Skip to contentSkip to conversation

fs/ext2: Replace kmap_atomic() with kmap_local_page()

PATCH In review Fabio M. De Francesco 3 participants 3 messages Dec 31, 2022 – Jan 3, 2023

Subsystems (2)
lore
Share & tools
lore
#0 lore raw

PATCH 1/1 fs/ext2: Replace kmap_atomic() with kmap_local_page()

kmap_atomic() is deprecated in favor of kmap_local_page(). Therefore, replace kmap_atomic() with kmap_local_page().

kmap_atomic() is implemented like a kmap_local_page() which also disables page-faults and preemption (the latter only for !PREEMPT_RT kernels).

However, the code within the mapping and un-mapping in ext2_make_empty() does not depend on the above-mentioned side effects.

Therefore, a mere replacement of the old API with the new one is all it is required (i.e., there is no need to explicitly add any calls to pagefault_disable() and/or preempt_disable()).

---

I tried my best to understand the code within mapping and un-mapping. However, I'm not an expert. Therefore, although I'm pretty confident, I cannot be 100% sure that the code between the mapping and the un-mapping does not depend on pagefault_disable() and/or preempt_disable().

Unfortunately, I cannot currently test this changes to check the above-mentioned assumptions. However, if I'm required to do the tests with (x)fstests, I have no problems with doing them in the next days.

If so, I'll test in a QEMU/KVM x86_32 VM, 6GB RAM, booting a kernel with HIGHMEM64GB enabled.

I'd like to hear whether or not the maintainers require these tests and/or other tests.

 fs/ext2/dir.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
fs/ext2/dir.c +2 -2
index e5cbc27ba459..0f144c5c7861 100644
@@ -646,7 +646,7 @@ int ext2_make_empty(struct inode *inode, struct inode *parent)
 		unlock_page(page);
 		goto fail;
 	}
-	kaddr = kmap_atomic(page);
+	kaddr = kmap_local_page(page);
 	memset(kaddr, 0, chunk_size);
 	de = (struct ext2_dir_entry_2 *)kaddr;
 	de->name_len = 1;
@@ -661,7 +661,7 @@ int ext2_make_empty(struct inode *inode, struct inode *parent)
 	de->inode = cpu_to_le32(parent->i_ino);
 	memcpy (de->name, "..\0", 4);
 	ext2_set_de_type (de, inode);
-	kunmap_atomic(kaddr);
+	kunmap_local(kaddr);
 	ext2_commit_chunk(page, 0, chunk_size);
 	err = ext2_handle_dirsync(inode);
 fail:
Signature · 1 line
2.39.0
reviewer
#1 lore raw
Quoted 9 lines · Fabio M. De Francesco

kmap_atomic() is deprecated in favor of kmap_local_page(). Therefore, replace kmap_atomic() with kmap_local_page().

kmap_atomic() is implemented like a kmap_local_page() which also disables page-faults and preemption (the latter only for !PREEMPT_RT kernels).

However, the code within the mapping and un-mapping in ext2_make_empty() does not depend on the above-mentioned side effects.

Quoted 6 lines

Therefore, a mere replacement of the old API with the new one is all it is required (i.e., there is no need to explicitly add any calls to pagefault_disable() and/or preempt_disable()).

Thanks, the patch looks good and I'll queue it in my tree. I'm not sure why it got missed during the initial conversion by Ira :).

Honza

Quoted 45 lines

---

I tried my best to understand the code within mapping and un-mapping. However, I'm not an expert. Therefore, although I'm pretty confident, I cannot be 100% sure that the code between the mapping and the un-mapping does not depend on pagefault_disable() and/or preempt_disable().

Unfortunately, I cannot currently test this changes to check the above-mentioned assumptions. However, if I'm required to do the tests with (x)fstests, I have no problems with doing them in the next days.

If so, I'll test in a QEMU/KVM x86_32 VM, 6GB RAM, booting a kernel with HIGHMEM64GB enabled.

