| Message ID | 1588200982-69492-2-git-send-email-yang.shi@linux.alibaba.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | [linux-next,1/2] mm: khugepaged: add exceed_max_ptes_* helpers | expand |
On 4/29/20 3:56 PM, Yang Shi wrote: > When khugepaged successfully isolated and copied data from base page to > collapsed THP, the base page is about to be freed. So putting the page > back to lru sounds not that productive since the page might be isolated > by vmscan but it can't be reclaimed by vmscan since it can't be unmapped > by try_to_unmap() at all. > > Actually khugepaged is the last user of this page so it can be freed > directly. So, clearing active and unevictable flags, unlocking and > dropping refcount from isolate instead of calling putback_lru_page(). Please disregard the patch. I just remembered Kirill added collapse shared pages support. If the pages are shared then they have to be put back to lru since they may be still mapped by other processes. So we need check the mapcount if we would like to skip lru. And I spotted the other issue. The release_pte_page() calls mod_node_page_state() unconditionally, it was fine before. But, due to the support for collapsing shared pages we need check if the last mapcount is gone or not. Andrew, would you please remove this patch from -mm tree? I will send one or two rectified patches. Sorry for the inconvenience. > > Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> > Cc: Hugh Dickins <hughd@google.com> > Cc: Andrea Arcangeli <aarcange@redhat.com> > Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com> > --- > mm/khugepaged.c | 15 +++++++++++++-- > 1 file changed, 13 insertions(+), 2 deletions(-) > > diff --git a/mm/khugepaged.c b/mm/khugepaged.c > index 0c8d30b..c131a90 100644 > --- a/mm/khugepaged.c > +++ b/mm/khugepaged.c > @@ -559,6 +559,17 @@ void __khugepaged_exit(struct mm_struct *mm) > static void release_pte_page(struct page *page) > { > mod_node_page_state(page_pgdat(page), > + NR_ISOLATED_ANON + page_is_file_lru(page), -compound_nr(page)); > + ClearPageActive(page); > + ClearPageUnevictable(page); > + unlock_page(page); > + /* Drop refcount from isolate */ > + put_page(page); > +} > + > +static void release_pte_page_to_lru(struct page *page) > +{ > + mod_node_page_state(page_pgdat(page), > NR_ISOLATED_ANON + page_is_file_lru(page), > -compound_nr(page)); > unlock_page(page); > @@ -576,12 +587,12 @@ static void release_pte_pages(pte_t *pte, pte_t *_pte, > page = pte_page(pteval); > if (!pte_none(pteval) && !is_zero_pfn(pte_pfn(pteval)) && > !PageCompound(page)) > - release_pte_page(page); > + release_pte_page_to_lru(page); > } > > list_for_each_entry_safe(page, tmp, compound_pagelist, lru) { > list_del(&page->lru); > - release_pte_page(page); > + release_pte_page_to_lru(page); > } > } >
On 4/29/20 5:41 PM, Yang Shi wrote: > > > On 4/29/20 3:56 PM, Yang Shi wrote: >> When khugepaged successfully isolated and copied data from base page to >> collapsed THP, the base page is about to be freed. So putting the page >> back to lru sounds not that productive since the page might be isolated >> by vmscan but it can't be reclaimed by vmscan since it can't be unmapped >> by try_to_unmap() at all. >> >> Actually khugepaged is the last user of this page so it can be freed >> directly. So, clearing active and unevictable flags, unlocking and >> dropping refcount from isolate instead of calling putback_lru_page(). > > Please disregard the patch. I just remembered Kirill added collapse > shared pages support. If the pages are shared then they have to be put > back to lru since they may be still mapped by other processes. So we > need check the mapcount if we would like to skip lru. > > And I spotted the