On 5/11/15, 3:21 PM, "Shuah Khan" <shuahkh@xxxxxxxxxxxxxxx> wrote: >>>>No need for a new pull request. Have you seen these errors before: >>>> >>>>make[2]: Entering directory >>>>'/mnt/data/lkml/linux-kselftest/tools/testing/selftests/futex' >>>>for DIR in functional; do make -C $DIR all ; done >>>>make[3]: Entering directory >>>>'/mnt/data/lkml/linux-kselftest/tools/testing/selftests/futex/functiona >>>>l' >>>>gcc -g -O2 -Wall -D_GNU_SOURCE -I../include -I../../ -lpthread -lrt >>>>futex_requeue_pi.c ../include/futextest.h -o futex_requeue_pi >>>>/tmp/cc2UgUVs.o: In function `create_rt_thread': >>>>/mnt/data/lkml/linux-kselftest/tools/testing/selftests/futex/functional >>>>/fu >>>>tex_requeue_pi.c:102: >>>>undefined reference to `pthread_create' >>>>/tmp/cc2UgUVs.o: In function `unit_test': >>>>/mnt/data/lkml/linux-kselftest/tools/testing/selftests/futex/functional >>>>/fu >>>>tex_requeue_pi.c:342: >>>>undefined reference to `pthread_join' >>>>/mnt/data/lkml/linux-kselftest/tools/testing/selftests/futex/functional >>>>/fu >>>>tex_requeue_pi.c:347: >>>>undefined reference to `pthread_join' >>>>/mnt/data/lkml/linux-kselftest/tools/testing/selftests/futex/functional >>>>/fu >>>>tex_requeue_pi.c:346: >>>>undefined reference to `pthread_join' >>>>collect2: error: ld returned 1 exit status >>>><builtin>: recipe for target 'futex_requeue_pi' failed >>>>make[3]: *** [futex_requeue_pi] Error 1 >>>>make[3]: Leaving directory >>>>'/mnt/data/lkml/linux-kselftest/tools/testing/selftests/futex/functiona >>>>l' >>>>Makefile:7: recipe for target 'all' failed >>>> >>>>I am running make kselftest target when I saw the above build failures. >>>>kselftest run doesn't fail which is good, however futex tests won't >>>>run with this failure. >>> >>>I have not seen these errors whilst developing with futextest for the >>>kernel. That looks like you may be missing the pthread development >>>headers >>>from your build machine. >>> >>>Those are provided by libc6-dev on my Debian systems. >>> >>I do have lib6-dev installed. timers uses pthread compiles fine. >>So does mqueue. I am not sure what is going on with futex tests >>though. I am guessing it has to do with library link order. >>The following fixed it for me: >>-LDFLAGS := $(LDFLAGS) -lpthread -lrt >>+LDFLAGS := $(LDFLAGS) -pthread -lrt >>Could you please make this change and resend the patch series. >>I prefer patch series over pull request. >>thanks, >>-- Shuah > >Sorry full diff: > > >diff --git a/tools/testing/selftests/futex/functional/Makefile >b/tools/testing/selftests/futex/functional/Makefile >index e64d43b..b8a2e9b 100644 >--- a/tools/testing/selftests/futex/functional/Makefile >+++ b/tools/testing/selftests/futex/functional/Makefile >@@ -1,6 +1,6 @@ >INCLUDES := -I../include -I../../ >CFLAGS := $(CFLAGS) -g -O2 -Wall -D_GNU_SOURCE $(INCLUDES) >-LDFLAGS := $(LDFLAGS) -lpthread -lrt >+LDFLAGS := $(LDFLAGS) -pthread -lrt I'm happy to do that, but I would like to make sure I'm doing the right thing. I'm building with: $ gcc --version gcc (Debian 4.9.2-10) 4.9.2 What's there now works for me: LDFLAGS := $(LDFLAGS) -lpthread -lrt What you propose also works for me: LDFLAGS := $(LDFLAGS) -pthread -lrt I notice that most other test cases list -lrt first, including timers: LDFLAGS += -lrt -lpthread Which works for me for timers - does that work for you? I assume so. It also works for futexes for me. I've found references out there suggesting -lrt should always come before -lpthread. I suspect you are correct as -pthread provides flags for both the preprocessor and the linker, however, I'm still concerned about the ordering. John - did you deliberately choose to use "-lrt -lpthread" ? If ordering matters and -pthread is preferred over -lpthread, then perhaps the "right way" is: LDFLAGS += -lrt -pthread Since -pthread also impacts the pre-processor, it seems that if used, it should be present in both LDFLAGS as well as CFLAGS. However, I've also read that -pthread is a non-standard compiler option and is best avoided. If so, and if the reversed order works for you, "-lrt -lpthread", perhaps that would be the "right fix". I'd like to answer this once and for all for all tests we add to selftest. -- Darren Hart Intel Open Source Technology Center -- To unsubscribe from this list: send the line "unsubscribe linux-api" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
