RE: Probing the expected state/support of DPDK@armhf

[Ruifeng Wang <Ruifeng.Wang@arm.com>]

> -----Original Message-----
> From: Juraj Linkeš <juraj.linkes@pantheon.tech>
> Sent: Tuesday, November 16, 2021 8:06 PM
> To: Christian Ehrhardt <christian.ehrhardt@canonical.com>; Ruifeng Wang
> <Ruifeng.Wang@arm.com>
> Cc: Jan Viktorin <viktorin@rehivetech.com>; dev <dev@dpdk.org>; Luca
> Boccassi <bluca@debian.org>; Honnappa Nagarahalli
> <Honnappa.Nagarahalli@arm.com>; nd <nd@arm.com>
> Subject: RE: Probing the expected state/support of DPDK@armhf
> 
> 
> 
> > -----Original Message-----
> > From: Christian Ehrhardt <christian.ehrhardt@canonical.com>
> > Sent: Wednesday, November 10, 2021 8:48 AM
> > To: Ruifeng Wang <ruifeng.wang@arm.com>
> > Cc: Jan Viktorin <viktorin@rehivetech.com>; dev <dev@dpdk.org>; Luca
> > Boccassi <bluca@debian.org>; Juraj Linkeš
> > <juraj.linkes@pantheon.tech>; Honnappa Nagarahalli
> > <Honnappa.Nagarahalli@arm.com>; nd <nd@arm.com>
> > Subject: Re: Probing the expected state/support of DPDK@armhf
> >
> > On Tue, Nov 9, 2021 at 8:17 AM Ruifeng Wang <Ruifeng.Wang@arm.com>
> > wrote:
> > >
> > > > -----Original Message-----
> > > > From: Christian Ehrhardt <christian.ehrhardt@canonical.com>
> > > > Sent: Monday, November 8, 2021 8:32 PM
> > > > To: Jan Viktorin <viktorin@rehivetech.com>; Ruifeng Wang
> > > > <Ruifeng.Wang@arm.com>; dev <dev@dpdk.org>
> > > > Cc: Luca Boccassi <bluca@debian.org>
> > > > Subject: Probing the expected state/support of DPDK@armhf
> > > >
> > > > Hi,
> > > > I wanted to ask about the current state of DPDK@armhf (not arm64,
> > > > that seems fine AFAICS).
> > > > Since there are too many arms today, I mean armhf as in [0].
> > > I think armhf in question refers to armv7.
> > >
> > > >
> > > > What I see when building DPDK 21.11 is
> > > > 2973 ../config/meson.build:364:1: ERROR: Problem encountered:
> > > > Number of CPU cores not specified.
> > > >
> > > > Right now this seems to be broken the same everywhere - Suse [1],
> > > > fedora [2], Debian/Ubuntu [3]
> > > Looks like this happens with native build on armv7 machine.
> > > RTE_MAX_LCORE
> > is not set for the build.
> >
> 
> What do we want to do with armv7 native build, Ruifeng? For aarch64, we
> detect which machine we're building on and we set everything accordingly,
> unless the generic build is enabled. Do we want to add support for just the
> generic build for armv7 (i.e. regardless of what's set in -Dplatform)? What
> values of RTE_MAX_LCORE and RTE_MAX_NUMA_NODES make sense for an
> armv7 generic build?

Yes, support just the generic build for armv7.
I think we should have RTE_MAX_LCORE=128, RTE_MAX_NUMA_NODES=8 for the build.
These values are consistent with values used in stable branch where armv7 still have Make build support.

> 
> > Thanks for the hint,
> > just FYI using the newer -Dplatform=generic without specifying the
> > machine as we did in the past yields the same issue.
> > I understand that every custom built project needs it's little special
> > twist, but that (ask for the lowest common denominator) is exactly
> > what generic builds in Distributions will need.
> >
> >
> > > > I'm not asking for a fix for this particular issue (although I
> > > > guess people would be happy), but more about the general state of
> DPDK@armhf.
> > > >
> > > > Debian and Ubuntu used to build it on armhf as well, but over the
> > > > recent years I feel (no hard data) that usage there was next to none.
> > > >
> > > > OTOH Thomas said that recently people cared about armv7 [4]
> > > >
> > > > My suggestion would be to disable the build on armhf in
> > > > Debian/Ubuntu
> > > > (+elsewhere?) until it reaches a more stable phase and real use-cases.
> > > > But maybe I missed some use-cases, therefore I wanted to reach out
> > > > to the mailing list to probe for more opinions on this.
> > > I'm not aware of amount of users that using DPDK on armv7. But [4]
> > > suggests
> > that there is real use case.
> > > I think the build on armhf does have value. It helps to maintain
> > > general status on armv7. That is one of the reason why user [4] can
> > > enable
> > armv7 cross compile smoothly.
> > > I suggest to keep the build and fix the failure.
> > >
> 
> That patch actually uses the aarch64 generic config with armhf compiler, so
> it's not really an armv7 build. I'll need to weigh in on that one.

OK. Thanks.
> 
> > > >
> > > > Thanks for your thoughts on this in advance!
> > > >
> > > > P.S. If it is meant to work and be supported, then we will need a
> > > > fix for that
> > > Hi Juraj,
> > > Can you have a look at the issue?
> > >
> > > >
> > > > [0]: https://wiki.debian.org/ArmHardFloatPort
> > > > [1]:
> > > >
> https://build.opensuse.org/public/build/home:bluca:dpdk/openSUSE_F
> > > > ac
> > > > to
> > > > ry_ARM/armv7l/dpdk/_log
> > > > [2]:
> > > > https://build.opensuse.org/public/build/home:bluca:dpdk/Fedora_35/
> > > > ar
> > > > mv
> > > > 7l/dpdk/_log
> > > > [3]:
> > > > https://launchpadlibrarian.net/567810935/buildlog_ubuntu-jammy-
> > > > armhf.dpdk_21.11~rc1-1u~ppa1_BUILDING.txt.gz
> > > > [4]:
> > > > https://patches.dpdk.org/project/dpdk/patch/20210610111839.7481-1-
> > > > s.chandrakant@globaledgesoft.com/
> > > >
> > > > --
> > > > Christian Ehrhardt
> > > > Staff Engineer, Ubuntu Server
> > > > Canonical Ltd
> >
> >
> >
> > --
> > Christian Ehrhardt
> > Staff Engineer, Ubuntu Server
> > Canonical Ltd