Skip to contentSkip to conversation

mm: fixes for userspace hugepage collapse, v7

PATCH In review Zach O'Keefe 3 participants 14 messages Jul 20, 2022 – Jul 21, 2022

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

cover mm: fixes for userspace hugepage collapse, v7

Hey Andrew,

These are a few requested cleanups for the "mm: userspace hugepage collapse, v7" series[1] currently in mm-unstable. Please considering squashing them into the v7 series. Note that https://lkml.kernel.org/r/[email protected] is still outstanding, and that the series is incomplete until a suitable resolution is reached there.

Thanks, and apologies for the multiple fixes / adjustments required here. Zach

[1] https://lore.kernel.org/linux-mm/[email protected]/

Zach O'Keefe (4):
mm/khugepaged: Use minimal bits to store num page < HPAGE_PMD_NR mm/khugepaged: consistently order cc->is_khugepaged and pte_* checks mm/khugepaged: delay computation of hpage boundaries until use Revert "mm/madvise: add huge_memory:mm_madvise_collapse tracepoint"

 include/trace/events/huge_memory.h | 22 ----------
 mm/khugepaged.c                    | 64 +++++++++++++++++-------------
 2 files changed, 37 insertions(+), 49 deletions(-)
Signature · 1 line
2.37.0.170.g444d1eabd0-goog
#1 lore raw

PATCH 1/4 mm/khugepaged: Use minimal bits to store num page < HPAGE_PMD_NR

Minimally, node_load[] entries just need to be able to hold the maximum value of HPAGE_PMD_NR, which is compile-time defined per-arch based on PMD_SHIFT and PAGE_SHIFT. node_load[] is only written either via memset(), or with via post-increment. struct collapse_control may be allocated via kmalloc() in other collapse contexts, and MAX_NUMNODES may be arbitrarily large. #define the underlying type of node_load[] based off HPAGE_PMD_NR to avoid excessive memory allocated for this struct.

Fixes 3b07f3bb225a ("mm/khugepaged: add struct collapse_control")
---
 mm/khugepaged.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)
mm/khugepaged.c +5 -2
index 69990dacde14..ecd28bfeab60 100644
@@ -92,8 +92,11 @@ struct collapse_control {
 	bool is_khugepaged;
 
 	/* Num pages scanned per node */
-	int node_load[MAX_NUMNODES];
-
+#if HPAGE_PMD_ORDER < 16
+	u16 node_load[MAX_NUMNODES];
+#else
+	u32 node_load[MAX_NUMNODES];
+#endif
 	/* Last target selected in hpage_collapse_find_target_node() */
 	int last_target_node;
 };
Signature · 1 line
2.37.0.170.g444d1eabd0-goog
Reviewed-by on v1 1/4
#2 lore raw

Re: [PATCH mm-unstable 1/4] mm/khugepaged: Use minimal bits to store num page < HPAGE_PMD_NR

Quoted 9 lines · Zach O'Keefe

Minimally, node_load[] entries just need to be able to hold the maximum value of HPAGE_PMD_NR, which is compile-time defined per-arch based on PMD_SHIFT and PAGE_SHIFT. node_load[] is only written either via memset(), or with via post-increment. struct collapse_control may be allocated via kmalloc() in other collapse contexts, and MAX_NUMNODES may be arbitrarily large. #define the underlying type of node_load[] based off HPAGE_PMD_NR to avoid excessive memory allocated for this struct.

Quoted 3 lines
Fixes 3b07f3bb225a ("mm/khugepaged: add struct collapse_control")
Quoted 25 lines
---
 mm/khugepaged.c | 7 +++++--
 1 file changed, 5 insertions(+), 2 deletions(-)
mm/khugepaged.c +5 -3
index 69990dacde14..ecd28bfeab60 100644
@@ -92,8 +92,11 @@ struct collapse_control {
        bool is_khugepaged;
 
        /* Num pages scanned per node */
-       int node_load[MAX_NUMNODES];
-
+#if HPAGE_PMD_ORDER < 16
+       u16 node_load[MAX_NUMNODES];
+#else
+       u32 node_load[MAX_NUMNODES];
+#endif
        /* Last target selected in hpage_collapse_find_target_node() */
        int last_target_node;
 };
--

