On 30.04.20 23:16, Andrew F. Davis wrote:
On 4/30/20 3:03 PM, Jan Kiszka wrote:
Hi all,
I've noticed that building am65x_evm_a53_defconfig causes the dtbs to be built twice, once for the main u-boot and once for the spl. This is because of an extra dependency in mach-k3/config_secure.mk added by 508369672ca3. Why should the produced dtbs depend on the target that produces them? Why not simply this?
diff --git a/arch/arm/mach-k3/config_secure.mk b/arch/arm/mach-k3/config_secure.mk index 6d63c57665..cbe9b684fb 100644 --- a/arch/arm/mach-k3/config_secure.mk +++ b/arch/arm/mach-k3/config_secure.mk @@ -34,11 +34,8 @@ MKIMAGEFLAGS_u-boot.img_HS = -f auto -A $(ARCH) -T firmware -C none -O u-boot \ -n "U-Boot $(UBOOTRELEASE) for $(BOARD) board" -E \ $(patsubst %,-b arch/$(ARCH)/dts/%.dtb_HS,$(subst ",,$(CONFIG_OF_LIST)))
-OF_LIST_TARGETS = $(patsubst %,arch/$(ARCH)/dts/%.dtb,$(subst ",,$(CONFIG_OF_LIST))) -$(OF_LIST_TARGETS): dtbs
- u-boot-nodtb.bin_HS: u-boot-nodtb.bin FORCE $(call if_changed,k3secureimg)
-u-boot.img_HS: u-boot-nodtb.bin_HS u-boot.img $(patsubst %.dtb,%.dtb_HS,$(OF_LIST_TARGETS)) FORCE +u-boot.img_HS: u-boot-nodtb.bin_HS u-boot.img dtbs FORCE $(call if_changed,mkimage)
Because then no one will depend on the _HS versions of the .dtb files, so they wont get built.
Indeed, missed that.
I agree it's a huge mess right now, the "correct" thing to do would be for the %.dtb_HS files to depend on their corresponding %.dtb files from which they are generated. Which they do, kinda. But Make doesn't seem smart enough to know that at the start, it checks for the .dtb files and fails instantly during the first round of parsing, due to the .dtb files not existing and no rules existing for them. The .dtb files are not being generated by a rule with their name so Make doesn't understand they will be generated later, because U-Boot uses scripting for dtbs and uses target "dtbs". So we are stuck depending on that until someone does some major rework of the U-Boot/Linux makefiles.. Feel free :D
loop_cmd = $(echo-cmd) $(cmd_$(1)) || exit;
dtbs_HS: dtbs FORCE $(foreach dtb, ..., $(call, loop_cmd, ...))
This avoids per-dtb rules and dependencies. But it will require some variant of k3secureimg that takes the source as a parameter and not from the rule's dependency list. This is the same pattern I currently play with for injecting the public key into dtbs for fit image verification. I'll see if I can get something building with am65x_hs_evm_a53_defconfig to validate a concrete fix proposal.
Again my question: What can be the other broken dependency that causes "make -j" failures?
Jan