Some cleanups for shmem

Move the suitable huge orders validation into shmem_suitable_orders() for tmpfs, which can reuse some code to simplify the logic.

In addition, we don't have special handling for the error code -E2BIG when checking for conflicts with PMD sized THP in the pagecache for tmpfs, instead, it will just fallback to order-0 allocations like this patch does, so this simplification will not add functional changes.

Signed-off-by Baolin Wang <[email protected]>

---
 mm/shmem.c | 39 +++++++++++++++------------------------
 1 file changed, 15 insertions(+), 24 deletions(-)

mm/shmem.c +15 -24

index f24dfbd387ba..db7e9808830f 100644
@@ -1685,19 +1685,29 @@ static unsigned long shmem_suitable_orders(struct inode *inode, struct vm_fault
 					   struct address_space *mapping, pgoff_t index,
 					   unsigned long orders)
 {
-	struct vm_area_struct *vma = vmf->vma;
+	struct vm_area_struct *vma = vmf ? vmf->vma : NULL;
 	unsigned long pages;
 	int order;
 
-	orders = thp_vma_suitable_orders(vma, vmf->address, orders);
-	if (!orders)
-		return 0;
+	if (vma) {
+		orders = thp_vma_suitable_orders(vma, vmf->address, orders);
+		if (!orders)
+			return 0;
+	}
 
 	/* Find the highest order that can add into the page cache */
 	order = highest_order(orders);
 	while (orders) {
 		pages = 1UL << order;
 		index = round_down(index, pages);
+		/*
+		 * Check for conflict before waiting on a huge allocation.
+		 * Conflict might be that a huge page has just been allocated
+		 * and added to page cache by a racing thread, or that there
+		 * is already at least one small page in the huge extent.
+		 * Be careful to retry when appropriate, but not forever!
+		 * Elsewhere -EEXIST would be the right code, but not here.
+		 */
 		if (!xa_find(&mapping->i_pages, &index,
 			     index + pages - 1, XA_PRESENT))
 			break;

@@ -1735,7 +1745,6 @@ static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf,
 {
 	struct address_space *mapping = inode->i_mapping;
 	struct shmem_inode_info *info = SHMEM_I(inode);
-	struct vm_area_struct *vma = vmf ? vmf->vma : NULL;
 	unsigned long suitable_orders = 0;
 	struct folio *folio = NULL;
 	long pages;

@@ -1745,26 +1754,8 @@ static struct folio *shmem_alloc_and_add_folio(struct vm_fault *vmf,
 		orders = 0;
 
 	if (orders > 0) {
-		if (vma && vma_is_anon_shmem(vma)) {
-			suitable_orders = shmem_suitable_orders(inode, vmf,
+		suitable_orders = shmem_suitable_orders(inode, vmf,
 							mapping, index, orders);
-		} else if (orders & BIT(HPAGE_PMD_ORDER)) {
-			pages = HPAGE_PMD_NR;
-			suitable_orders = BIT(HPAGE_PMD_ORDER);
-			index = round_down(index, HPAGE_PMD_NR);
-
-			/*
-			 * Check for conflict before waiting on a huge allocation.
-			 * Conflict might be that a huge page has just been allocated
-			 * and added to page cache by a racing thread, or that there
-			 * is already at least one small page in the huge extent.
-			 * Be careful to retry when appropriate, but not forever!
-			 * Elsewhere -EEXIST would be the right code, but not here.
-			 */
-			if (xa_find(&mapping->i_pages, &index,
-				    index + HPAGE_PMD_NR - 1, XA_PRESENT))
-				return ERR_PTR(-E2BIG);
-		}
 
 		order = highest_order(suitable_orders);
 		while (suitable_orders) {

Move the suitable huge orders validation into shmem_suitable_orders() for tmpfs, which can reuse some code to simplify the logic.