2.37.0.170.g444d1eabd0-goog

Acked-by on v1 1/4
#3 lore raw

Re: [PATCH mm-unstable 1/4] mm/khugepaged: Use minimal bits to store num page < HPAGE_PMD_NR

In reply to Zach O'Keefe
Quoted 11 lines · Zach O'Keefe

Minimally, node_load[] entries just need to be able to hold the maximum value of HPAGE_PMD_NR, which is compile-time defined per-arch based on PMD_SHIFT and PAGE_SHIFT. node_load[] is only written either via memset(), or with via post-increment. struct collapse_control may be allocated via kmalloc() in other collapse contexts, and MAX_NUMNODES may be arbitrarily large. #define the underlying type of node_load[] based off HPAGE_PMD_NR to avoid excessive memory allocated for this struct.

Fixes 3b07f3bb225a ("mm/khugepaged: add struct collapse_control")
#4 lore raw

PATCH 2/4 mm/khugepaged: consistently order cc->is_khugepaged and pte_* checks

In reply to Zach O'Keefe

cc->is_khugepaged is used to predicate the khugepaged-only behavior of enforcing khugepaged heuristics limited by the sysfs knobs khugepaged_max_ptes_[none|swap|shared].

In branches where khugepaged_max_ptes_* is checked, consistently check cc->is_khugepaged first. Also, local counters (for comparison vs khugepaged_max_ptes_* limits) were previously incremented in the comparison expression. Some of these counters (unmapped) are additionally used outside of khugepaged_max_ptes_* enforcement, and all counters are communicated in tracepoints. Move the correct accounting of these counters before branching statements to avoid future errors due to C's short-circuiting evaluation.

Fixes 9fab4752a181 ("mm/khugepaged: add flag to predicate khugepaged-only behavior")
---
 mm/khugepaged.c | 49 +++++++++++++++++++++++++++++--------------------
 1 file changed, 29 insertions(+), 20 deletions(-)
mm/khugepaged.c +29 -20
index ecd28bfeab60..290422577172 100644
@@ -574,9 +574,10 @@ static int __collapse_huge_page_isolate(struct vm_area_struct *vma,
 		pte_t pteval = *_pte;
 		if (pte_none(pteval) || (pte_present(pteval) &&
 				is_zero_pfn(pte_pfn(pteval)))) {
+			++none_or_zero;
 			if (!userfaultfd_armed(vma) &&
-			    (++none_or_zero <= khugepaged_max_ptes_none ||
-			     !cc->is_khugepaged)) {
+			    (!cc->is_khugepaged ||
+			     none_or_zero <= khugepaged_max_ptes_none)) {
 				continue;
 			} else {
 				result = SCAN_EXCEED_NONE_PTE;
@@ -596,11 +597,14 @@ static int __collapse_huge_page_isolate(struct vm_area_struct *vma,
 
 		VM_BUG_ON_PAGE(!PageAnon(page), page);
 
-		if (cc->is_khugepaged && page_mapcount(page) > 1 &&
-		    ++shared > khugepaged_max_ptes_shared) {
-			result = SCAN_EXCEED_SHARED_PTE;
-			count_vm_event(THP_SCAN_EXCEED_SHARED_PTE);
-			goto out;
+		if (page_mapcount(page) > 1) {
+			++shared;
+			if (cc->is_khugepaged &&
+			    shared > khugepaged_max_ptes_shared) {
+				result = SCAN_EXCEED_SHARED_PTE;
+				count_vm_event(THP_SCAN_EXCEED_SHARED_PTE);
+				goto out;
+			}
 		}
 
