On Wed, Aug 04, 2021 at 06:41:55PM +0200, Bert Vermeulen wrote:
> On 7/30/21 4:46 PM, Mark Rutland wrote:
> > On Fri, Jul 30, 2021 at 03:45:50PM +0200, Bert Vermeulen wrote:
> > > + timer {
> > > + compatible = "arm,armv8-timer";
> >
> > This should be "arm,armv7-timer".
> >
> > > + interrupt-parent = <&gic>;
> > > + interrupts = <GIC_PPI 13 (GIC_CPU_MASK_SIMPLE(4) | IRQ_TYPE_LEVEL_LOW)>,
> > > + <GIC_PPI 14 (GIC_CPU_MASK_SIMPLE(4) | IRQ_TYPE_LEVEL_LOW)>,
> > > + <GIC_PPI 11 (GIC_CPU_MASK_SIMPLE(4) | IRQ_TYPE_LEVEL_LOW)>,
> > > + <GIC_PPI 10 (GIC_CPU_MASK_SIMPLE(4) | IRQ_TYPE_LEVEL_LOW)>;
> >
> > GICv3 doesn't have a cpumask in its PPI description, so the
> > GIC_CPU_MASK_SIMPLE() bits should be removed.
>
> Ok, will fix.
>
> > > + clock-frequency = <25000000>;
> >
> > Please have your FW configure CNTFRQ on each CPU; the clock-frequency
> > property in the DT is a workaround for broken FW, and it's *vastly*
> > preferable for FW to configure this correctly (e.g. as it means VMs
> > should "just work").
>
> I've since got hold of the modified U-Boot that runs on my eval board, and
> indeed it doesn't set CNTFRQ. So the kernel does need this, for the moment.
Can't you write CNTFRQ in the u-boot shell/script?
> I may get a chance to upstream support for this SoC in U-Boot, but I can't
> control what people are going to ship with their board. Is it ok to leave
> this in?
If they want a working upstream Linux, then you can control it.
I seem to recall this being rejected in other cases. That may have been
on v8 which has taken stricter stances (but arguably any new v7 stuff
should too).
Rob
