From: ZhangPeng zhangpeng362@huawei.com
Support large folio for mlock. Backport from v6.7-rc1.
Yin Fengwei (3): mm: add functions folio_in_range() and folio_within_vma() mm: handle large folio when large folio in VM_LOCKED VMA range mm: mlock: update mlock_pte_range to handle large folio
mm/internal.h | 73 ++++++++++++++++++++++++++++++++++++++++++++------- mm/mlock.c | 66 ++++++++++++++++++++++++++++++++++++++++++++-- mm/rmap.c | 66 ++++++++++++++++++++++++++++++++++++++-------- 3 files changed, 182 insertions(+), 23 deletions(-)
From: Yin Fengwei fengwei.yin@intel.com
mainline inclusion from mainline-v6.7-rc1 commit 28e566572aacdc551e24649e57cc9f04ba880cd2 category: other bugzilla: https://gitee.com/openeuler/kernel/issues/I8YQMW
Reference: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?i...
--------------------------------
Patch series "support large folio for mlock", v3.
Yu mentioned at [1] about the mlock() can't be applied to large folio.
I leant the related code and here is my understanding:
- For RLIMIT_MEMLOCK related, there is no problem. Because the RLIMIT_MEMLOCK statistics is not related underneath page. That means underneath page mlock or munlock doesn't impact the RLIMIT_MEMLOCK statistics collection which is always correct.
- For keeping the page in RAM, there is no problem either. At least, during try_to_unmap_one(), once detect the VMA has VM_LOCKED bit set in vm_flags, the folio will be kept whatever the folio is mlocked or not.
So the function of mlock for large folio works. But it's not optimized because the page reclaim needs scan these large folio and may split them.
This series identified the large folio for mlock to four types: - The large folio is in VM_LOCKED range and fully mapped to the range
- The large folio is in the VM_LOCKED range but not fully mapped to the range
- The large folio cross VM_LOCKED VMA boundary
- The large folio cross last level page table boundary
For the first type, we mlock large folio so page reclaim will skip it.
For the second/third type, we don't mlock large folio. As the pages not mapped to VM_LOACKED range are mapped to none VM_LOCKED range, if system is in memory pressure situation, the large folio can be picked by page reclaim and split. Then the pages not mapped to VM_LOCKED range can be reclaimed.
For the fourth type, we don't mlock large folio because locking one page table lock can't prevent the part in another last level page table being unmapped. Thanks to Ryan for pointing this out.
To check whether the folio is fully mapped to the range, PTEs needs be checked to see whether the page of folio is associated. Which needs take page table lock and is heavy operation. So far, the only place needs this check is madvise and page reclaim. These functions already have their own PTE iterator.
patch1 introduce API to check whether large folio is in VMA range. patch2 make page reclaim/mlock_vma_folio/munlock_vma_folio support large folio mlock/munlock. patch3 make mlock/munlock syscall support large folio.
Yu also mentioned a race which can make folio unevictable after munlock during RFC v2 discussion [3]: We decided that race issue didn't block this series based on: - That race issue was not introduced by this series
- We had a looks-ok fix for that race issue. Need to wait for mlock_count fixing patch as Yosry Ahmed suggested [4]
[1] https://lore.kernel.org/linux-mm/CAOUHufbtNPkdktjt_5qM45GegVO-rCFOMkSh0HQmin... [2] https://lore.kernel.org/linux-mm/20230809061105.3369958-1-fengwei.yin@intel.... [3] https://lore.kernel.org/linux-mm/CAOUHufZ6=9P_=CAOQyw0xw-3q707q-1FVV09dBNDC-...
This patch (of 3):
folio_in_range() will be used to check whether the folio is mapped to specific VMA and whether the mapping address of folio is in the range.
Also a helper function folio_within_vma() to check whether folio is in the range of vma based on folio_in_range().
