Re: [RFC PATCH 2/3] mm/memory_hotplug: Create __shrink_pages and move it to offline_pages

From: David Hildenbrand
Date: Wed Aug 08 2018 - 03:51:57 EST


On 08.08.2018 09:38, Oscar Salvador wrote:
> On Tue, Aug 07, 2018 at 06:13:45PM -0400, Jerome Glisse wrote:
>>> And since we know for sure that memhotplug-code cannot call it with ZONE_DEVICE,
>>> I think this can be done easily.
>>
>> This might change down road but for now this is correct. They are
>> talks to enumerate device memory through standard platform mechanisms
>> and thus the kernel might see new types of resources down the road and
>> maybe we will want to hotplug them directly from regular hotplug path
>> as ZONE_DEVICE (lot of hypothetical at this point ;)).
>
> Well, I think that if that happens this whole thing will become
> much easier, since we will not have several paths for doing the same thing.
>
> Another thing that I realized is that while we want to move all operation-pages
> from remove_memory() path to offline_pages(), this can get tricky.
>
> Unless I am missing something, the devices from HMM and devm are not being registered
> against "memory_subsys" struct, and so, they never get to call memory_subsys_offline()
> and so offline_pages().
>
> Which means that we would have to call __remove_zone() from those paths.
> But this alone will not work.
>
> find_smallest/biggest_section_pfn are two functions that are being called from
>
> shrink_pgdat_span
> and
> shrink_zone_span
>
> to adjust zone_first_pfn/node_first_pfn and the spanned pages.
>
> Currently, find_smallest/biggest_section_pfn checks for the secion to be valid,
> and this is fine since we are removing those sections from the remove_memory path.
>
> But if we want to move __remove_zone() to offline_pages(), we have to use
> online_section() instead of valid_section().
>
> This is all fine from offline_pages because the sections get offlined in:
>
> __offline_pages
> offline_isolated_pages
> offline_isolated_pages_cb
> __offline_isolated_pages
> offline_mem_sections
>
>
> But this does not happen in HMM/devm path.
>
> I am pretty sure this is a dumb question, but why HMM/devm path
> do not call online_pages/offline_pages?

I think mainly because onlining/offlining (wild guesses)

- calls memory notifiers
- works with memory blocks

(and does some more things not applicable to ZONE_DEVICE memory)

>
> Thanks
>


--

Thanks,

David / dhildenb