| Message ID | 20240424211031.475756-1-zi.yan@sent.com (mailing list archive) |
|---|---|
| State | New |
| Headers | show |
| Series | [v2] mm/rmap: do not add fully unmapped large folio to deferred split list | expand |
On Wed, Apr 24, 2024 at 2:10 PM Zi Yan <zi.yan@sent.com> wrote: > > From: Zi Yan <ziy@nvidia.com> > > In __folio_remove_rmap(), a large folio is added to deferred split list > if any page in a folio loses its final mapping. It is possible that > the folio is unmapped fully, but it is unnecessary to add the folio > to deferred split list at all. Fix it by checking folio->_nr_pages_mapped > before adding a folio to deferred split list. If the folio is already > on the deferred split list, it will be skipped. > > Commit 98046944a159 ("mm: huge_memory: add the missing > folio_test_pmd_mappable() for THP split statistics") tried to exclude > mTHP deferred split stats from THP_DEFERRED_SPLIT_PAGE, but it does not > fix everything. A fully unmapped PTE-mapped order-9 THP was also added to > deferred split list and counted as THP_DEFERRED_SPLIT_PAGE, since nr is > 512 (non zero), level is RMAP_LEVEL_PTE, and inside deferred_split_folio() > the order-9 folio is folio_test_pmd_mappable(). However, this miscount > was present even earlier due to implementation, since PTEs are unmapped > individually and first PTE unmapping adds the THP into the deferred split > list. Shall you mention the miscounting for mTHP too? There is another patch series adding the counter support for mTHP. > > With commit b06dc281aa99 ("mm/rmap: introduce > folio_remove_rmap_[pte|ptes|pmd]()"), kernel is able to unmap PTE-mapped > folios in one shot without causing the miscount, hence this patch. > > Signed-off-by: Zi Yan <ziy@nvidia.com> > --- > mm/rmap.c | 8 +++++--- > 1 file changed, 5 insertions(+), 3 deletions(-) > > diff --git a/mm/rmap.c b/mm/rmap.c > index a7913a454028..220ad8a83589 100644 > --- a/mm/rmap.c > +++ b/mm/rmap.c > @@ -1553,9 +1553,11 @@ static __always_inline void __folio_remove_rmap(struct folio *folio, > * page of the folio is unmapped and at least one page > * is still mapped. > */ > - if (folio_test_large(folio) && folio_test_anon(folio)) > - if (level == RMAP_LEVEL_PTE || nr < nr_pmdmapped) > - deferred_split_folio(folio); > + if (folio_test_large(folio) && folio_test_anon(folio) && > + list_empty(&folio->_deferred_list) && Do we really need this check? deferred_split_folio() does the same check too. Bailing out earlier sounds ok too, but there may not be too much gain. > + ((level == RMAP_LEVEL_PTE && atomic_read(mapped)) || > + (level == RMAP_LEVEL_PMD && nr < nr_pmdmapped))) IIUC, this line is used to cover the case which has both partial PTE-mapping and PMD-mapping, then PMD mapping is unmapped fully. IIRC this case was not handled correctly before, the THP actually skipped deferred split queue. If so please add some description in the commit log. Otherwise the patch looks good to me. Reviewed-by: Yang Shi <shy828301@gmail.com> > + deferred_split_folio(folio); > } > > /* > > base-commit: 2541ee5668b019c486dd3e815114130e35c1495d > -- > 2.43.0 >
On 24 Apr 2024, at 18:32, Yang Shi wrote: > On Wed, Apr 24, 2024 at 2:10 PM Zi Yan <zi.yan@sent.com> wrote: >> >> From: Zi Yan <ziy@nvidia.com> >> >> In __folio_remove_rmap(), a large folio is added to deferred split list >> if any page in a folio loses its final mapping. It is possible that >> the folio is unmapped fully, but it is unnecessary to add the folio >> to deferred split list at all. Fix it by checking folio->_nr_pages_mapped >> before adding a folio to deferred split list. If the folio is already >> on the deferred split list, it will be skipped. >> >> Commit 98046944a159 ("mm: huge_memory: add the missing >> folio_test_pmd_mappable() for THP split statistics") tried to exclude >> mTHP deferred split stats from THP_DEFERRED_SPLIT_PAGE, but it does not >> fix everything. A fully unmapped PTE-mapped order-9 THP was also added to >> deferred split list and counted as THP_DEFERRED_SPLIT_PAGE, since nr is >> 512 (non zero), level is RMAP_LEVEL_PTE, and inside deferred_split_folio() >> the order-9 folio is folio_test_pmd_mappable(). However, this miscount >> was present even earlier due to implementation, since PTEs are unmapped >> individually and first PTE unmapping adds the THP into the deferred split >> list. > > Shall you mention the miscounting for mTHP too? There is another patch > series adding the counter support for mTHP. OK, will add it. > >> >> With commit b06dc281aa99 ("mm/rmap: introduce >> folio_remove_rmap_[pte|ptes|pmd]()"), kernel is able to unmap PTE-mapped >> folios in one shot without causing the miscount, hence this patch. >> >> Signed-off-by: Zi Yan <ziy@nvidia.com> >> --- >> mm/rmap.c | 8 +++++--- >> 1 file changed, 5 insertions(+), 3 deletions(-) >> >> diff --git a/mm/rmap.c b/mm/rmap.c >> index a7913a454028..220ad8a83589 100644 >> --- a/mm/rmap.c >> +++ b/mm/rmap.c >> @@ -1553,9 +1553,11 @@ static __always_inline void __folio_remove_rmap(struct folio *folio, >> * page of the folio is unmapped and at least one page >> * is still mapped. >> */ >> - if (folio_test_large(folio) && folio_test_anon(folio)) >> - if (level == RMAP_LEVEL_PTE || nr < nr_pmdmapped) >> - deferred_split_folio(folio); >> + if (folio_test_large(folio) && folio_test_anon(folio) && >> + list_empty(&folio->_deferred_list) && > > Do we really need this check? deferred_split_folio() does the same > check too. Bailing out earlier sounds ok too, but there may not be too > much gain. Sure, I can remove it. > >> + ((level == RMAP_LEVEL_PTE && atomic_read(mapped)) || >> + (level == RMAP_LEVEL_PMD && nr < nr_pmdmapped))) > > IIUC, this line is used to cover the case which has both partial > PTE-mapping and PMD-mapping, then PMD mapping is unmapped fully. IIRC > this case was not handled correctly before, the THP actually skipped > deferred split queue. If so please add some description in the commit > log. It is properly handled before, since the original code is (level == RMAP_LEVEL_PTE || nr < nr_pmdmapped), meaning if either level is RMAP_LEVEL_PTE or (level == RMAP_LEVEL_PMD && nr < nr_pmdmapped), the folio is added to the deferred split list. So only level == RMAP_LEVEL_PTE part of logic needs to be fixed. > > Otherwise the patch looks good to me. Reviewed-by: Yang Shi > <shy828301@gmail.com> > Thanks. >> + deferred_split_folio(folio); >> } >> >> /* >> >> base-commit: 2541ee5668b019c486dd3e815114130e35c1495d >> -- >> 2.43.0 >> -- Best Regards, Yan, Zi
On Wed, Apr 24, 2024 at 3:39 PM Zi Yan <ziy@nvidia.com> wrote: > > On 24 Apr 2024, at 18:32, Yang Shi wrote: > > > On Wed, Apr 24, 2024 at 2:10 PM Zi Yan <zi.yan@sent.com> wrote: > >> > >> From: Zi Yan <ziy@nvidia.com> > >> > >> In __folio_remove_rmap(), a large folio is added to deferred split list > >> if any page in a folio loses its final mapping. It is possible that > >> the folio is unmapped fully, but it is unnecessary to add the folio > >> to deferred split list at all. Fix it by checking folio->_nr_pages_mapped > >> before adding a folio to deferred split list. If the folio is already > >> on the deferred split list, it will be skipped. > >> > >> Commit 98046944a159 ("mm: huge_memory: add the missing > >> folio_test_pmd_mappable() for THP split statistics") tried to exclude > >> mTHP deferred split stats from THP_DEFERRED_SPLIT_PAGE, but it does not > >> fix everything. A fully unmapped PTE-mapped order-9 THP was also added to > >> deferred split list and counted as THP_DEFERRED_SPLIT_PAGE, since nr is > >> 512 (non zero), level is RMAP_LEVEL_PTE, and inside deferred_split_folio() > >> the order-9 folio is folio_test_pmd_mappable(). However, this miscount > >> was present even earlier due to implementation, since PTEs are unmapped > >> individually and first PTE unmapping adds the THP into the deferred split > >> list. > > > > Shall you mention the miscounting for mTHP too? There is another patch > > series adding the counter support for mTHP. > > OK, will add it. > > > >> > >> With commit b06dc281aa99 ("mm/rmap: introduce > >> folio_remove_rmap_[pte|ptes|pmd]()"), kernel is able to unmap PTE-mapped > >> folios in one shot without causing the miscount, hence this patch. > >> > >> Signed-off-by: Zi Yan <ziy@nvidia.com> > >> --- > >> mm/rmap.c | 8 +++++--- > >> 1 file changed, 5 insertions(+), 3 deletions(-) > >> > >> diff --git a/mm/rmap.c b/mm/rmap.c > >> index a7913a454028..220ad8a83589 100644 > >> --- a/mm/rmap.c > >> +++ b/mm/rmap.c > >> @@ -1553,9 +1553,11 @@ static __always_inline void __folio_remove_rmap(struct folio *folio, > >> * page of the folio is unmapped and at least one page > >> * is still