On Wed, Oct 18, 2017 at 03:16:34PM -0500, Rob Herring wrote:

Use devicetree-compiler list for dtc issues please.

On Wed, Oct 18, 2017 at 2:33 PM, Frank Rowand [off-list ref] wrote:

quoted

Hi Rob, Alan,

On 10/18/17 08:58, Alan Tull wrote:

quoted

Hi Rob,

I've noticed a problem compiling DT overlays and traced it back to
beginning in next-20171009

That tag adds the following in scripts/dtc

e9480c1 2017-10-09 16:17:32 +0100 : Mark Brown : Merge remote-tracking
branch 'devicetree/for-next'
4201d05 2017-10-03 15:03:47 -0500 : Rob Herring : scripts/dtc: Update
to upstream version v1.4.5-3-gb1a60033c110
4322323 2017-10-03 15:03:46 -0500 : Rob Herring : scripts/dtc: add
fdt_overlay.c and fdt_addresses.c to sync script

The error is:

dtc: /home/atull/repos/linux-socfpga/scripts/dtc/livetree.c:543:
get_node_by_phandle: Assertion `(phandle != 0) && (phandle != -1)'
failed.
arch/arm/boot/dts/socfpga_overlay.dtb: Warning (clocks_property):
Could not get phandle node for
/fragment@0/__overlay__/gpio@10040:clocks(cell 0)
Aborted (core dumped)
scripts/Makefile.lib:316: recipe for target
'arch/arm/boot/dts/socfpga_arria10_socdk_sdmmc_ghrd_ovl_ext_cfg.dtb'
failed
make[2]: *** [arch/arm/boot/dts/socfpga_arria10_socdk_sdmmc_ghrd_ovl_ext_cfg.dtb]
Error 134
arch/arm/Makefile:346: recipe for target 'dtbs' failed

Here's a simplified overlay that gets this error.  Taking out the line
"interrupt-parent = <&intc>;" fixes the build.

/dts-v1/;
/plugin/;
/ {
        fragment@0 {
                target-path = "/soc/base_fpga_region";
                #address-cells = <1>;
                #size-cells = <1>;

                __overlay__ {
                        ranges = <0x00000000 0x00000000 0xc0000000 0x00040000>,
                                 <0x00000001 0x00000000 0xff200000 0x00001000>;

                        external-fpga-config;

                        #address-cells = <2>;
                        #size-cells = <1>;

                        fpga_pr_region0 {
                                compatible = "fpga-region";
                                fpga-bridges = <&freeze_controller_0>;
                                ranges;
                        };

                        freeze_controller_0: freeze_controller@0x100000450 {
                                compatible = "altr,freeze-bridge-controller";
                                reg = <0x00000001 0x00000450 0x00000010>;
                                interrupt-parent = <&intc>;  /* <--
remove to fix build */
                                interrupts = <0 21 4>;
                        };
                };
        };
};

Alan

Phandle references in overlays are assigned the value of -1 (0xffffffff) in
the dtb, to be fixed up when loaded.  A new check sees this value and
triggers the assert.

It is this commit in the upstream dtc tools tree:

   commit ee3d26f6960bb5922d9a35fe266d9eac74a78ec0
   checks: add interrupts property check

There are a bunch of other new checks that call get_node_by_phandle(),
and thus could trigger the assertion.

I'm guessing that those checks would also trigger the assert if an
overlay contained something that would lead to one of the other checks
being processed.

You won't get an assert because I check for 0 or -1 and skip the check
in those cases. The interrupts check missed that condition.

However, as shown above, you will get an erroneous warning because it
just skips 1 cell and goes to the next to handle the case where the
phandle is optional and you want a fixed number of elements.

I guess we can't validate overlays which is unfortunate. I don't think
that's a solvable problem unless you have the base DT.

Right.  You can check some things in overlays, but when parsing
something requires information from the base tree, there is indeed not
much that can be done.

quoted

You can avoid the problem in your example dts with "-Wno-interrupts_property"

  dtc -Wno-interrupts_property fpga_01_a.dts

The larger set of other checks that might trigger the assert is too large
for me to want to add "-Wno-" flags for all of them to the command line
(as temporary workarounds).

David thought more switches were better.

Right, as a rule, I like fine configurability of the checks.

But.. there's a trick to make things easier to configure in batches:
when you enable a warning, you automatically enable all warnings that
warning has as prerequisites.  And when you disable a warning, every
other warning that has it as a prerequisite is also disabled.

So by adding some dummy checks with the right depencies, you can allow
enabling and disabling of groups of related checks with a single
command line option.

-- 
David Gibson			| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you.  NOT _the_ _other_
				| _way_ _around_!
http://www.ozlabs.org/~dgibson