| Message ID | 7a6c4814d8467999d9a9eb12906ff28a3f90b284.1536599879.git.dsterba@suse.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
| Series | Self-test cleanups | expand |
On Mon, Sep 10, 2018 at 07:22:31PM +0200, David Sterba wrote: > Avoid the inline ifdefs and use two sections for self-tests enabled and > disabled. > > Signed-off-by: David Sterba <dsterba@suse.com> > --- > fs/btrfs/ctree.h | 9 ++++++--- > 1 file changed, 6 insertions(+), 3 deletions(-) > > diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h > index 32d2fce4ac53..8dafc7bb6ad8 100644 > --- a/fs/btrfs/ctree.h > +++ b/fs/btrfs/ctree.h > @@ -3708,17 +3708,20 @@ static inline int btrfs_defrag_cancelled(struct btrfs_fs_info *fs_info) > #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS > void btrfs_test_inode_set_ops(struct inode *inode); > void btrfs_test_destroy_inode(struct inode *inode); > -#endif > > static inline int btrfs_is_testing(struct btrfs_fs_info *fs_info) > { > -#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS > if (unlikely(test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, > &fs_info->fs_state))) > return 1; > -#endif > return 0; How about just: return test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state); We can probably get away without the unlikely() considering that no one sane is going to run a kernel with CONFIG_BTRFS_FS_RUN_SANITY_TESTS in production.
On Mon, Sep 10, 2018 at 04:43:29PM -0700, Omar Sandoval wrote: > On Mon, Sep 10, 2018 at 07:22:31PM +0200, David Sterba wrote: > > Avoid the inline ifdefs and use two sections for self-tests enabled and > > disabled. > > > > Signed-off-by: David Sterba <dsterba@suse.com> > > --- > > fs/btrfs/ctree.h | 9 ++++++--- > > 1 file changed, 6 insertions(+), 3 deletions(-) > > > > diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h > > index 32d2fce4ac53..8dafc7bb6ad8 100644 > > --- a/fs/btrfs/ctree.h > > +++ b/fs/btrfs/ctree.h > > @@ -3708,17 +3708,20 @@ static inline int btrfs_defrag_cancelled(struct btrfs_fs_info *fs_info) > > #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS > > void btrfs_test_inode_set_ops(struct inode *inode); > > void btrfs_test_destroy_inode(struct inode *inode); > > -#endif > > > > static inline int btrfs_is_testing(struct btrfs_fs_info *fs_info) > > { > > -#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS > > if (unlikely(test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, > > &fs_info->fs_state))) > > return 1; > > -#endif > > return 0; > > How about just: > > return test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state); > > We can probably get away without the unlikely() considering that no one > sane is going to run a kernel with CONFIG_BTRFS_FS_RUN_SANITY_TESTS in > production. The unlikely can go away, sure. I would still like to remove the test_bit call when tests are compiled out. There are about 10 calls to btrfs_is_testing in various core functions, followed by further statements. This would have a (negligible) runtime penalty but generates effectively unused code on production builds. The static inline function returning 0 allows to optimize out the unused code, so smaller code, fewer inctructions, etc.
On Tue, Sep 11, 2018 at 11:22:51AM +0200, David Sterba wrote: > On Mon, Sep 10, 2018 at 04:43:29PM -0700, Omar Sandoval wrote: > > On Mon, Sep 10, 2018 at 07:22:31PM +0200, David Sterba wrote: > > > Avoid the inline ifdefs and use two sections for self-tests enabled and > > > disabled. > > > > > > Signed-off-by: David Sterba <dsterba@suse.com> > > > --- > > > fs/btrfs/ctree.h | 9 ++++++--- > > > 1 file changed, 6 insertions(+), 3 deletions(-) > > > > > > diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h > > > index 32d2fce4ac53..8dafc7bb6ad8 100644 > > > --- a/fs/btrfs/ctree.h > > > +++ b/fs/btrfs/ctree.h > > > @@ -3708,17 +3708,20 @@ static inline int btrfs_defrag_cancelled(struct btrfs_fs_info *fs_info) > > > #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS > > > void btrfs_test_inode_set_ops(struct inode *inode); > > > void btrfs_test_destroy_inode(struct inode *inode); > > > -#endif > > > > > > static inline int btrfs_is_testing(struct btrfs_fs_info *fs_info) > > > { > > > -#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS > > > if (unlikely(test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, > > > &fs_info->fs_state))) > > > return 1; > > > -#endif > > > return 0; > > > > How about just: > > > > return test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state); > > > > We can probably get away without the unlikely() considering that no one > > sane is going to run a kernel with CONFIG_BTRFS_FS_RUN_SANITY_TESTS in > > production. > > The unlikely can go away, sure. > > I would still like to remove the test_bit call when tests are compiled > out. There are about 10 calls to btrfs_is_testing in various core > functions, followed by further statements. This would have a > (negligible) runtime penalty but generates effectively unused code on > production builds. > > The static inline function returning 0 allows to optimize out the unused > code, so smaller code, fewer inctructions, etc. Absolutely, I just mean that the CONFIG_BTRFS_FS_RUN_SANITY_TESTS version can be cleaner: #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS static inline int btrfs_is_testing(struct btrfs_fs_info *fs_info) { return test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state); } #else static inline int btrfs_is_testing(struct btrfs_fs_info *fs_info) { return 0; } #endif I find `if (1) return 1; else return 0;` really icky.
On Tue, Sep 11, 2018 at 12:14:47PM -0700, Omar Sandoval wrote: > > The unlikely can go away, sure. > > > > I would still like to remove the test_bit call when tests are compiled > > out. There are about 10 calls to btrfs_is_testing in various core > > functions, followed by further statements. This would have a > > (negligible) runtime penalty but generates effectively unused code on > > production builds. > > > > The static inline function returning 0 allows to optimize out the unused > > code, so smaller code, fewer inctructions, etc. > > Absolutely, I just mean that the CONFIG_BTRFS_FS_RUN_SANITY_TESTS > version can be cleaner: > > #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS > static inline int btrfs_is_testing(struct btrfs_fs_info *fs_info) > { > return test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state); > } > #else > static inline int btrfs_is_testing(struct btrfs_fs_info *fs_info) > { > return 0; > } > #endif > > I find `if (1) return 1; else return 0;` really icky. I see what you mean, will fix it.
diff --git a/fs/btrfs/ctree.h b/fs/btrfs/ctree.h index 32d2fce4ac53..8dafc7bb6ad8 100644 --- a/fs/btrfs/ctree.h +++ b/fs/btrfs/ctree.h @@ -3708,17 +3708,20 @@ static inline int btrfs_defrag_cancelled(struct btrfs_fs_info *fs_info) #ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS void btrfs_test_inode_set_ops(struct inode *inode); void btrfs_test_destroy_inode(struct inode *inode); -#endif static inline int btrfs_is_testing(struct btrfs_fs_info *fs_info) { -#ifdef CONFIG_BTRFS_FS_RUN_SANITY_TESTS if (unlikely(test_bit(BTRFS_FS_STATE_DUMMY_FS_INFO, &fs_info->fs_state))) return 1; -#endif return 0; } +#else +static inline int btrfs_is_testing(struct btrfs_fs_info *fs_info) +{ + return 0; +} +#endif static inline void cond_wake_up(struct wait_queue_head *wq) {
Avoid the inline ifdefs and use two sections for self-tests enabled and disabled. Signed-off-by: David Sterba <dsterba@suse.com> --- fs/btrfs/ctree.h | 9 ++++++--- 1 file changed, 6 insertions(+), 3 deletions(-)