[RESEND PATCH 0/3] hfs: Replace kmap() with kmap_local_page()

[RESEND PATCH 0/3] hfs: Replace kmap() with kmap_local_page()

[Fabio M. De Francesco]

kmap() is being deprecated in favor of kmap_local_page().

There are two main problems with kmap(): (1) It comes with an overhead as
mapping space is restricted and protected by a global lock for
synchronization and (2) it also requires global TLB invalidation when the
kmap’s pool wraps and it might block when the mapping space is fully
utilized until a slot becomes available.

With kmap_local_page() the mappings are per thread, CPU local, can take
page faults, and can be called from any context (including interrupts).
It is faster than kmap() in kernels with HIGHMEM enabled. Furthermore,
the tasks can be preempted and, when they are scheduled to run again, the
kernel virtual addresses are restored and still valid.

Since its use in fs/hfs is safe everywhere, it should be preferred.

Therefore, replace kmap() with kmap_local_page() in fs/hfs. Where
possible, use the suited standard helpers (memzero_page(), memcpy_page())
instead of open coding kmap_local_page() plus memset() or memcpy().

Fix a bug due to a page being not unmapped if the code jumps to the
"fail_page" label (1/3).

Tested in a QEMU/KVM x86_32 VM, 6GB RAM, booting a kernel with
HIGHMEM64GB enabled.

Few days ago Andrew requested a resend of this series. In the meantime
I'm also forwarding a "Reviewed-by" tag from Viacheslav Dubeyko.

Fabio M. De Francesco (3):
  hfs: Unmap the page in the "fail_page" label
  hfs: Replace kmap() with kmap_local_page() in bnode.c
  hfs: Replace kmap() with kmap_local_page() in btree.c

 fs/hfs/bnode.c | 32 ++++++++++++--------------------
 fs/hfs/btree.c | 29 ++++++++++++++++-------------
 2 files changed, 28 insertions(+), 33 deletions(-)

-- 
2.37.1

[RESEND PATCH 1/3] hfs: Unmap the page in the "fail_page" label

[Fabio M. De Francesco]

Several paths within hfs_btree_open() jump to the "fail_page" label
where put_page() is called while the page is still mapped.

Call kunmap() to unmap the page soon before put_page().

Reviewed-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Viacheslav Dubeyko <slava@dubeyko.com>
Signed-off-by: Fabio M. De Francesco <fmdefrancesco@gmail.com>
---
 fs/hfs/btree.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/fs/hfs/btree.c b/fs/hfs/btree.c
index 19017d296173..56c6782436e9 100644
--- a/fs/hfs/btree.c
+++ b/fs/hfs/btree.c
@@ -124,6 +124,7 @@ struct hfs_btree *hfs_btree_open(struct super_block *sb, u32 id, btree_keycmp ke
 	return tree;
 
 fail_page:
+	kunmap(page);
 	put_page(page);
 free_inode:
 	tree->inode->i_mapping->a_ops = &hfs_aops;
-- 
2.37.1

[RESEND PATCH 2/3] hfs: Replace kmap() with kmap_local_page() in bnode.c

[Fabio M. De Francesco]

kmap() is being deprecated in favor of kmap_local_page().

Two main problems with kmap(): (1) It comes with an overhead as mapping
space is restricted and protected by a global lock for synchronization and
(2) it also requires global TLB invalidation when the kmap’s pool wraps
and it might block when the mapping space is fully utilized until a slot
becomes available.

With kmap_local_page() the mappings are per thread, CPU local, can take
page faults, and can be called from any context (including interrupts).
It is faster than kmap() in kernels with HIGHMEM enabled. Furthermore,
the tasks can be preempted and, when they are scheduled to run again, the
kernel virtual addresses are restored and still valid.

Since its use in bnode.c is safe everywhere, it should be preferred.

Therefore, replace kmap() with kmap_local_page() in bnode.c. Where
possible, use the suited standard helpers (memzero_page(), memcpy_page())
instead of open coding kmap_local_page() plus memset() or memcpy().

