[RFC PATCH 0/3] Support large folios for tmpfs

[RFC PATCH 0/3] Support large folios for tmpfs

[Baolin Wang]

Hi,

This RFC patch series attempts to support large folios for tmpfs. The first
two patches are based on Daniel's previous patches in [1], mainly using the
length in the write and fallocate paths to get a highest order hint for large
order allocation. The last patch adds mTHP filter control for tmpfs if mTHP
is set for the following reasons:

1. Maintain backward compatibility for the control interface. Tmpfs already
has a global 'huge=' mount option and '/sys/kernel/mm/transparent_hugepage/shmem_enabled'
interface to control large order allocations. mTHP extends this capability to a
per-size basis while maintaining good interface compatibility.

2. For the large order allocation of writable mmap() faults in tmpfs, we need
something like the mTHP interfaces to control large orders, as well as ensuring
consistent interfaces with shmem.

3. Ryan pointed out that large order allocations based on write length could
lead to memory fragmentation issue. Just quoting Ryan's comment [2]:
"And it's possible (likely even, in my opinion) that allocating lots of different
folio sizes will exacerbate memory fragmentation, leading to more order-0
fallbacks, which would hurt the overall system performance in the long run, vs
restricting to a couple of folio sizes."

4. Some hardware preferences, such as for the ARM64 architecture, can better
utilize the cont-pte feature to reduce TLB pressure and optimize performance with
a 64K size folio. Using mTHP can better leverage these hardware advantages.

Please correct me if I missed something. Thanks a lot.

[1] https://lore.kernel.org/all/20240515055719.32577-1-da.gomez@samsung.com/
[2] https://lore.kernel.org/all/e83e1687-3e3c-40d0-bf0e-225871647092@arm.com/

Baolin Wang (1):
  mm: shmem: use mTHP interface to control huge orders for tmpfs

Daniel Gomez (2):
  mm: shmem: add file length arg in shmem_get_folio() path
  mm: shmem: add large folio support to the write and fallocate paths

 fs/xfs/scrub/xfile.c     |   6 +--
 fs/xfs/xfs_buf_mem.c     |   3 +-
 include/linux/shmem_fs.h |   6 +--
 mm/huge_memory.c         |   2 +-
 mm/khugepaged.c          |   3 +-
 mm/memory.c              |   4 +-
 mm/shmem.c               | 101 +++++++++++++++++++++++++++++----------
 mm/userfaultfd.c         |   2 +-
 8 files changed, 91 insertions(+), 36 deletions(-)

-- 
2.39.3

[RFC PATCH 1/3] mm: shmem: add file length arg in shmem_get_folio() path

[Baolin Wang]

From: Daniel Gomez <da.gomez@samsung.com>

In preparation for large folio in the write and fallocate paths, add
file length argument in shmem_get_folio() path to be able to calculate
the folio order based on the file size. Use of order-0 (PAGE_SIZE) for
read, page cache read, and vm fault.

This enables high order folios in the write and fallocate path once the
folio order is calculated based on the length.

Signed-off-by: Daniel Gomez <da.gomez@samsung.com>
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
---
 fs/xfs/scrub/xfile.c     |  6 +++---
 fs/xfs/xfs_buf_mem.c     |  3 ++-
 include/linux/shmem_fs.h |  2 +-
 mm/khugepaged.c          |  3 ++-
 mm/shmem.c               | 28 ++++++++++++++++------------
 mm/userfaultfd.c         |  2 +-
 6 files changed, 25 insertions(+), 19 deletions(-)

diff --git a/fs/xfs/scrub/xfile.c b/fs/xfs/scrub/xfile.c
index d848222f802b..d814d9d786d3 100644
--- a/fs/xfs/scrub/xfile.c
+++ b/fs/xfs/scrub/xfile.c
@@ -127,7 +127,7 @@ xfile_load(
 		unsigned int	offset;
 
