Re: [PATCH 1/7] mm/sparse.c: Introduce new function fill_subsection_map()

[Baoquan He]

On 02/11/20 at 03:44pm, David Hildenbrand wrote:
> On 11.02.20 13:46, Baoquan He wrote:
> > On 02/10/20 at 10:49am, David Hildenbrand wrote:
> >> On 09.02.20 11:48, Baoquan He wrote:
> >>> Wrap the codes filling subsection map in section_activate() into
> >>> fill_subsection_map(), this makes section_activate() cleaner and
> >>> easier to follow.
> >>>
> >>> Signed-off-by: Baoquan He <bhe@xxxxxxxxxx>
> >>> ---
> >>>  mm/sparse.c | 45 ++++++++++++++++++++++++++++++++++-----------
> >>>  1 file changed, 34 insertions(+), 11 deletions(-)
> >>>
> >>> diff --git a/mm/sparse.c b/mm/sparse.c
> >>> index c184b69460b7..9ad741ccbeb6 100644
> >>> --- a/mm/sparse.c
> >>> +++ b/mm/sparse.c
> >>> @@ -788,24 +788,28 @@ static void section_deactivate(unsigned long pfn, unsigned long nr_pages,
> >>>  		depopulate_section_memmap(pfn, nr_pages, altmap);
> >>>  }
> >>>  
> >>> -static struct page * __meminit section_activate(int nid, unsigned long pfn,
> >>> -		unsigned long nr_pages, struct vmem_altmap *altmap)
> >>> +/**
> >>> + * fill_subsection_map - fill subsection map of a memory region
> >>> + * @pfn - start pfn of the memory range
> >>> + * @nr_pages - number of pfns to add in the region
> >>> + *
> >>> + * This clears the related subsection map inside one section, and only
> >>> + * intended for hotplug.
> >>> + *
> >>> + * Return:
> >>> + * * 0		- On success.
> >>> + * * -EINVAL	- Invalid memory region.
> >>> + * * -EEXIST	- Subsection map has been set.
> >>> + */
> >>> +static int fill_subsection_map(unsigned long pfn, unsigned long nr_pages)
> >>>  {
> >>> -	DECLARE_BITMAP(map, SUBSECTIONS_PER_SECTION) = { 0 };
> >>>  	struct mem_section *ms = __pfn_to_section(pfn);
> >>> -	struct mem_section_usage *usage = NULL;
> >>> +	DECLARE_BITMAP(map, SUBSECTIONS_PER_SECTION) = { 0 };
> >>>  	unsigned long *subsection_map;
> >>> -	struct page *memmap;
> >>>  	int rc = 0;
> >>>  
> >>>  	subsection_mask_set(map, pfn, nr_pages);
> >>>  
> >>> -	if (!ms->usage) {
> >>> -		usage = kzalloc(mem_section_usage_size(), GFP_KERNEL);
> >>> -		if (!usage)
> >>> -			return ERR_PTR(-ENOMEM);
> >>> -		ms->usage = usage;
> >>> -	}
> >>>  	subsection_map = &ms->usage->subsection_map[0];
> >>>  
> >>>  	if (bitmap_empty(map, SUBSECTIONS_PER_SECTION))
> >>> @@ -816,6 +820,25 @@ static struct page * __meminit section_activate(int nid, unsigned long pfn,
> >>>  		bitmap_or(subsection_map, map, subsection_map,
> >>>  				SUBSECTIONS_PER_SECTION);
> >>>  
> >>> +	return rc;
> >>> +}
> >>> +
> >>> +static struct page * __meminit section_activate(int nid, unsigned long pfn,
> >>> +		unsigned long nr_pages, struct vmem_altmap *altmap)
> >>> +{
> >>> +	struct mem_section *ms = __pfn_to_section(pfn);
> >>> +	struct mem_section_usage *usage = NULL;
> >>> +	struct page *memmap;
> >>> +	int rc = 0;
> >>> +
> >>> +	if (!ms->usage) {
> >>> +		usage = kzalloc(mem_section_usage_size(), GFP_KERNEL);
> >>> +		if (!usage)
> >>> +			return ERR_PTR(-ENOMEM);
> >>> +		ms->usage = usage;
> >>> +	}
> >>> +
> >>> +	rc = fill_subsection_map(pfn, nr_pages);
> >>>  	if (rc) {
> >>>  		if (usage)
> >>>  			ms->usage = NULL;
> >>>
> >>
> >> What about having two variants of
> >> section_activate()/section_deactivate() instead? Then we don't have any
> >> subsection related stuff in !subsection code.
> > 
> > Thanks for looking into this, David.
> > 
> > Having two variants of section_activate()/section_deactivate() is also
> > good. Just not like memmap handling which is very different between classic
> > sparse and vmemmap, makes having two variants very attractive, the code
> > and logic in section_activate()/section_deactivate() is not too much,
> > and both of them basically can share the most of code, these make the
> > variants way not so necessary. I personally prefer the current way, what
> > do you think?
> 
> I was looking at
> 
> if (nr_pages < PAGES_PER_SECTION && early_section(ms))
> 	return pfn_to_page(pfn);
> 
> and thought that it is also specific to sub-section handling. I wonder
> if we can simply move that into the VMEMMAP variant of
> populate_section_memmap()?
> 
> But apart from that I agree that the end result with the current
> approach is also nice.
> 
> Can you reshuffle the patches, moving the fixes to the very front so we
> can backport more easily?

Sure, I will move it as the 1st one. Thanks.