Re: [dpdk-dev] [PATCH] mk: Introduce NXP dpaa2 architecture based on armv8-a

[Jerin Jacob <jerin.jacob@caviumnetworks.com>]

On Tue, May 10, 2016 at 10:10:07AM +0800, Jianbo Liu wrote:
> On 10 May 2016 at 00:17, Jerin Jacob <jerin.jacob@caviumnetworks.com> wrote:
> > On Mon, May 09, 2016 at 11:22:15PM +0800, Jianbo Liu wrote:
> >> On 9 May 2016 at 20:11, Jerin Jacob <jerin.jacob@caviumnetworks.com> wrote:
> >> > On Mon, May 09, 2016 at 07:02:36PM +0800, Jianbo Liu wrote:
> >> >> On 9 May 2016 at 17:06, Jerin Jacob <jerin.jacob@caviumnetworks.com> wrote:
> >> >> > On Mon, May 09, 2016 at 07:18:22PM +0530, Hemant Agrawal wrote:
> >> >> >> This patch introduces dpaa2 machine target to address difference
> >> >> >> in cpu parameter, number of core to 8 and no numa support
> >> >> >> w.r.t default armv8-a machine
> >> >> >>
> >> >> >> Signed-off-by: Hemant Agrawal <hemant.agrawal@nxp.com>
> >> >> >> ---
> >
> > Snip
> >
> >> >> >> +#
> >> >> >> +# Compile Environment Abstraction Layer
> >> >> >> +#
> >> >> >> +CONFIG_RTE_MAX_LCORE=8
> >> >> >> +CONFIG_RTE_MAX_NUMA_NODES=1
> >> >> >> +CONFIG_RTE_EAL_IGB_UIO=n
> >> >> >
> >> >> > I think it makes sense to move this option to generic arm64 config
> >> >> > as upstream arm64 kernel does not have support for sysfs based PCI mmap
> >> >> > resource file,(/sys/bus/pci/devices/B:D:F/resource[_wc]X) need for
> >> >> > CONFIG_RTE_EAL_IGB_UIO to work) and use VFIO for all cases.
> >> >> >
> >> >> > Any objections?
> >> >> >
> >> >> Is there any conflict to keep both?
> >> >
> >> > I would like to avoid the case like below in dpdk.org ml.
> >> > http://dpdk.org/ml/archives/dev/2016-January/031313.html
> >> >
> >> So no conflict to enable both.
> >
> > IMO, Conflict part comes secondary, It does not even work with upstream kernel.
> > Why keep the broken configuration? Two main reasons I think it makes
> > sense to disable
> > - It is broken, I don't think arm64 kernel developers likes non VFIO approach
> I don't think DPDK user is kernel developer in most cases. They maybe
> like the traditional way.

But, the traditional way is _broken_. If we create the shortcut in
dpdk.org then upstream arm64 linux kernel will never get fixed.I attempted
twice to fix it, but arm64 kernel developers like VFIO approach.
if you think it makes sense to have traditional way then can you
upstream the patch to the arm64 linux kernel and add it in common
config

Technically, with VFIO and VFIONOIOMMU way we can replace the need for
uio in both host and virtualization use cases for arm64.

> > now. So mostly likely it will be broken
> > - Trying to avoid out of tree patches wherever is possible as
> > distribution folks like to work with upstream version.
> Agree. But there is possible that people/company maintain their own kernel tree.

This excactly same thing we would like to avoid in long term.

> >
> >> I'd rather keep as it is for armv8a defconfig, becasue it's the base,
> >> any change may affect existing user.
> > IMO, It makes sense to disable at armv8a defconfig otherwise all armv8
> > variants need add CONFIG_RTE_EAL_IGB_UIO=n in all the configs and its
> > arch specific issue.
> We don't have to do that.
> You didn't explictly disable this config in your current
> defconfig_arm64-thunderx-linuxapp-gcc, but you know which module to
> bind.
But not all the end users.