Skip to contentSkip to conversation

mm hugetlb x86: fix hugepage memory leak in mincore()

PATCH In review Naoya Horiguchi 2 participants 4 messages Dec 7, 2009 – Dec 8, 2009 1 message not in the archive

Subsystems (4)
lore
Share & tools
lore

A message that is not in the archive (sent off-list or not yet received): [email protected]

#1 lore raw

PATCH 1/1 mm hugetlb x86: fix hugepage memory leak in mincore()

Most callers of pmd_none_or_clear_bad() check whether the target page is in a hugepage or not, but mincore() and walk_page_range() do not check it. So if we use mincore() on a hugepage on x86 machine, the hugepage memory is leaked as shown below.
This patch fixes it by extending mincore() system call to support hugepages.

Details
=======
My test program (leak_mincore) works as follows:
- creat() and mmap() a file on hugetlbfs (file size is 200MB == 100 hugepages,)
- read()/write() something on it,
- call mincore() for first ten pages and printf() the values of *vec
- munmap() and unlink() the file on hugetlbfs

Without my patch
----------------

$ cat /proc/meminfo| grep "HugePage"
HugePages_Total:    1000
HugePages_Free:     1000
HugePages_Rsvd:        0
HugePages_Surp:        0
$ ./leak_mincore
vec[0] 0
vec[1] 0
vec[2] 0
vec[3] 0
vec[4] 0
vec[5] 0
vec[6] 0
vec[7] 0
vec[8] 0
vec[9] 0
$ cat /proc/meminfo |grep "HugePage"
HugePages_Total:    1000
HugePages_Free:      999
HugePages_Rsvd:        0
HugePages_Surp:        0
$ ls /hugetlbfs/
$

Return values in *vec from mincore() are set to 0, while the hugepage should be in memory, and 1 hugepage is still accounted as used while there is no file on hugetlbfs.

With my patch
-------------

$ cat /proc/meminfo| grep "HugePage"
HugePages_Total:    1000
HugePages_Free:     1000
HugePages_Rsvd:        0
HugePages_Surp:        0
$ ./leak_mincore
vec[0] 1
vec[1] 1
vec[2] 1
vec[3] 1
vec[4] 1
vec[5] 1
vec[6] 1
vec[7] 1
vec[8] 1
vec[9] 1
$ cat /proc/meminfo |grep "HugePage"
HugePages_Total:    1000
HugePages_Free:     1000
HugePages_Rsvd:        0
HugePages_Surp:        0
$ ls /hugetlbfs/
$

Return value in *vec set to 1 and no memory leaks.

---
 mm/mincore.c |   34 ++++++++++++++++++++++++++++++++++
 1 files changed, 34 insertions(+), 0 deletions(-)
mm/mincore.c +34 -0
index 8cb508f..f977e3e 100644
@@ -14,6 +14,7 @@
 #include <linux/syscalls.h>
 #include <linux/swap.h>
 #include <linux/swapops.h>
+#include <linux/hugetlb.h>
 
 #include <asm/uaccess.h>
 #include <asm/pgtable.h>
@@ -72,6 +73,39 @@ static long do_mincore(unsigned long addr, unsigned char *vec, unsigned long pag
 	if (!vma || addr < vma->vm_start)
 		return -ENOMEM;
 
+	if (is_vm_hugetlb_page(vma)) {
+		struct hstate *h;
+		unsigned long nr_huge;
+		unsigned char present;
+		i = 0;
+		nr = min(pages, (vma->vm_end - addr) >> PAGE_SHIFT);
+		h = hstate_vma(vma);
+		nr_huge = ((addr + pages * PAGE_SIZE - 1) >> huge_page_shift(h))
+			  - (addr >> huge_page_shift(h)) + 1;
+		nr_huge = min(nr_huge,
+			      (vma->vm_end - addr) >> huge_page_shift(h));
+		while (1) {
+			/* hugepage always in RAM for now,
+			 * but generally it needs to be check */
+			ptep = huge_pte_offset(current->mm,
+					       addr & huge_page_mask(h));
+			present = !!(ptep &&
+				     !huge_pte_none(huge_ptep_get(ptep)));
+			while (1) {
+				vec[i++] = present;
+				addr += PAGE_SIZE;
+				/* reach buffer limit */
+				if (i == nr)
+					return nr;
+				/* check hugepage border */
+				if (!((addr & ~huge_page_mask(h))
+				      >> PAGE_SHIFT))
+					break;
+			}
+		}
+		return nr;
+	}
+
 	/*
 	 * Calculate how many pages there are left in the last level of the
 	 * PTE array for our address.
Signature · 8 lines
1.6.0.6


