On Tue, Aug 20, 2024 at 8:16 AM Usama Arif <usamaarif642@xxxxxxxxx> wrote:
>
>
>
> On 19/08/2024 20:00, Barry Song wrote:
> > On Tue, Aug 20, 2024 at 2:17 AM Usama Arif <usamaarif642@xxxxxxxxx> wrote:
> >>
> >>
> >>
> >> On 19/08/2024 09:29, Barry Song wrote:
> >>> Hi Usama,
> >>>
> >>> I feel it is much better now! thanks!
> >>>
> >>> On Mon, Aug 19, 2024 at 2:31 PM Usama Arif <usamaarif642@xxxxxxxxx> wrote:
> >>>>
> >>>> Currently folio->_deferred_list is used to keep track of
> >>>> partially_mapped folios that are going to be split under memory
> >>>> pressure. In the next patch, all THPs that are faulted in and collapsed
> >>>> by khugepaged are also going to be tracked using _deferred_list.
> >>>>
> >>>> This patch introduces a pageflag to be able to distinguish between
> >>>> partially mapped folios and others in the deferred_list at split time in
> >>>> deferred_split_scan. Its needed as __folio_remove_rmap decrements
> >>>> _mapcount, _large_mapcount and _entire_mapcount, hence it won't be
> >>>> possible to distinguish between partially mapped folios and others in
> >>>> deferred_split_scan.
> >>>>
> >>>> Eventhough it introduces an extra flag to track if the folio is
> >>>> partially mapped, there is no functional change intended with this
> >>>> patch and the flag is not useful in this patch itself, it will
> >>>> become useful in the next patch when _deferred_list has non partially
> >>>> mapped folios.
> >>>>
> >>>> Signed-off-by: Usama Arif <usamaarif642@xxxxxxxxx>
> >>>> ---
> >>>> include/linux/huge_mm.h | 4 ++--
> >>>> include/linux/page-flags.h | 11 +++++++++++
> >>>> mm/huge_memory.c | 23 ++++++++++++++++-------
> >>>> mm/internal.h | 4 +++-
> >>>> mm/memcontrol.c | 3 ++-
> >>>> mm/migrate.c | 3 ++-
> >>>> mm/page_alloc.c | 5 +++--
> >>>> mm/rmap.c | 5 +++--
> >>>> mm/vmscan.c | 3 ++-
> >>>> 9 files changed, 44 insertions(+), 17 deletions(-)
> >>>>
> >>>> diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h
> >>>> index 4c32058cacfe..969f11f360d2 100644
> >>>> --- a/include/linux/huge_mm.h
> >>>> +++ b/include/linux/huge_mm.h
> >>>> @@ -321,7 +321,7 @@ static inline int split_huge_page(struct page *page)
> >>>> {
> >>>> return split_huge_page_to_list_to_order(page, NULL, 0);
> >>>> }
> >>>> -void deferred_split_folio(struct folio *folio);
> >>>> +void deferred_split_folio(struct folio *folio, bool partially_mapped);
> >>>>
> >>>> void __split_huge_pmd(struct vm_area_struct *vma, pmd_t *pmd,
> >>>> unsigned long address, bool freeze, struct folio *folio);
> >>>> @@ -495,7 +495,7 @@ static inline int split_huge_page(struct page *page)
> >>>> {
> >>>> return 0;
> >>>> }
> >>>> -static inline void deferred_split_folio(struct folio *folio) {}
> >>>> +static inline void deferred_split_folio(struct folio *folio, bool partially_mapped) {}
> >>>> #define split_huge_pmd(__vma, __pmd, __address) \
> >>>> do { } while (0)
> >>>>
> >>>> diff --git a/include/linux/page-flags.h b/include/linux/page-flags.h
> >>>> index a0a29bd092f8..c3bb0e0da581 100644
> >>>> --- a/include/linux/page-flags.h
> >>>> +++ b/include/linux/page-flags.h
> >>>> @@ -182,6 +182,7 @@ enum pageflags {
> >>>> /* At least one page in this folio has the hwpoison flag set */
> >>>> PG_has_hwpoisoned = PG_active,
> >>>> PG_large_rmappable = PG_workingset, /* anon or file-backed */
> >>>> + PG_partially_mapped = PG_reclaim, /* was identified to be partially mapped */
> >>>> };
> >>>>
> >>>> #define PAGEFLAGS_MASK ((1UL << NR_PAGEFLAGS) - 1)
> >>>> @@ -861,8 +862,18 @@ static inline void ClearPageCompound(struct page *page)
> >>>> ClearPageHead(page);
> >>>> }
> >>>> FOLIO_FLAG(large_rmappable, FOLIO_SECOND_PAGE)
> >>>> +FOLIO_TEST_FLAG(partially_mapped, FOLIO_SECOND_PAGE)
> >>>> +/*
> >>>> + * PG_partially_mapped is protected by deferred_split split_queue_lock,
> >>>> + * so its safe to use non-atomic set/clear.
