| Message ID | 1468499372-24687-1-git-send-email-dsterba@suse.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On 07/14/2016 08:29 AM, David Sterba wrote: > The calculation of extent_buffer::pages size was done for 4k PAGE_SIZE, > but this wastes 15 unused pointers on arches with large page size. Eg. > on ppc64 this gives 15 * 8 = 120 bytes. > > Signed-off-by: David Sterba <dsterba@suse.com> Reviewed-by: Chris Mason <clm@fb.com> -chris -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Thursday, July 14, 2016 02:29:32 PM David Sterba wrote: > The calculation of extent_buffer::pages size was done for 4k PAGE_SIZE, > but this wastes 15 unused pointers on arches with large page size. Eg. > on ppc64 this gives 15 * 8 = 120 bytes. > The non PAGE_SIZE aligned extent buffer usage in page straddling tests in test_eb_bitmaps() need atleast one more page. So how about the following ... #define INLINE_EXTENT_BUFFER_PAGES (BTRFS_MAX_METADATA_BLOCKSIZE / PAGE_SIZE + 1) > Signed-off-by: David Sterba <dsterba@suse.com> > --- > fs/btrfs/ctree.h | 6 ------ > fs/btrfs/extent_io.c | 2 ++ > fs/btrfs/extent_io.h | 8 +++++++- > 3 files changed, 9 insertions(+), 7 deletions(-) > > diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h > index 4274a7bfdaed..f914f6187753 100644 > --- a/fs/btrfs/ctree.h > +++ b/fs/btrfs/ctree.h > @@ -66,12 +66,6 @@ struct btrfs_ordered_sum; > #define BTRFS_COMPAT_EXTENT_TREE_V0 > > /* > - * the max metadata block size. This limit is somewhat artificial, > - * but the memmove costs go through the roof for larger blocks. > - */ > -#define BTRFS_MAX_METADATA_BLOCKSIZE 65536 > - > -/* > * we can actually store much bigger names, but lets not confuse the rest > * of linux > */ > diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c > index 75533adef998..6f468a1842e6 100644 > --- a/fs/btrfs/extent_io.c > +++ b/fs/btrfs/extent_io.c > @@ -4660,6 +4660,8 @@ __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start, > /* > * Sanity checks, currently the maximum is 64k covered by 16x 4k pages > */ > + BUILD_BUG_ON(INLINE_EXTENT_BUFFER_PAGES * PAGE_SIZE > + != BTRFS_MAX_METADATA_BLOCKSIZE); > BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE > > MAX_INLINE_EXTENT_BUFFER_SIZE); > BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE); > diff --git a/fs/btrfs/extent_io.h b/fs/btrfs/extent_io.h > index c0c1c4fef6ce..edfa1a0ab82b 100644 > --- a/fs/btrfs/extent_io.h > +++ b/fs/btrfs/extent_io.h > @@ -4,6 +4,12 @@ > #include <linux/rbtree.h> > #include "ulist.h" > > +/* > + * The maximum metadata block size. This limit is somewhat artificial, > + * but the memmove costs go through the roof for larger blocks. > + */ > +#define BTRFS_MAX_METADATA_BLOCKSIZE (65536U) > + > /* bits for the extent state */ > #define EXTENT_DIRTY (1U << 0) > #define EXTENT_WRITEBACK (1U << 1) > @@ -118,7 +124,7 @@ struct extent_state { > #endif > }; > > -#define INLINE_EXTENT_BUFFER_PAGES 16 > +#define INLINE_EXTENT_BUFFER_PAGES (BTRFS_MAX_METADATA_BLOCKSIZE / PAGE_SIZE) > #define MAX_INLINE_EXTENT_BUFFER_SIZE (INLINE_EXTENT_BUFFER_PAGES * PAGE_SIZE) > struct extent_buffer { > u64 start; >
On Fri, Jul 15, 2016 at 11:47:07AM +0530, Chandan Rajendra wrote: > On Thursday, July 14, 2016 02:29:32 PM David Sterba wrote: > > The calculation of extent_buffer::pages size was done for 4k PAGE_SIZE, > > but this wastes 15 unused pointers on arches with large page size. Eg. > > on ppc64 this gives 15 * 8 = 120 bytes. > > > > The non PAGE_SIZE aligned extent buffer usage in page straddling tests in > test_eb_bitmaps() need atleast one more page. So how about the following ... > > #define INLINE_EXTENT_BUFFER_PAGES (BTRFS_MAX_METADATA_BLOCKSIZE / PAGE_SIZE + 1) Could the extra page pointer be normally used? Ie. not just for the sake of the tests. I'd rather not waste the bytes. As a compromise, we can do +1 only if the tests are compiled in. -- To unsubscribe from this list: send the line "unsubscribe linux-btrfs" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html
