Re: [Patch v2] mm/sparsemem: get address to page struct instead of address to pfn

[David Hildenbrand]

On 12.02.20 03:28, Wei Yang wrote:
> On Tue, Feb 11, 2020 at 03:01:35PM +0100, David Hildenbrand wrote:
>> On 11.02.20 00:16, Wei Yang wrote:
>>> On Mon, Feb 10, 2020 at 10:00:47AM +0100, David Hildenbrand wrote:
>>>> On 10.02.20 01:50, Wei Yang wrote:
>>>>> memmap should be the address to page struct instead of address to pfn.
>>>>>
>>>>
>>>> "mm/sparsemem: fix wrong address in ms->section_mem_map with sub-sections
>>>>
>>>> We want to store the address of the memmap, not the address of the first
>>>> pfn.
>>>>
>>>> E.g., we can have both (boot) system memory and devmem residing in a
>>>> single section. Once we hot-add the devmem part, the address stored in
>>>> ms->section_mem_map would be wrong, and kdump would not be able to
>>>> dump the right memory.
>>>> "
>>>>
>>>> ? See below
>>>>
>>>>> As mentioned by David, if system memory and devmem sit within a
>>>>> section, the mismatch address would lead kdump to dump unexpected
>>>>> memory.
>>>>>
>>>>> Since sub-section only works for SPARSEMEM_VMEMMAP, pfn_to_page() is
>>>>> valid to get the page struct address at this point.
>>>>>
>>>>> Fixes: ba72b4c8cf60 ("mm/sparsemem: support sub-section hotplug")
>>>>> Signed-off-by: Wei Yang <richardw.yang@xxxxxxxxxxxxxxx>
>>>>> CC: Dan Williams <dan.j.williams@xxxxxxxxx>
>>>>> CC: David Hildenbrand <david@xxxxxxxxxx>
>>>>> CC: Baoquan He <bhe@xxxxxxxxxx>
>>>>>
>>>>> ---
>>>>> v2:
>>>>>   * adjust comment to mention the mismatch data would affect kdump
>>>>>
>>>>> ---
>>>>>  mm/sparse.c | 2 +-
>>>>>  1 file changed, 1 insertion(+), 1 deletion(-)
>>>>>
>>>>> diff --git a/mm/sparse.c b/mm/sparse.c
>>>>> index 586d85662978..4862ec2cfbc0 100644
>>>>> --- a/mm/sparse.c
>>>>> +++ b/mm/sparse.c
>>>>> @@ -887,7 +887,7 @@ int __meminit sparse_add_section(int nid, unsigned long start_pfn,
>>>>>  
>>>>>  	/* Align memmap to section boundary in the subsection case */
>>>>>  	if (section_nr_to_pfn(section_nr) != start_pfn)
>>>>> -		memmap = pfn_to_kaddr(section_nr_to_pfn(section_nr));
>>>>> +		memmap = pfn_to_page(section_nr_to_pfn(section_nr));
>>>>
>>>> I think this whole code should be reworked.
>>>>
>>>> Callee returns a pointer. Caller: Nah, I know it better.
>>>>
>>>> Just nasty.
>>>>
>>>>
>>>> Can we do something like this instead:
>>>>
>>>>
>>>> diff --git a/mm/sparse-vmemmap.c b/mm/sparse-vmemmap.c
>>>> index 200aef686722..c5091feef29e 100644
>>>> --- a/mm/sparse-vmemmap.c
>>>> +++ b/mm/sparse-vmemmap.c
>>>> @@ -266,5 +266,5 @@ struct page * __meminit
>>>> __populate_section_memmap(unsigned long pfn,
>>>>        if (vmemmap_populate(start, end, nid, altmap))
>>>>                return NULL;
>>>>
>>>> -       return pfn_to_page(pfn);
>>>> +       return pfn_to_page(SECTION_ALIGN_DOWN(pfn));
>>>> }
>>>> diff --git a/mm/sparse.c b/mm/sparse.c
>>>> index c184b69460b7..21902d7931e4 100644
>>>> --- a/mm/sparse.c
>>>> +++ b/mm/sparse.c
>>>> @@ -788,6 +788,10 @@ static void section_deactivate(unsigned long pfn,
>>>> unsigned long nr_pages,
>>>>                depopulate_section_memmap(pfn, nr_pages, altmap);
>>>> }
>>>>
>>>> +/*
>>>> + * Returns the memmap of the first pfn of the section (not of
>>>> + * sub-sections).
>>>> + */
>>>> static struct page * __meminit section_activate(int nid, unsigned long pfn,
>>>>                unsigned long nr_pages, struct vmem_altmap *altmap)
>>>> {
>>>> @@ -882,9 +886,6 @@ int __meminit sparse_add_section(int nid, unsigned
>>>> long start_pfn,
>>>>        set_section_nid(section_nr, nid);
>>>>        section_mark_present(ms);
>>>>
>>>> -       /* Align memmap to section boundary in the subsection case */
>>>> -       if (section_nr_to_pfn(section_nr) != start_pfn)
>>>> -               memmap = pfn_to_kaddr(section_nr_to_pfn(section_nr));
>>>>        sparse_init_one_section(ms, section_nr, memmap, ms->usage, 0);
>>>>
>>>>        return 0;
>>>>
>>>>
>>>> Untested, of course :)
>>>
>>> I think you get some point. As you mentioned in the following reply, we need
>>> to adjust poisoning after this change.
>>
>> We can just poison after setting up the section (IOW, move it further down).
>>
>>>
>>> This looks like a trade off between two options. I don't have a strong
>>> preference.
>>
>> I clearly prefer if *section*_activate() returns the memmap of the
>> section. This code is just confusing. But I can send a cleanup on top if
>> you want to keep it like that for now.
>>
> 
> Sure, a cleanup patch may help audience get more understanding about the
> change.
> 

For this simple fix for now.

Acked-by: David Hildenbrand <david@xxxxxxxxxx>

Will send a cleanup in case I don't forget :)

-- 
Thanks,

David / dhildenb