Re: [RFC] mm/hotplug: Make get_nid_for_pfn() work with HAVE_ARCH_PFN_VALID

[Michal Hocko]

On Thu 21-03-19 13:38:20, Anshuman Khandual wrote:
> Memory hot remove uses get_nid_for_pfn() while tearing down linked sysfs
> entries between memory block and node. It first checks pfn validity with
> pfn_valid_within() before fetching nid. With CONFIG_HOLES_IN_ZONE config
> (arm64 has this enabled) pfn_valid_within() calls pfn_valid().
> 
> pfn_valid() is an arch implementation on arm64 (CONFIG_HAVE_ARCH_PFN_VALID)
> which scans all mapped memblock regions with memblock_is_map_memory(). This
> creates a problem in memory hot remove path which has already removed given
> memory range from memory block with memblock_[remove|free] before arriving
> at unregister_mem_sect_under_nodes().

Could you be more specific on what is the actual problem please? It
would be also helpful to mention when is the memblock[remove|free]
called actually.

> During runtime memory hot remove get_nid_for_pfn() needs to validate that
> given pfn has a struct page mapping so that it can fetch required nid. This
> can be achieved just by looking into it's section mapping information. This
> adds a new helper pfn_section_valid() for this purpose. Its same as generic
> pfn_valid().

I have to say I do not like this. Having pfn_section_valid != pfn_valid_within
is just confusing as hell. pfn_valid_within should return true whenever
a struct page exists and it is sensible (same like pfn_valid). So it
seems that this is something to be solved on that arch specific side of
pfn_valid.
-- 
Michal Hocko
SUSE Labs