Messages in this thread | | | Subject | Re: [PATCH v2 2/4] selftests: Fix seccomp to support relocatable build (O=objdir) | From | Shuah Khan <> | Date | Tue, 10 Mar 2020 17:11:47 -0600 |
| |
On 3/5/20 10:42 AM, Kees Cook wrote: > On Thu, Mar 05, 2020 at 09:41:34AM -0700, Shuah Khan wrote: >> On 3/4/20 7:20 PM, Kees Cook wrote: >>> Instead of the TEST_CUSTOM_PROGS+all dance, you can just add an explicit >>> dependency, with the final seccomp/Makefile looking like this: >>> >>> >>> # SPDX-License-Identifier: GPL-2.0 >>> CFLAGS += -Wl,-no-as-needed -Wall >>> LDFLAGS += -lpthread >>> >>> TEST_GEN_PROGS := seccomp_bpf seccomp_benchmark >>> >> >> TEST_CUSTOM_PROGS is for differentiating test programs that >> can't use lib.mk generic rules. It is appropriate to use >> for seccomp_bpf > > I don't follow? This suggested Makefile works for me (i.e. it can use > the lib.mk generic rules since CFLAGS and LDFLAGS can be customized > first, and it just adds an additional dependency). >
Yeah. TEST_CUSTOM_PROGS isn't really needed for this custom case. I can refine it and get rid of the dependency.
>>> include ../lib.mk >>> >>> # Additional dependencies >>> $(OUTPUT)/seccomp_bpf: ../kselftest_harness.h > > BTW, I see a lot of other targets that use kselftest_harness.h appear to > be missing this Makefile dependency, but that's a different problem. :) > >>> (Though this fails in the same way as above when run from the top-level >>> directory.) >>> >> >> I didn't see this because I have been the same directory I used >> for relocated cross-build kernel. :( >> >> Thanks for testing this. I know the problem here. all is a dependency >> for install step and $(OUTPUT) is referencing the objdir before it >> gets created. It is a Makefile/lib.mk problem to fix. >>
I was way off with my analysis. :(
>> I will do a separate patch for this. This will show up in any test >> that is using $(OUTPUT) to relocate objects mainly the ones that >> require custom build rule like seeccomp. > > Okay, cool. It looked to me like it lost track of the top level source > directory (i.e. "make: entering $output" ... "can't find > ../other/files") >
Odd that you would have empty objdir in the cross-compile case.
In the cross-compile case, you would have cross-built kernel first in the object directory. Your objdir won't be empty.
This is no different from kselftest build dependency on kernel build even when srcdir=objdir
So for cross-build case, the following is the workflow to build kernel first and then the tests:
make O=/../objdir ARCH=arm64 HOSTCC=gcc CROSS_COMPILE=aarch64-linux-gnu- defconfig
make O=/../objdir ARCH=arm64 HOSTCC=gcc CROSS_COMPILE=aarch64-linux-gnu- all
make kselftest-install O=/../objdir ARCH=arm64 HOSTCC=gcc CROSS_COMPILE=aarch64-linux-gnu- TARGETS=seccomp
You can isolate a single test when you are do native build:
make kselftest-install O=/../objdir TARGETS=seccomp
The above won't fail even if objdir doesn't exist and/or empty.
thanks, -- Shuah
| |