mapped. > >> */ > >> - if (folio_test_large(folio) && folio_test_anon(folio)) > >> - if (level == RMAP_LEVEL_PTE || nr < nr_pmdmapped) > >> - deferred_split_folio(folio); > >> + if (folio_test_large(folio) && folio_test_anon(folio) && > >> + list_empty(&folio->_deferred_list) && > > > > Do we really need this check? deferred_split_folio() does the same > > check too. Bailing out earlier sounds ok too, but there may not be too > > much gain. > > Sure, I can remove it. > > > > >> + ((level == RMAP_LEVEL_PTE && atomic_read(mapped)) || > >> + (level == RMAP_LEVEL_PMD && nr < nr_pmdmapped))) > > > > IIUC, this line is used to cover the case which has both partial > > PTE-mapping and PMD-mapping, then PMD mapping is unmapped fully. IIRC > > this case was not handled correctly before, the THP actually skipped > > deferred split queue. If so please add some description in the commit > > log. > > It is properly handled before, since the original code is > (level == RMAP_LEVEL_PTE || nr < nr_pmdmapped), meaning > if either level is RMAP_LEVEL_PTE or > (level == RMAP_LEVEL_PMD && nr < nr_pmdmapped), the folio > is added to the deferred split list. So only level == RMAP_LEVEL_PTE > part of logic needs to be fixed. Oh, yes. I misread "||" to "&&". Thanks for correcting me and fixing the problem. > > > > > Otherwise the patch looks good to me. Reviewed-by: Yang Shi > > <shy828301@gmail.com> > > > > Thanks. > >> + deferred_split_folio(folio); > >> } > >> > >> /* > >> > >> base-commit: 2541ee5668b019c486dd3e815114130e35c1495d > >> -- > >> 2.43.0 > >> > > > -- > Best Regards, > Yan, Zi
On 25.04.24 00:39, Zi Yan wrote: > On 24 Apr 2024, at 18:32, Yang Shi wrote: > >> On Wed, Apr 24, 2024 at 2:10 PM Zi Yan <zi.yan@sent.com> wrote: >>> >>> From: Zi Yan <ziy@nvidia.com> >>> >>> In __folio_remove_rmap(), a large folio is added to deferred split list >>> if any page in a folio loses its final mapping. It is possible that >>> the folio is unmapped fully, but it is unnecessary to add the folio >>> to deferred split list at all. Fix it by checking folio->_nr_pages_mapped >>> before adding a folio to deferred split list. If the folio is already >>> on the deferred split list, it will be skipped. >>> >>> Commit 98046944a159 ("mm: huge_memory: add the missing >>> folio_test_pmd_mappable() for THP split statistics") tried to exclude >>> mTHP deferred split stats from THP_DEFERRED_SPLIT_PAGE, but it does not >>> fix everything. A fully unmapped PTE-mapped order-9 THP was also added to >>> deferred split list and counted as THP_DEFERRED_SPLIT_PAGE, since nr is >>> 512 (non zero), level is RMAP_LEVEL_PTE, and inside deferred_split_folio() >>> the order-9 folio is folio_test_pmd_mappable(). However, this miscount >>> was present even earlier due to implementation, since PTEs are unmapped >>> individually and first PTE unmapping adds the THP into the deferred split >>> list. >> >> Shall you mention the miscounting for mTHP too? There is another patch >> series adding the counter support for mTHP. > > OK, will add it. I thought I made it clear: this patch won't "fix" it. Misaccounting will still happen. Just less frequently. Please spell that out. >> >>> >>> With commit b06dc281aa99 ("mm/rmap: introduce >>> folio_remove_rmap_[pte|ptes|pmd]()"), kernel is able to unmap PTE-mapped >>> folios in one shot without causing the miscount, hence this patch. >>> >>> Signed-off-by: Zi Yan <ziy@nvidia.com> >>> --- >>> mm/rmap.c | 8 +++++--- >>> 1 file changed, 5 insertions(+), 3 deletions(-) >>> >>> diff --git a/mm/rmap.c b/mm/rmap.c >>> index a7913a454028..220ad8a83589 100644 >>> --- a/mm/rmap.c >>> +++ b/mm/rmap.c >>> @@ -1553,9 +1553,11 @@ static __always_inline void __folio_remove_rmap(struct folio *folio, >>> * page of the folio is unmapped and at least one page >>> * is still mapped. >>> */ >>> - if (folio_test_large(folio) && folio_test_anon(folio)) >>> - if (level == RMAP_LEVEL_PTE || nr < nr_pmdmapped) >>> - deferred_split_folio(folio); >>> + if (folio_test_large(folio) && folio_test_anon(folio) && >>> + list_empty(&folio->_deferred_list) && >> >> Do we really need this check? deferred_split_folio() does the same >> check too. Bailing out earlier sounds ok too, but there may not be too >> much gain. > > Sure, I can remove it. Please leave it. It's a function call that cannot be optimized out otherwise.