In addition, we don't have special handling for the error code -E2BIG when checking for conflicts with PMD sized THP in the pagecache for tmpfs, instead, it will just fallback to order-0 allocations like this patch does, so this simplification will not add functional changes.

Signed-off-by Baolin Wang <[email protected]>

---

The shmem_is_huge() is now used to check if the top-level huge page is enabled, thus rename it to reflect its usage.

Signed-off-by Baolin Wang <[email protected]>

---
 include/linux/shmem_fs.h |  9 +++++----
 mm/huge_memory.c         |  5 +++--
 mm/shmem.c               | 15 ++++++++-------
 3 files changed, 16 insertions(+), 13 deletions(-)

include/linux/shmem_fs.h +5 -4

index 1d06b1e5408a..405ee8d3589a 100644
@@ -111,14 +111,15 @@ extern void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end);
 int shmem_unuse(unsigned int type);
 
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
-extern bool shmem_is_huge(struct inode *inode, pgoff_t index, bool shmem_huge_force,
-			  struct mm_struct *mm, unsigned long vm_flags);
+extern bool shmem_huge_global_enabled(struct inode *inode, pgoff_t index, bool shmem_huge_force,
+				      struct mm_struct *mm, unsigned long vm_flags);
 unsigned long shmem_allowable_huge_orders(struct inode *inode,
 				struct vm_area_struct *vma, pgoff_t index,
 				bool global_huge);
 #else
-static __always_inline bool shmem_is_huge(struct inode *inode, pgoff_t index, bool shmem_huge_force,
-					  struct mm_struct *mm, unsigned long vm_flags)
+static __always_inline bool shmem_huge_global_enabled(struct inode *inode, pgoff_t index,
+						      bool shmem_huge_force, struct mm_struct *mm,
+						      unsigned long vm_flags)
 {
 	return false;
 }

mm/huge_memory.c +3 -2

index f9696c94e211..cc9bad12be75 100644
@@ -152,8 +152,9 @@ unsigned long __thp_vma_allowable_orders(struct vm_area_struct *vma,
 	 * own flags.
 	 */
 	if (!in_pf && shmem_file(vma->vm_file)) {
-		bool global_huge = shmem_is_huge(file_inode(vma->vm_file), vma->vm_pgoff,
-							!enforce_sysfs, vma->vm_mm, vm_flags);
+		bool global_huge = shmem_huge_global_enabled(file_inode(vma->vm_file),
+							     vma->vm_pgoff, !enforce_sysfs,
+							     vma->vm_mm, vm_flags);
 
 		if (!vma_is_anon_shmem(vma))
 			return global_huge ? orders : 0;

mm/shmem.c +8 -7

index db7e9808830f..1445dcd39b6f 100644
@@ -548,9 +548,9 @@ static bool shmem_confirm_swap(struct address_space *mapping,
 
 static int shmem_huge __read_mostly = SHMEM_HUGE_NEVER;
 
-static bool __shmem_is_huge(struct inode *inode, pgoff_t index,
-			    bool shmem_huge_force, struct mm_struct *mm,
-			    unsigned long vm_flags)
+static bool __shmem_huge_global_enabled(struct inode *inode, pgoff_t index,
+					bool shmem_huge_force, struct mm_struct *mm,
+					unsigned long vm_flags)
 {
 	loff_t i_size;
 

@@ -581,14 +581,15 @@ static bool __shmem_is_huge(struct inode *inode, pgoff_t index,
 	}
 }
 
-bool shmem_is_huge(struct inode *inode, pgoff_t index,
+bool shmem_huge_global_enabled(struct inode *inode, pgoff_t index,
 		   bool shmem_huge_force, struct mm_struct *mm,
 		   unsigned long vm_flags)
 {
 	if (HPAGE_PMD_ORDER > MAX_PAGECACHE_ORDER)
 		return false;
 
