configs/kontron_bl_imx8mm_defconfig: fix build failure with u-boot 2022.04
Subsystems (2)
PATCH 1/1 configs/kontron_bl_imx8mm_defconfig: fix build failure with u-boot 2022.04
With U-Boot 2022.04 libuuid is required for building the host tool mkeficapsule. The lib is included in the util-linux package. Thus the BR2_TARGET_UBOOT_NEEDS_UTIL_LINUX config is needed.
In addition an U-boot patch is required to fix an issue in U-Boot for linking the mkeficapsule tool against -luuid and -lgnutls.
--- ...le-use-pkg-config-to-get-luuid-and-l.patch | 33 +++++++++++++++++++ configs/kontron_bl_imx8mm_defconfig | 2 ++ 2 files changed, 35 insertions(+) create mode 100644 board/kontron/bl-imx8mm/patches/uboot/2022.04/0001-tools-mkeficapsule-use-pkg-config-to-get-luuid-and-l.patch
board/kontron/bl-imx8mm/patches/uboot/2022.04/0001-tools-mkeficapsule-use-pkg-config-to-get-luuid-and-l.patch new file +33 -0
@@ -0,0 +1,33 @@ +From f3523977e8f5f6b2173708777001332431ebc609 Mon Sep 17 00:00:00 2001 +From: Heiko Thiery <[email protected]> +Date: Tue, 19 Jul 2022 16:17:09 +0200 +Subject: [PATCH 1/2] tools: mkeficapsule: use pkg-config to get -luuid and -lgnutls + +Instead of hardcoding -luuid -lgnutls as the flags needed to build +mkeficapsule, use pkg-config when available. + +We gracefully fallback on the previous behavior of hardcoding -luuid +-lgnutls if pkg-config is not available or fails with an error. + +Signed-off-by: Heiko Thiery <[email protected]> +--- + tools/Makefile | 3 ++- + 1 file changed, 2 insertions(+), 1 deletion(-) + +diff --git a/tools/Makefile b/tools/Makefile +index 9f2339666a..9f6b282ad8 100644 +--- a/tools/Makefile ++++ b/tools/Makefile +@@ -242,7 +242,8 @@ hostprogs-$(CONFIG_MIPS) += mips-relocs + hostprogs-$(CONFIG_ASN1_COMPILER) += asn1_compiler + HOSTCFLAGS_asn1_compiler.o = -idirafter $(srctree)/include + +-HOSTLDLIBS_mkeficapsule += -lgnutls -luuid ++HOSTLDLIBS_mkeficapsule += \ ++ $(shell pkg-config --libs gnutls uuid 2> /dev/null || echo "-lgnutls -luuid") + hostprogs-$(CONFIG_TOOLS_MKEFICAPSULE) += mkeficapsule + + # We build some files with extra pedantic flags to try to minimize things +-- +2.30.2 +
configs/kontron_bl_imx8mm_defconfig +2 -0
@@ -5,6 +5,7 @@ BR2_ARM_FPU_VFPV3=y # System BR2_TARGET_GENERIC_GETTY_PORT="ttymxc2" +BR2_GLOBAL_PATCH_DIR="board/kontron/bl-imx8mm/patches" # Kernel BR2_LINUX_KERNEL=y
@@ -48,6 +49,7 @@ BR2_TARGET_UBOOT_NEEDS_GNUTLS=y BR2_TARGET_UBOOT_NEEDS_ATF_BL31=y BR2_TARGET_UBOOT_NEEDS_ATF_BL31_BIN=y BR2_TARGET_UBOOT_NEEDS_IMX_FIRMWARE=y +BR2_TARGET_UBOOT_NEEDS_UTIL_LINUX=y BR2_TARGET_UBOOT_FORMAT_CUSTOM=y BR2_TARGET_UBOOT_FORMAT_CUSTOM_NAME="flash.bin" BR2_TARGET_UBOOT_SPL=y
Signature · 6 lines
2.20.1 _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
Hello Heiko,
Thanks for the patch!
Quoted 3 lines · On Sun, 7 Aug 2022 23:03:18 +0200 Heiko Thiery <[email protected]> wrote:
+-HOSTLDLIBS_mkeficapsule += -lgnutls -luuid
++HOSTLDLIBS_mkeficapsule += \
++ $(shell pkg-config --libs gnutls uuid 2>/dev/null|| echo "-lgnutls -luuid")
Do you understand why this is needed?
In the build failure reported by Gitlab CI, it fails to find libgnutls, even though host-gnutls has been built before. So the -lgnutls that was in HOSTLDLIBS_mkeficapsule should have been sufficient, provided $(HOST_LDFLAGS) are properly used.
