Re: WTH is going on with memory hotplug sysf interface (was: Re: [RFC PATCH] mm, hotplug: get rid of auto_online_blocks)

[Michal Hocko <mhocko@kernel.org>]

Let's add Andi

On Fri 10-03-17 16:53:33, Michal Hocko wrote:
> On Fri 10-03-17 14:58:07, Michal Hocko wrote:
> [...]
> > This would explain why onlining from the last block actually works but
> > to me this sounds like a completely crappy behavior. All we need to
> > guarantee AFAICS is that Normal and Movable zones do not overlap. I
> > believe there is even no real requirement about ordering of the physical
> > memory in Normal vs. Movable zones as long as they do not overlap. But
> > let's keep it simple for the start and always enforce the current status
> > quo that Normal zone is physically preceeding Movable zone.
> > Can somebody explain why we cannot have a simple rule for Normal vs.
> > Movable which would be:
> > 	- block [pfn, pfn+block_size] can be Normal if
> > 	  !zone_populated(MOVABLE) || pfn+block_size < ZONE_MOVABLE->zone_start_pfn
> > 	- block [pfn, pfn+block_size] can be Movable if
> > 	  !zone_populated(NORMAL) || ZONE_NORMAL->zone_end_pfn < pfn
> 
> OK, so while I was playing with this setup some more I probably got why
> this is done this way. All new memblocks are added to the zone Normal
> where they are accounted as spanned but not present. When we do
> online_movable we just cut from the end of the Normal zone and move it
> to Movable zone. This sounds really awkward. What was the reason to go
> this way? Why cannot we simply add those pages to the zone at the online
> time?

Answering to myself. So the reason seems to be 9d99aaa31f59 ("[PATCH]
x86_64: Support memory hotadd without sparsemem") which is no longer
true because 
config MEMORY_HOTPLUG
        bool "Allow for memory hot-add"
        depends on SPARSEMEM || X86_64_ACPI_NUMA
        depends on ARCH_ENABLE_MEMORY_HOTPLUG
        depends on COMPILE_TEST || !KASAN

so it is either SPARSEMEM or X86_64_ACPI_NUMA that would have to be enabled.
config X86_64_ACPI_NUMA
        def_bool y
        prompt "ACPI NUMA detection"
        depends on X86_64 && NUMA && ACPI && PCI
        select ACPI_NUMA

But I do not see any way how to enable anything but SPARSEMEM for x86_64
choice
        prompt "Memory model"
        depends on SELECT_MEMORY_MODEL
        default DISCONTIGMEM_MANUAL if ARCH_DISCONTIGMEM_DEFAULT
        default SPARSEMEM_MANUAL if ARCH_SPARSEMEM_DEFAULT
        default FLATMEM_MANUAL

ARCH_SPARSEMEM_DEFAULT is 32b only
config ARCH_DISCONTIGMEM_DEFAULT
        def_bool y
        depends on NUMA && X86_32

and ARCH_SPARSEMEM_DEFAULT is enabeld on 64b. So I guess whatever was
the reason to add this code back in 2006 is not true anymore. So I am
really wondering. Do we absolutely need to assign pages which are not
onlined yet to the ZONE_NORMAL unconditionally? Why cannot we put them
out of any zone and wait for memory online operation to put them where
requested?
-- 
Michal Hocko
SUSE Labs