Link: https://lkml.kernel.org/r/20230918073318.1181104-1-fengwei.yin@intel.com Link: https://lkml.kernel.org/r/20230918073318.1181104-2-fengwei.yin@intel.com Signed-off-by: Yin Fengwei fengwei.yin@intel.com Cc: David Hildenbrand david@redhat.com Cc: Hugh Dickins hughd@google.com Cc: Matthew Wilcox (Oracle) willy@infradead.org Cc: Ryan Roberts ryan.roberts@arm.com Cc: Yang Shi shy828301@gmail.com Cc: Yosry Ahmed yosryahmed@google.com Cc: Yu Zhao yuzhao@google.com Signed-off-by: Andrew Morton akpm@linux-foundation.org Signed-off-by: ZhangPeng zhangpeng362@huawei.com --- mm/internal.h | 50 ++++++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 50 insertions(+)
diff --git a/mm/internal.h b/mm/internal.h index a266a08e0831..92ce2170016a 100644 --- a/mm/internal.h +++ b/mm/internal.h @@ -592,6 +592,56 @@ extern long faultin_vma_page_range(struct vm_area_struct *vma, bool write, int *locked); extern bool mlock_future_ok(struct mm_struct *mm, unsigned long flags, unsigned long bytes); + +/* + * NOTE: This function can't tell whether the folio is "fully mapped" in the + * range. + * "fully mapped" means all the pages of folio is associated with the page + * table of range while this function just check whether the folio range is + * within the range [start, end). Funcation caller nees to do page table + * check if it cares about the page table association. + * + * Typical usage (like mlock or madvise) is: + * Caller knows at least 1 page of folio is associated with page table of VMA + * and the range [start, end) is intersect with the VMA range. Caller wants + * to know whether the folio is fully associated with the range. It calls + * this function to check whether the folio is in the range first. Then checks + * the page table to know whether the folio is fully mapped to the range. + */ +static inline bool +folio_within_range(struct folio *folio, struct vm_area_struct *vma, + unsigned long start, unsigned long end) +{ + pgoff_t pgoff, addr; + unsigned long vma_pglen = (vma->vm_end - vma->vm_start) >> PAGE_SHIFT; + + VM_WARN_ON_FOLIO(folio_test_ksm(folio), folio); + if (start > end) + return false; + + if (start < vma->vm_start) + start = vma->vm_start; + + if (end > vma->vm_end) + end = vma->vm_end; + + pgoff = folio_pgoff(folio); + + /* if folio start address is not in vma range */ + if (!in_range(pgoff, vma->vm_pgoff, vma_pglen)) + return false; + + addr = vma->vm_start + ((pgoff - vma->vm_pgoff) << PAGE_SHIFT); + + return !(addr < start || end - addr < folio_size(folio)); +} + +static inline bool +folio_within_vma(struct folio *folio, struct vm_area_struct *vma) +{ + return folio_within_range(folio, vma, vma->vm_start, vma->vm_end); +} + /* * mlock_vma_folio() and munlock_vma_folio(): * should be called with vma's mmap_lock held for read or write,
From: Yin Fengwei fengwei.yin@intel.com
mainline inclusion from mainline-v6.7-rc1 commit 1acbc3f936146d1b34987294803ac131bc298ce8 category: other bugzilla: https://gitee.com/openeuler/kernel/issues/I8YQMW
Reference: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?i...
--------------------------------
If large folio is in the range of VM_LOCKED VMA, it should be mlocked to avoid being picked by page reclaim. Which may split the large folio and then mlock each pages again.
Mlock this kind of large folio to prevent them being picked by page reclaim.
For the large folio which cross the boundary of VM_LOCKED VMA or not fully mapped to VM_LOCKED VMA, we'd better not to mlock it. So if the system is under memory pressure, this kind of large folio will be split and the pages ouf of VM_LOCKED VMA can be reclaimed.
Ideally, for large folio, we should mlock it when the large folio is fully mapped to VMA and munlock it if any page are unmampped from VMA. But it's not easy to detect whether the large folio is fully mapped to VMA in some cases (like add/remove rmap). So we update mlock_vma_folio() and munlock_vma_folio() to mlock/munlock the folio according to vma->vm_flags. Let caller to decide whether they should call these two functions.
For add rmap, only mlock normal 4K folio and postpone large folio handling to page reclaim phase. It is possible to reuse page table iterator to detect whether folio is fully mapped or not during page reclaim phase. For remove rmap, invoke munlock_vma_folio() to munlock folio unconditionly because rmap makes folio not fully mapped to VMA.