Of course, I have nothing against using pkg-config, but I was just wondering if we have an explanation on what was going on. And is pkg-config fixing it because it's return a full path to the library?
Obviously, one issue with your patch is that it's going to fix the problem only for the kontron defconfig, which is why I was interested to see if there wasn't a better solution.
Thanks!
Thomas
Signature · 7 lines
Thomas Petazzoni, co-owner and CEO, Bootlin Embedded Linux and Kernel engineering and training https://bootlin.com _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
Hi Thomas,
Am Mo., 8. Aug. 2022 um 10:41 Uhr schrieb Thomas Petazzoni <[email protected]>:
Quoted 12 lines
Hello Heiko,
Thanks for the patch!
Quoted 3 lines · On Sun, 7 Aug 2022 23:03:18 +0200 Heiko Thiery <[email protected]> wrote:
+-HOSTLDLIBS_mkeficapsule += -lgnutls -luuid
++HOSTLDLIBS_mkeficapsule += \
++ $(shell pkg-config --libs gnutls uuid 2>/dev/null|| echo "-lgnutls -luuid")
Quoted 6 lines
Do you understand why this is needed?
In the build failure reported by Gitlab CI, it fails to find libgnutls, even though host-gnutls has been built before. So the -lgnutls that was in
HOSTLDLIBS_mkeficapsuleshould have been sufficient, provided $(HOST_LDFLAGS) are properly used.
Without the patch:
--- 8< ---
/usr/bin/gcc -O2 -isystem/home/hthiery/sources/mainline/buildroot/output/host/include -Wp,-MD,tools/.mkeficapsule.d -Wall -Wstrict-prototypes -O2
-fomit-frame-pointer -std=gnu11 -DCONFIG_FIT_SIGNATURE
-DCONFIG_FIT_SIGNATURE_MAX_SIZE=0xffffffff -DCONFIG_FIT_CIPHER -include ./include/compiler.h -idirafterinclude -idirafter./arch/arm/include -I./scripts/dtc/libfdt -I./tools -DUSE_HOSTCC -D__KERNEL_STRICT_NAMES -D_GNU_SOURCE -otools/mkeficapsule tools/mkeficapsule.c -lgnutls -luuid
--- 8< ---
With the patch:
--- 8< ---
/usr/bin/gcc -O2 -isystem/home/hthiery/sources/mainline/buildroot/output/host/include -Wp,-MD,tools/.mkeficapsule.d -Wall -Wstrict-prototypes -O2
-fomit-frame-pointer -std=gnu11 -DCONFIG_FIT_SIGNATURE
-DCONFIG_FIT_SIGNATURE_MAX_SIZE=0xffffffff -DCONFIG_FIT_CIPHER -include ./include/compiler.h -idirafterinclude -idirafter./arch/arm/include -I./scripts/dtc/libfdt -I./tools -DUSE_HOSTCC -D__KERNEL_STRICT_NAMES -D_GNU_SOURCE -o tools/mkeficapsule tools/mkeficapsule.c -L/home/hthiery/sources/mainline/buildroot/output/host/lib -lgnutls -luuid
--- 8< ---
The -lgnuts and -luuid seems not to be sufficient. Using pkg-config will also add the -L<PATH-TO-LIBS>.
Quoted 16 lines
Of course, I have nothing against using pkg-config, but I was just wondering if we have an explanation on what was going on. And is pkg-config fixing it because it's return a full path to the library?
Obviously, one issue with your patch is that it's going to fix the problem only for the kontron defconfig, which is why I was interested to see if there wasn't a better solution.
Thanks!