Tested in a QEMU/KVM x86_32 VM, 6GB RAM, booting a kernel with
HIGHMEM64GB enabled.

Suggested-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Viacheslav Dubeyko <slava@dubeyko.com>
Signed-off-by: Fabio M. De Francesco <fmdefrancesco@gmail.com>
---
 fs/hfs/bnode.c | 32 ++++++++++++--------------------
 1 file changed, 12 insertions(+), 20 deletions(-)

diff --git a/fs/hfs/bnode.c b/fs/hfs/bnode.c
index c83fd0e8404d..2015e42e752a 100644
--- a/fs/hfs/bnode.c
+++ b/fs/hfs/bnode.c
@@ -21,7 +21,6 @@ void hfs_bnode_read(struct hfs_bnode *node, void *buf, int off, int len)
 	int pagenum;
 	int bytes_read;
 	int bytes_to_read;
-	void *vaddr;
 
 	off += node->page_offset;
 	pagenum = off >> PAGE_SHIFT;
@@ -33,9 +32,7 @@ void hfs_bnode_read(struct hfs_bnode *node, void *buf, int off, int len)
 		page = node->page[pagenum];
 		bytes_to_read = min_t(int, len - bytes_read, PAGE_SIZE - off);
 
-		vaddr = kmap_atomic(page);
-		memcpy(buf + bytes_read, vaddr + off, bytes_to_read);
-		kunmap_atomic(vaddr);
+		memcpy_from_page(buf + bytes_read, page, off, bytes_to_read);
 
 		pagenum++;
 		off = 0; /* page offset only applies to the first page */
@@ -80,8 +77,7 @@ void hfs_bnode_write(struct hfs_bnode *node, void *buf, int off, int len)
 	off += node->page_offset;
 	page = node->page[0];
 
-	memcpy(kmap(page) + off, buf, len);
-	kunmap(page);
+	memcpy_to_page(page, off, buf, len);
 	set_page_dirty(page);
 }
 
@@ -105,8 +101,7 @@ void hfs_bnode_clear(struct hfs_bnode *node, int off, int len)
 	off += node->page_offset;
 	page = node->page[0];
 
-	memset(kmap(page) + off, 0, len);
-	kunmap(page);
+	memzero_page(page, off, len);
 	set_page_dirty(page);
 }
 
@@ -123,9 +118,7 @@ void hfs_bnode_copy(struct hfs_bnode *dst_node, int dst,
 	src_page = src_node->page[0];
 	dst_page = dst_node->page[0];
 
-	memcpy(kmap(dst_page) + dst, kmap(src_page) + src, len);
-	kunmap(src_page);
-	kunmap(dst_page);
+	memcpy_page(dst_page, dst, src_page, src, len);
 	set_page_dirty(dst_page);
 }
 
@@ -140,9 +133,9 @@ void hfs_bnode_move(struct hfs_bnode *node, int dst, int src, int len)
 	src += node->page_offset;
 	dst += node->page_offset;
 	page = node->page[0];
-	ptr = kmap(page);
+	ptr = kmap_local_page(page);
 	memmove(ptr + dst, ptr + src, len);
-	kunmap(page);
+	kunmap_local(ptr);
 	set_page_dirty(page);
 }
 
@@ -346,13 +339,14 @@ struct hfs_bnode *hfs_bnode_find(struct hfs_btree *tree, u32 num)
 	if (!test_bit(HFS_BNODE_NEW, &node->flags))
 		return node;
 