 		if (PageCompound(page)) {
@@ -1170,8 +1174,9 @@ static int hpage_collapse_scan_pmd(struct mm_struct *mm,
 	     _pte++, _address += PAGE_SIZE) {
 		pte_t pteval = *_pte;
 		if (is_swap_pte(pteval)) {
-			if (++unmapped <= khugepaged_max_ptes_swap ||
-			    !cc->is_khugepaged) {
+			++unmapped;
+			if (!cc->is_khugepaged ||
+			    unmapped <= khugepaged_max_ptes_swap) {
 				/*
 				 * Always be strict with uffd-wp
 				 * enabled swap entries.  Please see
@@ -1189,9 +1194,10 @@ static int hpage_collapse_scan_pmd(struct mm_struct *mm,
 			}
 		}
 		if (pte_none(pteval) || is_zero_pfn(pte_pfn(pteval))) {
+			++none_or_zero;
 			if (!userfaultfd_armed(vma) &&
-			    (++none_or_zero <= khugepaged_max_ptes_none ||
-			     !cc->is_khugepaged)) {
+			    (!cc->is_khugepaged ||
+			     none_or_zero <= khugepaged_max_ptes_none)) {
 				continue;
 			} else {
 				result = SCAN_EXCEED_NONE_PTE;
@@ -1221,12 +1227,14 @@ static int hpage_collapse_scan_pmd(struct mm_struct *mm,
 			goto out_unmap;
 		}
 
-		if (cc->is_khugepaged &&
-		    page_mapcount(page) > 1 &&
-		    ++shared > khugepaged_max_ptes_shared) {
-			result = SCAN_EXCEED_SHARED_PTE;
-			count_vm_event(THP_SCAN_EXCEED_SHARED_PTE);
-			goto out_unmap;
+		if (page_mapcount(page) > 1) {
+			++shared;
+			if (cc->is_khugepaged &&
+			    shared > khugepaged_max_ptes_shared) {
+				result = SCAN_EXCEED_SHARED_PTE;
+				count_vm_event(THP_SCAN_EXCEED_SHARED_PTE);
+				goto out_unmap;
+			}
 		}
 
 		page = compound_head(page);
@@ -1961,8 +1969,9 @@ static int khugepaged_scan_file(struct mm_struct *mm, struct file *file,
 			continue;
 
 		if (xa_is_value(page)) {
+			++swap;
 			if (cc->is_khugepaged &&
-			    ++swap > khugepaged_max_ptes_swap) {
+			    swap > khugepaged_max_ptes_swap) {
 				result = SCAN_EXCEED_SWAP_PTE;
 				count_vm_event(THP_SCAN_EXCEED_SWAP_PTE);
 				break;
@@ -2013,8 +2022,8 @@ static int khugepaged_scan_file(struct mm_struct *mm, struct file *file,
 	rcu_read_unlock();
 
 	if (result == SCAN_SUCCEED) {
-		if (present < HPAGE_PMD_NR - khugepaged_max_ptes_none &&
-		    cc->is_khugepaged) {
+		if (cc->is_khugepaged &&
+		    present < HPAGE_PMD_NR - khugepaged_max_ptes_none) {
 			result = SCAN_EXCEED_NONE_PTE;
 			count_vm_event(THP_SCAN_EXCEED_NONE_PTE);
 		} else {
Signature · 1 line
2.37.0.170.g444d1eabd0-goog
Reviewed-by on v1 2/4
#5 lore raw

Re: [PATCH mm-unstable 2/4] mm/khugepaged: consistently order cc->is_khugepaged and pte_* checks

Quoted 5 lines · Zach O'Keefe

cc->is_khugepaged is used to predicate the khugepaged-only behavior of enforcing khugepaged heuristics limited by the sysfs knobs khugepaged_max_ptes_[none|swap|shared].

Quoted 8 lines

In branches where khugepaged_max_ptes_* is checked, consistently check cc->is_khugepaged first. Also, local counters (for comparison vs khugepaged_max_ptes_* limits) were previously incremented in the comparison expression. Some of these counters (unmapped) are additionally used outside of khugepaged_max_ptes_* enforcement, and all counters are communicated in tracepoints. Move the correct accounting of these counters before branching statements to avoid future errors due to C's short-circuiting evaluation.

Yeah, it is safer to not depend on the order of branch statements to inc the counter.