Thomas
--
Thomas Petazzoni, co-owner and CEO, Bootlin Embedded Linux and Kernel engineering and training
https://bootlin.com
Signature · 5 lines
Heiko _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
Hello Heiko,
Quoted 28 lines · On Tue, 9 Aug 2022 08:45:09 +0200 Heiko Thiery <[email protected]> wrote:
Without the patch:
--- 8< ---
/usr/bin/gcc-O2 -isystem/home/hthiery/sources/mainline/buildroot/output/host/include(quoted diff) +0
-6-Wp,-MD,tools/.mkeficapsule.d -Wall -Wstrict-prototypes -O2 -fomit-frame-pointer -std=gnu11 -DCONFIG_FIT_SIGNATURE -DCONFIG_FIT_SIGNATURE_MAX_SIZE=0xffffffff -DCONFIG_FIT_CIPHER -include ./include/compiler.h -idirafterinclude -idirafter./arch/arm/include -I./scripts/dtc/libfdt -I./tools -DUSE_HOSTCC -D__KERNEL_STRICT_NAMES -D_GNU_SOURCE -o
tools/mkeficapsuletools/mkeficapsule.c-lgnutls -luuid
--- 8< ---With the patch:
--- 8< ---
/usr/bin/gcc-O2 -isystem/home/hthiery/sources/mainline/buildroot/output/host/include(quoted diff) +0
-6-Wp,-MD,tools/.mkeficapsule.d -Wall -Wstrict-prototypes -O2 -fomit-frame-pointer -std=gnu11 -DCONFIG_FIT_SIGNATURE -DCONFIG_FIT_SIGNATURE_MAX_SIZE=0xffffffff -DCONFIG_FIT_CIPHER -include ./include/compiler.h -idirafterinclude -idirafter./arch/arm/include -I./scripts/dtc/libfdt -I./tools -DUSE_HOSTCC -D__KERNEL_STRICT_NAMES -D_GNU_SOURCE -o
tools/mkeficapsuletools/mkeficapsule.c(quoted diff) +0
-3-L/home/hthiery/sources/mainline/buildroot/output/host/lib -lgnutls -luuid --- 8< ---
Quoted 2 lines
The -lgnuts and -luuid seems not to be sufficient. Using pkg-config will also add the -L<PATH-TO-LIBS>.
But this is not normal, because:
UBOOT_MAKE_OPTS += \
CROSS_COMPILE="$(TARGET_CROSS)" \
ARCH=$(UBOOT_ARCH) \
HOSTCC="$(HOSTCC) $(subst -I/,-isystem /,$(subst -I /,-isystem /,$(HOST_CFLAGS)))" \
HOSTLDFLAGS="$(HOST_LDFLAGS)" \
$(call qstrip,$(BR2_TARGET_UBOOT_CUSTOM_MAKEOPTS))see we're passing HOSTLDFLAGS here? It contains -L$(HOST_DIR)/lib. So it means there is also something wrong in U-Boot in that it doesn't use $(HOSTLDFLAGS).
Best regards,
Thomas
Signature · 7 lines
Thomas Petazzoni, co-owner and CEO, Bootlin Embedded Linux and Kernel engineering and training https://bootlin.com _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
Hello Heiko,
Quoted 9 lines · On Tue, 9 Aug 2022 10:06:02 +0200 Thomas Petazzoni via buildroot…
But this is not normal, because:
UBOOT_MAKE_OPTS += \ CROSS_COMPILE="$(TARGET_CROSS)" \ ARCH=$(UBOOT_ARCH) \ HOSTCC="$(HOSTCC) $(subst -I/,-isystem /,$(subst -I /,-isystem /,$(HOST_CFLAGS)))" \ HOSTLDFLAGS="$(HOST_LDFLAGS)" \ $(call qstrip,$(BR2_TARGET_UBOOT_CUSTOM_MAKEOPTS))
Quoted 3 lines
see we're passing HOSTLDFLAGS here? It contains -L$(HOST_DIR)/lib. So it means there is also something wrong in U-Boot in that it doesn't use $(HOSTLDFLAGS).
Could you try this change in U-Boot:
scripts/Makefile.host +1 -1
@@ -89,7 +89,7 @@ hostcxx_flags = -Wp,-MD,$(depfile) $(__hostcxx_flags) # Create executable from a single .c file # host-csingle -> Executable quiet_cmd_host-csingle = HOSTCC $@ - cmd_host-csingle = $(HOSTCC) $(hostc_flags) -o $@ $< \ + cmd_host-csingle = $(HOSTCC) $(hostc_flags) $(KBUILD_HOSTLDFLAGS) -o $@ $< \ $(KBUILD_HOSTLDLIBS) $(HOSTLDLIBS_$(@F)) $(host-csingle): $(obj)/%: $(src)/%.c FORCE $(call if_changed_dep,host-csingle)
(without your pkg-config patch, of course).
Indeed, scripts/Makefile.host is a copy of the kernel's scripts/Makefile.host, and the kernel copy does use KBUILD_HOSTLDFLAGS when building host tools that are based on a single source file, but for some reason the U-Boot copy does not have it, which doesn't make much sense.
I think it doesn't mean your pkg-config patch is bogus: it still makes sense to use pkg-config when possible, but I also think not passing KBUILD_HOSTLDFLAGS is incorrect.
Of course, if it works, I think both patches should be submitted to U-Boot upstream.
Could you have a look?
Thanks a lot!