-	desc = (struct hfs_bnode_desc *)(kmap(node->page[0]) + node->page_offset);
+	desc = (struct hfs_bnode_desc *)(kmap_local_page(node->page[0]) +
+					 node->page_offset);
 	node->prev = be32_to_cpu(desc->prev);
 	node->next = be32_to_cpu(desc->next);
 	node->num_recs = be16_to_cpu(desc->num_recs);
 	node->type = desc->type;
 	node->height = desc->height;
-	kunmap(node->page[0]);
+	kunmap_local(desc);
 
 	switch (node->type) {
 	case HFS_NODE_HEADER:
@@ -436,14 +430,12 @@ struct hfs_bnode *hfs_bnode_create(struct hfs_btree *tree, u32 num)
 	}
 
 	pagep = node->page;
-	memset(kmap(*pagep) + node->page_offset, 0,
-	       min((int)PAGE_SIZE, (int)tree->node_size));
+	memzero_page(*pagep, node->page_offset,
+		     min((int)PAGE_SIZE, (int)tree->node_size));
 	set_page_dirty(*pagep);
-	kunmap(*pagep);
 	for (i = 1; i < tree->pages_per_bnode; i++) {
-		memset(kmap(*++pagep), 0, PAGE_SIZE);
+		memzero_page(*++pagep, 0, PAGE_SIZE);
 		set_page_dirty(*pagep);
-		kunmap(*pagep);
 	}
 	clear_bit(HFS_BNODE_NEW, &node->flags);
 	wake_up(&node->lock_wq);
-- 
2.37.1

[RESEND PATCH 3/3] hfs: Replace kmap() with kmap_local_page() in btree.c

[Fabio M. De Francesco]

kmap() is being deprecated in favor of kmap_local_page().

Two main problems with kmap(): (1) It comes with an overhead as mapping
space is restricted and protected by a global lock for synchronization and
(2) it also requires global TLB invalidation when the kmap’s pool wraps
and it might block when the mapping space is fully utilized until a slot
becomes available.

With kmap_local_page() the mappings are per thread, CPU local, can take
page faults, and can be called from any context (including interrupts).
It is faster than kmap() in kernels with HIGHMEM enabled. Furthermore,
the tasks can be preempted and, when they are scheduled to run again, the
kernel virtual addresses are restored and still valid.

Since its use in btree.c is safe everywhere, it should be preferred.

Therefore, replace kmap() with kmap_local_page() in btree.c. Where
possible, use the suited standard helpers (memzero_page(), memcpy_page())
instead of open coding kmap_local_page() plus memset() or memcpy().

Tested in a QEMU/KVM x86_32 VM, 6GB RAM, booting a kernel with
HIGHMEM64GB enabled.

Suggested-by: Ira Weiny <ira.weiny@intel.com>
Reviewed-by: Viacheslav Dubeyko <slava@dubeyko.com>
Signed-off-by: Fabio M. De Francesco <fmdefrancesco@gmail.com>
---
 fs/hfs/btree.c | 30 ++++++++++++++++--------------
 1 file changed, 16 insertions(+), 14 deletions(-)

diff --git a/fs/hfs/btree.c b/fs/hfs/btree.c
index 56c6782436e9..2fa4b1f8cc7f 100644
--- a/fs/hfs/btree.c
+++ b/fs/hfs/btree.c
@@ -80,7 +80,8 @@ struct hfs_btree *hfs_btree_open(struct super_block *sb, u32 id, btree_keycmp ke
 		goto free_inode;
 
 	/* Load the header */
-	head = (struct hfs_btree_header_rec *)(kmap(page) + sizeof(struct hfs_bnode_desc));
+	head = (struct hfs_btree_header_rec *)(kmap_local_page(page) +
+					       sizeof(struct hfs_bnode_desc));
 	tree->root = be32_to_cpu(head->root);
 	tree->leaf_count = be32_to_cpu(head->leaf_count);
 	tree->leaf_head = be32_to_cpu(head->leaf_head);
@@ -119,12 +120,12 @@ struct hfs_btree *hfs_btree_open(struct super_block *sb, u32 id, btree_keycmp ke
 	tree->node_size_shift = ffs(size) - 1;
 	tree->pages_per_bnode = (tree->node_size + PAGE_SIZE - 1) >> PAGE_SHIFT;
 
