Re: [Xen-devel] [XEN PATCH v3 2/2] xen/arm: Configure early printk via Kconfig

[Anthony PERARD <anthony.perard@xxxxxxxxxx>]

On Tue, Mar 17, 2020 at 10:51:34AM -0700, Stefano Stabellini wrote:
> On Tue, 17 Mar 2020, Julien Grall wrote:
> > I noticed below you added "depends on ARM_64" on the Xilinx SoC. In 
> > general, platform specific options are tied to either arm32 or arm64, 
> > even if the UART "driver" is arch agnostic.
> > 
> > You could technically boot Xen on Arm 32-bit on Armv8 HW provided they 
> > support 32-bit at the hypervisor level, but we never supported this 
> > case. So I am wondering whether we should add depends on each 
> > earlyprintk. Stefano, any opinions?
> 
> Well spotted.
> 
> Xilinx doesn't support 32-bit Xen on their boards, "support" as in test,
> run or validate. So it would not be a problem from Xilinx point of view
> to add a "depends on ARM_64".
> 
> I take that you are suggesting adding "depends on ARM_64/32" under the
> legacy platform earlyprintk options, from EARLY_PRINTK_BRCM to
> EARLY_PRINTK_ZYNQMP right? If so, I am fine with it, and it seems like a
> good idea.

I don't have useful information on which Xen bitness each platform can
boot or support, so I can't really add those "depends on". But that
could be done in a follow-up.

> The other new generic earlyprintk options, the ones that only depend on
> the uart driver, from EARLY_UART_CHOICE_8250 to EARLY_UART_CHOICE_SCIF,
> it feels more natural to leave them without a specific arch dependency.

That would mean adding drivers for both arm32 and arm64. For example,
debug-cadence.inc is only available in arm64/. So if someone selects
arm32 and the cadence early uart driver, there's going to be a compile
error. That's the only reason on why I've added "depends on" on all
EARLY_UART_CHOICE_*.

Thanks,

-- 
Anthony PERARD