Reza Arbab <ar...@linux.vnet.ibm.com> writes:
> On Mon, Sep 19, 2016 at 11:59:35AM +0530, Aneesh Kumar K.V wrote:
>>Movable node also does.
>>What is the impact of that. Do we need changes equivalent to that ? Also
>>where are we marking the nodes which can be hotplugged, ie where do we
>>do memblock_mark_hotplug() ?
> These are related to the mechanism x86 uses to create movable nodes at
> boot. The SRAT is parsed to mark any hotplug memory. That marking is
> used later when initializing ZONE_MOVABLE for each node. 
> The bottom-up allocation is due to a timing issue . There is a window
> where kernel memory may be allocated before the SRAT is parsed. Any
> bottom-up allocations done during that time will likely be in the same
> (nonmovable) node as the kernel image.
> On power, I don't think we have a heuristic equivalent to that SRAT
> memory hotplug info. So, we'll be limited to dynamically adding movable
> nodes after boot.
> 1. http://events.linuxfoundation.org/sites/events/files/lcjp13_chen.pdf
> 2. commit 79442ed189ac ("mm/memblock.c: introduce bottom-up allocation
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().
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 ?