Thomas
Signature · 7 lines
Thomas Petazzoni, co-owner and CEO, Bootlin Embedded Linux and Kernel engineering and training https://bootlin.com _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot
Hi Thomas,
Am Mo., 15. Aug. 2022 um 12:14 Uhr schrieb Thomas Petazzoni <[email protected]>:
Quoted 35 lines
Hello Heiko,
Quoted 9 lines · On Tue, 9 Aug 2022 10:06:02 +0200 Thomas Petazzoni via buildroot…
But this is not normal, because:
UBOOT_MAKE_OPTS += \ CROSS_COMPILE="$(TARGET_CROSS)" \ ARCH=$(UBOOT_ARCH) \ HOSTCC="$(HOSTCC) $(subst -I/,-isystem /,$(subst -I /,-isystem /,$(HOST_CFLAGS)))" \ HOSTLDFLAGS="$(HOST_LDFLAGS)" \ $(call qstrip,$(BR2_TARGET_UBOOT_CUSTOM_MAKEOPTS))Quoted 3 lines
see we're passing HOSTLDFLAGS here? It contains -L$(HOST_DIR)/lib. So it means there is also something wrong in U-Boot in that it doesn't use $(HOSTLDFLAGS).
Could you try this change in U-Boot:
scripts/Makefile.host +1
-1@@ -89,7 +89,7 @@ hostcxx_flags = -Wp,-MD,$(depfile) $(__hostcxx_flags) # Create executable from a single .c file # host-csingle -> Executable quiet_cmd_host-csingle = HOSTCC $@ - cmd_host-csingle = $(HOSTCC) $(hostc_flags) -o $@ $< \ + cmd_host-csingle = $(HOSTCC) $(hostc_flags) $(KBUILD_HOSTLDFLAGS) -o $@ $< \ $(KBUILD_HOSTLDLIBS) $(HOSTLDLIBS_$(@F)) $(host-csingle): $(obj)/%: $(src)/%.c FORCE $(call if_changed_dep,host-csingle)
Quoted 7 lines
(without your pkg-config patch, of course).
Indeed,
scripts/Makefile.hostis a copy of the kernel'sscripts/Makefile.host, and the kernel copy does useKBUILD_HOSTLDFLAGSwhen building host tools that are based on a single source file, but for some reason the U-Boot copy does not have it, which doesn't make much sense.
Indeed ... I just added this change to my local copy of u-boot rebuild it without the previous patch and it works.
Quoted 7 lines
I think it doesn't mean your pkg-config patch is bogus: it still makes sense to use pkg-config when possible, but I also think not passing
KBUILD_HOSTLDFLAGSis incorrect.Of course, if it works, I think both patches should be submitted to U-Boot upstream.
I will prepare an U-Boot patch and send it to the mailing list. I saw that you cahnged the patch in patchwork to "changes requested". Would you accept it with the "new" patch?
Quoted 1 line
Could you have a look?
without the change:
--- 8< ---
/usr/bin/gcc -O2 -isystem/home/hthiery/sources/mainline/buildroot/output/host/include -Wp,-MD,tools/.mkeficapsule.d -Wall -Wstrict-prototypes -O2
-fomit-frame-pointer -std=gnu11 -DCONFIG_FIT_SIGNATURE
-DCONFIG_FIT_SIGNATURE_MAX_SIZE=0xffffffff -DCONFIG_FIT_CIPHER -include ./include/compiler.h -idirafterinclude -idirafter./arch/arm/include -I./scripts/dtc/libfdt -I./tools -DUSE_HOSTCC -D__KERNEL_STRICT_NAMES -D_GNU_SOURCE -otools/mkeficapsule tools/mkeficapsule.c -lgnutls -luuid
--- 8< ---
with the change you suggested:
--- 8< ---
/usr/bin/gcc -O2 -isystem/home/hthiery/sources/mainline/buildroot/output/host/include -Wp,-MD,tools/.mkeficapsule.d -Wall -Wstrict-prototypes -O2
-fomit-frame-pointer -std=gnu11 -DCONFIG_FIT_SIGNATURE
-DCONFIG_FIT_SIGNATURE_MAX_SIZE=0xffffffff -DCONFIG_FIT_CIPHER -include ./include/compiler.h -idirafterinclude -idirafter./arch/arm/include -I./scripts/dtc/libfdt -I./tools -DUSE_HOSTCC -D__KERNEL_STRICT_NAMES -D_GNU_SOURCE -L/home/hthiery/sources/mainline/buildroot/output/host/lib -Wl,-rpath,/home/hthiery/sources/mainline/buildroot/output/host/lib -otools/mkeficapsule tools/mkeficapsule.c -lgnutls -luuid
--- 8< ---
Quoted 7 lines
Thanks a lot!
Thomas
--
Thomas Petazzoni, co-owner and CEO, Bootlin Embedded Linux and Kernel engineering and training
https://bootlin.com
Signature · 5 lines
Heiko _______________________________________________ buildroot mailing list [email protected] https://lists.buildroot.org/mailman/listinfo/buildroot