 		if (shmem_get_folio(inode, pos >> PAGE_SHIFT, &folio,
-				SGP_READ) < 0)
+				SGP_READ, PAGE_SIZE) < 0)
 			break;
 		if (!folio) {
 			/*
@@ -197,7 +197,7 @@ xfile_store(
 		unsigned int	offset;
 
 		if (shmem_get_folio(inode, pos >> PAGE_SHIFT, &folio,
-				SGP_CACHE) < 0)
+				SGP_CACHE, PAGE_SIZE) < 0)
 			break;
 		if (filemap_check_wb_err(inode->i_mapping, 0)) {
 			folio_unlock(folio);
@@ -268,7 +268,7 @@ xfile_get_folio(
 
 	pflags = memalloc_nofs_save();
 	error = shmem_get_folio(inode, pos >> PAGE_SHIFT, &folio,
-			(flags & XFILE_ALLOC) ? SGP_CACHE : SGP_READ);
+			(flags & XFILE_ALLOC) ? SGP_CACHE : SGP_READ, PAGE_SIZE);
 	memalloc_nofs_restore(pflags);
 	if (error)
 		return ERR_PTR(error);
diff --git a/fs/xfs/xfs_buf_mem.c b/fs/xfs/xfs_buf_mem.c
index 9bb2d24de709..784c81d35a1f 100644
--- a/fs/xfs/xfs_buf_mem.c
+++ b/fs/xfs/xfs_buf_mem.c
@@ -149,7 +149,8 @@ xmbuf_map_page(
 		return -ENOMEM;
 	}
 
-	error = shmem_get_folio(inode, pos >> PAGE_SHIFT, &folio, SGP_CACHE);
+	error = shmem_get_folio(inode, pos >> PAGE_SHIFT, &folio, SGP_CACHE,
+				PAGE_SIZE);
 	if (error)
 		return error;
 
diff --git a/include/linux/shmem_fs.h b/include/linux/shmem_fs.h
index 1564d7d3ca61..34beaca2f853 100644
--- a/include/linux/shmem_fs.h
+++ b/include/linux/shmem_fs.h
@@ -144,7 +144,7 @@ enum sgp_type {
 };
 
 int shmem_get_folio(struct inode *inode, pgoff_t index, struct folio **foliop,
-		enum sgp_type sgp);
+		enum sgp_type sgp, size_t len);
 struct folio *shmem_read_folio_gfp(struct address_space *mapping,
 		pgoff_t index, gfp_t gfp);
 
diff --git a/mm/khugepaged.c b/mm/khugepaged.c
index a5ec03ef8722..3c9dbebbdf38 100644
--- a/mm/khugepaged.c
+++ b/mm/khugepaged.c
@@ -1867,7 +1867,8 @@ static int collapse_file(struct mm_struct *mm, unsigned long addr,
 				xas_unlock_irq(&xas);
 				/* swap in or instantiate fallocated page */
 				if (shmem_get_folio(mapping->host, index,
-						&folio, SGP_NOALLOC)) {
+						&folio, SGP_NOALLOC,
+						PAGE_SIZE)) {
 					result = SCAN_FAIL;
 					goto xa_unlocked;
 				}
diff --git a/mm/shmem.c b/mm/shmem.c
index db8f74cac1a2..92ed09527682 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -980,7 +980,7 @@ static struct folio *shmem_get_partial_folio(struct inode *inode, pgoff_t index)
 	 * (although in some cases this is just a waste of time).
 	 */
 	folio = NULL;
-	shmem_get_folio(inode, index, &folio, SGP_READ);
+	shmem_get_folio(inode, index, &folio, SGP_READ, PAGE_SIZE);
 	return folio;
 }
 
