[PATCH] perf: bpf: Fix build breakage due to libbpf
Naveen N. Rao
naveen.n.rao at linux.vnet.ibm.com
Tue Dec 15 02:17:31 AEDT 2015
On 2015/12/14 07:53PM, Wang Nan wrote:
> Hi Naveen,
>
> On 2015/12/14 18:50, Naveen N. Rao wrote:
> >perf build is currently (v4.4-rc5) broken on powerpc:
> >
> >bpf.c:28:4: error: #error __NR_bpf not defined. libbpf does not support
> >your arch.
> > # error __NR_bpf not defined. libbpf does not support your arch.
> > ^
> >
> >Fix this by including tools/perf/config/Makefile.arch for the proper
> >$ARCH macro. While at it, remove redundant LP64 macro definition.
> >
> >Signed-off-by: Naveen N. Rao <naveen.n.rao at linux.vnet.ibm.com>
> >---
> > tools/lib/bpf/Makefile | 3 ++-
> > 1 file changed, 2 insertions(+), 1 deletion(-)
> >
> >diff --git a/tools/lib/bpf/Makefile b/tools/lib/bpf/Makefile
> >index 636e3dd..050e0e8 100644
> >--- a/tools/lib/bpf/Makefile
> >+++ b/tools/lib/bpf/Makefile
> >@@ -31,7 +31,8 @@ INSTALL = install
> > DESTDIR ?=
> > DESTDIR_SQ = '$(subst ','\'',$(DESTDIR))'
> >-LP64 := $(shell echo __LP64__ | ${CC} ${CFLAGS} -E -x c - | tail -n 1)
> >+include $(srctree)/tools/perf/config/Makefile.arch
> >+
> > ifeq ($(LP64), 1)
> > libdir_relative = lib64
> > else
>
> Are you doing cross compiling? In this case you should provide an 'ARCH' to
> make
> through cmdline. For example, this is how yocto help me build perf on
> aarch64:
>
> $ make -C /patch/to/kernel/tools/perf O=/path/to/output \
> CROSS_COMPILE=aarch64-oe-linux- ARCH=arm64 \
> CC=aarch64-oe-linux-gcc --sysroot=... AR=aarch64-oe-linux-ar ...
I am not cross-compiling - the build error was with trying to build perf
on ppc64le. Even with cross-compiling (and explicitly specifying $ARCH),
Makefile.arch should still work.
>
> If you include Makefile.arch, I think you are seeking for a 'uname -m'
> result,
> then you are not doing cross compiling, right? In this case what you need is
> providing
> a __NR_bpf entry for your platform, because in some cases that value is
> overrided because
> we have
>
> $ find ./tools/ -name "unistd*"
> ./tools/perf/util/include/asm/unistd_64.h
> ./tools/perf/util/include/asm/unistd_32.h
>
> You can find the reason of these two files through:
> eae7a755ee81129370c8f555b0d5672e6673735d
I thought of just defining __NR_bpf for powerpc, but it will still break
perf build for most other architectures. Moreover,
tools/lib/bpf/Makefile explicitly includes headers from the linux kernel
build tree (rather than the system headers) and I thought this was
specifically to pull in __NR_bpf, among others -- just that we were not
properly including the right headers since $ARCH was not defined. More
specifically:
$ make V=1
<snip>
gcc -Wp,-MD,./.bpf.o.d,-MT,bpf.o -g -Wall -DHAVE_LIBELF_MMAP_SUPPORT
-DHAVE_ELF_GETPHDRNUM_SUPPORT -Wbad-function-cast
-Wdeclaration-after-statement -Wformat-security -Wformat-y2k
-Winit-self -Wmissing-declarations -Wmissing-prototypes
-Wnested-externs -Wno-system-headers -Wold-style-definition -Wpacked
-Wredundant-decls -Wshadow -Wstrict-aliasing=3 -Wstrict-prototypes
-Wswitch-default -Wswitch-enum -Wundef -Wwrite-strings -Wformat
-Werror -Wall -fPIC -I. -I/root/linux/tools/include
-I/root/linux/arch//include/uapi -I/root/linux/include/uapi
-D"BUILD_STR(s)=#s" -c -o bpf.o bpf.c
Notice // in -I/root/linux/arch//include/uapi. Hence, my patch to ensure
the right headers get included.
I'm thinking the explicit __NR_bpf definitions are only needed for x86
-- since I don't see the definitions in the kernel tree headers.
Regards,
Naveen
>
> I posted a patch by replying this mail. I think you can find it through [1]
> soon.
> I don't have PPC environment to check it. Could you please help me check
> this patch
> in your environment?
>
> [1] http://lkml.kernel.org/g/1450092932-123588-1-git-send-email-wangnan0@huawei.com
>
>
> _______________________________________________
> Linuxppc-dev mailing list
> Linuxppc-dev at lists.ozlabs.org
> https://lists.ozlabs.org/listinfo/linuxppc-dev
More information about the Linuxppc-dev
mailing list