Re: [RFC v2 1/7] PCI: Introduce pci_host_bridge::domain_nr

On Tue, May 4, 2021 at 12:16 AM Mike Rapoport [off-list ref] wrote:

On Mon, May 03, 2021 at 10:46:29PM +0800, Boqun Feng wrote:

quoted

Currently we retrieve the PCI domain number of the host bridge from the
bus sysdata (or pci_config_window if PCI_DOMAINS_GENERIC=y). Actually
we have the information at PCI host bridge probing time, and it makes
sense that we store it into pci_host_bridge. One benefit of doing so is
the requirement for supporting PCI on Hyper-V for ARM64, because the
host bridge of Hyper-V doesnt' have pci_config_window, whereas ARM64 is
a PCI_DOMAINS_GENERIC=y arch, so we cannot retrieve the PCI domain
number from pci_config_window on ARM64 Hyper-V guest.

As the preparation for ARM64 Hyper-V PCI support, we introduce the
domain_nr in pci_host_bridge, and set it properly at probing time, then
for PCI_DOMAINS_GENERIC=y archs, bus domain numbers are set by the
bridge domain_nr.

Signed-off-by: Boqun Feng <redacted>
---
 arch/arm/kernel/bios32.c              |  2 ++
 arch/arm/mach-dove/pcie.c             |  2 ++
 arch/arm/mach-mv78xx0/pcie.c          |  2 ++
 arch/arm/mach-orion5x/pci.c           |  2 ++
 arch/arm64/kernel/pci.c               |  3 +--
 arch/mips/pci/pci-legacy.c            |  2 ++
 arch/mips/pci/pci-xtalk-bridge.c      |  2 ++
 drivers/pci/controller/pci-ftpci100.c |  2 ++
 drivers/pci/controller/pci-mvebu.c    |  2 ++
 drivers/pci/pci.c                     |  4 ++--
 drivers/pci/probe.c                   |  7 ++++++-
 include/linux/pci.h                   | 11 ++++++++---
 12 files changed, 33 insertions(+), 8 deletions(-)

diff --git a/arch/arm/kernel/bios32.c b/arch/arm/kernel/bios32.c
index e7ef2b5bea9c..4942cd681e41 100644
--- a/arch/arm/kernel/bios32.c
+++ b/arch/arm/kernel/bios32.c

@@ -471,6 +471,8 @@ static void pcibios_init_hw(struct device *parent, struct hw_pci *hw,
                              bridge->sysdata = sys;
                              bridge->busnr = sys->busnr;
                              bridge->ops = hw->ops;
+                             if (IS_ENABLED(CONFIG_PCI_DOMAINS_GENERIC))
+                                     bridge->domain_nr = pci_bus_find_domain_nr(sys, parent);

                              ret = pci_scan_root_bus_bridge(bridge);
                      }

diff --git a/arch/arm/mach-dove/pcie.c b/arch/arm/mach-dove/pcie.c
index ee91ac6b5ebf..92eb8484b49b 100644
--- a/arch/arm/mach-dove/pcie.c
+++ b/arch/arm/mach-dove/pcie.c

@@ -167,6 +167,8 @@ dove_pcie_scan_bus(int nr, struct pci_host_bridge *bridge)
      bridge->sysdata = sys;
      bridge->busnr = sys->busnr;
      bridge->ops = &pcie_ops;
+     if (IS_ENABLED(CONFIG_PCI_DOMAINS_GENERIC))
+             bridge->domain_nr = pci_bus_find_domain_nr(sys, NULL);

The check for CONFIG_PCI_DOMAINS_GENERIC is excessive because there is a
stub for pci_bus_find_domain_nr().

I'm not an expert in PCI, but maybe the repeated assignment of
bridge->domain_nr can live in the generic code, say, in
pci_scan_root_bus_bridge(). E.g. it will set the domain_nr when it is zero.

Yes. There's zero reason h/w drivers should care what the domain_nr is.

There's another issue with domains you should be aware of:

https://lore.kernel.org/linux-pci/20210425152155.mstuxndsoqdbdape@pali/ (local)

That may need to be fixed first because deferred probing could cause
the domain to increment each time you retry probe.

Rob