PCI: keystone: ks_pcie_v3_65_add_bus()

PCI: keystone: ks_pcie_v3_65_add_bus()

[Bjorn Helgaas]

Not sure whether anybody cares about the keystone driver any more.  It
seems basically unmaintained after 57e1d8206e48 ("MAINTAINERS: move
Murali Karicheri to credits") [1].

Anyway, ks_pcie_v3_65_add_bus() [2] looks unusual to me.  It's an
.add_bus() method that is called whenever we create a new PCI bus:

  ks_pcie_v3_65_add_bus(...)
  {
    ks_pcie_set_dbi_mode
    dw_pcie_writel_dbi(pci, PCI_BASE_ADDRESS_0, 1)
    dw_pcie_writel_dbi(pci, PCI_BASE_ADDRESS_0, SZ_4K - 1)
    ks_pcie_clear_dbi_mode(ks_pcie)
    dw_pcie_writel_dbi(pci, PCI_BASE_ADDRESS_0, ks_pcie->app.start)
  }

This seems like something that should be done when the Root Port is
enumerated, not when we set up its secondary bus.  Maybe somewhere in
ks_pcie_host_init() or ks_pcie_config_msi_irq()?

I don't think we should use .add_bus() unless it's actually something
related to adding a bus.

[1] https://git.kernel.org/linus/57e1d8206e48
[2] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/drivers/pci/controller/dwc/pci-keystone.c?id=v5.18#n452

Re: PCI: keystone: ks_pcie_v3_65_add_bus()

[Rob Herring]

On Tue, May 24, 2022 at 3:15 PM Bjorn Helgaas <helgaas@kernel.org> wrote:
>
> Not sure whether anybody cares about the keystone driver any more.  It
> seems basically unmaintained after 57e1d8206e48 ("MAINTAINERS: move
> Murali Karicheri to credits") [1].
>
> Anyway, ks_pcie_v3_65_add_bus() [2] looks unusual to me.  It's an
> .add_bus() method that is called whenever we create a new PCI bus:
>
>   ks_pcie_v3_65_add_bus(...)
>   {
>     ks_pcie_set_dbi_mode
>     dw_pcie_writel_dbi(pci, PCI_BASE_ADDRESS_0, 1)
>     dw_pcie_writel_dbi(pci, PCI_BASE_ADDRESS_0, SZ_4K - 1)
>     ks_pcie_clear_dbi_mode(ks_pcie)
>     dw_pcie_writel_dbi(pci, PCI_BASE_ADDRESS_0, ks_pcie->app.start)
>   }
>
> This seems like something that should be done when the Root Port is
> enumerated, not when we set up its secondary bus.  Maybe somewhere in
> ks_pcie_host_init() or ks_pcie_config_msi_irq()?

See these commits for the reasoning:

1df793054859 PCI: dwc: Convert to use pci_host_probe()
6ab15b5e7057 PCI: dwc: keystone: Convert .scan_bus() callback to use add_bus
6e8e104d2196 PCI: Also call .add_bus() callback for root bus

The reason was to keep the initialization in the same order as when we
had the DWC specific .scan_bus() op. There's not a driver hook in
between the h/w setup and scanning devices.


> I don't think we should use .add_bus() unless it's actually something
> related to adding a bus.

I think I was inspired from how pcibios_add_bus() is (ab)used.

Rob