On 10 October 2017 at 01:13, Bjorn Helgaas [off-list ref] wrote:

On Mon, Oct 09, 2017 at 05:46:07PM +0100, Will Deacon wrote:

quoted

On Fri, Oct 06, 2017 at 06:21:59PM -0500, Bjorn Helgaas wrote:

quoted

On Fri, Oct 06, 2017 at 05:39:18PM +0100, Ard Biesheuvel wrote:

quoted

diff --git a/drivers/pci/host/pci-host-generic.c b/drivers/pci/host/pci-host-generic.c
index 7d709a7e0aa8..01e81a30e303 100644
--- a/drivers/pci/host/pci-host-generic.c
+++ b/drivers/pci/host/pci-host-generic.c

@@ -35,6 +35,43 @@ static struct pci_ecam_ops gen_pci_cfg_cam_bus_ops = {
  }
 };

+static int pci_dw_ecam_config_read(struct pci_bus *bus, u32 devfn, int where,
+                            int size, u32 *val)
+{
+ struct pci_config_window *cfg = bus->sysdata;
+
+ /*
+  * The Synopsys DesignWare PCIe controller in ECAM mode will not filter
+  * type 0 config TLPs sent to devices 1 and up on its downstream port,
+  * resulting in devices appearing multiple times on bus 0 unless we
+  * filter out those accesses here.
+  */
+ if (bus->number == cfg->busr.start && PCI_SLOT(devfn) > 0)
+         return PCIBIOS_DEVICE_NOT_FOUND;

I think we should return 0xffffffff data here, as we do in other
similar accessors (dw_pcie_rd_conf(), altera_pcie_cfg_read(),
rockchip_pcie_rd_conf()).

Actually, xilinx-nwl has a nice trick: it implements
nwl_pcie_map_bus(), which returns NULL for invalid devices.  Then it
can use the generic accessors, and pci_generic_config_read() already
fills in ~0 for this case.

What do you think of the following?  I put it on my pci/host-generic
branch for now (pending your response and an ack from Will).

I'd probably just inline the device check in pci_dw_ecam_map_bus directly,
but either way:

Acked-by: Will Deacon <redacted>

Thanks.  The reason I split out pci_dw_valid_device() is because there are
several other *_valid_device() functions in other drivers that do basically
the same thing, and I think it's useful to be able to see that they're all
similar, as opposed to being buried in the config accessor or map
functions, which vary a little more.

quoted

I'm assuming this is so small that there's no point in having a Kconfig
option for these guys that depends on the generic host driver?

Good question; I don't care either way.

In this case, may I suggest we leave the patch as is? Neither of you
seem to mind, and it will make /my/ life much easier, given that I
don't have to chase distros to get this driver enabled in their kernel
configs (unless we make if 'def_bool y', in which case it makes even
less sense to add the Kconfig option in the first place).