On Tue, Sep 23, 2014 at 7:12 PM, Liviu Dudau [off-list ref] wrote:

On Tue, Sep 23, 2014 at 06:22:53PM -0600, Bjorn Helgaas wrote:

quoted

[+cc Andrew]

On Tue, Sep 23, 2014 at 08:01:07PM +0100, Liviu Dudau wrote:

quoted

The ranges property for a host bridge controller in DT describes
the mapping between the PCI bus address and the CPU physical address.
The resources framework however expects that the IO resources start
at a pseudo "port" address 0 (zero) and have a maximum size of IO_SPACE_LIMIT.
The conversion from pci ranges to resources failed to take that into account,
returning a CPU physical address instead of a port number.

Also fix all the drivers that depend on the old behaviour by fetching
the CPU physical address based on the port number where it is being needed.

Cc: Grant Likely <redacted>
Cc: Rob Herring <robh+dt@kernel.org>
Cc: Arnd Bergmann <arnd@arndb.de>
Acked-by: Linus Walleij <redacted>
Cc: Thierry Reding <redacted>
Cc: Simon Horman <horms@verge.net.au>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Signed-off-by: Liviu Dudau <Liviu.Dudau@arm.com>
---
 arch/arm/mach-integrator/pci_v3.c | 23 ++++++++++----------
 drivers/of/address.c              | 44 +++++++++++++++++++++++++++++++++++----
 drivers/pci/host/pci-tegra.c      | 10 ++++++---
 drivers/pci/host/pcie-rcar.c      | 21 +++++++++++++------
 include/linux/of_address.h        | 15 ++++++-------
 5 files changed, 82 insertions(+), 31 deletions(-)
...

The of_pci_range_to_resource() implementation in drivers/of/address.c is
always compiled when CONFIG_OF_ADDRESS=y, but when CONFIG_OF_ADDRESS=y and
CONFIG_PCI is not set, we get the static inline version from
include/linux/of_address.h as well, causing a redefinition error.

quoted

diff --git a/drivers/of/address.c b/drivers/of/address.c

@@ -957,12 +957,48 @@ bool of_dma_is_coherent(struct device_node *np)

...
+int of_pci_range_to_resource(struct of_pci_range *range,
+   struct device_node *np, struct resource *res)

quoted

diff --git a/include/linux/of_address.h b/include/linux/of_address.h

...
 #else /* CONFIG_OF_ADDRESS && CONFIG_PCI */
 static inline int of_pci_address_to_resource(struct device_node *dev, int bar,
                                         struct resource *r)

@@ -144,6 +139,12 @@ static inline int of_pci_address_to_resource(struct device_node *dev, int bar,
    return -ENOSYS;
 }

+static inline int of_pci_range_to_resource(struct of_pci_range *range,
+           struct device_node *np, struct resource *res)
+{
+   return -ENOSYS;
+}

My proposal to fix it is the following three patches.  The first moves the
inline version of of_pci_range_to_resource() into the existing "#if
defined(CONFIG_OF_ADDRESS) && defined(CONFIG_PCI)" block.

Andrew added it (and some other PCI-related things) with 29b635c00f3e
("of/pci: Provide support for parsing PCI DT ranges property") to
of_address.h outside of any ifdefs, so it's always available.  Maybe
there's a reason that's needed in the non-CONFIG_PCI case, but I didn't see
it with a quick look.

The second moves of_pci_range_to_resource() to address.c, still inside the
"#ifdef CONFIG_PCI" block.

Hi Bjorn,

Looks good to me. Sorry for messing this up, I've tested a combination of
CONFIG_PCI=y and CONFIG_PCI=n but it looks like I've only tested
CONFIG_OF_ADDRESS=CONFIG_PCI (both enabled or both disabled).

You probably saw the subsequent [pci:pci/host-generic 10/13]
drivers/of/of_pci.c:202:3: error: implicit declaration of function
'of_pci_range_to_resource' error.

I looked at that a bit, but gave up.  Apparently drivers/of/of_pci.c
can be compiled with CONFIG_PCI=y but CONFIG_OF_ADDRESS not set..

Bjorn