-	return __shmem_is_huge(inode, index, shmem_huge_force, mm, vm_flags);
+	return __shmem_huge_global_enabled(inode, index, shmem_huge_force,
+					   mm, vm_flags);
 }
 
 #if defined(CONFIG_SYSFS)

@@ -1156,7 +1157,7 @@ static int shmem_getattr(struct mnt_idmap *idmap,
 			STATX_ATTR_NODUMP);
 	generic_fillattr(idmap, request_mask, inode, stat);
 
-	if (shmem_is_huge(inode, 0, false, NULL, 0))
+	if (shmem_huge_global_enabled(inode, 0, false, NULL, 0))
 		stat->blksize = HPAGE_PMD_SIZE;
 
 	if (request_mask & STATX_BTIME) {

@@ -2153,7 +2154,7 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,
 		return 0;
 	}
 
-	huge = shmem_is_huge(inode, index, false, fault_mm,
+	huge = shmem_huge_global_enabled(inode, index, false, fault_mm,
 			     vma ? vma->vm_flags : 0);
 	/* Find hugepage orders that are allowed for anonymous shmem. */
 	if (vma && vma_is_anon_shmem(vma))

The shmem_is_huge() is now used to check if the top-level huge page is enabled, thus rename it to reflect its usage.

Signed-off-by Baolin Wang <[email protected]>

---

Move shmem_huge_global_enabled() into the shmem_allowable_huge_orders() function, so that shmem_allowable_huge_orders() can also help to find the allowable huge orders for tmpfs. Moreover the shmem_huge_global_enabled() can become static.

No functional changes.

Signed-off-by Baolin Wang <[email protected]>

---
 include/linux/shmem_fs.h | 12 ++----------
 mm/huge_memory.c         | 12 +++---------
 mm/shmem.c               | 41 ++++++++++++++++++++++++++--------------
 3 files changed, 32 insertions(+), 33 deletions(-)

include/linux/shmem_fs.h +2 -10

index 405ee8d3589a..1564d7d3ca61 100644
@@ -111,21 +111,13 @@ extern void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end);
 int shmem_unuse(unsigned int type);
 
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
-extern bool shmem_huge_global_enabled(struct inode *inode, pgoff_t index, bool shmem_huge_force,
-				      struct mm_struct *mm, unsigned long vm_flags);
 unsigned long shmem_allowable_huge_orders(struct inode *inode,
 				struct vm_area_struct *vma, pgoff_t index,
-				bool global_huge);
+				bool shmem_huge_force);
 #else
-static __always_inline bool shmem_huge_global_enabled(struct inode *inode, pgoff_t index,
-						      bool shmem_huge_force, struct mm_struct *mm,
-						      unsigned long vm_flags)
-{
-	return false;
-}
 static inline unsigned long shmem_allowable_huge_orders(struct inode *inode,
 				struct vm_area_struct *vma, pgoff_t index,
-				bool global_huge)
+				bool shmem_huge_force)
 {
 	return 0;
 }

mm/huge_memory.c +3 -9

index cc9bad12be75..f69980b5b5fc 100644
@@ -151,16 +151,10 @@ unsigned long __thp_vma_allowable_orders(struct vm_area_struct *vma,
 	 * Must be done before hugepage flags check since shmem has its
 	 * own flags.
 	 */
-	if (!in_pf && shmem_file(vma->vm_file)) {
-		bool global_huge = shmem_huge_global_enabled(file_inode(vma->vm_file),
-							     vma->vm_pgoff, !enforce_sysfs,
-							     vma->vm_mm, vm_flags);
-
-		if (!vma_is_anon_shmem(vma))
-			return global_huge ? orders : 0;
+	if (!in_pf && shmem_file(vma->vm_file))
 		return shmem_allowable_huge_orders(file_inode(vma->vm_file),
-							vma, vma->vm_pgoff, global_huge);
-	}
+						   vma, vma->vm_pgoff,
+						   !enforce_sysfs);
 
 	if (!vma_is_anonymous(vma)) {
 		/*

mm/shmem.c +8 -1

index 1445dcd39b6f..4d274f5a17d9 100644
@@ -581,7 +581,7 @@ static bool __shmem_huge_global_enabled(struct inode *inode, pgoff_t index,
 	}
 }
 