@@ -2094,7 +2094,7 @@ static int shmem_swapin_folio(struct inode *inode, pgoff_t index,
  */
 static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,
 		struct folio **foliop, enum sgp_type sgp, gfp_t gfp,
-		struct vm_fault *vmf, vm_fault_t *fault_type)
+		struct vm_fault *vmf, vm_fault_t *fault_type, size_t len)
 {
 	struct vm_area_struct *vma = vmf ? vmf->vma : NULL;
 	struct mm_struct *fault_mm;
@@ -2297,10 +2297,10 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,
  * Return: 0 if successful, else a negative error code.
  */
 int shmem_get_folio(struct inode *inode, pgoff_t index, struct folio **foliop,
-		enum sgp_type sgp)
+		enum sgp_type sgp, size_t len)
 {
 	return shmem_get_folio_gfp(inode, index, foliop, sgp,
-			mapping_gfp_mask(inode->i_mapping), NULL, NULL);
+			mapping_gfp_mask(inode->i_mapping), NULL, NULL, len);
 }
 EXPORT_SYMBOL_GPL(shmem_get_folio);
 
@@ -2395,7 +2395,7 @@ static vm_fault_t shmem_fault(struct vm_fault *vmf)
 
 	WARN_ON_ONCE(vmf->page != NULL);
 	err = shmem_get_folio_gfp(inode, vmf->pgoff, &folio, SGP_CACHE,
-				  gfp, vmf, &ret);
+				  gfp, vmf, &ret, PAGE_SIZE);
 	if (err)
 		return vmf_error(err);
 	if (folio) {
@@ -2895,6 +2895,9 @@ shmem_write_begin(struct file *file, struct address_space *mapping,
 	struct folio *folio;
 	int ret = 0;
 
+	if (!mapping_large_folio_support(mapping))
+		len = min_t(size_t, len, PAGE_SIZE - offset_in_page(pos));
+
 	/* i_rwsem is held by caller */
 	if (unlikely(info->seals & (F_SEAL_GROW |
 				   F_SEAL_WRITE | F_SEAL_FUTURE_WRITE))) {
@@ -2904,7 +2907,7 @@ shmem_write_begin(struct file *file, struct address_space *mapping,
 			return -EPERM;
 	}
 
-	ret = shmem_get_folio(inode, index, &folio, SGP_WRITE);
+	ret = shmem_get_folio(inode, index, &folio, SGP_WRITE, len);
 	if (ret)
 		return ret;
 
@@ -2975,7 +2978,7 @@ static ssize_t shmem_file_read_iter(struct kiocb *iocb, struct iov_iter *to)
 				break;
 		}
 
-		error = shmem_get_folio(inode, index, &folio, SGP_READ);
+		error = shmem_get_folio(inode, index, &folio, SGP_READ, PAGE_SIZE);
 		if (error) {
 			if (error == -EINVAL)
 				error = 0;
@@ -3152,7 +3155,7 @@ static ssize_t shmem_file_splice_read(struct file *in, loff_t *ppos,
 			break;
 
 		error = shmem_get_folio(inode, *ppos / PAGE_SIZE, &folio,
-					SGP_READ);
+					SGP_READ, PAGE_SIZE);
 		if (error) {
 			if (error == -EINVAL)
 				error = 0;
@@ -3339,7 +3342,8 @@ static long shmem_fallocate(struct file *file, int mode, loff_t offset,
 			error = -ENOMEM;
 		else
 			error = shmem_get_folio(inode, index, &folio,
-						SGP_FALLOC);
+						SGP_FALLOC,
+						(end - index) << PAGE_SHIFT);
 		if (error) {
 			info->fallocend = undo_fallocend;
 			/* Remove the !uptodate folios we added */
@@ -3690,7 +3694,7 @@ static int shmem_symlink(struct mnt_idmap *idmap, struct inode *dir,
 	} else {
 		inode_nohighmem(inode);
 		inode->i_mapping->a_ops = &shmem_aops;
-		error = shmem_get_folio(inode, 0, &folio, SGP_WRITE);
+		error = shmem_get_folio(inode, 0, &folio, SGP_WRITE, PAGE_SIZE);
 		if (error)
 			goto out_remove_offset;
 		inode->i_op = &shmem_symlink_inode_operations;
@@ -3736,7 +3740,7 @@ static const char *shmem_get_link(struct dentry *dentry, struct inode *inode,
 			return ERR_PTR(-ECHILD);
 		}
 	} else {
-		error = shmem_get_folio(inode, 0, &folio, SGP_READ);
+		error = shmem_get_folio(inode, 0, &folio, SGP_READ, PAGE_SIZE);
 		if (error)
 			return ERR_PTR(error);
 		if (!folio)
@@ -5209,7 +5213,7 @@ struct folio *shmem_read_folio_gfp(struct address_space *mapping,
 	int error;
 
