Re: [RFC PATCH] selftests: use INSTALL_HDR_PATH instead of relative paths

On 14/01/16 03:37, Michael Ellerman wrote:

On Mon, 2016-01-11 at 14:22 +0000, Sudeep Holla wrote:

quoted

The selftests build fails or picks up wrong headers when cross-compiling
and build directory is different from source as the headers are referred
using the relative paths.

This patch use $INSTALL_HDR_PATH to fix the compilation. It also removes
installing the headers.

Cc: Shuah Khan <redacted>
Signed-off-by: Sudeep Holla <redacted>
---
  tools/testing/selftests/membarrier/Makefile | 2 +-
  tools/testing/selftests/memfd/Makefile      | 2 +-
  tools/testing/selftests/net/Makefile        | 2 +-
  tools/testing/selftests/vm/Makefile         | 7 ++-----
  4 files changed, 5 insertions(+), 8 deletions(-)

Hi,

When I was cross-compiling these tests for arm{32,64}, I faced issues.
With limited knowledge I have on this build system, I came up with this
patch. Please treat this as bug report if this approach makes no-sense
and suggest on how to fix it.

Yeah sorry that doesn't work.

If $(INSTALL_HDR_PATH) isn't defined you end up with:

   gcc -Wall -I /include  -o compaction_test compaction_test.c -lrt
                ^

Ah OK, I got carried away by below in the main Makefile.

#Default location for installed headers
export INSTALL_HDR_PATH = $(objtree)/usr

Which is wrong.

INSTALL_HDR_PATH will not be defined when the tests are built standalone, ie.
like:

   $ make -C tools/testing/selftests

Thanks for pointing this out. I had totally forgotten about that.

This has been on my todo list to fix, but it doesn't actually bother me so I
haven't got around to it.

Does the patch below work for you? It uses INSTALL_HDR_PATH if it is defined,
or otherwise KERNEL_HEADERS if it is defined, or finally it guesses at
../../../usr/include (ie. the base of the source tree).

Yes it does work fine. So feel free to add my tested-by when you post
the patch. Thanks for the quick and clean fix.

-- 
Regards,
Sudeep