-bool shmem_huge_global_enabled(struct inode *inode, pgoff_t index,
+static bool shmem_huge_global_enabled(struct inode *inode, pgoff_t index,
 		   bool shmem_huge_force, struct mm_struct *mm,
 		   unsigned long vm_flags)
 {

@@ -772,6 +772,13 @@ static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
 {
 	return 0;
 }
+
+static bool shmem_huge_global_enabled(struct inode *inode, pgoff_t index,
+		bool shmem_huge_force, struct mm_struct *mm,
+		unsigned long vm_flags)
+{
+	return false;
+}
 #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 
 /*

@@ -1625,27 +1632,39 @@ static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp)
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE

mm/shmem.c +16 -11

 	/*
 	 * Check all the (large) orders below HPAGE_PMD_ORDER + 1 that
 	 * are enabled for this vma.
 	 */
 	unsigned long orders = BIT(PMD_ORDER + 1) - 1;
+	bool global_huge;
 	loff_t i_size;
 	int order;
 
-	if ((vm_flags & VM_NOHUGEPAGE) ||
-	    test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags))
+	if (vma && ((vm_flags & VM_NOHUGEPAGE) ||
+	    test_bit(MMF_DISABLE_THP, &vma->vm_mm->flags)))
 		return 0;
 
 	/* If the hardware/firmware marked hugepage support disabled. */
 	if (transparent_hugepage_flags & (1 << TRANSPARENT_HUGEPAGE_UNSUPPORTED))
 		return 0;
 
+	global_huge = shmem_huge_global_enabled(inode, index, shmem_huge_force,
+						fault_mm, vm_flags);
+	if (!vma || !vma_is_anon_shmem(vma)) {
+		/*
+		 * For tmpfs, we now only support PMD sized THP if huge page
+		 * is enabled, otherwise fallback to order 0.
+		 */
+		return global_huge ? BIT(HPAGE_PMD_ORDER) : 0;
+	}
+
 	/*
 	 * Following the 'deny' semantics of the top level, force the huge
 	 * option off from all mounts.

@@ -2081,7 +2100,7 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,
 	struct mm_struct *fault_mm;
 	struct folio *folio;
 	int error;
-	bool alloced, huge;
+	bool alloced;
 	unsigned long orders = 0;
 
 	if (WARN_ON_ONCE(!shmem_mapping(inode->i_mapping)))

@@ -2154,14 +2173,8 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,
 		return 0;
 	}
 
-	huge = shmem_huge_global_enabled(inode, index, false, fault_mm,
-			     vma ? vma->vm_flags : 0);
-	/* Find hugepage orders that are allowed for anonymous shmem. */
-	if (vma && vma_is_anon_shmem(vma))
-		orders = shmem_allowable_huge_orders(inode, vma, index, huge);
-	else if (huge)
-		orders = BIT(HPAGE_PMD_ORDER);
-
+	/* Find hugepage orders that are allowed for anonymous shmem and tmpfs. */
+	orders = shmem_allowable_huge_orders(inode, vma, index, false);
 	if (orders > 0) {
 		gfp_t huge_gfp;
 

Quoted 7 lines · Baolin Wang

Move shmem_huge_global_enabled() into the shmem_allowable_huge_orders() function, so that shmem_allowable_huge_orders() can also help to find the allowable huge orders for tmpfs. Moreover the shmem_huge_global_enabled() can become static.

No functional changes.