 	error = shmem_get_folio_gfp(inode, index, &folio, SGP_CACHE,
-				    gfp, NULL, NULL);
+				    gfp, NULL, NULL, PAGE_SIZE);
 	if (error)
 		return ERR_PTR(error);
 
diff --git a/mm/userfaultfd.c b/mm/userfaultfd.c
index e54e5c8907fa..c275e34c435a 100644
--- a/mm/userfaultfd.c
+++ b/mm/userfaultfd.c
@@ -391,7 +391,7 @@ static int mfill_atomic_pte_continue(pmd_t *dst_pmd,
 	struct page *page;
 	int ret;
 
-	ret = shmem_get_folio(inode, pgoff, &folio, SGP_NOALLOC);
+	ret = shmem_get_folio(inode, pgoff, &folio, SGP_NOALLOC, PAGE_SIZE);
 	/* Our caller expects us to return -EFAULT if we failed to find folio */
 	if (ret == -ENOENT)
 		ret = -EFAULT;
-- 
2.39.3

[RFC PATCH 2/3] mm: shmem: add large folio support to the write and fallocate paths

[Baolin Wang]

From: Daniel Gomez <da.gomez@samsung.com>

Add large folio support for shmem write and fallocate paths matching the
same high order preference mechanism used in the iomap buffered IO path
as used in __filemap_get_folio().

Add shmem_mapping_size_order() to get a hint for the order of the folio
based on the file size which takes care of the mapping requirements.

Swap does not support high order folios for now, so make it order-0 in
case swap is enabled.

If the top level huge page (controlled by '/sys/kernel/mm/transparent_hugepage/shmem_enabled')
is enabled, we just allow PMD sized THP to keep interface backward
compatibility.

Co-developed-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Signed-off-by: Daniel Gomez <da.gomez@samsung.com>
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
---
 include/linux/shmem_fs.h |  4 +--
 mm/huge_memory.c         |  2 +-
 mm/shmem.c               | 57 ++++++++++++++++++++++++++++++++++++----
 3 files changed, 55 insertions(+), 8 deletions(-)

diff --git a/include/linux/shmem_fs.h b/include/linux/shmem_fs.h
index 34beaca2f853..fb0771218f1b 100644
--- a/include/linux/shmem_fs.h
+++ b/include/linux/shmem_fs.h
@@ -113,11 +113,11 @@ int shmem_unuse(unsigned int type);
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 unsigned long shmem_allowable_huge_orders(struct inode *inode,
 				struct vm_area_struct *vma, pgoff_t index,
-				bool shmem_huge_force);
+				bool shmem_huge_force, size_t len);
 #else
 static inline unsigned long shmem_allowable_huge_orders(struct inode *inode,
 				struct vm_area_struct *vma, pgoff_t index,
-				bool shmem_huge_force)
+				bool shmem_huge_force, size_t len)
 {
 	return 0;
 }
diff --git a/mm/huge_memory.c b/mm/huge_memory.c
index e555fcdd19d4..a8fc3b9e4034 100644
--- a/mm/huge_memory.c
+++ b/mm/huge_memory.c
@@ -162,7 +162,7 @@ unsigned long __thp_vma_allowable_orders(struct vm_area_struct *vma,
 	if (!in_pf && shmem_file(vma->vm_file))
 		return shmem_allowable_huge_orders(file_inode(vma->vm_file),
 						   vma, vma->vm_pgoff,
-						   !enforce_sysfs);
+						   !enforce_sysfs, PAGE_SIZE);
 
 	if (!vma_is_anonymous(vma)) {
 		/*
diff --git a/mm/shmem.c b/mm/shmem.c
index 92ed09527682..cc0c1b790267 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -1630,10 +1630,47 @@ static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp)
 	return result;
 }
 
