Re: [PATCH v2 2/4] tools/nolibc: Provide vfork()

["Thomas Weißschuh" <linux@weissschuh.net>]

On 2025-06-10 13:29:45+0100, Mark Brown wrote:
> To allow testing of vfork() support in the arm64 basic-gcs test provide an
> implementation for nolibc, using the vfork() syscall if one is available
> and otherwise clone3(). We implement in terms of clone3() since the order
> of the arguments for clone() varies between architectures.
> 
> Signed-off-by: Mark Brown <broonie@kernel.org>
> ---
>  tools/include/nolibc/sys.h | 27 +++++++++++++++++++++++++++
>  1 file changed, 27 insertions(+)
> 
> diff --git a/tools/include/nolibc/sys.h b/tools/include/nolibc/sys.h
> index aabac97a7fb0..5932ae8828a1 100644
> --- a/tools/include/nolibc/sys.h
> +++ b/tools/include/nolibc/sys.h
> @@ -22,6 +22,7 @@
>  #include <linux/time.h>
>  #include <linux/auxvec.h>
>  #include <linux/fcntl.h> /* for O_* and AT_* */
> +#include <linux/sched.h> /* for clone_args */
>  #include <linux/stat.h>  /* for statx() */
>  
>  #include "errno.h"
> @@ -340,6 +341,32 @@ pid_t fork(void)
>  	return __sysret(sys_fork());
>  }
>  
> +static __attribute__((unused))
> +pid_t sys_vfork(void)
> +{
> +#if defined(__NR_vfork)
> +	return my_syscall0(__NR_vfork);
> +#elif defined(__NR_clone3)
> +	/*
> +	 * clone() could be used but has different argument orders per
> +	 * architecture.
> +	 */
> +	struct clone_args args = {
> +		.flags		= CLONE_VM | CLONE_VFORK,
> +		.exit_signal	= SIGCHLD,
> +	};
> +
> +	return my_syscall2(__NR_clone3, &args, sizeof(args));
> +#else
> +	return __nolibc_enosys(__func__);
> +#endif
> +}

It seems that on SPARC __NR_vfork has the same non-standard return value
as __NR_fork. It therefore needs the same special handling in arch-sparc.h.
Maybe with a helper to avoid the duplication.

If you want I can fix this up when applying.
Otherwise the series looks good.

> +
> +static __attribute__((unused))
> +pid_t vfork(void)
> +{
> +	return __sysret(sys_vfork());
> +}
>  
>  /*
>   * int fsync(int fd);
> 
> -- 
> 2.39.5
>