32.c |  6 ++++++
> > >>  2 files changed, 17 insertions(+)
> > >>
> > >> diff --git a/arch/powerpc/Kconfig b/arch/powerpc/Kconfig
> > >> index be68c1f02b79..f66084bc1dfe 100644
> > >> --- a/arch/powerpc/Kconfig
> > >> +++ b/arch/powerpc/Kconfig
> > >> @@ -370,6 +370,17 @@ config PPC_DCR
> > >>  	depends on PPC_DCR_NATIVE || PPC_DCR_MMIO
> > >>  	default y
> > >>
> > >> +config PPC_PCI_BUS_NUM_DOMAIN_DEPENDENT
> > >> +	depends on PPC32
> > >> +	depends on !PPC_PMAC && !PPC_CHRP
> > >> +	bool "Assign PCI bus numbers from zero individually for each PCI domain"
> > >> +	help
> > >> +	  By default on PPC32 were PCI bus numbers unique across all PCI domains.
> > >> +	  So system could have only 256 PCI buses independently of available
> > >> +	  PCI domains. When this option is enabled then PCI bus numbers are
> > >> +	  PCI domain dependent and each PCI controller on own domain can have
> > >> +	  256 PCI buses, like it is on other Linux architectures.
> > >> +
> > >
> > > What do you think, would it be possible to set default value of this
> > > option to enabled?
> > 
> > My preference would be to not have the option at all, just make it the
> > default behaviour. Every new CONFIG option adds more combinations that
> > need testing, or more likely don't get well tested.
> > 
> > But I don't have a good feel for what could break if we turn it on, so
> > honestly I don't really know.
> > 
> > Also I do most of my 32-bit testing on pmacs, which are not affected by
> > the change.
> 
> It is because this change is incompatible with deprecated pci-OF-bus-map
> which pmac uses. I do not have any pmac box for testing or development,
> so I let this part as is.
> 
> If one day pci-OF-bus-map would be possible to disable on pmac then this
> pci bus number change can be enabled also for pmac.

Hello! I have created this patch which allows to disable deprecated
pci-OF-bus-map on powermac and allow to enable this new config option
PPC_PCI_BUS_NUM_DOMAIN_DEPENDENT also on powermac.

So you can test this option too on your powermac boxes.

I'm really not sure if that pci-OF-bus-map is required and for which
platforms or software...

Patch for allowing to disable pci-OF-bus-map is here:
https://patchwork.ozlabs.org/project/linuxppc-dev/patch/20220817163927.24453-1-pali@kernel.org/

> > So I'll probably take the series as-is, and then we can do some more
> > widespread testing and possibly flip the default to enabled, and then
> > maybe remove the option entirely in future.
> > 
> > cheers
> 
> I have tested it on P2020 board with 3 PCIe devices, each on own bus
> where each bus is connected to different PCIe controller / domain and it
> works correctly. Every PCIe device is on bus 1 bus but on different
> domains.