When I attempt to build perf on a system with slang but without libelf,
'make' would wrongly complain that the slang library could not be found.
It turns out that this was happening because we are not filtering -lelf
from EXTLIBS early enough. As a result, the library test for slang
(ditto for gtk, libaudit, etc) erroneously passes -lelf to try-cc, which
of course fails on a system without libelf.
This patch makes the filtering of -lelf from EXTLIBS occur right after
testing for libelf support, so that the subsequent library tests will
not erroneously pass -lelf to try-cc when building without libelf
support.
Cc: Peter Zijlstra <[email protected]>
Cc: Paul Mackerras <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Arnaldo Carvalho de Melo <[email protected]>
Cc: Jiri Olsa <[email protected]>
Signed-off-by: Patrick Palka <[email protected]>
---
tools/perf/Makefile | 2 --
tools/perf/config/Makefile | 4 ++++
2 files changed, 4 insertions(+), 2 deletions(-)
diff --git a/tools/perf/Makefile b/tools/perf/Makefile
index 64c043b..94635c8 100644
--- a/tools/perf/Makefile
+++ b/tools/perf/Makefile
@@ -443,8 +443,6 @@ ifneq ($(OUTPUT),)
endif
ifdef NO_LIBELF
-EXTLIBS := $(filter-out -lelf,$(EXTLIBS))
-
# Remove ELF/DWARF dependent codes
LIB_OBJS := $(filter-out $(OUTPUT)util/symbol-elf.o,$(LIB_OBJS))
LIB_OBJS := $(filter-out $(OUTPUT)util/dwarf-aux.o,$(LIB_OBJS))
diff --git a/tools/perf/config/Makefile b/tools/perf/config/Makefile
index 5f6f9b3..1748767 100644
--- a/tools/perf/config/Makefile
+++ b/tools/perf/config/Makefile
@@ -174,6 +174,10 @@ else
endif # SOURCE_LIBELF
endif # NO_LIBELF
+ifdef NO_LIBELF
+EXTLIBS := $(filter-out -lelf,$(EXTLIBS))
+endif
+
ifndef NO_LIBELF
CFLAGS += -DLIBELF_SUPPORT
FLAGS_LIBELF=$(CFLAGS) $(LDFLAGS) $(EXTLIBS)
--
1.8.4.rc3
Em Wed, Oct 23, 2013 at 10:51:09PM -0400, Patrick Palka escreveu:
> When I attempt to build perf on a system with slang but without libelf,
> 'make' would wrongly complain that the slang library could not be found.
>
> It turns out that this was happening because we are not filtering -lelf
> from EXTLIBS early enough. As a result, the library test for slang
> (ditto for gtk, libaudit, etc) erroneously passes -lelf to try-cc, which
> of course fails on a system without libelf.
>
> This patch makes the filtering of -lelf from EXTLIBS occur right after
> testing for libelf support, so that the subsequent library tests will
> not erroneously pass -lelf to try-cc when building without libelf
> support.
Can you please check with the perf/core branch in my tree or Ingo's?
git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux.git perf/core
or Ingo's:
git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git perf/core
The feature detection was reworked and perhaps it is fixed there.
- Arnaldo
On Thu, Oct 24, 2013 at 4:13 PM, Arnaldo Carvalho de Melo
<[email protected]> wrote:
> Em Wed, Oct 23, 2013 at 10:51:09PM -0400, Patrick Palka escreveu:
>> When I attempt to build perf on a system with slang but without libelf,
>> 'make' would wrongly complain that the slang library could not be found.
>>
>> It turns out that this was happening because we are not filtering -lelf
>> from EXTLIBS early enough. As a result, the library test for slang
>> (ditto for gtk, libaudit, etc) erroneously passes -lelf to try-cc, which
>> of course fails on a system without libelf.
>>
>> This patch makes the filtering of -lelf from EXTLIBS occur right after
>> testing for libelf support, so that the subsequent library tests will
>> not erroneously pass -lelf to try-cc when building without libelf
>> support.
>
> Can you please check with the perf/core branch in my tree or Ingo's?
>
> git://git.kernel.org/pub/scm/linux/kernel/git/acme/linux.git perf/core
>
> or Ingo's:
>
> git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip.git perf/core
>
> The feature detection was reworked and perhaps it is fixed there.
>
> - Arnaldo
I can't reproduce the detection failure in your branch, and a quick inspection
shows that the new feature-detection code is not subject to the kind of bug
that my patch fixed. So consider my patch obsolete.
Thanks,
Patrick
Em Thu, Oct 24, 2013 at 04:36:47PM -0400, Patrick Palka escreveu:
> On Thu, Oct 24, 2013 at 4:13 PM, Arnaldo Carvalho de Melo
> > The feature detection was reworked and perhaps it is fixed there.
> I can't reproduce the detection failure in your branch, and a quick inspection
> shows that the new feature-detection code is not subject to the kind of bug
> that my patch fixed. So consider my patch obsolete.
Thanks for checking!
- Arnaldo