[dpdk-dev] [PATCH v3 01/17] mk: Introduce ARMv7 architecture

Hunt, David david.hunt at intel.com
Wed Oct 28 14:44:50 CET 2015


On 28/10/2015 10:56, Jan Viktorin wrote:
> On Wed, 28 Oct 2015 11:09:21 +0100
> David Marchand <david.marchand at 6wind.com> wrote:
>
>> Hello Jan,
>>
>> On Tue, Oct 27, 2015 at 8:13 PM, Jan Viktorin <viktorin at rehivetech.com>
>> wrote:
>>
>> +# PCI is usually not used on ARM
>>> +CONFIG_RTE_EAL_IGB_UIO=n
>>>
>>
>> Not sure "usually not used" is a good reason to disable something.
>> Is there a real issue on arm with igb_uio code (compilation, pci accesses) ?
>>
>
> Well, it requires to set some options in Linux Kernel (at least PCI
> support) which are usually disabled by the in-kernel *arm*_defconfigs.
> Moreover, it seems I cannot enable it for some ARM architectures (I've
> tried Altera SoC FPGA). That's because you hardly find an ARMv7 system
> with a PCI bus. I suppose that if somebody _really_ needs this, she would
> enable it by hand.
>
> At the moment, it breaks my common builds... The driver is mostly
> useless on ARMv7 and just takes space in the filesystem.

I have an ARMv8 board here that I've built a new kernel for the purposes 
of an ARMv8 port, and it took quite a while to get the PCI
functionality all working, including implementing a fix to the kernel 
PCI driver to expose the mmap resources in sysfs properly. But after 
that, igb_uio compiles fine (on the ARMv8 patch) and works with a 
Niantic to pass traffic between ports.

If the majority of ARMv7 boards don't have a PCI bus, then I'd suggest 
leaving igb_uio disabled. Those few boards with PCI will most likely 
have a correctly kernel (and source) ready to go, so enabling igb_uio 
for them will be easy, but disabling seems a more sensible default for 
the majority of ARMv7 users.

Rgds,
Dave.




More information about the dev mailing list