+/**
+ * shmem_mapping_size_order - Get maximum folio order for the given file size.
+ * @mapping: Target address_space.
+ * @index: The page index.
+ * @size: The suggested size of the folio to create.
+ *
+ * This returns a high order for folios (when supported) based on the file size
+ * which the mapping currently allows at the given index. The index is relevant
+ * due to alignment considerations the mapping might have. The returned order
+ * may be less than the size passed.
+ *
+ * Like __filemap_get_folio order calculation.
+ *
+ * Return: The order.
+ */
+static inline unsigned int
+shmem_mapping_size_order(struct address_space *mapping, pgoff_t index,
+			 size_t size, struct shmem_sb_info *sbinfo)
+{
+	unsigned int order = ilog2(size);
+
+	if ((order <= PAGE_SHIFT) ||
+	    (!mapping_large_folio_support(mapping) || !sbinfo->noswap))
+		return 0;
+
+	order -= PAGE_SHIFT;
+
+	/* If we're not aligned, allocate a smaller folio */
+	if (index & ((1UL << order) - 1))
+		order = __ffs(index);
+
+	order = min_t(size_t, order, MAX_PAGECACHE_ORDER);
+
+	/* Order-1 not supported due to THP dependency */
+	return (order == 1) ? 0 : order;
+}
+
 #ifdef CONFIG_TRANSPARENT_HUGEPAGE
 unsigned long shmem_allowable_huge_orders(struct inode *inode,
 				struct vm_area_struct *vma, pgoff_t index,
-				bool shmem_huge_force)
+				bool shmem_huge_force, size_t len)
 {
 	unsigned long mask = READ_ONCE(huge_shmem_orders_always);
 	unsigned long within_size_orders = READ_ONCE(huge_shmem_orders_within_size);
@@ -1659,10 +1696,20 @@ unsigned long shmem_allowable_huge_orders(struct inode *inode,
 						vma, 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.
+		 * For tmpfs, if top level huge page is enabled, we just allow
+		 * PMD size THP to keep interface backward compatibility.
+		 */
+		if (global_huge)
+			return BIT(HPAGE_PMD_ORDER);
+
+		/*
+		 * Otherwise, get a highest order hint based on the size of
+		 * write and fallocate paths, then will try each allowable
+		 * huge orders.
 		 */
-		return global_huge ? BIT(HPAGE_PMD_ORDER) : 0;
+		order = shmem_mapping_size_order(inode->i_mapping, index,
+						 len, SHMEM_SB(inode->i_sb));
+		return BIT(order + 1) - 1;
 	}
 
 	/*
@@ -2174,7 +2221,7 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,
 	}
 
 	/* Find hugepage orders that are allowed for anonymous shmem and tmpfs. */
-	orders = shmem_allowable_huge_orders(inode, vma, index, false);
+	orders = shmem_allowable_huge_orders(inode, vma, index, false, len);
 	if (orders > 0) {
 		gfp_t huge_gfp;
 
-- 
2.39.3

[RFC PATCH 3/3] mm: shmem: use mTHP interface to control huge orders for tmpfs

[Baolin Wang]

For the huge orders allowed by writable mmap() faults on tmpfs,
the mTHP interface is used to control the allowable huge orders,
while 'huge_shmem_orders_inherit' maintains backward compatibility
with top-level interface.

For the huge orders allowed by write() and fallocate() paths on tmpfs,
getting a highest order hint based on the size of write and fallocate
paths, then will try each allowable huge orders filtered by the mTHP
interfaces if set.

Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
---
 mm/memory.c |  4 ++--
 mm/shmem.c  | 42 ++++++++++++++++++++++--------------------
 2 files changed, 24 insertions(+), 22 deletions(-)

diff --git a/mm/memory.c b/mm/memory.c
index 802d0d8a40f9..3a7f43c66db7 100644
--- a/mm/memory.c
+++ b/mm/memory.c
@@ -4877,10 +4877,10 @@ vm_fault_t finish_fault(struct vm_fault *vmf)
 
 	/*
 	 * Using per-page fault to maintain the uffd semantics, and same
-	 * approach also applies to non-anonymous-shmem faults to avoid
+	 * approach also applies to non shmem/tmpfs faults to avoid
 	 * inflating the RSS of the process.
 	 */
-	if (!vma_is_anon_shmem(vma) || unlikely(userfaultfd_armed(vma))) {
+	if (!vma_is_shmem(vma) || unlikely(userfaultfd_armed(vma))) {
 		nr_pages = 1;
 	} else if (nr_pages > 1) {
 		pgoff_t idx = folio_page_idx(folio, page);
diff --git a/mm/shmem.c b/mm/shmem.c
index cc0c1b790267..8e60cc566196 100644
--- a/mm/shmem.c
+++ b/mm/shmem.c
@@ -1692,26 +1692,6 @@ unsigned long shmem_allowable_huge_orders(struct inode *inode,
 	if (transparent_hugepage_flags & (1 << TRANSPARENT_HUGEPAGE_UNSUPPORTED))
 		return 0;
 
-	global_huge = shmem_huge_global_enabled(inode, index, shmem_huge_force,
-						vma, vm_flags);
-	if (!vma || !vma_is_anon_shmem(vma)) {
-		/*
-		 * For tmpfs, if top level huge page is enabled, we just allow
-		 * PMD size THP to keep interface backward compatibility.
-		 */
-		if (global_huge)
-			return BIT(HPAGE_PMD_ORDER);
-
-		/*
-		 * Otherwise, get a highest order hint based on the size of
-		 * write and fallocate paths, then will try each allowable
-		 * huge orders.
-		 */
-		order = shmem_mapping_size_order(inode->i_mapping, index,
-						 len, SHMEM_SB(inode->i_sb));
-		return BIT(order + 1) - 1;
-	}
-
 	/*
 	 * Following the 'deny' semantics of the top level, force the huge
 	 * option off from all mounts.
@@ -1742,9 +1722,31 @@ unsigned long shmem_allowable_huge_orders(struct inode *inode,
 	if (vm_flags & VM_HUGEPAGE)
 		mask |= READ_ONCE(huge_shmem_orders_madvise);
 
+	global_huge = shmem_huge_global_enabled(inode, index, shmem_huge_force,
+						vma, vm_flags);
 	if (global_huge)
 		mask |= READ_ONCE(huge_shmem_orders_inherit);
 
+	/*
+	 * For the huge orders allowed by writable mmap() faults on tmpfs,
+	 * the mTHP interface is used to control the allowable huge orders,
+	 * while 'huge_shmem_orders_inherit' maintains backward compatibility
+	 * with top-level interface.
+	 *
+	 * For the huge orders allowed by write() and fallocate() paths on tmpfs,
+	 * get a highest order hint based on the size of write and fallocate
+	 * paths, then will try each allowable huge orders filtered by the mTHP
+	 * interfaces if set.
+	 */
+	if (!vma && !global_huge) {
+		int highest_order = shmem_mapping_size_order(inode->i_mapping, index, len,
+							     SHMEM_SB(inode->i_sb));
+
+		if (!mask)
+			return highest_order > 0 ? BIT(highest_order + 1) - 1 : 0;
+
+		mask &= BIT(highest_order + 1) - 1;
+	}
 	return orders & mask;
 }
 
-- 
2.39.3

Re: [RFC PATCH 2/3] mm: shmem: add large folio support to the write and fallocate paths

[Daniel Gomez]

