Re: [PATCH v2 2/2] mm: memory_hotplug: introduce SECTION_CANNOT_OPTIMIZE_VMEMMAP

[David Hildenbrand <david@redhat.com>]

On 17.06.22 11:54, Oscar Salvador wrote:
> On Fri, Jun 17, 2022 at 09:43:33AM +0200, David Hildenbrand wrote:
>> VmemmapSelfHosted, then the function names get nicer.
> 
> Definitely.
> 
>>
>>> +#endif
>>> +
>>>  /*
>>>   * On an anonymous page mapped into a user virtual memory area,
>>>   * page->mapping points to its anon_vma, not to a struct address_space;
>>> diff --git a/mm/hugetlb_vmemmap.c b/mm/hugetlb_vmemmap.c
>>> index 1089ea8a9c98..e2de7ed27e9e 100644
>>> --- a/mm/hugetlb_vmemmap.c
>>> +++ b/mm/hugetlb_vmemmap.c
>>> @@ -101,6 +101,14 @@ void hugetlb_vmemmap_free(struct hstate *h, struct page *head)
>>>  {
>>>  	unsigned long vmemmap_addr = (unsigned long)head;
>>>  	unsigned long vmemmap_end, vmemmap_reuse, vmemmap_pages;
>>> +	struct mem_section *ms = __pfn_to_section(page_to_pfn(head));
>>> +	struct page *memmap;
>>> +
>>> +	memmap = sparse_decode_mem_map(ms->section_mem_map,
>>> +				       pfn_to_section_nr(page_to_pfn(head)));
>>
>> Why can't we check the head page directly? Either I need more coffee or
>> this can be simplified.
> 
> Uhm, maybe I'm the one who needs coffe here but we have something like:
> 
> [    hot-plugges section   ]
> [memmap pages][normal pages]

Oh, right, we need the memmap for our hugetlb page (which resides in the
reserved section), but these are not marked. We need the memmap of that
memmap.

I assume one could directly via the page address. Because we want the
memmap of the memmap.

vmmemmap_page = virt_to_page(head);

Not sure, though, if that would work with that function.

-- 
Thanks,

David / dhildenb