Hi Rob,

On Wed, Mar 07, 2018 at 10:08:28AM -0600, Rob Herring wrote:
> Please compile with W=1 and fix any issues like this one which is a
> unit-address without a reg property. Drop the unit-address.

I was just giving the BMIPS W=1 DT warnings a look, and a few look
spurious. I'd value your opinion on their legitimacy (its hard to care
about W=1 if spurious or seemingly pedantic warnings are going to be
common). e.g.


1)
arch/mips/boot/dts/brcm/bcm9ejtagprb.dtb: Warning (unit_address_vs_reg): Node /ubus/syscon-reboot@10000068 has a unit name, but no reg property          

due to:

periph_cntl: syscon@fff8c000 {
	compatible = "syscon";
	reg = <0xfff8c000 0xc>;
	native-endian;
};

reboot: syscon-reboot@fff8c008 {
	compatible = "syscon-reboot";
	regmap = <&periph_cntl>;
	offset = <0x8>;
	mask = <0x1>;
};

That doesn't seem to take regmap into account. Would you strictly drop
the unit-address in this case, or is there a way the DT compiler can be
fixed (i presume offset and mask are binding specific, so the best it
could do is probably to allow the unit-address due to the regmap without
checking the actual address)?


2)
arch/mips/boot/dts/brcm/bcm9ejtagprb.dtb: Warning (simple_bus_reg): Node /ubus/syscon-reboot@10000068 missing or empty reg/ranges property

Same code as above. Should syscon-reboot be outside of the simple-bus
that both nodes are in, or is it fine there? There's a similar warning
from a DTS which has a syscon property instead of regmap.


3)
arch/mips/boot/dts/brcm/bcm97425svmb.dtb: Warning (simple_bus_reg): Node /rdb@10000000/spi@41c000 simple-bus unit address format error, expected "419920"

qspi: spi@41c000 {
	#address-cells = <0x1>;
	#size-cells = <0x0>;
	compatible = "brcm,spi-bcm-qspi",
		     "brcm,spi-brcmstb-qspi";
	clocks = <&upg_clk>;
	reg = <0x419920 0x4 0x41c200 0x188 0x41c000 0x50>;
	reg-names = "cs_reg", "hif_mspi", "bspi";
	...

Well 41c000 is one of the reg entries, just not the first. I presume
bspi is the "main" one, perhaps that should come first since we have
reg-names, but even that could potentially confuse driver code if it
didn't find reg resources by name (in this case it does appear to, so
perhaps that would fine)?


Thanks
James