[PATCH 5/6] ptrace: introduce PTRACE_SET_SYSCALL_INFO request

[Oleg Nesterov]

On 01/08, Dmitry V. Levin wrote:
>
> +ptrace_set_syscall_info_entry(struct task_struct *child, struct pt_regs *regs,
> +			      struct ptrace_syscall_info *info)
> +{
...
> +	syscall_set_nr(child, regs, nr);
> +	syscall_set_arguments(child, regs, args);
> +	if (nr == -1) {
> +		/*
> +		 * When the syscall number is set to -1, the syscall will be
> +		 * skipped.  In this case also set the syscall return value to
> +		 * -ENOSYS, otherwise on some architectures the corresponding
> +		 * struct pt_regs field will remain unchanged.
> +		 *
> +		 * Note that on some architectures syscall_set_return_value()
> +		 * modifies one of the struct pt_regs fields also modified by
> +		 * syscall_set_arguments(), so the former should be called
> +		 * after the latter.
> +		 */
> +		syscall_set_return_value(child, regs, -ENOSYS, 0);
> +	}

This doesn't look nice to me...

We don't need this syscall_set_return_value(ENOSYS) on x86, right?

So perhaps we should move this "if (nr == -1) code  into
syscall_set_nr/syscall_set_arguments on those "some architectures" which
actually need it ?

Oleg.