Link: https://lkml.kernel.org/r/20230918073318.1181104-3-fengwei.yin@intel.com Signed-off-by: Yin Fengwei fengwei.yin@intel.com Cc: David Hildenbrand david@redhat.com Cc: Hugh Dickins hughd@google.com Cc: Matthew Wilcox (Oracle) willy@infradead.org Cc: Ryan Roberts ryan.roberts@arm.com Cc: Yang Shi shy828301@gmail.com Cc: Yosry Ahmed yosryahmed@google.com Cc: Yu Zhao yuzhao@google.com Signed-off-by: Andrew Morton akpm@linux-foundation.org
Conflicts: mm/rmap.c
Signed-off-by: ZhangPeng zhangpeng362@huawei.com --- mm/internal.h | 23 ++++++++++-------- mm/rmap.c | 66 ++++++++++++++++++++++++++++++++++++++++++--------- 2 files changed, 68 insertions(+), 21 deletions(-)
diff --git a/mm/internal.h b/mm/internal.h index 92ce2170016a..d02db3e9eda9 100644 --- a/mm/internal.h +++ b/mm/internal.h @@ -650,14 +650,10 @@ folio_within_vma(struct folio *folio, struct vm_area_struct *vma) * mlock is usually called at the end of page_add_*_rmap(), munlock at * the end of page_remove_rmap(); but new anon folios are managed by * folio_add_lru_vma() calling mlock_new_folio(). - * - * @compound is used to include pmd mappings of THPs, but filter out - * pte mappings of THPs, which cannot be consistently counted: a pte - * mapping of the THP head cannot be distinguished by the page alone. */ void mlock_folio(struct folio *folio); static inline void mlock_vma_folio(struct folio *folio, - struct vm_area_struct *vma, bool compound) + struct vm_area_struct *vma) { /* * The VM_SPECIAL check here serves two purposes. @@ -667,17 +663,24 @@ static inline void mlock_vma_folio(struct folio *folio, * file->f_op->mmap() is using vm_insert_page(s), when VM_LOCKED may * still be set while VM_SPECIAL bits are added: so ignore it then. */ - if (unlikely((vma->vm_flags & (VM_LOCKED|VM_SPECIAL)) == VM_LOCKED) && - (compound || !folio_test_large(folio))) + if (unlikely((vma->vm_flags & (VM_LOCKED|VM_SPECIAL)) == VM_LOCKED)) mlock_folio(folio); }
void munlock_folio(struct folio *folio); static inline void munlock_vma_folio(struct folio *folio, - struct vm_area_struct *vma, bool compound) + struct vm_area_struct *vma) { - if (unlikely(vma->vm_flags & VM_LOCKED) && - (compound || !folio_test_large(folio))) + /* + * munlock if the function is called. Ideally, we should only + * do munlock if any page of folio is unmapped from VMA and + * cause folio not fully mapped to VMA. + * + * But it's not easy to confirm that's the situation. So we + * always munlock the folio and page reclaim will correct it + * if it's wrong. + */ + if (unlikely(vma->vm_flags & VM_LOCKED)) munlock_folio(folio); }
diff --git a/mm/rmap.c b/mm/rmap.c index 93ea81fe5180..770f5d25046a 100644 --- a/mm/rmap.c +++ b/mm/rmap.c @@ -798,6 +798,7 @@ struct folio_referenced_arg { unsigned long vm_flags; struct mem_cgroup *memcg; }; + /* * arg: folio_referenced_arg will be passed */ @@ -807,17 +808,33 @@ static bool folio_referenced_one(struct folio *folio, struct folio_referenced_arg *pra = arg; DEFINE_FOLIO_VMA_WALK(pvmw, folio, vma, address, 0); int referenced = 0; + unsigned long start = address, ptes = 0;
while (page_vma_mapped_walk(&pvmw)) { address = pvmw.address;
