[Devel] Re: [RFC v14-rc3][PATCH 15/36] c/r of restart-blocks

[Oren Laadan]

Dan Smith wrote:
> OL> +int cr_retval_restart(struct cr_ctx *ctx)
> OL> +{
> OL> +	struct pt_regs *regs = task_pt_regs(current);
> OL> +	int ret = 0;
> OL> +
> OL> +	/*
> OL> +	 * The retval should be either zero if the checkpointed task
> OL> +	 * had been in user-space when frozen, or the retval from the
> OL> +	 * syscall that had been interrupted then.
> OL> +	 *
> OL> +	 * In the latter, if the syscall succeeded (perhaps partially)
> OL> +	 * then the retval is non-negative. If it failed, the error
> OL> +	 * may be one of -ERESTART... gang, interpreted in the signal
> OL> +	 * handling code. In restart it must happen, too.
> OL> +	 *
> OL> +	 * To force execution of the signal handler now, too, we fake
> OL> +	 * a signal to ourselves (a la freeze/thaw) when ret < 0.
> OL> +	 */
> OL> +
> OL> +	/* were we from a system call?  if so, get old error/retval */
> OL> +	if (syscall_get_nr(current, regs) >= 0)
> OL> +		ret = syscall_get_error(current, regs);

The test "were we from a system call ?" is implemented differently on
the s390, for example. Compare the code in handle_signal(), whose logic
I follow.

> OL> +	/* old error ?  if so, make sure signal handling kicks in */
> OL> +	if (ret < 0)
> OL> +		set_tsk_thread_flag(current, TIF_SIGPENDING);
> OL> +	return ret;

> OL> +}
> 
> Does this need to be arch-specific?  At first glance, I can't think of
> anything I'd need to change for s390, for example.
> 

So the code above is similar to that in handle_signal() of x86_32 arch.
The same code in other architectures looks differently, and I didn't
investigate whether the code in other architectures can be transformed
to a common form.

Oren.
_______________________________________________
Containers mailing list
Containers at lists.linux-foundation.org
https://lists.linux-foundation.org/mailman/listinfo/containers