On Wed, Sep 21, 2016 at 12:39:51PM +0530, Aneesh Kumar K.V wrote:
What I was checking was how will one mark a node movable in ppc64 ? I
don't see ppc64 code doing the equivalent of memblock_mark_hotplug().


Post boot, the marking mechanism is not necessary. You can create a movable node by putting all of the node's memory into ZONE_MOVABLE during the hotplug.

So when you say "Onlining memory into ZONE_MOVABLE requires
CONFIG_MOVABLE_NODE" where is that restriction ?. IIUC,
should_add_memory_movable() will only return ZONE_MOVABLE only if it is
non empty and MOVABLE_NODE will create a ZONE_MOVABLE zone by default
only if it finds a memblock marked hotpluggable. So wondering if we
are not calling memblock_mark_hotplug() how is it working. Or am I
missing something ?

You are looking at the addition step of hotplug. You're correct there, the memory is added to the default zone, not ZONE_MOVABLE. The transition to ZONE_MOVABLE takes place during the onlining step. In online_pages():

        zone = move_pfn_range(zone_shift, pfn, pfn + nr_pages);

The reason we need CONFIG_MOVABLE_NODE is right before that:

        if ((zone_idx(zone) > ZONE_NORMAL ||
            online_type == MMOP_ONLINE_MOVABLE) &&
            !can_online_high_movable(zone))
                return -EINVAL;

where can_online_high_movable() is defined like this:

        #ifdef CONFIG_MOVABLE_NODE
        /*
         * When CONFIG_MOVABLE_NODE, we permit onlining of a node which doesn't 
have
         * normal memory.
         */
        static bool can_online_high_movable(struct zone *zone)
        {
                return true;
        }
        #else /* CONFIG_MOVABLE_NODE */
        /* ensure every online node has NORMAL memory */
        static bool can_online_high_movable(struct zone *zone)
        {
                return node_state(zone_to_nid(zone), N_NORMAL_MEMORY);
        }
        #endif /* CONFIG_MOVABLE_NODE */

To be more clear, I can change the commit log to say "Onlining all of a node's memory into ZONE_MOVABLE requires CONFIG_MOVABLE_NODE".

--
Reza Arbab

Reply via email to