Re: [PATCH] KVM: selftests: Guard execinfo.h inclusion for non-glibc builds

[Hisam Mehboob]

On 4/1/26 19:01, Sean Christopherson wrote:
> On Tue, Mar 31, 2026, Shuah Khan wrote:
> > On 3/25/26 12:47, Hisam Mehboob wrote:
> > > On 3/25/26 23:03, Shuah Khan wrote:
> > > > On 3/24/26 12:02, Shuah Khan wrote:
> > > > > On 3/18/26 18:08, Hisam Mehboob wrote:
> > > > > > The backtrace() function and execinfo.h are GNU extensions available
> > > > > > in glibc but not in non-glibc C libraries such as musl. Building KVM
> > > > > > selftests with musl-gcc fails with:
> > > > > >
> > > > > >     lib/assert.c:9:10: fatal error: execinfo.h: No such file or directory
> > > > > >
> > > > > > Guard the inclusion of execinfo.h under #ifdef __GLIBC__, and wrap
> > > > > > all backtrace() usage under the same guard with a fallback message
> > > > > > for non-glibc builds indicating that stack traces are not available.
> > > > > >
> > > > > > Unlike the approach of adding a weak stub for backtrace(), this
> > > > > > explicitly handles the non-glibc case rather than silently providing
> > > > > > an empty implementation.
> > > > > >
> > > > > > Link: https://lore.kernel.org/kvm/20250829142556.72577-7- aqibaf@xxxxxxxxxx/
> > > > > >
> > > > > > Suggested-by: Aqib Faruqui <aqibaf@xxxxxxxxxx>
> > > > > > Signed-off-by: Hisam Mehboob <hisamshar@xxxxxxxxx>
> > > > > > ---
> > > > > >    tools/testing/selftests/kvm/lib/assert.c | 7 +++++++
> > > > > >    1 file changed, 7 insertions(+)
> > > > > >
> > > > > > diff --git a/tools/testing/selftests/kvm/lib/assert.c b/tools/ testing/selftests/kvm/lib/assert.c
> > > > > > index b49690658c60..3442b80c37c1 100644
> > > > > > --- a/tools/testing/selftests/kvm/lib/assert.c
> > > > > > +++ b/tools/testing/selftests/kvm/lib/assert.c
> > > > > > @@ -6,7 +6,9 @@
> > > > > >     */
> > > > > >    #include "test_util.h"
> > > > > > +#ifdef __GLIBC__
> > > > > >    #include <execinfo.h>
> > > > > > +#endif
> > > > > > Is __GLIBC__ defined in musl-gcc? Looks like that is the case with the
> > > > > error?
> > > >
> > > > If __GLIBC__ isn't there you shouldn't see this error because the include
> > > > is in - this error doesn't make sense if __GLIBC__ isn't defined. What
> > > > am I missing?
> > >
> > > To clarify the compiler error you mentioned: the error log in the commit
> > > message shows the failure that occurs before this patch is applied. Because
> > > musl-gcc doesn't define __GLIBC__, the original unconditional <execinfo.h>
> > > inclusion causes the build to fail. The #ifdef in my patch was intended to
> > > fix that exact failure.
> > >
> > > > +#ifdef __GLIBC__
> > > >    #include <execinfo.h>
> > > > +#endif
> > > >
> > > > Also check tools/testing/selftests/bpf/test_progs.c - I think backtrace()
> > > > stub needs be defined only for the !__GLIBC__ case
> > >
> > > Looking at how bpf/test_progs.c handles it, I agree the weak stub approach
> > > is much cleaner. I will implement it so that it still prints an explicit
> > > warning message when a trace is unavailable.
>
> I disagree.  _If_ we didn't need the __GLIBC__ #ifdef, then I would be in favor
> of __weak, but since the #ifdeffery is needed, using an #ifdef and a __weak symbol
> is double the ugliness.
>
> IMO, the way to make this less ugly is to using a single #ifdef and a local stub.
>
> diff --git a/tools/testing/selftests/kvm/lib/assert.c b/tools/testing/selftests/kvm/lib/assert.c
> index b49690658c60..315175ca49f1 100644
> --- a/tools/testing/selftests/kvm/lib/assert.c
> +++ b/tools/testing/selftests/kvm/lib/assert.c
> @@ -6,11 +6,13 @@
>    */
>   #include "test_util.h"
>   
> -#include <execinfo.h>
>   #include <sys/syscall.h>
>   
>   #include "kselftest.h"
>   
> +#ifdef __GLIBC__
> +#include <execinfo.h>
> +
>   /* Dumps the current stack trace to stderr. */
>   static void __attribute__((noinline)) test_dump_stack(void);
>   static void test_dump_stack(void)
> @@ -57,6 +59,9 @@ static void test_dump_stack(void)
>          system(cmd);
>   #pragma GCC diagnostic pop
>   }
> +#else
> +static void test_dump_stack(void) {}
> +#endif
>   
>   static pid_t _gettid(void)
>   {

Thanks for the suggestion. I will send a v2 implementing this approach.