From: Ira Weiny <[email protected]>
Use a simple coccinelle script to help convert the most common
kmap()/kunmap() patterns to kmap_local_page()/kunmap_local().
Note that some kmaps which were caught by this script needed to be
handled by hand because of the strict unmapping order of kunmap_local()
so they are not included in this patch. But this script got us started.
The development of this patch was aided by the follow script:
// <smpl>
// SPDX-License-Identifier: GPL-2.0-only
// Find kmap and replace with kmap_local_page then mark kunmap
//
// Confidence: Low
// Copyright: (C) 2021 Intel Corporation
// URL: http://coccinelle.lip6.fr/
@ catch_all @
expression e, e2;
@@
(
-kmap(e)
+kmap_local_page(e)
)
...
(
-kunmap(...)
+kunmap_local()
)
// </smpl>
Signed-off-by: Ira Weiny <[email protected]>
---
fs/btrfs/compression.c | 4 ++--
fs/btrfs/inode.c | 4 ++--
fs/btrfs/lzo.c | 9 ++++-----
fs/btrfs/raid56.c | 4 ++--
4 files changed, 10 insertions(+), 11 deletions(-)
diff --git a/fs/btrfs/compression.c b/fs/btrfs/compression.c
index a219dcdb749e..21833a2fe8c6 100644
--- a/fs/btrfs/compression.c
+++ b/fs/btrfs/compression.c
@@ -1578,7 +1578,7 @@ static void heuristic_collect_sample(struct inode *inode, u64 start, u64 end,
curr_sample_pos = 0;
while (index < index_end) {
page = find_get_page(inode->i_mapping, index);
- in_data = kmap(page);
+ in_data = kmap_local_page(page);
/* Handle case where the start is not aligned to PAGE_SIZE */
i = start % PAGE_SIZE;
while (i < PAGE_SIZE - SAMPLING_READ_SIZE) {
@@ -1591,7 +1591,7 @@ static void heuristic_collect_sample(struct inode *inode, u64 start, u64 end,
start += SAMPLING_INTERVAL;
curr_sample_pos += SAMPLING_READ_SIZE;
}
- kunmap(page);
+ kunmap_local(in_data);
put_page(page);
index++;
diff --git a/fs/btrfs/inode.c b/fs/btrfs/inode.c
index 641f21a11722..66c6f1185de2 100644
--- a/fs/btrfs/inode.c
+++ b/fs/btrfs/inode.c
@@ -6845,7 +6845,7 @@ struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
if (ret)
goto out;
} else {
- map = kmap(page);
+ map = kmap_local_page(page);
read_extent_buffer(leaf, map + pg_offset, ptr,
copy_size);
if (pg_offset + copy_size < PAGE_SIZE) {
@@ -6853,7 +6853,7 @@ struct extent_map *btrfs_get_extent(struct btrfs_inode *inode,
PAGE_SIZE - pg_offset -
copy_size);
}
- kunmap(page);
+ kunmap_local(map);
}
flush_dcache_page(page);
}
diff --git a/fs/btrfs/lzo.c b/fs/btrfs/lzo.c
index 9084a950dc09..cd042c7567a4 100644
--- a/fs/btrfs/lzo.c
+++ b/fs/btrfs/lzo.c
@@ -118,7 +118,7 @@ int lzo_compress_pages(struct list_head *ws, struct address_space *mapping,
struct workspace *workspace = list_entry(ws, struct workspace, list);
int ret = 0;
char *data_in;
- char *cpage_out;
+ char *cpage_out, *sizes_ptr;
int nr_pages = 0;
struct page *in_page = NULL;
struct page *out_page = NULL;
@@ -258,10 +258,9 @@ int lzo_compress_pages(struct list_head *ws, struct address_space *mapping,
}
/* store the size of all chunks of compressed data */
- cpage_out = kmap(pages[0]);
- write_compress_length(cpage_out, tot_out);
-
- kunmap(pages[0]);
+ sizes_ptr = kmap_local_page(pages[0]);
+ write_compress_length(sizes_ptr, tot_out);
+ kunmap_local(sizes_ptr);
ret = 0;
*total_out = tot_out;
diff --git a/fs/btrfs/raid56.c b/fs/btrfs/raid56.c
index 7c3f6dc918c1..9759fb31b73e 100644
--- a/fs/btrfs/raid56.c
+++ b/fs/btrfs/raid56.c
@@ -2391,13 +2391,13 @@ static noinline void finish_parity_scrub(struct btrfs_raid_bio *rbio,
/* Check scrubbing parity and repair it */
p = rbio_stripe_page(rbio, rbio->scrubp, pagenr);
- parity = kmap(p);
+ parity = kmap_local_page(p);
if (memcmp(parity, pointers[rbio->scrubp], PAGE_SIZE))
copy_page(parity, pointers[rbio->scrubp]);
else
/* Parity is right, needn't writeback */
bitmap_clear(rbio->dbitmap, pagenr, 1);
- kunmap(p);
+ kunmap_local(parity);
for (stripe = 0; stripe < nr_data; stripe++)
kunmap(page_in_rbio(rbio, stripe, pagenr, 0));
--
2.28.0.rc0.12.gb6a658bd00c9
On Tue, Feb 16, 2021 at 06:48:23PM -0800, [email protected] wrote:
> --- a/fs/btrfs/lzo.c
> +++ b/fs/btrfs/lzo.c
> @@ -118,7 +118,7 @@ int lzo_compress_pages(struct list_head *ws, struct address_space *mapping,
> struct workspace *workspace = list_entry(ws, struct workspace, list);
> int ret = 0;
> char *data_in;
> - char *cpage_out;
> + char *cpage_out, *sizes_ptr;
> int nr_pages = 0;
> struct page *in_page = NULL;
> struct page *out_page = NULL;
> @@ -258,10 +258,9 @@ int lzo_compress_pages(struct list_head *ws, struct address_space *mapping,
> }
>
> /* store the size of all chunks of compressed data */
> - cpage_out = kmap(pages[0]);
> - write_compress_length(cpage_out, tot_out);
> -
> - kunmap(pages[0]);
> + sizes_ptr = kmap_local_page(pages[0]);
> + write_compress_length(sizes_ptr, tot_out);
> + kunmap_local(sizes_ptr);
Why is not cpage_out reused for this mapping? I don't see any reason for
another temporary variable, cpage_out is not used at this point.
