[RFC PATCH v2 0/2] Add DSA MultiCPU port support

Ansuel Smith ansuelsmth at gmail.com
Fri Jan 28 08:44:08 PST 2022


Il giorno dom 23 gen 2022 alle ore 02:39 Ansuel Smith
<ansuelsmth at gmail.com> ha scritto:
>
> Il giorno dom 23 gen 2022 alle ore 02:29 Daniel Golle
> <daniel at makrotopia.org> ha scritto:
> >
> > On Sun, Jan 23, 2022 at 01:35:24AM +0100, Ansuel Smith wrote:
> > > This adds the hack patches for DSA multicpu support.
> > > I still have to clean patch 1, 3, 4 but considering this is still a bit WIP
> > > I decided to clean and provide a correct patches for the final version.
> > >
> > > This version won't change the logic by DSA that assing every port to the first
> > > cpu port. A init script is required to change the cpu port at runtime.
> >
> > Imho we should also add patch
> >
> >  From: LGA1150 <dqfext at gmail.com>
> >  Date: Mon, 17 May 2021 10:34:58 +0200
> >  Subject: [PATCH] net: dsa: add dts-property default_cpu
> >
> > so this can be done in device-tree rather than using an init-script.
> >
>
> Don't know if you check the pr but the additional binding was rejected upstream.
> We need to understand if we can accept this kind of patch or not.
> IMHO considering it should be a small patch, we can totally accept
> that but we have
> to be conscious that it won't ever be merged and it will stay in hack forever.
>
> > > This change was done for the only reason that a round-robin way can't be trusted
> > > and is too random. Some cpu port in some switch (brcm) for example doesn't
> > > behave the same way and randomly assigning the cpu port would cause
> > > problems/malfunctions.
> >
> > I agree that round-robin is not such a good idea, the commit message of
> > the patch itself will also have to be updated to reflect that change.
> >
>
> Sure as I said, I still have to rework this and produce clear patches.
> (I assume many changes are still needed for this)
>

I will send v3 in the next few days. Any other comments?

> > >
> > > v2:
> > > - Rework iproute logic to not pollute link
> > > - Rework the round-robin cpu port assign logic
> > >
> > > Ansuel Smith (2):
> > >   linux: introduce multi-cpu dsa patch
> > >   iproute2: add support for cpu set
> > >
> > >  ...101-iplink_allow_to_change_cpu_value.patch |  81 ++++++++++
> > >  ...net-dsa-allow_for_multiple_CPU_ports.patch | 151 ++++++++++++++++++
> > >  ...add_ndo_for_setting_the_cpu_property.patch | 113 +++++++++++++
> > >  ..._set_cpu_for_changing_ports_CPU_port.patch |  89 +++++++++++
> > >  ...clude-net-add-dsa_cpu_ports-function.patch |  34 ++++
> > >  5 files changed, 468 insertions(+)
> > >  create mode 100644 package/network/utils/iproute2/patches/101-iplink_allow_to_change_cpu_value.patch
> > >  create mode 100644 target/linux/generic/hack-5.10/780-1-net-dsa-allow_for_multiple_CPU_ports.patch
> > >  create mode 100644 target/linux/generic/hack-5.10/780-2-net-add_ndo_for_setting_the_cpu_property.patch
> > >  create mode 100644 target/linux/generic/hack-5.10/780-3-net-dsa-implement_ndo_set_cpu_for_changing_ports_CPU_port.patch
> > >  create mode 100644 target/linux/generic/hack-5.10/780-4-include-net-add-dsa_cpu_ports-function.patch
> > >
> > > --
> > > 2.33.1
> > >
> > >
> > > _______________________________________________
> > > openwrt-devel mailing list
> > > openwrt-devel at lists.openwrt.org
> > > https://lists.openwrt.org/mailman/listinfo/openwrt-devel



More information about the openwrt-devel mailing list