[PATCH] [RFC] ARM: imx: add smp support for imx7d

[Shawn Guo]

On Tue, Sep 22, 2020 at 04:59:44PM +0200, Jan Kiszka wrote:
> On 22.09.20 13:34, Marek Vasut wrote:
> > On 9/22/20 8:27 AM, Shawn Guo wrote:
> > > On Tue, Sep 22, 2020 at 2:22 PM Shawn Guo <shawnguo at kernel.org> wrote:
> > > > 
> > > > On Thu, Sep 17, 2020 at 06:08:14PM +0200, Marek Vasut wrote:
> > > > > From: Anson Huang <b20788 at freescale.com>
> > > > > 
> > > > > Add SMP support for i.MX7D, including CPU hotplug support, for
> > > > > systems where TFA is not present.
> > > > 
> > > > These systems are not supported by upstream kernel.  Sorry.
> > > 
> > > I meant for systems without PSCI support actually.
> > 
> > Is there any specific reason for that ?
> > 
> > The SoC works fully well with mainline U-Boot and without TFA, except
> > the code for bringing up the second core is missing from mainline and
> > that is all that is missing. PSCI is unnecessary extra complexity here.
> > 
> 
> We are coming from vendor kernels and would like to base our products on
> mainline for $countless-good-reasons. With the vendor kernels, this
> "classic" way of booting worked fine, with historic bootloaders and also
> with current mainline U-Boot.
> 
> If PSCI support is mandated, we would now be unable to migrate devices in
> the field that should not or cannot receive a bootloader update because
> existing deployments generally do not ship TF-A or any other PSCI
> implementation on ARMv7 - there was mostly no use for it (and there will
> likely be none, except for CPU onlining). Would be a shame.

Thanks for sharing such user story that we would love to hear.  We like
such transition from vendor kernel to upstream for sure.  I would hope
my upstream maintainers would also agree this is a worthy compromise.
I will try to help get this in.

> So I'd also like to understand what speaks against a merge, provided this
> patch does not break other cases or make the code significantly more complex
> and harder to maintain.

The patch itself is indeed not a maintenance burden.  What scares me is
the suspend and idle code in vendor kernel, which we hope that PSCI can
take care of for upstream kernel.  For record, I reserve the right to
say NO to those code even if we get this patch in :)

Shawn