--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
maintainer
#2 lore raw
Quoted 5 lines · On Mon, 07 Dec 2009 16:59:14 +0900 Naoya Horiguchi <[email protected]>…

Most callers of pmd_none_or_clear_bad() check whether the target page is in a hugepage or not, but mincore() and walk_page_range() do not check it. So if we use mincore() on a hugepage on x86 machine, the hugepage memory is leaked as shown below.
This patch fixes it by extending mincore() system call to support hugepages.

This bug is fairly embarrassing. I tagged the patch for a -stable backport.

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to [email protected]. For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>

#3 lore raw

PATCH 1/2 mm hugetlb x86: fix hugepage memory leak in walk_page_range()

In reply to a message not in the archive
---
 mm/pagewalk.c |   15 ++++++++++++++-
 1 files changed, 14 insertions(+), 1 deletions(-)
mm/pagewalk.c +14 -1
index d5878be..3d88824 100644
@@ -1,6 +1,7 @@
 #include <linux/mm.h>
 #include <linux/highmem.h>
 #include <linux/sched.h>
+#include <linux/hugetlb.h>
 
 static int walk_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 			  struct mm_walk *walk)
@@ -107,6 +108,7 @@ int walk_page_range(unsigned long addr, unsigned long end,
 	pgd_t *pgd;
 	unsigned long next;
 	int err = 0;
+	struct vm_area_struct *vma;
 
 	if (addr >= end)
 		return err;
@@ -117,11 +119,21 @@ int walk_page_range(unsigned long addr, unsigned long end,
 	pgd = pgd_offset(walk->mm, addr);
 	do {
 		next = pgd_addr_end(addr, end);
+
+		/* skip hugetlb vma to avoid hugepage PMD being cleared
+		 * in pmd_none_or_clear_bad(). */
+		vma = find_vma(walk->mm, addr);
+		if (is_vm_hugetlb_page(vma)) {
+			next = (vma->vm_end < next) ? vma->vm_end : next;
+			continue;
+		}
+
 		if (pgd_none_or_clear_bad(pgd)) {
 			if (walk->pte_hole)
 				err = walk->pte_hole(addr, next, walk);
 			if (err)
 				break;
+			pgd++;
 			continue;
 		}
 		if (walk->pgd_entry)
@@ -131,7 +143,8 @@ int walk_page_range(unsigned long addr, unsigned long end,
 			err = walk_pud_range(pgd, addr, next, walk);
 		if (err)
 			break;
-	} while (pgd++, addr = next, addr != end);
+		pgd++;
+	} while (addr = next, addr != end);
 
 	return err;
 }
Signature · 8 lines
1.6.0.6


--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>
#4 lore raw

PATCH 2/2 mm hugetlb: add hugepage support to pagemap

In reply to a message not in the archive

This patch enables to extract pfn of the hugepage from /proc/pid/pagemap in architecture independent manner.

Cc Andi Kleen <[email protected]>
Cc Wu Fengguang <[email protected]>
---
 fs/proc/task_mmu.c |   43 +++++++++++++++++++++++++++++++++++++++++++
 include/linux/mm.h |    3 +++
 mm/pagewalk.c      |   18 ++++++++++++++++--
 3 files changed, 62 insertions(+), 2 deletions(-)
fs/proc/task_mmu.c +43 -0
index 2a1bef9..5d8e86b 100644
@@ -650,6 +650,48 @@ static int pagemap_pte_range(pmd_t *pmd, unsigned long addr, unsigned long end,
 	return err;
 }
 
