grub: Allow build on armv7ve/a with softfp
PATCH 1/1 grub: Allow build on armv7ve/a with softfp
From: Andrei Gherzan <[email protected]>
Grub tries to force the compilation in softfp mode for some known target CPU. That is because it doesn't use floats or doubles and there are known limitations of using a configuration with hardfp. There are though target CPUs that grub won't know how to configure failing with something similar to:
| checking for options to get soft-float... no
| configure: error: could not force soft-float
This change relaxes a bit the COMPATIBLE_HOST restriction to only apply for hardfp configuration even for arm7ve/a.
--- meta/recipes-bsp/grub/grub2.inc | 15 ++++++++++++--- 1 file changed, 12 insertions(+), 3 deletions(-)
meta/recipes-bsp/grub/grub2.inc +12 -3
@@ -43,9 +43,18 @@ CVE_CHECK_IGNORE += "CVE-2021-46705" DEPENDS = "flex-native bison-native gettext-native" -COMPATIBLE_HOST = '(x86_64.*|i.86.*|arm.*|aarch64.*|riscv.*)-(linux.*|freebsd.*)' -COMPATIBLE_HOST:armv7a = 'null' -COMPATIBLE_HOST:armv7ve = 'null' +GRUB_COMPATIBLE_HOST = '(x86_64.*|i.86.*|arm.*|aarch64.*|riscv.*)-(linux.*|freebsd.*)' +COMPATIBLE_HOST = "${GRUB_COMPATIBLE_HOST}" +# Grub doesn't support hard float toolchain and won't be able to forcefully +# disable it on some of the target CPUs. See 'configure.ac' for +# supported/unsupported CPUs in hardfp. +COMPATIBLE_HOST:armv7a = "${@'null' if d.getVar('TUNE_CCARGS_MFLOAT') == 'hardfp' else d.getVar('GRUB_COMPATIBLE_HOST')}" +COMPATIBLE_HOST:armv7ve = "${@'null' if d.getVar('TUNE_CCARGS_MFLOAT') == 'hardfp' else d.getVar('GRUB_COMPATIBLE_HOST')}" + +# Grub doesn't use float or doubles and tries to disable forcefully floating +# point at configure time. That dynamic flags configuration doesn't always work +# (for example on arm 32bit) so force the compilation in softfp. +TUNE_CCARGS_MFLOAT:class-target = "softfp" # configure.ac has code to set this automagically from the target tuple # but the OE freeform one (core2-foo-bar-linux) don't work with that.
Signature · 1 line
2.25.1
Hi,
Quoted 47 lines · Andrei Gherzan
From: Andrei Gherzan <[email protected]>
Grub tries to force the compilation in softfp mode for some known target CPU. That is because it doesn't use floats or doubles and there are known limitations of using a configuration with hardfp. There are though target CPUs that grub won't know how to configure failing with something similar to:
| checking for options to get soft-float... no
| configure: error: could not force soft-floatThis change relaxes a bit the
COMPATIBLE_HOSTrestriction to only apply for hardfp configuration even for arm7ve/a.Signed-off-by Andrei Gherzan <[email protected]>--- meta/recipes-bsp/grub/grub2.inc | 15 ++++++++++++--- 1 file changed, 12 insertions(+), 3 deletions(-)meta/recipes-bsp/grub/grub2.inc +0
-1@@ -43,9 +43,18 @@ CVE_CHECK_IGNORE += "CVE-2021-46705" DEPENDS = "flex-native bison-native gettext-native" -COMPATIBLE_HOST ='(x86_64.*|i.86.*|arm.*|aarch64.*|riscv.*)-(linux.*|freebsd.*)'
meta/recipes-bsp/grub/grub2.inc +1
-2-COMPATIBLE_HOST:armv7a = 'null' -COMPATIBLE_HOST:armv7ve = 'null' +GRUB_COMPATIBLE_HOST ='(x86_64.*|i.86.*|arm.*|aarch64.*|riscv.*)-(linux.*|freebsd.*)'
meta/recipes-bsp/grub/grub2.inc +2
-0+COMPATIBLE_HOST = "${GRUB_COMPATIBLE_HOST}" +# Grub doesn't support hard float toolchain and won't be able toforcefully
meta/recipes-bsp/grub/grub2.inc +3
-0+# disable it on some of the target CPUs. See 'configure.ac' for +# supported/unsupported CPUs in hardfp. +COMPATIBLE_HOST:armv7a = "${@'null' if d.getVar('TUNE_CCARGS_MFLOAT')== 'hardfp' else
d.getVar('GRUB_COMPATIBLE_HOST')}" +COMPATIBLE_HOST:armv7ve = "${@'null' ifd.getVar('TUNE_CCARGS_MFLOAT')== 'hardfp' elsed.getVar('GRUB_COMPATIBLE_HOST')}"meta/recipes-bsp/grub/grub2.inc +2
-0+ +# Grub doesn't use float or doubles and tries to disable forcefullyfloating
+# point at configure time. That dynamic flags configuration doesn't always work
meta/recipes-bsp/grub/grub2.inc CVE_CHECK_IGNORE += "CVE-2021-46705" · line 57
Quoted 2 lines
meta/recipes-bsp/grub/grub2.inc +2
-0+# (for example on arm 32bit) so force the compilation in softfp. +TUNE_CCARGS_MFLOAT:class-target = "softfp"
Please ignore this v1 patch as the above was not supposed to be part of the patch. I've sent a cleaned up v2: https://lists.openembedded.org/g/openembedded-core/message/170695
Thanks,
Andrei