On 25 Apr 2024, at 3:15, David Hildenbrand wrote: > On 25.04.24 00:39, Zi Yan wrote: >> On 24 Apr 2024, at 18:32, Yang Shi wrote: >> >>> On Wed, Apr 24, 2024 at 2:10 PM Zi Yan <zi.yan@sent.com> wrote: >>>> >>>> From: Zi Yan <ziy@nvidia.com> >>>> >>>> In __folio_remove_rmap(), a large folio is added to deferred split list >>>> if any page in a folio loses its final mapping. It is possible that >>>> the folio is unmapped fully, but it is unnecessary to add the folio >>>> to deferred split list at all. Fix it by checking folio->_nr_pages_mapped >>>> before adding a folio to deferred split list. If the folio is already >>>> on the deferred split list, it will be skipped. >>>> >>>> Commit 98046944a159 ("mm: huge_memory: add the missing >>>> folio_test_pmd_mappable() for THP split statistics") tried to exclude >>>> mTHP deferred split stats from THP_DEFERRED_SPLIT_PAGE, but it does not >>>> fix everything. A fully unmapped PTE-mapped order-9 THP was also added to >>>> deferred split list and counted as THP_DEFERRED_SPLIT_PAGE, since nr is >>>> 512 (non zero), level is RMAP_LEVEL_PTE, and inside deferred_split_folio() >>>> the order-9 folio is folio_test_pmd_mappable(). However, this miscount >>>> was present even earlier due to implementation, since PTEs are unmapped >>>> individually and first PTE unmapping adds the THP into the deferred split >>>> list. >>> >>> Shall you mention the miscounting for mTHP too? There is another patch >>> series adding the counter support for mTHP. >> >> OK, will add it. > > I thought I made it clear: this patch won't "fix" it. Misaccounting will still happen. Just less frequently. > > Please spell that out. Sure. Sorry I did not make that clear. > >>> >>>> >>>> With commit b06dc281aa99 ("mm/rmap: introduce >>>> folio_remove_rmap_[pte|ptes|pmd]()"), kernel is able to unmap PTE-mapped >>>> folios in one shot without causing the miscount, hence this patch. >>>> >>>> Signed-off-by: Zi Yan <ziy@nvidia.com> >>>> --- >>>> mm/rmap.c | 8 +++++--- >>>> 1 file changed, 5 insertions(+), 3 deletions(-) >>>> >>>> diff --git a/mm/rmap.c b/mm/rmap.c >>>> index a7913a454028..220ad8a83589 100644 >>>> --- a/mm/rmap.c >>>> +++ b/mm/rmap.c >>>> @@ -1553,9 +1553,11 @@ static __always_inline void __folio_remove_rmap(struct folio *folio, >>>> * page of the folio is unmapped and at least one page >>>> * is still mapped. >>>> */ >>>> - if (folio_test_large(folio) && folio_test_anon(folio)) >>>> - if (level == RMAP_LEVEL_PTE || nr < nr_pmdmapped) >>>> - deferred_split_folio(folio); >>>> + if (folio_test_large(folio) && folio_test_anon(folio) && >>>> + list_empty(&folio->_deferred_list) && >>> >>> Do we really need this check? deferred_split_folio() does the same >>> check too. Bailing out earlier sounds ok too, but there may not be too >>> much gain. >> >> Sure, I can remove it. > > Please leave it. It's a function call that cannot be optimized out otherwise. OK. If you think it is worth optimizing that function call, I will keep it. -- Best Regards, Yan, Zi
diff --git a/mm/rmap.c b/mm/rmap.c index a7913a454028..220ad8a83589 100644 --- a/mm/rmap.c +++ b/mm/rmap.c @@ -1553,9 +1553,11 @@ static __always_inline void __folio_remove_rmap(struct folio *folio, * page of the folio is unmapped and at least one page * is still mapped. */ - if (folio_test_large(folio) && folio_test_anon(folio)) - if (level == RMAP_LEVEL_PTE || nr < nr_pmdmapped) - deferred_split_folio(folio); + if (folio_test_large(folio) && folio_test_anon(folio) && + list_empty(&folio->_deferred_list) && + ((level == RMAP_LEVEL_PTE && atomic_read(mapped)) || + (level == RMAP_LEVEL_PMD && nr < nr_pmdmapped))) + deferred_split_folio(folio); } /*