Quoted 116 lines
Fixes 9fab4752a181 ("mm/khugepaged: add flag to predicate khugepaged-only behavior")
---
 mm/khugepaged.c | 49 +++++++++++++++++++++++++++++--------------------
 1 file changed, 29 insertions(+), 20 deletions(-)
mm/khugepaged.c +29 -21
index ecd28bfeab60..290422577172 100644
@@ -574,9 +574,10 @@ static int __collapse_huge_page_isolate(struct vm_area_struct *vma,
                pte_t pteval = *_pte;
                if (pte_none(pteval) || (pte_present(pteval) &&
                                is_zero_pfn(pte_pfn(pteval)))) {
+                       ++none_or_zero;
                        if (!userfaultfd_armed(vma) &&
-                           (++none_or_zero <= khugepaged_max_ptes_none ||
-                            !cc->is_khugepaged)) {
+                           (!cc->is_khugepaged ||
+                            none_or_zero <= khugepaged_max_ptes_none)) {
                                continue;
                        } else {
                                result = SCAN_EXCEED_NONE_PTE;
@@ -596,11 +597,14 @@ static int __collapse_huge_page_isolate(struct vm_area_struct *vma,
 
                VM_BUG_ON_PAGE(!PageAnon(page), page);
 
-               if (cc->is_khugepaged && page_mapcount(page) > 1 &&
-                   ++shared > khugepaged_max_ptes_shared) {
-                       result = SCAN_EXCEED_SHARED_PTE;
-                       count_vm_event(THP_SCAN_EXCEED_SHARED_PTE);
-                       goto out;
+               if (page_mapcount(page) > 1) {
+                       ++shared;
+                       if (cc->is_khugepaged &&
+                           shared > khugepaged_max_ptes_shared) {
+                               result = SCAN_EXCEED_SHARED_PTE;
+                               count_vm_event(THP_SCAN_EXCEED_SHARED_PTE);
+                               goto out;
+                       }
                }
 
                if (PageCompound(page)) {
@@ -1170,8 +1174,9 @@ static int hpage_collapse_scan_pmd(struct mm_struct *mm,
             _pte++, _address += PAGE_SIZE) {
                pte_t pteval = *_pte;
                if (is_swap_pte(pteval)) {
-                       if (++unmapped <= khugepaged_max_ptes_swap ||
-                           !cc->is_khugepaged) {
+                       ++unmapped;
+                       if (!cc->is_khugepaged ||
+                           unmapped <= khugepaged_max_ptes_swap) {
                                /*
                                 * Always be strict with uffd-wp
                                 * enabled swap entries.  Please see
@@ -1189,9 +1194,10 @@ static int hpage_collapse_scan_pmd(struct mm_struct *mm,
                        }
                }
                if (pte_none(pteval) || is_zero_pfn(pte_pfn(pteval))) {
+                       ++none_or_zero;
                        if (!userfaultfd_armed(vma) &&
-                           (++none_or_zero <= khugepaged_max_ptes_none ||
-                            !cc->is_khugepaged)) {
+                           (!cc->is_khugepaged ||
+                            none_or_zero <= khugepaged_max_ptes_none)) {
                                continue;
                        } else {
                                result = SCAN_EXCEED_NONE_PTE;
@@ -1221,12 +1227,14 @@ static int hpage_collapse_scan_pmd(struct mm_struct *mm,
                        goto out_unmap;
                }
 
-               if (cc->is_khugepaged &&
-                   page_mapcount(page) > 1 &&
-                   ++shared > khugepaged_max_ptes_shared) {
-                       result = SCAN_EXCEED_SHARED_PTE;
-                       count_vm_event(THP_SCAN_EXCEED_SHARED_PTE);
-                       goto out_unmap;
+               if (page_mapcount(page) > 1) {
+                       ++shared;
+                       if (cc->is_khugepaged &&
+                           shared > khugepaged_max_ptes_shared) {
+                               result = SCAN_EXCEED_SHARED_PTE;
+                               count_vm_event(THP_SCAN_EXCEED_SHARED_PTE);
+                               goto out_unmap;
+                       }
                }
 
                page = compound_head(page);
@@ -1961,8 +1969,9 @@ static int khugepaged_scan_file(struct mm_struct *mm, struct file *file,
                        continue;
 
