[PATCH] tools: Override makefile ARCH variable if defined, but empty

Quentin Monnet qmo at kernel.org
Thu Nov 21 03:29:59 PST 2024


2024-11-20 22:04 UTC-0800 ~ Namhyung Kim <namhyung at kernel.org>
> On Wed, Nov 20, 2024 at 02:25:22PM +0100, Björn Töpel wrote:
>> Björn Töpel <bjorn at kernel.org> writes:
>>
>>> From: Björn Töpel <bjorn at rivosinc.com>
>>>
>>> There are a number of tools (bpftool, selftests), that require a
>>> "bootstrap" build. Here, a bootstrap build is a build host variant of
>>> a target. E.g., assume that you're performing a bpftool cross-build on
>>> x86 to riscv, a bootstrap build would then be an x86 variant of
>>> bpftool. The typical way to perform the host build variant, is to pass
>>> "ARCH=" in a sub-make. However, if a variable has been set with a
>>> command argument, then ordinary assignments in the makefile are
>>> ignored.
>>>
>>> This side-effect results in that ARCH, and variables depending on ARCH
>>> are not set.
>>>
>>> Workaround by overriding ARCH to the host arch, if ARCH is empty.
>>>
>>> Fixes: 8859b0da5aac ("tools/bpftool: Fix cross-build")
>>> Signed-off-by: Björn Töpel <bjorn at rivosinc.com>
> 
> Reviewed-by: Namhyung Kim <namhyung at kernel.org>


Acked-by: Quentin Monnet <qmo at kernel.org>


>> Arnaldo/Palmer/Quentin:
>>
>> A bit unsure what tree this patch should go. It's very important for the
>> RISC-V builds, so maybe via Palmer's RISC-V tree?
> 
> I think it'd be best to route this through the bpf tree as it seems the
> main target is bpftool.  But given the size and the scope of the change,
> it should be fine with perf-tools or RISC-V tree.


The bpf tree would make sense to me as well (but I don't merge patches
myself; let me Cc BPF maintainers).

Quentin



More information about the linux-riscv mailing list