Signed-off-by Baolin Wang <[email protected]>

one nit below, but either way:

Reviewed-by Ryan Roberts <[email protected]>

Quoted 95 lines

---
  include/linux/shmem_fs.h | 12 ++----------
  mm/huge_memory.c         | 12 +++---------
  mm/shmem.c               | 41 ++++++++++++++++++++++++++--------------
  3 files changed, 32 insertions(+), 33 deletions(-)

include/linux/shmem_fs.h +2 -10

index 405ee8d3589a..1564d7d3ca61 100644
@@ -111,21 +111,13 @@ extern void shmem_truncate_range(struct inode *inode, loff_t start, loff_t end);
  int shmem_unuse(unsigned int type);
 
  #ifdef CONFIG_TRANSPARENT_HUGEPAGE
-extern bool shmem_huge_global_enabled(struct inode *inode, pgoff_t index, bool shmem_huge_force,
-				      struct mm_struct *mm, unsigned long vm_flags);
  unsigned long shmem_allowable_huge_orders(struct inode *inode,
  				struct vm_area_struct *vma, pgoff_t index,
-				bool global_huge);
+				bool shmem_huge_force);
  #else
-static __always_inline bool shmem_huge_global_enabled(struct inode *inode, pgoff_t index,
-						      bool shmem_huge_force, struct mm_struct *mm,
-						      unsigned long vm_flags)
-{
-	return false;
-}
  static inline unsigned long shmem_allowable_huge_orders(struct inode *inode,
  				struct vm_area_struct *vma, pgoff_t index,
-				bool global_huge)
+				bool shmem_huge_force)
  {
  	return 0;
  }

mm/huge_memory.c +3 -9

index cc9bad12be75..f69980b5b5fc 100644
@@ -151,16 +151,10 @@ unsigned long __thp_vma_allowable_orders(struct vm_area_struct *vma,
  	 * Must be done before hugepage flags check since shmem has its
  	 * own flags.
  	 */
-	if (!in_pf && shmem_file(vma->vm_file)) {
-		bool global_huge = shmem_huge_global_enabled(file_inode(vma->vm_file),
-							     vma->vm_pgoff, !enforce_sysfs,
-							     vma->vm_mm, vm_flags);
-
-		if (!vma_is_anon_shmem(vma))
-			return global_huge ? orders : 0;
+	if (!in_pf && shmem_file(vma->vm_file))
  		return shmem_allowable_huge_orders(file_inode(vma->vm_file),
-							vma, vma->vm_pgoff, global_huge);
-	}
+						   vma, vma->vm_pgoff,
+						   !enforce_sysfs);
 
  	if (!vma_is_anonymous(vma)) {
  		/*

mm/shmem.c +9 -3

index 1445dcd39b6f..4d274f5a17d9 100644
@@ -581,7 +581,7 @@ static bool __shmem_huge_global_enabled(struct inode *inode, pgoff_t index,
  	}
  }
 
-bool shmem_huge_global_enabled(struct inode *inode, pgoff_t index,
+static bool shmem_huge_global_enabled(struct inode *inode, pgoff_t index,
  		   bool shmem_huge_force, struct mm_struct *mm,
  		   unsigned long vm_flags)
  {

@@ -772,6 +772,13 @@ static unsigned long shmem_unused_huge_shrink(struct shmem_sb_info *sbinfo,
  {
  	return 0;
  }
+
+static bool shmem_huge_global_enabled(struct inode *inode, pgoff_t index,
+		bool shmem_huge_force, struct mm_struct *mm,
+		unsigned long vm_flags)
+{
+	return false;
+}
  #endif /* CONFIG_TRANSPARENT_HUGEPAGE */
 
  /*

@@ -1625,27 +1632,39 @@ static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp)
  #ifdef CONFIG_TRANSPARENT_HUGEPAGE
  unsigned long shmem_allowable_huge_orders(struct inode *inode,
  				struct vm_area_struct *vma, pgoff_t index,
-				bool global_huge)
+				bool shmem_huge_force)
  {
  	unsigned long mask = READ_ONCE(huge_shmem_orders_always);
  	unsigned long within_size_orders = READ_ONCE(huge_shmem_orders_within_size);
-	unsigned long vm_flags = vma->vm_flags;

mm/shmem.c static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp) · line 1640