                if (xa_is_value(page)) {
+                       ++swap;
                        if (cc->is_khugepaged &&
-                           ++swap > khugepaged_max_ptes_swap) {
+                           swap > khugepaged_max_ptes_swap) {
                                result = SCAN_EXCEED_SWAP_PTE;
                                count_vm_event(THP_SCAN_EXCEED_SWAP_PTE);
                                break;
@@ -2013,8 +2022,8 @@ static int khugepaged_scan_file(struct mm_struct *mm, struct file *file,
        rcu_read_unlock();
 
        if (result == SCAN_SUCCEED) {
-               if (present < HPAGE_PMD_NR - khugepaged_max_ptes_none &&
-                   cc->is_khugepaged) {
+               if (cc->is_khugepaged &&
+                   present < HPAGE_PMD_NR - khugepaged_max_ptes_none) {
                        result = SCAN_EXCEED_NONE_PTE;
                        count_vm_event(THP_SCAN_EXCEED_NONE_PTE);
                } else {
--

2.37.0.170.g444d1eabd0-goog

#6 lore raw

Re: [PATCH mm-unstable 2/4] mm/khugepaged: consistently order cc->is_khugepaged and pte_* checks

Quoted 15 lines · Yang Shi
Quoted 5 lines · Zach O'Keefe

cc->is_khugepaged is used to predicate the khugepaged-only behavior of enforcing khugepaged heuristics limited by the sysfs knobs khugepaged_max_ptes_[none|swap|shared].

Quoted 8 lines

In branches where khugepaged_max_ptes_* is checked, consistently check cc->is_khugepaged first. Also, local counters (for comparison vs khugepaged_max_ptes_* limits) were previously incremented in the comparison expression. Some of these counters (unmapped) are additionally used outside of khugepaged_max_ptes_* enforcement, and all counters are communicated in tracepoints. Move the correct accounting of these counters before branching statements to avoid future errors due to C's short-circuiting evaluation.

Quoted 3 lines

Yeah, it is safer to not depend on the order of branch statements to inc the counter.

Only cost me a couple hours when I got bit by this after naively moving checks around :) Hopefully I can save the next person.

Also, thanks for the reviews, Yang!

Quoted 118 lines
Quoted 116 lines
Fixes 9fab4752a181 ("mm/khugepaged: add flag to predicate khugepaged-only behavior")
---
 mm/khugepaged.c | 49 +++++++++++++++++++++++++++++--------------------
 1 file changed, 29 insertions(+), 20 deletions(-)
mm/khugepaged.c +29 -21
index ecd28bfeab60..290422577172 100644
@@ -574,9 +574,10 @@ static int __collapse_huge_page_isolate(struct vm_area_struct *vma,
                pte_t pteval = *_pte;
                if (pte_none(pteval) || (pte_present(pteval) &&
                                is_zero_pfn(pte_pfn(pteval)))) {
+                       ++none_or_zero;
                        if (!userfaultfd_armed(vma) &&
-                           (++none_or_zero <= khugepaged_max_ptes_none ||
-                            !cc->is_khugepaged)) {
+                           (!cc->is_khugepaged ||
+                            none_or_zero <= khugepaged_max_ptes_none)) {
                                continue;
                        } else {
                                result = SCAN_EXCEED_NONE_PTE;
@@ -596,11 +597,14 @@ static int __collapse_huge_page_isolate(struct vm_area_struct *vma,
 
                VM_BUG_ON_PAGE(!PageAnon(page), page);
 
-               if (cc->is_khugepaged && page_mapcount(page) > 1 &&
-                   ++shared > khugepaged_max_ptes_shared) {
-                       result = SCAN_EXCEED_SHARED_PTE;
-                       count_vm_event(THP_SCAN_EXCEED_SHARED_PTE);
-                       goto out;
+               if (page_mapcount(page) > 1) {
+                       ++shared;
+                       if (cc->is_khugepaged &&
+                           shared > khugepaged_max_ptes_shared) {
+                               result = SCAN_EXCEED_SHARED_PTE;
+                               count_vm_event(THP_SCAN_EXCEED_SHARED_PTE);
+                               goto out;
+                       }
                }
 
