Re: [PATCH v4 4/8] hugetlb: perform vmemmap restoration on a list of pages

From: Muchun Song
Date: Thu Sep 21 2023 - 13:24:29 EST




On 2023/9/21 17:31, Muchun Song wrote:


On 2023/9/21 09:12, Mike Kravetz wrote:
On 09/20/23 11:03, Muchun Song wrote:
On Sep 20, 2023, at 10:56, Muchun Song <muchun.song@xxxxxxxxx> wrote:
On Sep 20, 2023, at 04:57, Mike Kravetz <mike.kravetz@xxxxxxxxxx> wrote:
On 09/19/23 17:52, Muchun Song wrote:
On 2023/9/19 07:01, Mike Kravetz wrote:

I still think we should free a non-optimized HugeTLB page if we
encounter an OOM situation instead of continue to restore
vemmmap pages. Restoring vmemmmap pages will only aggravate
the OOM situation. The suitable appraoch is to free a non-optimized
HugeTLB page to satisfy our allocation of vmemmap pages, what's
your opinion, Mike?
I agree.

As you mentioned previously, this may complicate this code path a bit.
I will rewrite to make this happen.
Maybe we could introduced two list passed to update_and_free_pages_bulk (this
will be easy for the callers of it), one is for non-optimized huge page,
another is optimized one. In update_and_free_pages_bulk, we could first
free those non-optimized huge page, and then restore vemmmap pages for
those optimized ones, in which case, the code could be simple.
hugetlb_vmemmap_restore_folios() dose not need to add complexity, which
still continue to restore vmemmap pages and will stop once we encounter
an OOM situation.
I am not sure if passing in optimized and non-optimized lists to
update_and_free_pages_bulk will help much.  IIUC, it will almost always
be the case where only one list has entries.  Is that mostly accurate?

I think you are right. It will be less helpful since most of
pages will be not optimized when HVO is enabled.

Sorry, correction: **not** should be deleted.


BTW, maybe we should try again iff there are some non-optimized huge page
whose vmemmap pages are restored successfully previously and could be freed
first, then continue to restore the vmemmap pages of the remaining huge pages.
I think the retry code could be done in update_and_free_pages_bulk() as well.
I came up with a new routine to handle these ENOMEM returns from
hugetlb_vmemmap_restore_folios.  I 'think' it handles these situations.
Here is an updated version of this patch.  Sorry, diff makes it a bit
hard to read.

 From b13bdccb01730f995191944769f87d0725c289ad Mon Sep 17 00:00:00 2001
From: Mike Kravetz <mike.kravetz@xxxxxxxxxx>
Date: Sun, 10 Sep 2023 16:14:50 -0700
Subject: [PATCH] hugetlb: perform vmemmap restoration on a list of pages

The routine update_and_free_pages_bulk already performs vmemmap
restoration on the list of hugetlb pages in a separate step.  In
preparation for more functionality to be added in this step, create a
new routine hugetlb_vmemmap_restore_folios() that will restore
vmemmap for a list of folios.

This new routine must provide sufficient feedback about errors and
actual restoration performed so that update_and_free_pages_bulk can
perform optimally.

Special care must be taken when encountering a ENOMEM error from
hugetlb_vmemmap_restore_folios.  We want to continue making as much
forward progress as possible.  A new routine bulk_vmemmap_restore_enomem
handles this specific situation.

Signed-off-by: Mike Kravetz <mike.kravetz@xxxxxxxxxx>
---
  mm/hugetlb.c         | 83 ++++++++++++++++++++++++++++++++++----------
  mm/hugetlb_vmemmap.c | 39 +++++++++++++++++++++
  mm/hugetlb_vmemmap.h | 11 ++++++
  3 files changed, 115 insertions(+), 18 deletions(-)

diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 70fedf8682c4..52abe56cf38a 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -1834,38 +1834,85 @@ static void update_and_free_hugetlb_folio(struct hstate *h, struct folio *folio,
          schedule_work(&free_hpage_work);
  }
  -static void update_and_free_pages_bulk(struct hstate *h, struct list_head *list)
+static void bulk_vmemmap_restore_enomem(struct hstate *h,
+                        struct list_head *list,
+                        unsigned long restored)
  {
      struct folio *folio, *t_folio;
-    bool clear_dtor = false;
  -    /*
-     * First allocate required vmemmmap (if necessary) for all folios on
-     * list.  If vmemmap can not be allocated, we can not free folio to
-     * lower level allocator, so add back as hugetlb surplus page.
-     * add_hugetlb_folio() removes the page from THIS list.
-     * Use clear_dtor to note if vmemmap was successfully allocated for
-     * ANY page on the list.
-     */
-    list_for_each_entry_safe(folio, t_folio, list, lru) {
-        if (folio_test_hugetlb_vmemmap_optimized(folio)) {
+    if (restored) {
+        /*
+         * On ENOMEM error, free any restored hugetlb pages so that
+         * restore of the entire list can be retried.
+         * The idea is that by freeing hugetlb pages with vmemmap
+         * (those previously restored) we will free up memory so that
+         * we can allocate vmemmap for more hugetlb pages.
+         * We must examine and possibly free EVERY hugetlb page on list
+         * in order to call hugetlb_vmemmap_restore_folios again.
+         * This is not optimal, but is an error case that should not
+         * happen frequently.
+         */
+        list_for_each_entry_safe(folio, t_folio, list, lru)
+            if (!folio_test_hugetlb_vmemmap_optimized(folio)) {
+                list_del(&folio->lru);
+                spin_lock_irq(&hugetlb_lock);
+                __clear_hugetlb_destructor(h, folio);
+                spin_unlock_irq(&hugetlb_lock);
+                update_and_free_hugetlb_folio(h, folio, false);
+                cond_resched();
+            }
+    } else {
+        /*
+         * In the case where vmemmap was not restored for ANY folios,
+         * we loop through them trying to restore individually in the
+         * hope that someone elsewhere may free enough memory.
+         * If unable to restore a page, the hugetlb page is made a
+         * surplus page and removed from the list.
+         * If are able to restore vmemmap for one hugetlb page, we free
+         * it and quit processing the list to retry the bulk operation.
+         */
+        list_for_each_entry_safe(folio, t_folio, list, lru)
              if (hugetlb_vmemmap_restore(h, &folio->page)) {
                  spin_lock_irq(&hugetlb_lock);
                  add_hugetlb_folio(h, folio, true);
                  spin_unlock_irq(&hugetlb_lock);
-            } else
-                clear_dtor = true;
-        }
+            } else {
+                list_del(&folio->lru);
+                spin_lock_irq(&hugetlb_lock);
+                __clear_hugetlb_destructor(h, folio);
+                spin_unlock_irq(&hugetlb_lock);
+                update_and_free_hugetlb_folio(h, folio, false);
+                break;
+            }
      }
+}
+
+static void update_and_free_pages_bulk(struct hstate *h, struct list_head *list)
+{
+    int ret;
+    unsigned long restored;
+    struct folio *folio, *t_folio;
        /*
-     * If vmemmmap allocation was performed on any folio above, take lock
-     * to clear destructor of all folios on list.  This avoids the need to
+     * First allocate required vmemmmap (if necessary) for all folios.
+     * Carefully handle ENOMEM errors and free up any available hugetlb
+     * pages in order to make forward progress.
+     */
+retry:
+    ret = hugetlb_vmemmap_restore_folios(h, list, &restored);
+    if (ret == -ENOMEM) {
+        bulk_vmemmap_restore_enomem(h, list, restored);
+        goto retry;
+    }
+
+    /*
+     * If vmemmmap allocation was performed on ANY folio , take lock to
+     * clear destructor of all folios on list.  This avoids the need to
       * lock/unlock for each individual folio.
       * The assumption is vmemmap allocation was performed on all or none
       * of the folios on the list.  This is true expect in VERY rare cases.
       */
-    if (clear_dtor) {
+    if (restored) {
          spin_lock_irq(&hugetlb_lock);
          list_for_each_entry(folio, list, lru)
              __clear_hugetlb_destructor(h, folio);
diff --git a/mm/hugetlb_vmemmap.c b/mm/hugetlb_vmemmap.c
index 4558b814ffab..cc91edbfb68b 100644
--- a/mm/hugetlb_vmemmap.c
+++ b/mm/hugetlb_vmemmap.c
@@ -480,6 +480,45 @@ int hugetlb_vmemmap_restore(const struct hstate *h, struct page *head)
      return ret;
  }
  +/**
+ * hugetlb_vmemmap_restore_folios - restore vmemmap for every folio on the list.
+ * @h:        struct hstate.
+ * @folio_list:    list of folios.
+ * @restored:    Set to number of folios for which vmemmap was restored
+ *        successfully if caller passes a non-NULL pointer.
+ *
+ * Return: %0 if vmemmap exists for all folios on the list.  If an error is
+ *        encountered restoring vmemmap for ANY folio, an error code
+ *        will be returned to the caller.  It is then the responsibility
+ *        of the caller to check the hugetlb vmemmap optimized flag of
+ *        each folio to determine if vmemmap was actually restored.
+ *        Note that processing is stopped when first error is encountered.
+ */
+int hugetlb_vmemmap_restore_folios(const struct hstate *h,
+                    struct list_head *folio_list,
+                    unsigned long *restored)