On Fri, Mar 12, 2021 at 07:58:39PM +0100, David Sterba wrote:
> On Tue, Feb 16, 2021 at 06:48:23PM -0800, [email protected] wrote:
> > --- a/fs/btrfs/lzo.c
> > +++ b/fs/btrfs/lzo.c
> > @@ -118,7 +118,7 @@ int lzo_compress_pages(struct list_head *ws, struct address_space *mapping,
> > struct workspace *workspace = list_entry(ws, struct workspace, list);
> > int ret = 0;
> > char *data_in;
> > - char *cpage_out;
> > + char *cpage_out, *sizes_ptr;
> > int nr_pages = 0;
> > struct page *in_page = NULL;
> > struct page *out_page = NULL;
> > @@ -258,10 +258,9 @@ int lzo_compress_pages(struct list_head *ws, struct address_space *mapping,
> > }
> >
> > /* store the size of all chunks of compressed data */
> > - cpage_out = kmap(pages[0]);
> > - write_compress_length(cpage_out, tot_out);
> > -
> > - kunmap(pages[0]);
> > + sizes_ptr = kmap_local_page(pages[0]);
> > + write_compress_length(sizes_ptr, tot_out);
> > + kunmap_local(sizes_ptr);
>
> Why is not cpage_out reused for this mapping? I don't see any reason for
> another temporary variable, cpage_out is not used at this point.
For this patch that is true. However, I'm trying to convert the other kmaps as
well. To do that I'll need cpage_out preserved for the final kunmap_local().
Unfortunately, the required nesting ordering of kmap_local_page() makes
converting the other calls hacky at best. I'm not sure what to do about them.
The best I've come up with is doing a hacky extra unmap/remap to preserve the
nesting.
Anyway, I'd prefer to leave this additional temp variable but I can certainly
change if it you want. The other conversions may never work/land. :-/
Ira
On Fri, Mar 12, 2021 at 12:03:14PM -0800, Ira Weiny wrote:
> On Fri, Mar 12, 2021 at 07:58:39PM +0100, David Sterba wrote:
> > On Tue, Feb 16, 2021 at 06:48:23PM -0800, [email protected] wrote:
> > > --- a/fs/btrfs/lzo.c
> > > +++ b/fs/btrfs/lzo.c
> > > @@ -118,7 +118,7 @@ int lzo_compress_pages(struct list_head *ws, struct address_space *mapping,
> > > struct workspace *workspace = list_entry(ws, struct workspace, list);
> > > int ret = 0;
> > > char *data_in;
> > > - char *cpage_out;
> > > + char *cpage_out, *sizes_ptr;
> > > int nr_pages = 0;
> > > struct page *in_page = NULL;
> > > struct page *out_page = NULL;
> > > @@ -258,10 +258,9 @@ int lzo_compress_pages(struct list_head *ws, struct address_space *mapping,
> > > }
> > >
> > > /* store the size of all chunks of compressed data */
> > > - cpage_out = kmap(pages[0]);
> > > - write_compress_length(cpage_out, tot_out);
> > > -
> > > - kunmap(pages[0]);
> > > + sizes_ptr = kmap_local_page(pages[0]);
> > > + write_compress_length(sizes_ptr, tot_out);
> > > + kunmap_local(sizes_ptr);
> >
> > Why is not cpage_out reused for this mapping? I don't see any reason for
> > another temporary variable, cpage_out is not used at this point.
>
> For this patch that is true. However, I'm trying to convert the other kmaps as
> well. To do that I'll need cpage_out preserved for the final kunmap_local().
>
> Unfortunately, the required nesting ordering of kmap_local_page() makes
> converting the other calls hacky at best. I'm not sure what to do about them.
> The best I've come up with is doing a hacky extra unmap/remap to preserve the
> nesting.
>
> Anyway, I'd prefer to leave this additional temp variable but I can certainly
> change if it you want. The other conversions may never work/land. :-/
Ok, no problem keeping the variable then. I've added a note to changelog
why it's there. The whole conversion sounds tricky so adding trivial
helper code is no big deal.