                if (PageCompound(page)) {
@@ -1170,8 +1174,9 @@ static int hpage_collapse_scan_pmd(struct mm_struct *mm,
             _pte++, _address += PAGE_SIZE) {
                pte_t pteval = *_pte;
                if (is_swap_pte(pteval)) {
-                       if (++unmapped <= khugepaged_max_ptes_swap ||
-                           !cc->is_khugepaged) {
+                       ++unmapped;
+                       if (!cc->is_khugepaged ||
+                           unmapped <= khugepaged_max_ptes_swap) {
                                /*
                                 * Always be strict with uffd-wp
                                 * enabled swap entries.  Please see
@@ -1189,9 +1194,10 @@ static int hpage_collapse_scan_pmd(struct mm_struct *mm,
                        }
                }
                if (pte_none(pteval) || is_zero_pfn(pte_pfn(pteval))) {
+                       ++none_or_zero;
                        if (!userfaultfd_armed(vma) &&
-                           (++none_or_zero <= khugepaged_max_ptes_none ||
-                            !cc->is_khugepaged)) {
+                           (!cc->is_khugepaged ||
+                            none_or_zero <= khugepaged_max_ptes_none)) {
                                continue;
                        } else {
                                result = SCAN_EXCEED_NONE_PTE;
@@ -1221,12 +1227,14 @@ static int hpage_collapse_scan_pmd(struct mm_struct *mm,
                        goto out_unmap;
                }
 
-               if (cc->is_khugepaged &&
-                   page_mapcount(page) > 1 &&
-                   ++shared > khugepaged_max_ptes_shared) {
-                       result = SCAN_EXCEED_SHARED_PTE;
-                       count_vm_event(THP_SCAN_EXCEED_SHARED_PTE);
-                       goto out_unmap;
+               if (page_mapcount(page) > 1) {
+                       ++shared;
+                       if (cc->is_khugepaged &&
+                           shared > khugepaged_max_ptes_shared) {
+                               result = SCAN_EXCEED_SHARED_PTE;
+                               count_vm_event(THP_SCAN_EXCEED_SHARED_PTE);
+                               goto out_unmap;
+                       }
                }
 
                page = compound_head(page);
@@ -1961,8 +1969,9 @@ static int khugepaged_scan_file(struct mm_struct *mm, struct file *file,
                        continue;
 
                if (xa_is_value(page)) {
+                       ++swap;
                        if (cc->is_khugepaged &&
-                           ++swap > khugepaged_max_ptes_swap) {
+                           swap > khugepaged_max_ptes_swap) {
                                result = SCAN_EXCEED_SWAP_PTE;
                                count_vm_event(THP_SCAN_EXCEED_SWAP_PTE);
                                break;
@@ -2013,8 +2022,8 @@ static int khugepaged_scan_file(struct mm_struct *mm, struct file *file,
        rcu_read_unlock();
 
        if (result == SCAN_SUCCEED) {
-               if (present < HPAGE_PMD_NR - khugepaged_max_ptes_none &&
-                   cc->is_khugepaged) {
+               if (cc->is_khugepaged &&
+                   present < HPAGE_PMD_NR - khugepaged_max_ptes_none) {
                        result = SCAN_EXCEED_NONE_PTE;
                        count_vm_event(THP_SCAN_EXCEED_NONE_PTE);
                } else {
--

2.37.0.170.g444d1eabd0-goog

Acked-by on v1 2/4
#7 lore raw

Re: [PATCH mm-unstable 2/4] mm/khugepaged: consistently order cc->is_khugepaged and pte_* checks

In reply to Zach O'Keefe
Quoted 13 lines · Zach O'Keefe

cc->is_khugepaged is used to predicate the khugepaged-only behavior of enforcing khugepaged heuristics limited by the sysfs knobs khugepaged_max_ptes_[none|swap|shared].

In branches where khugepaged_max_ptes_* is checked, consistently check cc->is_khugepaged first. Also, local counters (for comparison vs khugepaged_max_ptes_* limits) were previously incremented in the comparison expression. Some of these counters (unmapped) are additionally used outside of khugepaged_max_ptes_* enforcement, and all counters are communicated in tracepoints. Move the correct accounting of these counters before branching statements to avoid future errors due to C's short-circuiting evaluation.

