Re: [PATCH v7 4/4] arm64: dts: Add Caninos Loucos Labrador v3

From: Catalin Marinas
Date: Tue Sep 22 2020 - 06:26:13 EST

On Tue, Sep 22, 2020 at 10:32:06AM +0200, Arnd Bergmann wrote:
> On Tue, Sep 22, 2020 at 8:15 AM Manivannan Sadhasivam
> <manivannan.sadhasivam@xxxxxxxxxx> wrote:
> > On Mon, Sep 21, 2020 at 11:43:02PM -0300, Matheus Castello wrote:
> > > + /* Labrador v3 firmware does not support PSCI */
> >
> > Oops. This is unfortunate... I'm not sure if this is even acceptable for
> > ARM64 machines.
> >
> > Let me add Olof and Arnd...
> Adding Catalin and Will for additional input as well, this is more their
> area than ours.
> I don't think we have formalized this as a policy, but we clearly don't
> want new boards to use the spin table hack. As there are other
> boards using psci on the same chip, I don't think this is a
> hardware bug.

I fully agree, we shouldn't allow new boards to use the spin-table
method unless EL3 is missing on the CPU implementation (not the case
here; only the APM hardware has this issue). Unfortunately we missed
another platform with A53, see commit bc66392d8258 ("arm64: dts: fsl:
Add device tree for S32V234-EVB").

The kernel relies on firmware for other things (power management, errata
workarounds), so an SMC calling convention compliant firmware is highly
recommended. I also don't see why it would be that hard to add PSCI.
Even if you don't port something like Trusted Firmware, U-Boot has basic
support for PSCI.

So from my perspective, NAK on this patch.

I'm tempted to also modify smp_spin_table_cpu_init() to print a big
warning and return an error if this is attempted on new platforms. IOW,
we make it a policy from now on.