Re: [PATCH v3] x86/cpu: Use SERIALIZE in sync_core() when available

[Dave Hansen]

On 8/6/20 12:25 PM, Ricardo Neri wrote:
>  static inline void sync_core(void)
>  {
>  	/*
> -	 * There are quite a few ways to do this.  IRET-to-self is nice
> +	 * Hardware can do this for us if SERIALIZE is available. Otherwise,
> +	 * there are quite a few ways to do this.  IRET-to-self is nice

This seems to imply that things other than SERIALIZE aren't the hardware
doing this.  All of these methods are equally architecturally
serializing *and* provided by the hardware.

I also don't quite get the point of separating the comments from the
code.  Shouldn't this be:

	/*
	 * The SERIALIZE instruction is the most straightforward way to
	 * do this but it not universally available.
	 */
	if (static_cpu_has(X86_FEATURE_SERIALIZE)) {
		asm volatile(__ASM_SERIALIZE ::: "memory");
		return;
	}

	/*
	 * For all other processors, IRET-to-self is nice ...
	 */
	iret_to_self();