On Wed, May 1, 2019 at 11:07 PM Dan Williams <dan.j.williams@xxxxxxxxx> wrote:
>
> On Wed, May 1, 2019 at 4:25 PM Pavel Tatashin <pasha.tatashin@xxxxxxxxxx> wrote:
> >
> > On 19-04-17 11:39:00, Dan Williams wrote:
> > > Towards enabling memory hotplug to track partial population of a
> > > section, introduce 'struct mem_section_usage'.
> > >
> > > A pointer to a 'struct mem_section_usage' instance replaces the existing
> > > pointer to a 'pageblock_flags' bitmap. Effectively it adds one more
> > > 'unsigned long' beyond the 'pageblock_flags' (usemap) allocation to
> > > house a new 'map_active' bitmap. The new bitmap enables the memory
> > > hot{plug,remove} implementation to act on incremental sub-divisions of a
> > > section.
> > >
> > > The primary motivation for this functionality is to support platforms
> > > that mix "System RAM" and "Persistent Memory" within a single section,
> > > or multiple PMEM ranges with different mapping lifetimes within a single
> > > section. The section restriction for hotplug has caused an ongoing saga
> > > of hacks and bugs for devm_memremap_pages() users.
> > >
> > > Beyond the fixups to teach existing paths how to retrieve the 'usemap'
> > > from a section, and updates to usemap allocation path, there are no
> > > expected behavior changes.
> > >
> > > Cc: Michal Hocko <mhocko@xxxxxxxx>
> > > Cc: Vlastimil Babka <vbabka@xxxxxxx>
> > > Cc: Logan Gunthorpe <logang@xxxxxxxxxxxx>
> > > Signed-off-by: Dan Williams <dan.j.williams@xxxxxxxxx>
> > > ---
> > > include/linux/mmzone.h | 23 ++++++++++++--
> > > mm/memory_hotplug.c | 18 ++++++-----
> > > mm/page_alloc.c | 2 +
> > > mm/sparse.c | 81 ++++++++++++++++++++++++------------------------
> > > 4 files changed, 71 insertions(+), 53 deletions(-)
> > >
> > > diff --git a/include/linux/mmzone.h b/include/linux/mmzone.h
> > > index 70394cabaf4e..f0bbd85dc19a 100644
> > > --- a/include/linux/mmzone.h
> > > +++ b/include/linux/mmzone.h
> > > @@ -1160,6 +1160,19 @@ static inline unsigned long section_nr_to_pfn(unsigned long sec)
> > > #define SECTION_ALIGN_UP(pfn) (((pfn) + PAGES_PER_SECTION - 1) & PAGE_SECTION_MASK)
> > > #define SECTION_ALIGN_DOWN(pfn) ((pfn) & PAGE_SECTION_MASK)
> > >
> > > +#define SECTION_ACTIVE_SIZE ((1UL << SECTION_SIZE_BITS) / BITS_PER_LONG)
> > > +#define SECTION_ACTIVE_MASK (~(SECTION_ACTIVE_SIZE - 1))
> > > +
> > > +struct mem_section_usage {
> > > + /*
> > > + * SECTION_ACTIVE_SIZE portions of the section that are populated in
> > > + * the memmap
> > > + */
> > > + unsigned long map_active;
> >
> > I think this should be proportional to section_size / subsection_size.
> > For example, on intel section size = 128M, and subsection is 2M, so
> > 64bits work nicely. But, on arm64 section size if 1G, so subsection is
> > 16M.
> >
> > On the other hand 16M is already much better than what we have: with 1G
> > section size and 2M pmem alignment we guaranteed to loose 1022M. And
> > with 16M subsection it is only 14M.
>
> I'm ok with it being 16M for now unless it causes a problem in
> practice, i.e. something like the minimum hardware mapping alignment
> for physical memory being less than 16M.
On second thought, arbitrary differences across architectures is a bit
sad. The most common nvdimm namespace alignment granularity is
PMD_SIZE, so perhaps the default sub-section size should try to match
that default.