Quoted 2 lines

mm/shmem.c +2 -0

+	unsigned long vm_flags = vma ? vma->vm_flags : 0;
+	struct mm_struct *fault_mm = vma ? vma->vm_mm : NULL;

nit: rather than deriving the fault_mm here, I wonder if its cleaner to just pass vma to shmem_huge_global_enabled()? shmem_huge_global_enabled() is just using it as a guard to access vm_flags, which you can just as easily do by testing the vma for non-NULL. And you can access mm flags with vma->vm_mm->flags after testing the vma too.

Quoted 11 lines

Quoted 11 lines

   /*

mm/shmem.c +1 -2

@@ -1625,27 +1632,39 @@ static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp)
   #ifdef CONFIG_TRANSPARENT_HUGEPAGE
   unsigned long shmem_allowable_huge_orders(struct inode *inode,
   				struct vm_area_struct *vma, pgoff_t index,
-				bool global_huge)
+				bool shmem_huge_force)
   {
   	unsigned long mask = READ_ONCE(huge_shmem_orders_always);
   	unsigned long within_size_orders = READ_ONCE(huge_shmem_orders_within_size);
-	unsigned long vm_flags = vma->vm_flags;

Quoted 8 lines

mm/shmem.c static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp) · line 1640

Quoted 2 lines

mm/shmem.c +2 -0

+	unsigned long vm_flags = vma ? vma->vm_flags : 0;
+	struct mm_struct *fault_mm = vma ? vma->vm_mm : NULL;

nit: rather than deriving the fault_mm here, I wonder if its cleaner to just pass vma to shmem_huge_global_enabled()? shmem_huge_global_enabled() is just using it as a guard to access vm_flags, which you can just as easily do by testing the vma for non-NULL. And you can access mm flags with vma->vm_mm->flags after testing the vma too.

Make sense to me, and will do in next version.

Quoted 21 lines

Quoted 20 lines

Quoted 4 lines

   /*

mm/shmem.c +0 -0

@@ -1625,27 +1632,39 @@ static gfp_t limit_gfp_mask(gfp_t huge_gfp, 
 gfp_t limit_gfp)
    #ifdef CONFIG_TRANSPARENT_HUGEPAGE

mm/shmem.c static gfp_t limit_gfp_mask(gfp_t huge_gfp, · line 1641

Quoted 10 lines

mm/shmem.c +3 -2

    unsigned long shmem_allowable_huge_orders(struct inode *inode,
                    struct vm_area_struct *vma, pgoff_t index,
-                bool global_huge)
+                bool shmem_huge_force)
    {
        unsigned long mask = READ_ONCE(huge_shmem_orders_always);
        unsigned long within_size_orders = READ_ONCE(huge_shmem_orders_within_size);
-    unsigned long vm_flags = vma->vm_flags;
+    unsigned long vm_flags = vma ? vma->vm_flags : 0;
+    struct mm_struct *fault_mm = vma ? vma->vm_mm : NULL;

nit: rather than deriving the fault_mm here, I wonder if its cleaner to just pass vma to shmem_huge_global_enabled()? shmem_huge_global_enabled() is just using it as a guard to access vm_flags, which you can just as easily do by testing the vma for non-NULL. And you can access mm flags with vma->vm_mm->flags after testing the vma too.

Quoted 1 line

Make sense to me, and will do in next version.

Feel free to add my

Acked-by David Hildenbrand <[email protected]>

Changes from v1:
- Add a dummy function in case CONFIG_TRANSPARENT_HUGEPAGE is not enabled, which fixes a building error reported by kernel test robot.

The only difference I'm seeing from the v1 series is the below update to [3/3]: