Re: [PATCH RFC 18/18] ARM: dts: Add minimal Raspberry Pi 4 support

[Marc Zyngier <marc.zyngier@arm.com>]

Hi Stephan,

On Sun, 28 Jul 2019 11:58:58 +0100,
Stefan Wahren <stefan.wahren@i2se.com> wrote:
> 
> Hi Marc,
> 
> Am 23.07.19 um 20:13 schrieb Marc Zyngier:
> > On Wed, 17 Jul 2019 23:16:04 +0200
> > Stefan Wahren <wahrenst@gmx.net> wrote:
> >
> >> This adds minimal support for the new Raspberry Pi 4 without the
> >> fancy stuff like GENET, PCIe, XHCI, 40 bit DMA and V3D. The RPi 4 is available
> >> in 3 different variants (1, 2 and 4 GB RAM), so leave the memory size to zero
> >> and let the bootloader take care of it.
> >>
> >> Other differences to the Raspberry Pi 3:
> >> - additional GIC 400 Interrupt controller
> >> - new thermal IP and HWRNG
> >> - additional MMC interface (emmc2)
> >> - additional UART, I2C, SPI and PWM interfaces
> >>
> >> Signed-off-by: Stefan Wahren <wahrenst@gmx.net>
> >>
> >> ...
> >> +	};
> >> +
> >> +	timer {
> >> +		compatible = "arm,armv7-timer";

BTW, this should be "arm,armv8-timer".

> >> +		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)>;
> >> +		arm,cpu-registers-not-fw-configured;
> > Really? :-(
> i asked about that [1] and they said that the arm stubs [2] only set the
> timer frequency and zeroing CNTVOFF_EL2.

Who is "they"? As long as it is done on all CPUs, that's the right
thing to do, and this property is pretty useless in that case.

> >
> >> +		always-on;
> > I'm pretty sure that's not the case. This indicates that you can power
> > the CPU off, bring it back on, and yet have your timer alive and
> > kicking. The only case this actually exists is for VMs. I can see it
> > being cargo-culted from previous RPi revisions, but that doesn't make
> > it true.
> 
> They said that they used this to suppress high interrupt load for the
> timers.

Indeed, and that's the only thing the kernel can do if it isn't
provided with a global timer that can be used as a backup then the
CPUs transition to a deep idle state where the timers are not actually
running.

If you have such a global timer, enable it and you'll see the
interrupt rate dropping like a rock. Don't you have a
brcm,bcm2835-system-timer stashed somewhere?

If you don't have such a timer, then the question is whether the CPUs
can go into a deep enough sleep state where the timer comparator is
disabled, leading to missed events. If this system can only reach very
shallow sleep states, then always-on is OK. Otherwise, this is a total
lie and you're likely to miss timer events.

Thanks,

	M.

-- 
Jazz is not dead, it just smells funny.

_______________________________________________
linux-arm-kernel mailing list
linux-arm-kernel@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-arm-kernel