Messages in this thread | | | Date | Thu, 24 Sep 2015 15:45:41 -0700 | From | David Daney <> | Subject | Re: [PATCH] net: mdio-octeon: Add PCI driver binding. |
| |
On 09/24/2015 03:14 PM, David Miller wrote: > From: David Daney <ddaney@caviumnetworks.com> > Date: Thu, 24 Sep 2015 15:04:23 -0700 > >> On 09/24/2015 02:52 PM, David Miller wrote: >>> From: David Daney <ddaney.cavm@gmail.com> >>> Date: Tue, 22 Sep 2015 17:41:36 -0700 >>> >>>> From: David Daney <david.daney@cavium.com> >>>> >>>> When the Cavium mdio-octeon devices appear in the Thunder family of >>>> arm64 based SoCs, they show up as PCI devices. Add PCI driver >>>> wrapping so the driver is bound in the standard PCI device scan. >>>> >>>> When in this form, a single PCI device may have more than a single >>>> bus, we call this a "nexus" of buses. The standard firmware >>>> device_for_each_child_node() iterator is used to find the individual >>>> buses underneath the "nexus". >>>> >>>> Update the device tree binding documentation for the new PCI driver >>>> binding. >>>> >>>> Signed-off-by: David Daney <david.daney@cavium.com> >>> >>> This patch breaks the build: >> >> For which architecture? >> >> I tested it on mips and arm64. I will try x86, as I guess that is >> where you tried your test build. > > x86-64. > >> There is, somewhat of, a method behind the madness here. >> >> In order to use MSI-X interrupts, we need a corresponding PCI >> device. Now, this driver doesn't currently use interrupts, but other >> devices in the SoC do, so they must be PCI devices. > > "I need this thing, which isn't needed, therefore I'm making this > change." >
That is not the exact argument. It is really more like this:
1) The firmware is supplying a uniform view of the devices by making them all PCI devices. This is done because:
a) Devices that use interrupts must be PCI.
b) Uniformity is good.
2) The OF device tree nodes for PCI devices do not result in the creation of a platform device.
3) If there is no platform device, platform device driver binding does not occur, and the device is useless to the kernel.
So, we think it is a good change. I don't really want to argue about the semantics of it being a "needed change".
> Sorry, that's not a good argument. > > ACPI nodes have names and whatnot as well. > > So I haven't heard a compelling argument so far. > > So why not just implement this cleanly and using the existing > framework now, and then when you have a legitimate reason for making a > major change to the probing scheme you can do it then. >
| |