-	kunmap(page);
+	kunmap_local(head);
 	put_page(page);
 	return tree;
 
 fail_page:
-	kunmap(page);
+	kunmap_local(head);
 	put_page(page);
 free_inode:
 	tree->inode->i_mapping->a_ops = &hfs_aops;
@@ -170,7 +171,8 @@ void hfs_btree_write(struct hfs_btree *tree)
 		return;
 	/* Load the header */
 	page = node->page[0];
-	head = (struct hfs_btree_header_rec *)(kmap(page) + sizeof(struct hfs_bnode_desc));
+	head = (struct hfs_btree_header_rec *)(kmap_local_page(page) +
+					       sizeof(struct hfs_bnode_desc));
 
 	head->root = cpu_to_be32(tree->root);
 	head->leaf_count = cpu_to_be32(tree->leaf_count);
@@ -181,7 +183,7 @@ void hfs_btree_write(struct hfs_btree *tree)
 	head->attributes = cpu_to_be32(tree->attributes);
 	head->depth = cpu_to_be16(tree->depth);
 
-	kunmap(page);
+	kunmap_local(head);
 	set_page_dirty(page);
 	hfs_bnode_put(node);
 }
@@ -269,7 +271,7 @@ struct hfs_bnode *hfs_bmap_alloc(struct hfs_btree *tree)
 
 	off += node->page_offset;
 	pagep = node->page + (off >> PAGE_SHIFT);
-	data = kmap(*pagep);
+	data = kmap_local_page(*pagep);
 	off &= ~PAGE_MASK;
 	idx = 0;
 
@@ -282,7 +284,7 @@ struct hfs_bnode *hfs_bmap_alloc(struct hfs_btree *tree)
 						idx += i;
 						data[off] |= m;
 						set_page_dirty(*pagep);
-						kunmap(*pagep);
+						kunmap_local(data);
 						tree->free_nodes--;
 						mark_inode_dirty(tree->inode);
 						hfs_bnode_put(node);
@@ -291,14 +293,14 @@ struct hfs_bnode *hfs_bmap_alloc(struct hfs_btree *tree)
 				}
 			}
 			if (++off >= PAGE_SIZE) {
-				kunmap(*pagep);
-				data = kmap(*++pagep);
+				kunmap_local(data);
+				data = kmap_local_page(*++pagep);
 				off = 0;
 			}
 			idx += 8;
 			len--;
 		}
-		kunmap(*pagep);
+		kunmap_local(data);
 		nidx = node->next;
 		if (!nidx) {
 			printk(KERN_DEBUG "create new bmap node...\n");
@@ -314,7 +316,7 @@ struct hfs_bnode *hfs_bmap_alloc(struct hfs_btree *tree)
 		off = off16;
 		off += node->page_offset;
 		pagep = node->page + (off >> PAGE_SHIFT);
-		data = kmap(*pagep);
+		data = kmap_local_page(*pagep);
 		off &= ~PAGE_MASK;
 	}
 }
@@ -361,20 +363,20 @@ void hfs_bmap_free(struct hfs_bnode *node)
 	}
 	off += node->page_offset + nidx / 8;
 	page = node->page[off >> PAGE_SHIFT];
-	data = kmap(page);
+	data = kmap_local_page(page);
 	off &= ~PAGE_MASK;
 	m = 1 << (~nidx & 7);
 	byte = data[off];
 	if (!(byte & m)) {
 		pr_crit("trying to free free bnode %u(%d)\n",
 			node->this, node->type);
-		kunmap(page);
+		kunmap_local(data);
 		hfs_bnode_put(node);
 		return;
 	}
 	data[off] = byte & ~m;
 	set_page_dirty(page);
-	kunmap(page);
+	kunmap_local(data);
 	hfs_bnode_put(node);
 	tree->free_nodes++;
 	mark_inode_dirty(tree->inode);
-- 
2.37.1