other issue. The release_pte_page() calls > mod_node_page_state() unconditionally, it was fine before. But, due to > the support for collapsing shared pages we need check if the last > mapcount is gone or not. Hmm... this is false. I mixed up NR_ISOLATED_ANON and NR_ANON_MAPPED. > > Andrew, would you please remove this patch from -mm tree? I will send > one or two rectified patches. Sorry for the inconvenience. > >> >> Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> >> Cc: Hugh Dickins <hughd@google.com> >> Cc: Andrea Arcangeli <aarcange@redhat.com> >> Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com> >> --- >> mm/khugepaged.c | 15 +++++++++++++-- >> 1 file changed, 13 insertions(+), 2 deletions(-) >> >> diff --git a/mm/khugepaged.c b/mm/khugepaged.c >> index 0c8d30b..c131a90 100644 >> --- a/mm/khugepaged.c >> +++ b/mm/khugepaged.c >> @@ -559,6 +559,17 @@ void __khugepaged_exit(struct mm_struct *mm) >> static void release_pte_page(struct page *page) >> { >> mod_node_page_state(page_pgdat(page), >> + NR_ISOLATED_ANON + page_is_file_lru(page), -compound_nr(page)); >> + ClearPageActive(page); >> + ClearPageUnevictable(page); >> + unlock_page(page); >> + /* Drop refcount from isolate */ >> + put_page(page); >> +} >> + >> +static void release_pte_page_to_lru(struct page *page) >> +{ >> + mod_node_page_state(page_pgdat(page), >> NR_ISOLATED_ANON + page_is_file_lru(page), >> -compound_nr(page)); >> unlock_page(page); >> @@ -576,12 +587,12 @@ static void release_pte_pages(pte_t *pte, pte_t >> *_pte, >> page = pte_page(pteval); >> if (!pte_none(pteval) && !is_zero_pfn(pte_pfn(pteval)) && >> !PageCompound(page)) >> - release_pte_page(page); >> + release_pte_page_to_lru(page); >> } >> list_for_each_entry_safe(page, tmp, compound_pagelist, lru) { >> list_del(&page->lru); >> - release_pte_page(page); >> + release_pte_page_to_lru(page); >> } >> } >
diff --git a/mm/khugepaged.c b/mm/khugepaged.c index 0c8d30b..c131a90 100644 --- a/mm/khugepaged.c +++ b/mm/khugepaged.c @@ -559,6 +559,17 @@ void __khugepaged_exit(struct mm_struct *mm) static void release_pte_page(struct page *page) { mod_node_page_state(page_pgdat(page), + NR_ISOLATED_ANON + page_is_file_lru(page), -compound_nr(page)); + ClearPageActive(page); + ClearPageUnevictable(page); + unlock_page(page); + /* Drop refcount from isolate */ + put_page(page); +} + +static void release_pte_page_to_lru(struct page *page) +{ + mod_node_page_state(page_pgdat(page), NR_ISOLATED_ANON + page_is_file_lru(page), -compound_nr(page)); unlock_page(page); @@ -576,12 +587,12 @@ static void release_pte_pages(pte_t *pte, pte_t *_pte, page = pte_page(pteval); if (!pte_none(pteval) && !is_zero_pfn(pte_pfn(pteval)) && !PageCompound(page)) - release_pte_page(page); + release_pte_page_to_lru(page); } list_for_each_entry_safe(page, tmp, compound_pagelist, lru) { list_del(&page->lru); - release_pte_page(page); + release_pte_page_to_lru(page); } }
When khugepaged successfully isolated and copied data from base page to collapsed THP, the base page is about to be freed. So putting the page back to lru sounds not that productive since the page might be isolated by vmscan but it can't be reclaimed by vmscan since it can't be unmapped by try_to_unmap() at all. Actually khugepaged is the last user of this page so it can be freed directly. So, clearing active and unevictable flags, unlocking and dropping refcount from isolate instead of calling putback_lru_page(). Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com> Cc: Hugh Dickins <hughd@google.com> Cc: Andrea Arcangeli <aarcange@redhat.com> Signed-off-by: Yang Shi <yang.shi@linux.alibaba.com> --- mm/khugepaged.c | 15 +++++++++++++-- 1 file changed, 13 insertions(+), 2 deletions(-)