Quoted 3 lines
Fixes 9fab4752a181 ("mm/khugepaged: add flag to predicate khugepaged-only behavior")
#8 lore raw

PATCH 3/4 mm/khugepaged: delay computation of hpage boundaries until use

In reply to Zach O'Keefe

Only compute hstart/hend once we've passed all checks that would cause early return in madvise_collapse().

Fixes c9d968ffd9ba ("mm/madvise: introduce MADV_COLLAPSE sync hugepage collapse")
---
 mm/khugepaged.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
mm/khugepaged.c +3 -3
index 290422577172..70e9d9950415 100644
@@ -2417,9 +2417,6 @@ int madvise_collapse(struct vm_area_struct *vma, struct vm_area_struct **prev,
 	if (!vma->anon_vma || !vma_is_anonymous(vma))
 		return -EINVAL;
 
-	hstart = (start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK;
-	hend = end & HPAGE_PMD_MASK;
-
 	if (!hugepage_vma_check(vma, vma->vm_flags, false, false, false))
 		return -EINVAL;
 
@@ -2432,6 +2429,9 @@ int madvise_collapse(struct vm_area_struct *vma, struct vm_area_struct **prev,
 	mmgrab(mm);
 	lru_add_drain_all();
 
+	hstart = (start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK;
+	hend = end & HPAGE_PMD_MASK;
+
 	for (addr = hstart; addr < hend; addr += HPAGE_PMD_SIZE) {
 		int result = SCAN_FAIL;
 
Signature · 1 line
2.37.0.170.g444d1eabd0-goog
Reviewed-by on v1 3/4
#9 lore raw

Re: [PATCH mm-unstable 3/4] mm/khugepaged: delay computation of hpage boundaries until use

Quoted 6 lines · Zach O'Keefe

Only compute hstart/hend once we've passed all checks that would cause early return in madvise_collapse().

Fixes c9d968ffd9ba ("mm/madvise: introduce MADV_COLLAPSE sync hugepage collapse")
Quoted 31 lines
---
 mm/khugepaged.c | 6 +++---
 1 file changed, 3 insertions(+), 3 deletions(-)
mm/khugepaged.c +3 -4
index 290422577172..70e9d9950415 100644
@@ -2417,9 +2417,6 @@ int madvise_collapse(struct vm_area_struct *vma, struct vm_area_struct **prev,
        if (!vma->anon_vma || !vma_is_anonymous(vma))
                return -EINVAL;
 
-       hstart = (start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK;
-       hend = end & HPAGE_PMD_MASK;
-
        if (!hugepage_vma_check(vma, vma->vm_flags, false, false, false))
                return -EINVAL;
 
@@ -2432,6 +2429,9 @@ int madvise_collapse(struct vm_area_struct *vma, struct vm_area_struct **prev,
        mmgrab(mm);
        lru_add_drain_all();
 
+       hstart = (start + ~HPAGE_PMD_MASK) & HPAGE_PMD_MASK;
+       hend = end & HPAGE_PMD_MASK;
+
        for (addr = hstart; addr < hend; addr += HPAGE_PMD_SIZE) {
                int result = SCAN_FAIL;
 
--

2.37.0.170.g444d1eabd0-goog

#11 lore raw

PATCH 4/4 Revert "mm/madvise: add huge_memory:mm_madvise_collapse tracepoint"

In reply to Zach O'Keefe

In the anonymous collapse path, huge_memory:mm_khugepaged_scan_pmd can be used to get roughly the same information as this proposed tracepoint. Remove it.

Fixes 0fff8a0de881 ("mm/madvise: add huge_memory:mm_madvise_collapse tracepoint")
---
 include/trace/events/huge_memory.h | 22 ----------------------
 mm/khugepaged.c                    |  2 --
 2 files changed, 24 deletions(-)
include/trace/events/huge_memory.h +0 -22
index 38d339ffdb16..55392bf30a03 100644
@@ -167,27 +167,5 @@ TRACE_EVENT(mm_collapse_huge_page_swapin,
 		__entry->ret)
 );
 