On Friday, July 15, 2016 11:44:06 AM David Sterba wrote: > On Fri, Jul 15, 2016 at 11:47:07AM +0530, Chandan Rajendra wrote: > > On Thursday, July 14, 2016 02:29:32 PM David Sterba wrote: > > > The calculation of extent_buffer::pages size was done for 4k PAGE_SIZE, > > > but this wastes 15 unused pointers on arches with large page size. Eg. > > > on ppc64 this gives 15 * 8 = 120 bytes. > > > > > > > The non PAGE_SIZE aligned extent buffer usage in page straddling tests in > > test_eb_bitmaps() need atleast one more page. So how about the following ... > > > > #define INLINE_EXTENT_BUFFER_PAGES (BTRFS_MAX_METADATA_BLOCKSIZE / PAGE_SIZE + 1) > > Could the extra page pointer be normally used? Ie. not just for the sake > of the tests. I'd rather not waste the bytes. As a compromise, we can do +1 > only if the tests are compiled in. > I don't see any other scenario where the extra page pointer gets used. Also, I just executed fstests with your patch applied and disabling self-tests from the kernel configuration. The tests ran fine.
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h index 4274a7bfdaed..f914f6187753 100644 --- a/fs/btrfs/ctree.h +++ b/fs/btrfs/ctree.h @@ -66,12 +66,6 @@ struct btrfs_ordered_sum; #define BTRFS_COMPAT_EXTENT_TREE_V0 /* - * the max metadata block size. This limit is somewhat artificial, - * but the memmove costs go through the roof for larger blocks. - */ -#define BTRFS_MAX_METADATA_BLOCKSIZE 65536 - -/* * we can actually store much bigger names, but lets not confuse the rest * of linux */ diff --git a/fs/btrfs/extent_io.c b/fs/btrfs/extent_io.c index 75533adef998..6f468a1842e6 100644 --- a/fs/btrfs/extent_io.c +++ b/fs/btrfs/extent_io.c @@ -4660,6 +4660,8 @@ __alloc_extent_buffer(struct btrfs_fs_info *fs_info, u64 start, /* * Sanity checks, currently the maximum is 64k covered by 16x 4k pages */ + BUILD_BUG_ON(INLINE_EXTENT_BUFFER_PAGES * PAGE_SIZE + != BTRFS_MAX_METADATA_BLOCKSIZE); BUILD_BUG_ON(BTRFS_MAX_METADATA_BLOCKSIZE > MAX_INLINE_EXTENT_BUFFER_SIZE); BUG_ON(len > MAX_INLINE_EXTENT_BUFFER_SIZE); diff --git a/fs/btrfs/extent_io.h b/fs/btrfs/extent_io.h index c0c1c4fef6ce..edfa1a0ab82b 100644 --- a/fs/btrfs/extent_io.h +++ b/fs/btrfs/extent_io.h @@ -4,6 +4,12 @@ #include <linux/rbtree.h> #include "ulist.h" +/* + * The maximum metadata block size. This limit is somewhat artificial, + * but the memmove costs go through the roof for larger blocks. + */ +#define BTRFS_MAX_METADATA_BLOCKSIZE (65536U) + /* bits for the extent state */ #define EXTENT_DIRTY (1U << 0) #define EXTENT_WRITEBACK (1U << 1) @@ -118,7 +124,7 @@ struct extent_state { #endif }; -#define INLINE_EXTENT_BUFFER_PAGES 16 +#define INLINE_EXTENT_BUFFER_PAGES (BTRFS_MAX_METADATA_BLOCKSIZE / PAGE_SIZE) #define MAX_INLINE_EXTENT_BUFFER_SIZE (INLINE_EXTENT_BUFFER_PAGES * PAGE_SIZE) struct extent_buffer { u64 start;
The calculation of extent_buffer::pages size was done for 4k PAGE_SIZE, but this wastes 15 unused pointers on arches with large page size. Eg. on ppc64 this gives 15 * 8 = 120 bytes. Signed-off-by: David Sterba <dsterba@suse.com> --- fs/btrfs/ctree.h | 6 ------ fs/btrfs/extent_io.c | 2 ++ fs/btrfs/extent_io.h | 8 +++++++- 3 files changed, 9 insertions(+), 7 deletions(-)