Message ID | 1452869536-13918-1-git-send-email-arnout@mind.be |
---|---|
State | Superseded |
Headers | show |
>>>>> "Arnout" == Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be> writes: > libnspr currently passes --enable-thumb2 if the CPU has thumb > instructions. This option will pass -mthumb to the compiler. However, > if an external multilib toolchain is used that has a thumb-specific > variant, it will try to use that one. But we only copy a single variant > to the sysroot, so the build will fail with: > In order to fix this, only enable thumb2 when we are really using > thumb2. Note that it is still necessary to pass this option, cfr. > http://autobuild.buildroot.org/results/d7323831372050e425a34f5104a46d8cbd6be214 > We could have chosen to still enable thumb2 with the internal toolchain. > However, that would mean we create a discrepancy between the result of > building with the same (buildroot) toolchain when it is used as an > internal one or as an external one. To avoid potentially surprising > results, just stick to the option chosen for the rest of the rootfs. Why do we actually ever pass --enable-thumb2 / --disable-thumb2? As far as I can see, the only thing this does is adding -marm / -mthumb to CFLAGS, and we already pass the correct -m<arm/thumb> option in the toolchain wrapper.
On 15-01-16 21:43, Peter Korsgaard wrote: >>>>>> "Arnout" == Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be> writes: > > > libnspr currently passes --enable-thumb2 if the CPU has thumb > > instructions. This option will pass -mthumb to the compiler. However, > > if an external multilib toolchain is used that has a thumb-specific > > variant, it will try to use that one. But we only copy a single variant > > to the sysroot, so the build will fail with: > > > In order to fix this, only enable thumb2 when we are really using > > thumb2. Note that it is still necessary to pass this option, cfr. > > http://autobuild.buildroot.org/results/d7323831372050e425a34f5104a46d8cbd6be214 > > > We could have chosen to still enable thumb2 with the internal toolchain. > > However, that would mean we create a discrepancy between the result of > > building with the same (buildroot) toolchain when it is used as an > > internal one or as an external one. To avoid potentially surprising > > results, just stick to the option chosen for the rest of the rootfs. > > Why do we actually ever pass --enable-thumb2 / --disable-thumb2? As far > as I can see, the only thing this does is adding -marm / -mthumb to > CFLAGS, and we already pass the correct -m<arm/thumb> option in the > toolchain wrapper. Well, if you don't pass anything, it will check if the _compiler_ supports thumb by compiling (not linking) something which checks for __thumb2__. So we have to pass something. Er, on second look, it uses that compile to set MOZ_THUMB2 and then doesn't do anything with it... Still, there is a good chance that upstream will in the future do something with that symbol, so I think it's safer to explicitly en/disable thumb2. Regards, Arnout
diff --git a/package/libnspr/libnspr.mk b/package/libnspr/libnspr.mk index 8e58986..c9bfa0c 100644 --- a/package/libnspr/libnspr.mk +++ b/package/libnspr/libnspr.mk @@ -50,7 +50,7 @@ LIBNSPR_INSTALL_STAGING_OPTS = DESTDIR=$(STAGING_DIR) LIBRARY= install endif ifeq ($(BR2_arm),y) -ifeq ($(BR2_ARM_CPU_HAS_THUMB2),y) +ifeq ($(BR2_ARM_INSTRUCTIONS_THUMB2),y) LIBNSPR_CONF_OPTS += --enable-thumb2 else LIBNSPR_CONF_OPTS += --disable-thumb2
libnspr currently passes --enable-thumb2 if the CPU has thumb instructions. This option will pass -mthumb to the compiler. However, if an external multilib toolchain is used that has a thumb-specific variant, it will try to use that one. But we only copy a single variant to the sysroot, so the build will fail with: In order to fix this, only enable thumb2 when we are really using thumb2. Note that it is still necessary to pass this option, cfr. http://autobuild.buildroot.org/results/d7323831372050e425a34f5104a46d8cbd6be214 We could have chosen to still enable thumb2 with the internal toolchain. However, that would mean we create a discrepancy between the result of building with the same (buildroot) toolchain when it is used as an internal one or as an external one. To avoid potentially surprising results, just stick to the option chosen for the rest of the rootfs. Signed-off-by: Arnout Vandecappelle (Essensium/Mind) <arnout@mind.be> --- Build-tested with various combinations of thumb, thumb2, floating point, and external toolchains. I didn't test with an internal toolchain but what could go wrong there? :-) --- package/libnspr/libnspr.mk | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)