-TRACE_EVENT(mm_madvise_collapse,
-
-	TP_PROTO(struct mm_struct *mm, unsigned long addr, int result),
-
-	TP_ARGS(mm, addr, result),
-
-	TP_STRUCT__entry(__field(struct mm_struct *, mm)
-			 __field(unsigned long, addr)
-			 __field(int, result)
-	),
-
-	TP_fast_assign(__entry->mm = mm;
-		       __entry->addr = addr;
-		       __entry->result = result;
-	),
-
-	TP_printk("mm=%p addr=%#lx result=%s",
-		  __entry->mm,
-		  __entry->addr,
-		  __print_symbolic(__entry->result, SCAN_STATUS))
-);
-
 #endif /* __HUGE_MEMORY_H */
 #include <trace/define_trace.h>
mm/khugepaged.c +0 -2
index 70e9d9950415..28cb8429dad4 100644
@@ -2452,8 +2452,6 @@ int madvise_collapse(struct vm_area_struct *vma, struct vm_area_struct **prev,
 		if (!mmap_locked)
 			*prev = NULL;  /* Tell caller we dropped mmap_lock */
 
-		trace_mm_madvise_collapse(mm, addr, result);
-
 		switch (result) {
 		case SCAN_SUCCEED:
 		case SCAN_PMD_MAPPED:
Signature · 1 line
2.37.0.170.g444d1eabd0-goog
Reviewed-by on v1 4/4
#12 lore raw

Re: [PATCH mm-unstable 4/4] Revert "mm/madvise: add huge_memory:mm_madvise_collapse tracepoint"

Quoted 8 lines · Zach O'Keefe

In the anonymous collapse path, huge_memory:mm_khugepaged_scan_pmd can be used to get roughly the same information as this proposed tracepoint. Remove it.

Fixes 0fff8a0de881 ("mm/madvise: add huge_memory:mm_madvise_collapse tracepoint")
Quoted 53 lines
---
 include/trace/events/huge_memory.h | 22 ----------------------
 mm/khugepaged.c                    |  2 --
 2 files changed, 24 deletions(-)
include/trace/events/huge_memory.h +0 -22
index 38d339ffdb16..55392bf30a03 100644
@@ -167,27 +167,5 @@ TRACE_EVENT(mm_collapse_huge_page_swapin,
                __entry->ret)
 );
 
-TRACE_EVENT(mm_madvise_collapse,
-
-       TP_PROTO(struct mm_struct *mm, unsigned long addr, int result),
-
-       TP_ARGS(mm, addr, result),
-
-       TP_STRUCT__entry(__field(struct mm_struct *, mm)
-                        __field(unsigned long, addr)
-                        __field(int, result)
-       ),
-
-       TP_fast_assign(__entry->mm = mm;
-                      __entry->addr = addr;
-                      __entry->result = result;
-       ),
-
-       TP_printk("mm=%p addr=%#lx result=%s",
-                 __entry->mm,
-                 __entry->addr,
-                 __print_symbolic(__entry->result, SCAN_STATUS))
-);
-
 #endif /* __HUGE_MEMORY_H */
 #include <trace/define_trace.h>
mm/khugepaged.c +0 -3
index 70e9d9950415..28cb8429dad4 100644
@@ -2452,8 +2452,6 @@ int madvise_collapse(struct vm_area_struct *vma, struct vm_area_struct **prev,
                if (!mmap_locked)
                        *prev = NULL;  /* Tell caller we dropped mmap_lock */
 
-               trace_mm_madvise_collapse(mm, addr, result);
-
                switch (result) {
                case SCAN_SUCCEED:
                case SCAN_PMD_MAPPED:
--

2.37.0.170.g444d1eabd0-goog

Acked-by on v1 4/4
#13 lore raw

Re: [PATCH mm-unstable 4/4] Revert "mm/madvise: add huge_memory:mm_madvise_collapse tracepoint"

In reply to Zach O'Keefe
Quoted 7 lines · Zach O'Keefe

In the anonymous collapse path, huge_memory:mm_khugepaged_scan_pmd can be used to get roughly the same information as this proposed tracepoint. Remove it.

Fixes 0fff8a0de881 ("mm/madvise: add huge_memory:mm_madvise_collapse tracepoint")