perf tests: Fix lib path detection for non-x86 architectures

[PATCH] perf tests: Fix lib path detection for non-x86 architectures

Posted by Suchit Karunakaran 2 months, 2 weeks ago

The existing Makefile logic for determining library paths was
hardcoded for x86 architectures, causing incorrect behavior
on other platforms. This patch implements a portable solution
using system bit detection.

Signed-off-by: Suchit Karunakaran <suchitkarunakaran@gmail.com>
---
 tools/perf/tests/make | 9 ++++++---
 1 file changed, 6 insertions(+), 3 deletions(-)

diff --git a/tools/perf/tests/make b/tools/perf/tests/make
index 0ee94caf9ec1..565522408f99 100644
--- a/tools/perf/tests/make
+++ b/tools/perf/tests/make
@@ -53,9 +53,12 @@ endif
 
 include $(srctree)/tools/scripts/Makefile.arch
 
-# FIXME looks like x86 is the only arch running tests ;-)
-# we need some IS_(32/64) flag to make this generic
-ifeq ($(ARCH)$(IS_64_BIT), x861)
+BITS := $(shell getconf LONG_BIT)
+IS_64_BIT ?= 1
+ifeq ($(BITS), 32)
+IS_64_BIT = 0
+endif
+ifeq ($(IS_64_BIT), 1)
 lib = lib64
 else
 lib = lib
-- 
2.50.1

Re: [PATCH] perf tests: Fix lib path detection for non-x86 architectures

Posted by Namhyung Kim 2 months, 2 weeks ago

On Sun, Jul 20, 2025 at 02:29:05PM +0530, Suchit Karunakaran wrote:
> The existing Makefile logic for determining library paths was
> hardcoded for x86 architectures, causing incorrect behavior
> on other platforms. This patch implements a portable solution
> using system bit detection.
> 
> Signed-off-by: Suchit Karunakaran <suchitkarunakaran@gmail.com>
> ---
>  tools/perf/tests/make | 9 ++++++---
>  1 file changed, 6 insertions(+), 3 deletions(-)
> 
> diff --git a/tools/perf/tests/make b/tools/perf/tests/make
> index 0ee94caf9ec1..565522408f99 100644
> --- a/tools/perf/tests/make
> +++ b/tools/perf/tests/make
> @@ -53,9 +53,12 @@ endif
>  
>  include $(srctree)/tools/scripts/Makefile.arch
>  
> -# FIXME looks like x86 is the only arch running tests ;-)
> -# we need some IS_(32/64) flag to make this generic
> -ifeq ($(ARCH)$(IS_64_BIT), x861)
> +BITS := $(shell getconf LONG_BIT)
> +IS_64_BIT ?= 1
> +ifeq ($(BITS), 32)
> +IS_64_BIT = 0
> +endif
> +ifeq ($(IS_64_BIT), 1)
>  lib = lib64
>  else
>  lib = lib

A dummy question: Does all other architectures have lib64 vs lib
separation?

Thanks,
Namhyung

> -- 
> 2.50.1
>

Re: [PATCH] perf tests: Fix lib path detection for non-x86 architectures

Posted by Suchit K 2 months, 2 weeks ago

>
> A dummy question: Does all other architectures have lib64 vs lib
> separation?
>

I had assumed there would always be symlinks, but thanks for pointing
that out. After your question, I checked various architectures like
x86, ARM, SPARC, s390x, etc and only x86 had both lib and lib64 (with
symlinks). On the others, even for 64-bit systems, only a lib
directory existed. I also realized this behavior seems to depend on
the distro. For example, multiarch distros like Debian use separate
directories for lib32 and lib64, and a lib symlink pointing to
/usr/lib. On the other hand, Arch Linux has both lib and lib64 as
symlinks to /usr/lib. Would it be reasonable if we create a symlink
named lib64 for non-x86 architectures? I'd appreciate your thoughts on
this. Thanks!

Re: [PATCH] perf tests: Fix lib path detection for non-x86 architectures

Posted by Namhyung Kim 2 months, 2 weeks ago

On Mon, Jul 21, 2025 at 11:10:18AM +0530, Suchit K wrote:
> >
> > A dummy question: Does all other architectures have lib64 vs lib
> > separation?
> >
> 
> I had assumed there would always be symlinks, but thanks for pointing
> that out. After your question, I checked various architectures like
> x86, ARM, SPARC, s390x, etc and only x86 had both lib and lib64 (with
> symlinks). On the others, even for 64-bit systems, only a lib
> directory existed. I also realized this behavior seems to depend on
> the distro. For example, multiarch distros like Debian use separate
> directories for lib32 and lib64, and a lib symlink pointing to
> /usr/lib. On the other hand, Arch Linux has both lib and lib64 as
> symlinks to /usr/lib. Would it be reasonable if we create a symlink
> named lib64 for non-x86 architectures? I'd appreciate your thoughts on
> this. Thanks!

I'd be intrusive if we create a new symlink.  Probably we need to check
if there's lib64 directory first and use it for 64 bit build.  But I
feel like this needs more testing.

Can you share what's the problem exactly?

Thanks,
Namhyung

Re: [PATCH] perf tests: Fix lib path detection for non-x86 architectures

Posted by Suchit K 2 months, 2 weeks ago

>
> I'd be intrusive if we create a new symlink.  Probably we need to check
> if there's lib64 directory first and use it for 64 bit build.  But I
> feel like this needs more testing.
>
> Can you share what's the problem exactly?
>
> Thanks,
> Namhyung
>

Yeah I agree it needs more testing. And yes we can check if the lib64
folder exists and then create a symlink. If there's no lib64 folder,
we can create a symlink named lib64 to the lib folder. I assume this
will cover most of the architectures since the lib folder existed in
almost all the architectures that I tested.