On Wed, Jul 24, 2024 at 03:03:59PM GMT, Baolin Wang wrote:
> From: Daniel Gomez <da.gomez@samsung.com>
> 
> Add large folio support for shmem write and fallocate paths matching the
> same high order preference mechanism used in the iomap buffered IO path
> as used in __filemap_get_folio().
> 
> Add shmem_mapping_size_order() to get a hint for the order of the folio
> based on the file size which takes care of the mapping requirements.
> 
> Swap does not support high order folios for now, so make it order-0 in
> case swap is enabled.
> 
> If the top level huge page (controlled by '/sys/kernel/mm/transparent_hugepage/shmem_enabled')
> is enabled, we just allow PMD sized THP to keep interface backward
> compatibility.
> 
> Co-developed-by: Baolin Wang <baolin.wang@linux.alibaba.com>
> Signed-off-by: Daniel Gomez <da.gomez@samsung.com>
> Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
> ---
>  include/linux/shmem_fs.h |  4 +--
>  mm/huge_memory.c         |  2 +-
>  mm/shmem.c               | 57 ++++++++++++++++++++++++++++++++++++----
>  3 files changed, 55 insertions(+), 8 deletions(-)
> 
> diff --git a/include/linux/shmem_fs.h b/include/linux/shmem_fs.h
> index 34beaca2f853..fb0771218f1b 100644
> --- a/include/linux/shmem_fs.h
> +++ b/include/linux/shmem_fs.h
> @@ -113,11 +113,11 @@ int shmem_unuse(unsigned int type);
>  #ifdef CONFIG_TRANSPARENT_HUGEPAGE
>  unsigned long shmem_allowable_huge_orders(struct inode *inode,
>  				struct vm_area_struct *vma, pgoff_t index,
> -				bool shmem_huge_force);
> +				bool shmem_huge_force, size_t len);
>  #else
>  static inline unsigned long shmem_allowable_huge_orders(struct inode *inode,
>  				struct vm_area_struct *vma, pgoff_t index,
> -				bool shmem_huge_force)
> +				bool shmem_huge_force, size_t len)
>  {
>  	return 0;
>  }
> diff --git a/mm/huge_memory.c b/mm/huge_memory.c
> index e555fcdd19d4..a8fc3b9e4034 100644
> --- a/mm/huge_memory.c
> +++ b/mm/huge_memory.c
> @@ -162,7 +162,7 @@ unsigned long __thp_vma_allowable_orders(struct vm_area_struct *vma,
>  	if (!in_pf && shmem_file(vma->vm_file))
>  		return shmem_allowable_huge_orders(file_inode(vma->vm_file),
>  						   vma, vma->vm_pgoff,
> -						   !enforce_sysfs);
> +						   !enforce_sysfs, PAGE_SIZE);
>  
>  	if (!vma_is_anonymous(vma)) {
>  		/*
> diff --git a/mm/shmem.c b/mm/shmem.c
> index 92ed09527682..cc0c1b790267 100644
> --- a/mm/shmem.c
> +++ b/mm/shmem.c
> @@ -1630,10 +1630,47 @@ static gfp_t limit_gfp_mask(gfp_t huge_gfp, gfp_t limit_gfp)
>  	return result;
>  }
>  
> +/**
> + * shmem_mapping_size_order - Get maximum folio order for the given file size.
> + * @mapping: Target address_space.
> + * @index: The page index.
> + * @size: The suggested size of the folio to create.
> + *
> + * This returns a high order for folios (when supported) based on the file size
> + * which the mapping currently allows at the given index. The index is relevant
> + * due to alignment considerations the mapping might have. The returned order
> + * may be less than the size passed.
> + *
> + * Like __filemap_get_folio order calculation.
> + *
> + * Return: The order.
> + */
> +static inline unsigned int
> +shmem_mapping_size_order(struct address_space *mapping, pgoff_t index,
> +			 size_t size, struct shmem_sb_info *sbinfo)
> +{
> +	unsigned int order = ilog2(size);
> +
> +	if ((order <= PAGE_SHIFT) ||
> +	    (!mapping_large_folio_support(mapping) || !sbinfo->noswap))
> +		return 0;
> +
> +	order -= PAGE_SHIFT;
> +
> +	/* If we're not aligned, allocate a smaller folio */
> +	if (index & ((1UL << order) - 1))
> +		order = __ffs(index);
> +
> +	order = min_t(size_t, order, MAX_PAGECACHE_ORDER);
> +
> +	/* Order-1 not supported due to THP dependency */
> +	return (order == 1) ? 0 : order;
> +}