+static u64 huge_pte_to_pagemap_entry(pte_t pte, int offset)
+{
+	u64 pme = 0;
+	if (pte_present(pte))
+		pme = PM_PFRAME(pte_pfn(pte) + offset)
+			| PM_PSHIFT(PAGE_SHIFT) | PM_PRESENT;
+	return pme;
+}
+
+static int pagemap_hugetlb_range(pte_t *pte, unsigned long addr,
+				 unsigned long end, struct mm_walk *walk)
+{
+	struct vm_area_struct *vma;
+	struct pagemapread *pm = walk->private;
+	struct hstate *hs = NULL;
+	int err = 0;
+
+	vma = find_vma(walk->mm, addr);
+	hs = hstate_vma(vma);
+	for (; addr != end; addr += PAGE_SIZE) {
+		u64 pfn = PM_NOT_PRESENT;
+
+		if (vma && (addr >= vma->vm_end)) {
+			vma = find_vma(walk->mm, addr);
+			hs = hstate_vma(vma);
+		}
+
+		if (vma && (vma->vm_start <= addr) && is_vm_hugetlb_page(vma)) {
+			/* calculate pfn of the "raw" page in the hugepage. */
+			int offset = (addr & ~huge_page_mask(hs)) >> PAGE_SHIFT;
+			pfn = huge_pte_to_pagemap_entry(*pte, offset);
+		}
+		err = add_to_pagemap(addr, pfn, pm);
+		if (err)
+			return err;
+	}
+
+	cond_resched();
+
+	return err;
+}
+
 /*
  * /proc/pid/pagemap - an array mapping virtual pages to pfns
  *
@@ -742,6 +784,7 @@ static ssize_t pagemap_read(struct file *file, char __user *buf,
 
 	pagemap_walk.pmd_entry = pagemap_pte_range;
 	pagemap_walk.pte_hole = pagemap_pte_hole;
+	pagemap_walk.hugetlb_entry = pagemap_hugetlb_range;
 	pagemap_walk.mm = mm;
 	pagemap_walk.private = &pm;
 
include/linux/mm.h +3 -0
index 4d33403..14835f0 100644
@@ -758,6 +758,7 @@ unsigned long unmap_vmas(struct mmu_gather **tlb,
  * @pmd_entry: if set, called for each non-empty PMD (3rd-level) entry
  * @pte_entry: if set, called for each non-empty PTE (4th-level) entry
  * @pte_hole: if set, called for each hole at all levels
+ * @hugetlb_entry: if set, called for each hugetlb entry
  *
  * (see walk_page_range for more details)
  */
@@ -767,6 +768,8 @@ struct mm_walk {
 	int (*pmd_entry)(pmd_t *, unsigned long, unsigned long, struct mm_walk *);
 	int (*pte_entry)(pte_t *, unsigned long, unsigned long, struct mm_walk *);
 	int (*pte_hole)(unsigned long, unsigned long, struct mm_walk *);
+	int (*hugetlb_entry)(pte_t *, unsigned long, unsigned long,
+			     struct mm_walk *);
 	struct mm_struct *mm;
 	void *private;
 };
mm/pagewalk.c +16 -2
index 3d88824..2d27a23 100644
@@ -106,9 +106,11 @@ int walk_page_range(unsigned long addr, unsigned long end,
 		    struct mm_walk *walk)
 {
 	pgd_t *pgd;
+	pte_t *pte;
 	unsigned long next;
 	int err = 0;
 	struct vm_area_struct *vma;
+	struct hstate *hs;
 
 	if (addr >= end)
 		return err;
@@ -120,11 +122,23 @@ int walk_page_range(unsigned long addr, unsigned long end,
 	do {
 		next = pgd_addr_end(addr, end);
 
-		/* skip hugetlb vma to avoid hugepage PMD being cleared
-		 * in pmd_none_or_clear_bad(). */
+		/*
+		 * handle hugetlb vma individually because pagetable walk for
+		 * the hugetlb page is dependent on the architecture and
+		 * we can't handled it in the same manner as non-huge pages.
+		 */
 		vma = find_vma(walk->mm, addr);
 		if (is_vm_hugetlb_page(vma)) {
 			next = (vma->vm_end < next) ? vma->vm_end : next;
+			hs = hstate_vma(vma);
+			pte = huge_pte_offset(walk->mm,
+					      addr & huge_page_mask(hs));
+			if (pte && !huge_pte_none(huge_ptep_get(pte))
+			    && walk->hugetlb_entry)
+				err = walk->hugetlb_entry(pte, addr,
+							  next, walk);
+			if (err)
+				break;
 			continue;
 		}
 
Signature · 8 lines
1.6.0.6


--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to [email protected].  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"[email protected]"> [email protected] </a>