- if ((vma->vm_flags & VM_LOCKED) && - (!folio_test_large(folio) || !pvmw.pte)) { - /* Restore the mlock which got missed */ - mlock_vma_folio(folio, vma, !pvmw.pte); - page_vma_mapped_walk_done(&pvmw); - pra->vm_flags |= VM_LOCKED; - return false; /* To break the loop */ + if (vma->vm_flags & VM_LOCKED) { + if (!folio_test_large(folio) || !pvmw.pte) { + /* Restore the mlock which got missed */ + mlock_vma_folio(folio, vma); + page_vma_mapped_walk_done(&pvmw); + pra->vm_flags |= VM_LOCKED; + return false; /* To break the loop */ + } + /* + * For large folio fully mapped to VMA, will + * be handled after the pvmw loop. + * + * For large folio cross VMA boundaries, it's + * expected to be picked by page reclaim. But + * should skip reference of pages which are in + * the range of VM_LOCKED vma. As page reclaim + * should just count the reference of pages out + * the range of VM_LOCKED vma. + */ + ptes++; + pra->mapcount--; + continue; }
if (pvmw.pte) { @@ -842,6 +859,23 @@ static bool folio_referenced_one(struct folio *folio, pra->mapcount--; }
+ if ((vma->vm_flags & VM_LOCKED) && + folio_test_large(folio) && + folio_within_vma(folio, vma)) { + unsigned long s_align, e_align; + + s_align = ALIGN_DOWN(start, PMD_SIZE); + e_align = ALIGN_DOWN(start + folio_size(folio) - 1, PMD_SIZE); + + /* folio doesn't cross page table boundary and fully mapped */ + if ((s_align == e_align) && (ptes == folio_nr_pages(folio))) { + /* Restore the mlock which got missed */ + mlock_vma_folio(folio, vma); + pra->vm_flags |= VM_LOCKED; + return false; /* To break the loop */ + } + } + if (referenced) folio_clear_idle(folio); if (folio_test_clear_young(folio)) @@ -1253,7 +1287,14 @@ void page_add_anon_rmap(struct page *page, struct vm_area_struct *vma, __page_check_anon_rmap(folio, page, vma, address); }
- mlock_vma_folio(folio, vma, compound); + /* + * For large folio, only mlock it if it's fully mapped to VMA. It's + * not easy to check whether the large folio is fully mapped to VMA + * here. Only mlock normal 4K folio and leave page reclaim to handle + * large folio. + */ + if (!folio_test_large(folio)) + mlock_vma_folio(folio, vma); }
/** @@ -1352,7 +1393,9 @@ void folio_add_file_rmap_range(struct folio *folio, struct page *page, if (nr) __lruvec_stat_mod_folio(folio, NR_FILE_MAPPED, nr);
- mlock_vma_folio(folio, vma, compound); + /* See comments in page_add_anon_rmap() */ + if (!folio_test_large(folio)) + mlock_vma_folio(folio, vma); }
/** @@ -1463,7 +1506,7 @@ void page_remove_rmap(struct page *page, struct vm_area_struct *vma, * it's only reliable while mapped. */
- munlock_vma_folio(folio, vma, compound); + munlock_vma_folio(folio, vma); }
/* @@ -1528,7 +1571,8 @@ static bool try_to_unmap_one(struct folio *folio, struct vm_area_struct *vma, if (!(flags & TTU_IGNORE_MLOCK) && (vma->vm_flags & VM_LOCKED)) { /* Restore the mlock which got missed */ - mlock_vma_folio(folio, vma, false); + if (!folio_test_large(folio)) + mlock_vma_folio(folio, vma); page_vma_mapped_walk_done(&pvmw); ret = false; break;
From: Yin Fengwei fengwei.yin@intel.com
mainline inclusion from mainline-v6.7-rc1 commit dc68badcede4ec3b4e5cdfcb8f678670220ac2ca category: other bugzilla: https://gitee.com/openeuler/kernel/issues/I8YQMW
Reference: https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?i...
--------------------------------
Current kernel only lock base size folio during mlock syscall. Add large folio support with following rules: - Only mlock large folio when it's in VM_LOCKED VMA range and fully mapped to page table.
fully mapped folio is required as if folio is not fully mapped to a VM_LOCKED VMA, if system is in memory pressure, page reclaim is allowed to pick up this folio, split it and reclaim the pages which are not in VM_LOCKED VMA.