I have an updated version of shmem_mapping_size_order() that I didn't posted but
uses get_order() instead as suggested in [1]:

[1] https://lore.kernel.org/all/ZT7rd3CSr+VnKj7v@casper.infradead.org/

/**
 * shmem_mapping_size_order - Get maximum folio order for the given file size.
 * @mapping: Target address_space.
 * @index: The page index.
 * @size: The suggested size of the folio to create.
 *
 * This returns a high order for folios (when supported) based on the file size
 * which the mapping currently allows at the given index. The index is relevant
 * due to alignment considerations the mapping might have. The returned order
 * may be less than the size passed.
 *
 * Like __filemap_get_folio order calculation.
 *
 * Return: The order.
 */
static inline unsigned int
shmem_mapping_size_order(struct address_space *mapping, pgoff_t index,
                        size_t size)
 * Return: The order.
 */
static inline unsigned int
shmem_mapping_size_order(struct address_space *mapping, pgoff_t index,
                        size_t size)
{
       unsigned int order = get_order(max_t(size_t, size, PAGE_SIZE));

       if (!mapping_large_folio_support(mapping))
               return 0;

       /* If we're not aligned, allocate a smaller folio */
       if (index & ((1UL << order) - 1))
               order = __ffs(index);

       return min_t(size_t, order, MAX_PAGECACHE_ORDER);
}

order-1 is already supported by commit [2], so I've removed that condition as
well.

[2] 8897277acfef7f70fdecc054073bea2542fc7a1b ("mm: support order-1 folios in the
page cache").

> +
>  #ifdef CONFIG_TRANSPARENT_HUGEPAGE
>  unsigned long shmem_allowable_huge_orders(struct inode *inode,
>  				struct vm_area_struct *vma, pgoff_t index,
> -				bool shmem_huge_force)
> +				bool shmem_huge_force, size_t len)
>  {
>  	unsigned long mask = READ_ONCE(huge_shmem_orders_always);
>  	unsigned long within_size_orders = READ_ONCE(huge_shmem_orders_within_size);
> @@ -1659,10 +1696,20 @@ unsigned long shmem_allowable_huge_orders(struct inode *inode,
>  						vma, 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.
> +		 * For tmpfs, if top level huge page is enabled, we just allow
> +		 * PMD size THP to keep interface backward compatibility.
> +		 */
> +		if (global_huge)
> +			return BIT(HPAGE_PMD_ORDER);
> +
> +		/*
> +		 * Otherwise, get a highest order hint based on the size of
> +		 * write and fallocate paths, then will try each allowable
> +		 * huge orders.
>  		 */
> -		return global_huge ? BIT(HPAGE_PMD_ORDER) : 0;
> +		order = shmem_mapping_size_order(inode->i_mapping, index,
> +						 len, SHMEM_SB(inode->i_sb));
> +		return BIT(order + 1) - 1;
>  	}
>  
>  	/*
> @@ -2174,7 +2221,7 @@ static int shmem_get_folio_gfp(struct inode *inode, pgoff_t index,
>  	}
>  
>  	/* Find hugepage orders that are allowed for anonymous shmem and tmpfs. */
> -	orders = shmem_allowable_huge_orders(inode, vma, index, false);
> +	orders = shmem_allowable_huge_orders(inode, vma, index, false, len);
>  	if (orders > 0) {
>  		gfp_t huge_gfp;
>  
> -- 
> 2.39.3
>