Re: [PATCH v3] zsmalloc: fix migrate_zspage-zs_free race condition

[Sergey Senozhatsky]

On (01/18/16 17:20), Minchan Kim wrote:
[..]
> > > oh... good find! lost release semantic of unpin_tag()...
> > 
> > Ah, release semantic, good point indeed. OK then we need the v2 approach again,
> > with WRITE_ONCE() in record_obj(). Or some kind of record_obj_release() with
> > release semantic, which would be a bit more effective, but I guess migration is
> > not that critical path to be worth introducing it.
> 
> WRITE_ONCE in record_obj would add more memory operations in obj_malloc
> but I don't feel it's too heavy in this phase so,
> 
> How about this? Junil, Could you resend patch if others agree this?
> Thanks.
> 
> +/*
> + * record_obj updates handle's value to free_obj and it shouldn't
> + * invalidate lock bit(ie, HANDLE_PIN_BIT) of handle, otherwise
> + * it breaks synchronization using pin_tag(e,g, zs_free) so let's
> + * keep the lock bit.
> + */
>  static void record_obj(unsigned long handle, unsigned long obj)
>  {
> -	*(unsigned long *)handle = obj;
> +	int locked = (*(unsigned long *)handle) & (1<<HANDLE_PIN_BIT);
> +	unsigned long val = obj | locked;
> +
> +	/*
> +	 * WRITE_ONCE could prevent store tearing like below
> +	 * *(unsigned long *)handle = free_obj
> +	 * *(unsigned long *)handle |= locked;
> +	 */
> +	WRITE_ONCE(*(unsigned long *)handle, val);
>  }

given that memory barriers are also compiler barriers, wouldn't

	record_obj()
	{
		barrier
		*(unsigned long *)handle) = new
	}

suffice?

	-ss