- munlock will apply to the large folio which is in VMA range or cross the VMA boundary.
This is required to handle the case that the large folio is mlocked, later the VMA is split in the middle of large folio.
Link: https://lkml.kernel.org/r/20230918073318.1181104-4-fengwei.yin@intel.com Signed-off-by: Yin Fengwei fengwei.yin@intel.com Cc: David Hildenbrand david@redhat.com Cc: Hugh Dickins hughd@google.com Cc: Matthew Wilcox (Oracle) willy@infradead.org Cc: Ryan Roberts ryan.roberts@arm.com Cc: Yang Shi shy828301@gmail.com Cc: Yosry Ahmed yosryahmed@google.com Cc: Yu Zhao yuzhao@google.com Signed-off-by: Andrew Morton akpm@linux-foundation.org Signed-off-by: ZhangPeng zhangpeng362@huawei.com --- mm/mlock.c | 66 ++++++++++++++++++++++++++++++++++++++++++++++++++++-- 1 file changed, 64 insertions(+), 2 deletions(-)
diff --git a/mm/mlock.c b/mm/mlock.c index 06bdfab83b58..42b6865f8f82 100644 --- a/mm/mlock.c +++ b/mm/mlock.c @@ -305,6 +305,58 @@ void munlock_folio(struct folio *folio) local_unlock(&mlock_fbatch.lock); }
+static inline unsigned int folio_mlock_step(struct folio *folio, + pte_t *pte, unsigned long addr, unsigned long end) +{ + unsigned int count, i, nr = folio_nr_pages(folio); + unsigned long pfn = folio_pfn(folio); + pte_t ptent = ptep_get(pte); + + if (!folio_test_large(folio)) + return 1; + + count = pfn + nr - pte_pfn(ptent); + count = min_t(unsigned int, count, (end - addr) >> PAGE_SHIFT); + + for (i = 0; i < count; i++, pte++) { + pte_t entry = ptep_get(pte); + + if (!pte_present(entry)) + break; + if (pte_pfn(entry) - pfn >= nr) + break; + } + + return i; +} + +static inline bool allow_mlock_munlock(struct folio *folio, + struct vm_area_struct *vma, unsigned long start, + unsigned long end, unsigned int step) +{ + /* + * For unlock, allow munlock large folio which is partially + * mapped to VMA. As it's possible that large folio is + * mlocked and VMA is split later. + * + * During memory pressure, such kind of large folio can + * be split. And the pages are not in VM_LOCKed VMA + * can be reclaimed. + */ + if (!(vma->vm_flags & VM_LOCKED)) + return true; + + /* folio not in range [start, end), skip mlock */ + if (!folio_within_range(folio, vma, start, end)) + return false; + + /* folio is not fully mapped, skip mlock */ + if (step != folio_nr_pages(folio)) + return false; + + return true; +} + static int mlock_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end, struct mm_walk *walk)
@@ -314,6 +366,8 @@ static int mlock_pte_range(pmd_t *pmd, unsigned long addr, pte_t *start_pte, *pte; pte_t ptent; struct folio *folio; + unsigned int step = 1; + unsigned long start = addr;
ptl = pmd_trans_huge_lock(pmd, vma); if (ptl) { @@ -334,6 +388,7 @@ static int mlock_pte_range(pmd_t *pmd, unsigned long addr, walk->action = ACTION_AGAIN; return 0; } + for (pte = start_pte; addr != end; pte++, addr += PAGE_SIZE) { ptent = ptep_get(pte); if (!pte_present(ptent)) @@ -341,12 +396,19 @@ static int mlock_pte_range(pmd_t *pmd, unsigned long addr, folio = vm_normal_folio(vma, addr, ptent); if (!folio || folio_is_zone_device(folio)) continue; - if (folio_test_large(folio)) - continue; + + step = folio_mlock_step(folio, pte, addr, end); + if (!allow_mlock_munlock(folio, vma, start, end, step)) + goto next_entry; + if (vma->vm_flags & VM_LOCKED) mlock_folio(folio); else munlock_folio(folio); + +next_entry: + pte += step - 1; + addr += (step - 1) << PAGE_SHIFT; } pte_unmap(start_pte); out:
-
Peng Zhang