How about changing parameter of @restored to a list_head type which
returns the non-optimized (previously) or vmemmap-restored-sucessful huge pages?
In which case, the caller could traverse this returned list to free
them first like you have implemented in bulk_vmemmap_restore_enomem(),
it will be more efficient. The meaning of returned value should also
be changed accordingly since update_and_free_pages_bulk() wants to
whether there is a vmemmap-optimized huge page being restored sucessfully
to determine if it should clear hugetlb flag. So hugetlb_vmemmap_restore_folios()
could return how many huge pages being restored successful, if a negative
number is returned meaning there is some error in the process of restoring
of vmemmap.

Thanks.

+{
+    unsigned long num_restored;
+    struct folio *folio;
+    int ret = 0;
+
+    num_restored = 0;
+    list_for_each_entry(folio, folio_list, lru) {
+        if (folio_test_hugetlb_vmemmap_optimized(folio)) {
+            ret = hugetlb_vmemmap_restore(h, &folio->page);
+            if (ret)
+                goto out;
+            else
+                num_restored++;
+        }
+    }
+
+out:
+    if (*restored)
+        *restored = num_restored;
+    return ret;
+}
+
  /* Return true iff a HugeTLB whose vmemmap should and can be optimized. */
  static bool vmemmap_should_optimize(const struct hstate *h, const struct page *head)
  {
diff --git a/mm/hugetlb_vmemmap.h b/mm/hugetlb_vmemmap.h
index c512e388dbb4..bb58453c3cc0 100644
--- a/mm/hugetlb_vmemmap.h
+++ b/mm/hugetlb_vmemmap.h
@@ -19,6 +19,8 @@
    #ifdef CONFIG_HUGETLB_PAGE_OPTIMIZE_VMEMMAP
  int hugetlb_vmemmap_restore(const struct hstate *h, struct page *head);
+int hugetlb_vmemmap_restore_folios(const struct hstate *h,
+            struct list_head *folio_list, unsigned long *restored);
  void hugetlb_vmemmap_optimize(const struct hstate *h, struct page *head);
  void hugetlb_vmemmap_optimize_folios(struct hstate *h, struct list_head *folio_list);
  @@ -45,6 +47,15 @@ static inline int hugetlb_vmemmap_restore(const struct hstate *h, struct page *h
      return 0;
  }
  +static inline int hugetlb_vmemmap_restore_folios(const struct hstate *h,
+                    struct list_head *folio_list,
+                    unsigned long *restored)
+{
+    if (restored)
+        *restored = 0;
+    return 0;
+}
+
  static inline void hugetlb_vmemmap_optimize(const struct hstate *h, struct page *head)
  {
  }