I'd like to hear whether or not the maintainers require these tests and/or other tests.

 fs/ext2/dir.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
fs/ext2/dir.c +2 -2
index e5cbc27ba459..0f144c5c7861 100644
@@ -646,7 +646,7 @@ int ext2_make_empty(struct inode *inode, struct inode *parent)
 		unlock_page(page);
 		goto fail;
 	}
-	kaddr = kmap_atomic(page);
+	kaddr = kmap_local_page(page);
 	memset(kaddr, 0, chunk_size);
 	de = (struct ext2_dir_entry_2 *)kaddr;
 	de->name_len = 1;
@@ -661,7 +661,7 @@ int ext2_make_empty(struct inode *inode, struct inode *parent)
 	de->inode = cpu_to_le32(parent->i_ino);
 	memcpy (de->name, "..\0", 4);
 	ext2_set_de_type (de, inode);
-	kunmap_atomic(kaddr);
+	kunmap_local(kaddr);
 	ext2_commit_chunk(page, 0, chunk_size);
 	err = ext2_handle_dirsync(inode);
 fail:
Signature · 1 line
2.39.0
Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR
Reviewed-by
#2 lore raw
Quoted 14 lines · Jan Kara
Quoted 13 lines · Fabio M. De Francesco

kmap_atomic() is deprecated in favor of kmap_local_page(). Therefore, replace kmap_atomic() with kmap_local_page().

kmap_atomic() is implemented like a kmap_local_page() which also disables page-faults and preemption (the latter only for !PREEMPT_RT kernels).

However, the code within the mapping and un-mapping in ext2_make_empty() does not depend on the above-mentioned side effects.

Therefore, a mere replacement of the old API with the new one is all it is required (i.e., there is no need to explicitly add any calls to pagefault_disable() and/or preempt_disable()).

Quoted 6 lines
Quoted 2 lines

Thanks, the patch looks good and I'll queue it in my tree. I'm not sure why it got missed during the initial conversion by Ira :).

I could claim something smart like I was focused on kmap() calls only. But let's just call it laziness. ;-)

LGTM

Quoted 51 lines

Honza

Quoted 45 lines

---

I tried my best to understand the code within mapping and un-mapping. However, I'm not an expert. Therefore, although I'm pretty confident, I cannot be 100% sure that the code between the mapping and the un-mapping does not depend on pagefault_disable() and/or preempt_disable().

Unfortunately, I cannot currently test this changes to check the above-mentioned assumptions. However, if I'm required to do the tests with (x)fstests, I have no problems with doing them in the next days.

If so, I'll test in a QEMU/KVM x86_32 VM, 6GB RAM, booting a kernel with HIGHMEM64GB enabled.

I'd like to hear whether or not the maintainers require these tests and/or other tests.

 fs/ext2/dir.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
fs/ext2/dir.c +2 -2
index e5cbc27ba459..0f144c5c7861 100644
@@ -646,7 +646,7 @@ int ext2_make_empty(struct inode *inode, struct inode *parent)
 		unlock_page(page);
 		goto fail;
 	}
-	kaddr = kmap_atomic(page);
+	kaddr = kmap_local_page(page);
 	memset(kaddr, 0, chunk_size);
 	de = (struct ext2_dir_entry_2 *)kaddr;
 	de->name_len = 1;
@@ -661,7 +661,7 @@ int ext2_make_empty(struct inode *inode, struct inode *parent)
 	de->inode = cpu_to_le32(parent->i_ino);
 	memcpy (de->name, "..\0", 4);
 	ext2_set_de_type (de, inode);
-	kunmap_atomic(kaddr);
+	kunmap_local(kaddr);
 	ext2_commit_chunk(page, 0, chunk_size);
 	err = ext2_handle_dirsync(inode);
 fail:
Signature · 1 line
2.39.0
Signature · 2 lines
Jan Kara <[email protected]>
SUSE Labs, CR