> >>>> + */
> >>>> +__FOLIO_SET_FLAG(partially_mapped, FOLIO_SECOND_PAGE)
> >>>> +__FOLIO_CLEAR_FLAG(partially_mapped, FOLIO_SECOND_PAGE)
> >>>> #else
> >>>> FOLIO_FLAG_FALSE(large_rmappable)
> >>>> +FOLIO_TEST_FLAG_FALSE(partially_mapped)
> >>>> +__FOLIO_SET_FLAG_NOOP(partially_mapped)
> >>>> +__FOLIO_CLEAR_FLAG_NOOP(partially_mapped)
> >>>> #endif
> >>>>
> >>>> #define PG_head_mask ((1UL << PG_head))
> >>>> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> >>>> index 2d77b5d2291e..70ee49dfeaad 100644
> >>>> --- a/mm/huge_memory.c
> >>>> +++ b/mm/huge_memory.c
> >>>> @@ -3398,6 +3398,7 @@ int split_huge_page_to_list_to_order(struct page *page, struct list_head *list,
> >>>> * page_deferred_list.
> >>>> */
> >>>> list_del_init(&folio->_deferred_list);
> >>>> + __folio_clear_partially_mapped(folio);
> >>>> }
> >>>> spin_unlock(&ds_queue->split_queue_lock);
> >>>> if (mapping) {
> >>>> @@ -3454,11 +3455,13 @@ void __folio_undo_large_rmappable(struct folio *folio)
> >>>> if (!list_empty(&folio->_deferred_list)) {
> >>>> ds_queue->split_queue_len--;
> >>>> list_del_init(&folio->_deferred_list);
> >>>> + __folio_clear_partially_mapped(folio);
> >>>
> >>> is it possible to make things clearer by
> >>>
> >>> if (folio_clear_partially_mapped)
> >>> __folio_clear_partially_mapped(folio);
> >>>
> >>> While writing without conditions isn't necessarily wrong, adding a condition
> >>> will improve the readability of the code and enhance the clarity of my mTHP
> >>> counters series. also help decrease smp cache sync if we can avoid
> >>> unnecessary writing?
> >>>
> >>
> >> Do you mean if(folio_test_partially_mapped(folio))?
> >>
> >> I don't like this idea. I think it makes the readability worse? If I was looking at if (test) -> clear for the first time, I would become confused why its being tested if its going to be clear at the end anyways?
> >
> > In the pmd-order case, the majority of folios are not partially mapped.
> > Unconditional writes will trigger cache synchronization across all
> > CPUs (related to the MESI protocol), making them more costly. By
> > using conditional writes, such as "if(test) write," we can avoid
> > most unnecessary writes, which is much more efficient. Additionally,
> > we only need to manage nr_split_deferred when the condition
> > is met. We are carefully evaluating all scenarios to determine
> > if modifications to the partially_mapped flag are necessary.
> >
>
>
> Hmm okay, as you said its needed for nr_split_deferred anyways. Something like below is ok to fold in?
>
> commit 4ae9e2067346effd902b342296987b97dee29018 (HEAD)
> Author: Usama Arif <usamaarif642@xxxxxxxxx>
> Date: Mon Aug 19 21:07:16 2024 +0100
>
> mm: Introduce a pageflag for partially mapped folios fix
>
> Test partially_mapped flag before clearing it. This should
> avoid unnecessary writes and will be needed in the nr_split_deferred
> series.
>
> Signed-off-by: Usama Arif <usamaarif642@xxxxxxxxx>
>
> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index 5d67d3b3c1b2..ccde60aaaa0f 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> @@ -3479,7 +3479,8 @@ void __folio_undo_large_rmappable(struct folio *folio)
> if (!list_empty(&folio->_deferred_list)) {
> ds_queue->split_queue_len--;
> list_del_init(&folio->_deferred_list);
> - __folio_clear_partially_mapped(folio);
> + if (folio_test_partially_mapped(folio))
> + __folio_clear_partially_mapped(folio);
> }
> spin_unlock_irqrestore(&ds_queue->split_queue_lock, flags);
> }
> @@ -3610,7 +3611,8 @@ static unsigned long deferred_split_scan(struct shrinker *shrink,
> } else {
> /* We lost race with folio_put() */
> list_del_init(&folio->_deferred_list);
> - __folio_clear_partially_mapped(folio);
> + if (folio_test_partially_mapped(folio))
> + __folio_clear_partially_mapped(folio);
> ds_queue->split_queue_len--;
> }
> if (!--sc->nr_to_scan)
>
Do we also need if (folio_test_partially_mapped(folio)) in
split_huge_page_to_list_to_order()?
I recall that in Yu Zhao's TAO, there’s a chance of splitting (shattering)
non-partially-mapped folios. To be future-proof, we might want to handle
both cases equally.
By the way, we might not need to clear the flag for a new folio. This differs
from the init_list, which is necessary. If a new folio has the partially_mapped
flag, it indicates that we failed to clear it when freeing the folio to
the buddy system, which is a bug we need to fix in the free path.
Thanks
Barry
