Re: [PATCH 2/2] ARM: pci: Allow passing per-controller private data

From: Thierry Reding <thierry.reding@avionic-design.de>
To: Bjorn Helgaas <bhelgaas@google.com>
Cc: Russell King <linux@arm.linux.org.uk>,
	linux-arm-kernel@lists.infradead.org,
	linux-kernel@vger.kernel.org
Subject: Re: [PATCH 2/2] ARM: pci: Allow passing per-controller private data
Date: Tue, 18 Sep 2012 20:34:28 +0200	[thread overview]
Message-ID: <20120918183428.GC29360@avionic-0098.mockup.avionic-design.de> (raw)
In-Reply-To: <CAErSpo754W2TKr46V5FKwX43eusoOVjxkAWJ=kkB0BvRt_bJBw@mail.gmail.com>

[-- Attachment #1: Type: text/plain, Size: 2217 bytes --]

On Tue, Sep 18, 2012 at 11:21:21AM -0600, Bjorn Helgaas wrote:
> On Fri, Sep 14, 2012 at 3:11 PM, Thierry Reding
> <thierry.reding@avionic-design.de> wrote:
> > In order to allow drivers to specify private data for each controller,
> > this commit adds a private_data field to the struct hw_pci. This field
> > is an array of nr_controllers pointers that will be used to initialize
> > the private_data field of the corresponding controller's pci_sys_data
> > structure.
> 
> I guess you aren't changing the design here because struct hw_pci
> already includes "nr_controllers," but having nr_controllers and a
> private_data[] array sounds like something that might make it hard to
> hot-add a host bridge after boot.

What I do in the Tegra PCIe driver is to pass each of the root ports to
pci_common_init() individually because they can be enabled or disabled
by device tree data. I suppose to some degree you can consider that hot-
adding. Not that Tegra is likely to ever need to support that. I don't
know how likely it is for any ARM platform to ever need support for hot-
adding a host bridge.

Eventually I think it would be advantageous for this to be generalized
further such that PCI initialization can be shared across architectures.
That's probably not an easy task so I was going to start by making
incremental changes that enable the Tegra code to work and, if time
allows, help further with subsequent improvements.

It also seems that parts of the PCI core aren't ready yet for hot-adding
host bridges. One thing I came across while working on the Tegra code is
that MSI setup and teardown needs to be done by the arch_setup_msi_irq()
and arch_teardown_msi_irq() respectively, which are expected to be
builtin. That was also the last issue that keeps the Tegra PCIe driver
from being built as a module. I think that will also make it impossible
to hot-add host bridges. On x86 this seems to be handled by platform
code, but on Tegra for example MSI setup and teardown is tightly coupled
to the PCIe controller. That was one of the things I thought I could
take a look at eventually, but getting Tegra support cleaned up is
higher priority right now.

Thierry

[-- Attachment #2: Type: application/pgp-signature, Size: 836 bytes --]