[PATCH v2 RESEND] arm64: mm: hugetlb: add support for free vmemmap pages of HugeTLB

[Barry Song]

On Wed, Mar 30, 2022 at 7:53 PM Muchun Song <songmuchun at bytedance.com> wrote:
>
> On Tue, Mar 29, 2022 at 7:44 PM Barry Song <21cnbao at gmail.com> wrote:
> >
> > On Tue, Mar 29, 2022 at 5:57 PM Muchun Song <songmuchun at bytedance.com> wrote:
> > >
> > > The feature of minimizing overhead of struct page associated with each
> > > HugeTLB page aims to free its vmemmap pages (used as struct page) to
> > > save memory, where is ~14GB/16GB per 1TB HugeTLB pages (2MB/1GB type).
> > > In short, when a HugeTLB page is allocated or freed, the vmemmap array
> > > representing the range associated with the page will need to be remapped.
> > > When a page is allocated, vmemmap pages are freed after remapping.
> > > When a page is freed, previously discarded vmemmap pages must be
> > > allocated before remapping.  More implementations and details can be
> > > found here [1].
> > >
> > > The preparation of freeing vmemmap pages associated with each HugeTLB
> > > page is ready, so we can support this feature for arm64 now.  The
> > > flush_dcache_page() need to be adapted to operate on the head page's
> > > flags since the tail vmemmap pages are mapped with read-only after
> > > the feature is enabled (clear operation is not permitted).
> > >
> > > There was some discussions about this in the thread [2], but there was
> > > no conclusion in the end.  And I copied the concern proposed by Anshuman
> > > to here.
> > >
> > > 1st concern:
> > > '''
> > > But what happens when a hot remove section's vmemmap area (which is
> > > being teared down) is nearby another vmemmap area which is either created
> > > or being destroyed for HugeTLB alloc/free purpose. As you mentioned
> > > HugeTLB pages inside the hot remove section might be safe. But what about
> > > other HugeTLB areas whose vmemmap area shares page table entries with
> > > vmemmap entries for a section being hot removed ? Massive HugeTLB alloc
> > > /use/free test cycle using memory just adjacent to a memory hotplug area,
> > > which is always added and removed periodically, should be able to expose
> > > this problem.
> > > '''
> > >
> > > Answer: At the time memory is removed, all HugeTLB pages either have been
> > > migrated away or dissolved.  So there is no race between memory hot remove
> > > and free_huge_page_vmemmap().  Therefore, HugeTLB pages inside the hot
> > > remove section is safe.  Let's talk your question "what about other
> > > HugeTLB areas whose vmemmap area shares page table entries with vmemmap
> > > entries for a section being hot removed ?", the question is not
> > > established.  The minimal granularity size of hotplug memory 128MB (on
> > > arm64, 4k base page), any HugeTLB smaller than 128MB is within a section,
> > > then, there is no share PTE page tables between HugeTLB in this section
> > > and ones in other sections and a HugeTLB page could not cross two
> > > sections.  In this case, the section cannot be freed.  Any HugeTLB bigger
> > > than 128MB (section size) whose vmemmap pages is an integer multiple of
> > > 2MB (PMD-mapped).  As long as:
> > >
> > >   1) HugeTLBs are naturally aligned, power-of-two sizes
> > >   2) The HugeTLB size >= the section size
> > >   3) The HugeTLB size >= the vmemmap leaf mapping size
> > >
> > > Then a HugeTLB will not share any leaf page table entries with *anything
> > > else*, but will share intermediate entries.  In this case, at the time memory
> > > is removed, all HugeTLB pages either have been migrated away or dissolved.
> > > So there is also no race between memory hot remove and
> > > free_huge_page_vmemmap().
> > >
> > > 2nd concern:
> > > '''
> > > differently, not sure if ptdump would require any synchronization.
> > >
> > > Dumping an wrong value is probably okay but crashing because a page table
> > > entry is being freed after ptdump acquired the pointer is bad. On arm64,
> > > ptdump() is protected against hotremove via [get|put]_online_mems().
> > > '''
> > >
> > > Answer: The ptdump should be fine since vmemmap_remap_free() only exchanges
> > > PTEs or split the PMD entry (which means allocating a PTE page table).  Both
> > > operations do not free any page tables (PTE), so ptdump cannot run into a
> > > UAF on any page tables.  The wrost case is just dumping an wrong value.
> > >
> > > [1] https://lore.kernel.org/all/20210510030027.56044-1-songmuchun@bytedance.com/
> > > [2] https://lore.kernel.org/all/20210518091826.36937-1-songmuchun@bytedance.com/
> > >
> > > Signed-off-by: Muchun Song <songmuchun at bytedance.com>
> > > ---
> > > Changes in v2:
> > >  - Update commit message (Mark Rutland).
> > >  - Fix flush_dcache_page().
> > >
> > >  arch/arm64/mm/flush.c | 14 ++++++++++++++
> > >  fs/Kconfig            |  2 +-
> > >  2 files changed, 15 insertions(+), 1 deletion(-)
> > >
> > > diff --git a/arch/arm64/mm/flush.c b/arch/arm64/mm/flush.c
> > > index a06c6ac770d4..705484a9b9df 100644
> > > --- a/arch/arm64/mm/flush.c
> > > +++ b/arch/arm64/mm/flush.c
> > > @@ -75,6 +75,20 @@ EXPORT_SYMBOL_GPL(__sync_icache_dcache);
> > >   */
> > >  void flush_dcache_page(struct page *page)
> > >  {
> > > +#ifdef CONFIG_HUGETLB_PAGE_FREE_VMEMMAP
> > > +       /*
> > > +        * Only the head page's flags of HugeTLB can be cleared since the tail
> > > +        * vmemmap pages associated with each HugeTLB page are mapped with
> > > +        * read-only when CONFIG_HUGETLB_PAGE_FREE_VMEMMAP is enabled (more
> > > +        * details can refer to vmemmap_remap_pte()).  Although
> > > +        * __sync_icache_dcache() only set PG_dcache_clean flag on the head
> > > +        * page struct, some tail page structs still can see the flag since
> > > +        * the head vmemmap page frame is reused (more details can refer to
> > > +        * the comments above page_fixed_fake_head()).
> >
> > Is this still true if hugetlb_free_vmemmap_enabled() is false?
>
> No.  Do you think it is better to add hugetlb_free_vmemmap_enabled()
> into the if block? Something like the following?

yep, with if (hugetlb_free_vmemmap_enabled() && PageHuge(page)), i guess
we won't need the "ifdef" any more?

>
> +       if (hugetlb_free_vmemmap_enabled() && PageHuge(page))
> +               page = compound_head(page);
>
> >
> > btw, the subject is a bit confusing as it seems it is not bringing up
> > HUGETLB_PAGE_FREE_VMEMMAP and it seems the feature
> > has been already there, but we are lacking some fixes for some
> > functions to make it work.
>
> Right.
>
> > could we explain this clear in commit
> > log? maybe we need a better subject for the commit as well.
>
> Will do.

Thanks
Barry